def __init__(self, **configs):
log.debug("Starting KafkaAdminClient with configuration: %s", configs)
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError(
"Unrecognized configs: {}".format(extra_configs))
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(
samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [
reporter() for reporter in self.config['metric_reporters']
]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug("KafkaAdminClient started.")
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
# TODO _metrics likely needs to be passed to KafkaClient, etc.
# api_version was previously a str. Accept old format for now
if isinstance(self.config['api_version'], str):
str_version = self.config['api_version']
if str_version == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, str_version.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), str_version)
self._client = KafkaClient(metrics=self._metrics, **self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest' }
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
reporters.append(DictReporter('kafka.consumer'))
self._metrics = Metrics(metric_config, reporters)
metric_group_prefix = 'consumer'
# TODO _metrics likely needs to be passed to KafkaClient, etc.
self._client = KafkaClient(**self.config)
# Check Broker Version if not set explicitly
if self.config['api_version'] == 'auto':
self.config['api_version'] = self._client.check_version()
assert self.config['api_version'] in ('0.9', '0.8.2', '0.8.1', '0.8.0'), 'Unrecognized api version'
# Convert api_version config to tuple for easy comparisons
self.config['api_version'] = tuple(
map(int, self.config['api_version'].split('.')))
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, metric_group_prefix, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics, metric_group_prefix,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def __init__(self, **configs):
log.debug("Starting KafkaAdminClient with configuration: %s", configs)
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError("Unrecognized configs: {}".format(extra_configs))
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug("KafkaAdminClient started.")
def fetcher(client, subscription_state):
subscription_state.subscribe(topics=['foobar'])
assignment = [TopicPartition('foobar', i) for i in range(3)]
subscription_state.assign_from_subscribed(assignment)
for tp in assignment:
subscription_state.seek(tp, 0)
return Fetcher(client, subscription_state, Metrics(), 'test_fetcher')
def test_maybe_auto_commit_offsets_sync(mocker, api_version, group_id, enable,
error, has_auto_commit, commit_offsets,
warn, exc):
mock_warn = mocker.patch('kafka.coordinator.consumer.log.warning')
mock_exc = mocker.patch('kafka.coordinator.consumer.log.exception')
coordinator = ConsumerCoordinator(KafkaClient(),
SubscriptionState(),
Metrics(),
'consumer',
api_version=api_version,
enable_auto_commit=enable,
group_id=group_id)
commit_sync = mocker.patch.object(coordinator,
'commit_offsets_sync',
side_effect=error)
if has_auto_commit:
assert coordinator._auto_commit_task is not None
coordinator._auto_commit_task.enable()
assert coordinator._auto_commit_task._enabled is True
else:
assert coordinator._auto_commit_task is None
assert coordinator._maybe_auto_commit_offsets_sync() is None
if has_auto_commit:
assert coordinator._auto_commit_task is not None
assert coordinator._auto_commit_task._enabled is False
assert commit_sync.call_count == (1 if commit_offsets else 0)
assert mock_warn.call_count == (1 if warn else 0)
assert mock_exc.call_count == (1 if exc else 0)
def test_maybe_auto_commit_offsets_sync(mocker, api_version, group_id, enable,
error, has_auto_commit, commit_offsets,
warn, exc):
mock_warn = mocker.patch('kafka.coordinator.consumer.log.warning')
mock_exc = mocker.patch('kafka.coordinator.consumer.log.exception')
client = KafkaClient(api_version=api_version)
coordinator = ConsumerCoordinator(client, SubscriptionState(),
Metrics(),
api_version=api_version,
session_timeout_ms=30000,
max_poll_interval_ms=30000,
enable_auto_commit=enable,
group_id=group_id)
commit_sync = mocker.patch.object(coordinator, 'commit_offsets_sync',
side_effect=error)
if has_auto_commit:
assert coordinator.next_auto_commit_deadline is not None
else:
assert coordinator.next_auto_commit_deadline is None
assert coordinator._maybe_auto_commit_offsets_sync() is None
if has_auto_commit:
assert coordinator.next_auto_commit_deadline is not None
assert commit_sync.call_count == (1 if commit_offsets else 0)
assert mock_warn.call_count == (1 if warn else 0)
assert mock_exc.call_count == (1 if exc else 0)
def fetcher(client, subscription_state, topic):
subscription_state.subscribe(topics=[topic])
assignment = [TopicPartition(topic, i) for i in range(3)]
subscription_state.assign_from_subscribed(assignment)
for tp in assignment:
subscription_state.seek(tp, 0)
return Fetcher(client, subscription_state, Metrics())
def test_init(conn):
cli = KafkaClient()
coordinator = ConsumerCoordinator(cli, SubscriptionState(), Metrics(),
'consumer')
# metadata update on init
assert cli.cluster._need_update is True
assert WeakMethod(
coordinator._handle_metadata_update) in cli.cluster._listeners
def __init__(self, **configs):
log.debug("Starting Kafka administration interface")
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError("Unrecognized configs: %s" %
extra_configs)
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(
map(int, deprecated.split('.')))
log.warning(
'use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(
samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [
reporter() for reporter in self.config['metric_reporters']
]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug('Kafka administration interface started')
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
metric_group_prefix = 'consumer'
# TODO _metrics likely needs to be passed to KafkaClient, etc.
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
str_version = self.config['api_version']
if str_version == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, str_version.split('.')))
log.warning('use api_version=%s (%s is deprecated)',
str(self.config['api_version']), str_version)
self._client = KafkaClient(**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, metric_group_prefix, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics, metric_group_prefix,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def test_autocommit_enable_api_version(client, api_version):
coordinator = ConsumerCoordinator(client, SubscriptionState(),
Metrics(),
enable_auto_commit=True,
group_id='foobar',
api_version=api_version)
if api_version < (0, 8, 1):
assert coordinator._auto_commit_task is None
assert coordinator.config['enable_auto_commit'] is False
else:
assert coordinator._auto_commit_task is not None
assert coordinator.config['enable_auto_commit'] is True
def test_autocommit_enable_api_version(client, api_version):
coordinator = ConsumerCoordinator(client, SubscriptionState(),
Metrics(),
enable_auto_commit=True,
session_timeout_ms=30000, # session_timeout_ms and max_poll_interval_ms
max_poll_interval_ms=30000, # should be the same to avoid KafkaConfigurationError
group_id='foobar',
api_version=api_version)
if api_version < (0, 8, 1):
assert coordinator.config['enable_auto_commit'] is False
else:
assert coordinator.config['enable_auto_commit'] is True
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest' }
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
reporters.append(DictReporter('kafka.consumer'))
self._metrics = Metrics(metric_config, reporters)
metric_group_prefix = 'consumer'
# TODO _metrics likely needs to be passed to KafkaClient, etc.
client = self.config.pop('client', None) or KafkaClient(**self.config)
self._client = client
# Check Broker Version if not set explicitly
if self.config['api_version'] == 'auto':
self.config['api_version'] = self._client.check_version()
assert self.config['api_version'] in ('0.10', '0.9', '0.8.2', '0.8.1', '0.8.0'), 'Unrecognized api version'
# Convert api_version config to tuple for easy comparisons
self.config['api_version'] = tuple(
map(int, self.config['api_version'].split('.')))
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, metric_group_prefix, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics, metric_group_prefix,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def __init__(self, config):
Thread.__init__(self)
self.config = config
self.timeout_ms = 10000
self.kafka_client = None
self.running = True
self.sc = None
metrics_tags = {"client-id": self.config["client_id"]}
metric_config = MetricConfig(samples=2, time_window_ms=30000, tags=metrics_tags)
self._metrics = Metrics(metric_config, reporters=[])
self.lock = Lock()
self.lock.acquire()
self.log = logging.getLogger("MasterCoordinator")
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
other value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): called as
callback(offsets, response) response will be either an Exception
or a OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used.
Default: [RangePartitionAssignor, RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
max_poll_records (int): The maximum number of records returned in a
single call to poll().
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). The java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). The java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
consumer_timeout_ms (int): number of milliseconds to block during
message iteration before raising StopIteration (i.e., ending the
iterator). Default block forever [float('inf')].
skip_double_compressed_messages (bool): A bug in KafkaProducer <= 1.2.4
caused some messages to be corrupted via double-compression.
By default, the fetcher will return these messages as a compressed
blob of bytes with a single offset, i.e. how the message was
actually published to the cluster. If you prefer to have the
fetcher automatically detect corrupt messages and skip them,
set this option to True. Default: False.
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
default: true.
ssl_cafile (str): optional filename of ca file to use in certificate
verification. default: none.
ssl_certfile (str): optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. default: none.
ssl_keyfile (str): optional filename containing the client private key.
default: none.
ssl_password (str): optional password to be used when loading the
certificate chain. default: None.
ssl_crlfile (str): optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
default: none.
api_version (tuple): specify which kafka API version to use.
If set to None, the client will attempt to infer the broker version
by probing various APIs. Default: None
Examples:
(0, 9) enables full group coordination features with automatic
partition assignment and rebalancing,
(0, 8, 2) enables kafka-storage offset commits with manual
partition assignment only,
(0, 8, 1) enables zookeeper-storage offset commits with manual
partition assignment only,
(0, 8, 0) enables basic functionality but requires manual
partition assignment and offset management.
For a full list of supported versions, see KafkaClient.API_VERSIONS
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to 'auto'
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The maximum age in milliseconds of
samples used to compute metrics. Default: 30000
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
exclude_internal_topics (bool): Whether records from internal topics
(such as offsets) should be exposed to the consumer. If set to True
the only way to receive records from an internal topic is
subscribing to it. Requires 0.10+ Default: True
sasl_mechanism (str): string picking sasl mechanism when security_protocol
is SASL_PLAINTEXT or SASL_SSL. Currently only PLAIN is supported.
Default: None
sasl_plain_username (str): username for sasl PLAIN authentication.
Default: None
sasl_plain_password (str): password for sasl PLAIN authentication.
Default: None
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/0100/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'group_id': 'kafka-python-default-group',
'key_deserializer': None,
'value_deserializer': None,
'fetch_max_wait_ms': 500,
'fetch_min_bytes': 1,
'max_partition_fetch_bytes': 1 * 1024 * 1024,
'request_timeout_ms': 40 * 1000,
'retry_backoff_ms': 100,
'reconnect_backoff_ms': 50,
'max_in_flight_requests_per_connection': 5,
'auto_offset_reset': 'latest',
'enable_auto_commit': True,
'auto_commit_interval_ms': 5000,
'default_offset_commit_callback': lambda offsets, response: True,
'check_crcs': True,
'metadata_max_age_ms': 5 * 60 * 1000,
'partition_assignment_strategy': (RangePartitionAssignor, RoundRobinPartitionAssignor),
'heartbeat_interval_ms': 3000,
'session_timeout_ms': 30000,
'max_poll_records': sys.maxsize,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'consumer_timeout_ms': float('inf'),
'skip_double_compressed_messages': False,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_crlfile': None,
'ssl_password': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'connections_max_idle_ms': 9 * 60 * 1000, # not implemented yet
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
'metric_group_prefix': 'consumer',
'selector': selectors.DefaultSelector,
'exclude_internal_topics': True,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
# TODO _metrics likely needs to be passed to KafkaClient, etc.
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
str_version = self.config['api_version']
if str_version == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, str_version.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), str_version)
self._client = KafkaClient(metrics=self._metrics, **self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
self._metrics.close()
self._client.close()
try:
self.config['key_deserializer'].close()
except AttributeError:
pass
try:
self.config['value_deserializer'].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
callback (callable, optional): called as callback(offsets, response)
with response as either an Exception or a OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(
offsets, callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = self._client.cluster
if self._client._metadata_refresh_in_progress and self._client._topics:
future = cluster.request_update()
self._client.poll(future=future)
stash = cluster.need_all_topic_metadata
cluster.need_all_topic_metadata = True
future = cluster.request_update()
self._client.poll(future=future)
cluster.need_all_topic_metadata = stash
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0, max_records=None):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): milliseconds spent waiting in poll if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
max_records (int, optional): The maximum number of records returned
in a single call to :meth:`poll`. Default: Inherit value from
max_poll_records.
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
"""
assert timeout_ms >= 0, 'Timeout must not be negative'
if max_records is None:
max_records = self.config['max_poll_records']
# poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining, max_records)
if records:
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms, max_records):
"""
Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block
Returns:
dict: map of topic to list of records (may be empty)
"""
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(self._subscription.missing_fetch_positions())
# if data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
records, partial = self._fetcher.fetched_records(max_records)
if records:
# before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
if not partial:
self._fetcher.send_fetches()
return records
# send any new fetches (won't resend pending fetches)
self._fetcher.send_fetches()
self._client.poll(timeout_ms=timeout_ms, sleep=True)
records, _ = self._fetcher.fetched_records(max_records)
return records
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions([partition])
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest available message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if no FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): partitions to pause
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def paused(self):
"""Get the partitions that were previously paused by a call to pause().
Returns:
set: {partition (TopicPartition), ...}
"""
return self._subscription.paused_partitions()
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): partitions to resume
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0; or if partition is not
currently assigned.
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), 'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_all_topic_metadata = True
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics([])
log.debug("Unsubscribed all topics or patterns and assigned partitions")
def metrics(self, raw=False):
"""Warning: this is an unstable interface.
It may change in future releases without warning"""
if raw:
return self._metrics.metrics
metrics = {}
for k, v in self._metrics.metrics.items():
if k.group not in metrics:
metrics[k.group] = {}
if k.name not in metrics[k.group]:
metrics[k.group][k.name] = {}
metrics[k.group][k.name] = v.value()
return metrics
def _use_consumer_group(self):
"""Return True iff this consumer can/should join a broker-coordinated group."""
if self.config['api_version'] < (0, 9):
return False
elif self.config['group_id'] is None:
return False
elif not self._subscription.partitions_auto_assigned():
return False
return True
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if (self.config['api_version'] >= (0, 8, 1)
and self.config['group_id'] is not None):
# refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# then do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription() is not None, 'No topic subscription or manual partition assignment'
while time.time() < self._consumer_timeout:
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch offsets for any subscribed partitions that we arent tracking yet
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
poll_ms = 1000 * (self._consumer_timeout - time.time())
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(timeout_ms=poll_ms, sleep=True)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = self._next_timeout()
# Because the consumer client poll does not sleep unless blocking on
# network IO, we need to explicitly sleep when we know we are idle
# because we haven't been assigned any partitions to fetch / consume
if self._use_consumer_group() and not self.assignment():
sleep_time = max(timeout_at - time.time(), 0)
if sleep_time > 0 and not self._client.in_flight_request_count():
log.debug('No partitions assigned; sleeping for %s', sleep_time)
time.sleep(sleep_time)
continue
# Short-circuit the fetch iterator if we are already timed out
# to avoid any unintentional interaction with fetcher setup
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# an else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# and we assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.send_fetches()
def _next_timeout(self):
timeout = min(self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time())
# Although the delayed_tasks timeout above should cover processing
# HeartbeatRequests, it is still possible that HeartbeatResponses
# are left unprocessed during a long _fetcher iteration without
# an intermediate poll(). And because tasks are responsible for
# rescheduling themselves, an unprocessed response will prevent
# the next heartbeat from being sent. This check should help
# avoid that.
if self._use_consumer_group():
heartbeat = time.time() + self._coordinator.heartbeat.ttl()
timeout = min(timeout, heartbeat)
return timeout
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config['consumer_timeout_ms'] >= 0:
self._consumer_timeout = time.time() + (
self.config['consumer_timeout_ms'] / 1000.0)
# old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError(
'deprecated -- initialize a new consumer')
def set_topic_partitions(self, *topics):
raise NotImplementedError(
'deprecated -- use subscribe() or assign()')
def fetch_messages(self):
raise NotImplementedError(
'deprecated -- use poll() or iterator interface')
def get_partition_offsets(self, topic, partition,
request_time_ms, max_num_offsets):
raise NotImplementedError(
'deprecated -- send an OffsetRequest with KafkaClient')
def offsets(self, group=None):
raise NotImplementedError('deprecated -- use committed(partition)')
def task_done(self, message):
raise NotImplementedError(
'deprecated -- commit offsets manually if needed')
def metrics(request, config, reporter):
metrics = Metrics(config, [reporter], enable_expiration=True)
yield metrics
metrics.close()
def __init__(self, **configs):
log.debug("Starting the Kafka producer") # trace
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % (configs,)
if self.config['client_id'] is None:
self.config['client_id'] = 'kafka-python-producer-%s' % \
(PRODUCER_CLIENT_ID_SEQUENCE.increment(),)
if self.config['acks'] == 'all':
self.config['acks'] = -1
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, deprecated.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
client = KafkaClient(metrics=self._metrics, metric_group_prefix='producer',
wakeup_timeout_ms=self.config['max_block_ms'],
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = client.config['api_version']
if self.config['compression_type'] == 'lz4':
assert self.config['api_version'] >= (0, 8, 2), 'LZ4 Requires >= Kafka 0.8.2 Brokers'
# Check compression_type for library support
ct = self.config['compression_type']
if ct not in self._COMPRESSORS:
raise ValueError("Not supported codec: {}".format(ct))
else:
checker, compression_attrs = self._COMPRESSORS[ct]
assert checker(), "Libraries for {} compression codec not found".format(ct)
self.config['compression_attrs'] = compression_attrs
message_version = self._max_usable_produce_magic()
self._accumulator = RecordAccumulator(message_version=message_version, metrics=self._metrics, **self.config)
self._metadata = client.cluster
guarantee_message_order = bool(self.config['max_in_flight_requests_per_connection'] == 1)
self._sender = Sender(client, self._metadata,
self._accumulator, self._metrics,
guarantee_message_order=guarantee_message_order,
**self.config)
self._sender.daemon = True
self._sender.start()
self._closed = False
self._cleanup = self._cleanup_factory()
atexit.register(self._cleanup)
log.debug("Kafka producer started")
class KafkaProducer(object):
"""A Kafka client that publishes records to the Kafka cluster.
The producer is thread safe and sharing a single producer instance across
threads will generally be faster than having multiple instances.
The producer consists of a pool of buffer space that holds records that
haven't yet been transmitted to the server as well as a background I/O
thread that is responsible for turning these records into requests and
transmitting them to the cluster.
:meth:`~kafka.KafkaProducer.send` is asynchronous. When called it adds the
record to a buffer of pending record sends and immediately returns. This
allows the producer to batch together individual records for efficiency.
The 'acks' config controls the criteria under which requests are considered
complete. The "all" setting will result in blocking on the full commit of
the record, the slowest but most durable setting.
If the request fails, the producer can automatically retry, unless
'retries' is configured to 0. Enabling retries also opens up the
possibility of duplicates (see the documentation on message
delivery semantics for details:
https://kafka.apache.org/documentation.html#semantics
).
The producer maintains buffers of unsent records for each partition. These
buffers are of a size specified by the 'batch_size' config. Making this
larger can result in more batching, but requires more memory (since we will
generally have one of these buffers for each active partition).
By default a buffer is available to send immediately even if there is
additional unused space in the buffer. However if you want to reduce the
number of requests you can set 'linger_ms' to something greater than 0.
This will instruct the producer to wait up to that number of milliseconds
before sending a request in hope that more records will arrive to fill up
the same batch. This is analogous to Nagle's algorithm in TCP. Note that
records that arrive close together in time will generally batch together
even with linger_ms=0 so under heavy load batching will occur regardless of
the linger configuration; however setting this to something larger than 0
can lead to fewer, more efficient requests when not under maximal load at
the cost of a small amount of latency.
The buffer_memory controls the total amount of memory available to the
producer for buffering. If records are sent faster than they can be
transmitted to the server then this buffer space will be exhausted. When
the buffer space is exhausted additional send calls will block.
The key_serializer and value_serializer instruct how to turn the key and
value objects the user provides into bytes.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the producer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client.
Default: 'kafka-python-producer-#' (appended with a unique number
per instance)
key_serializer (callable): used to convert user-supplied keys to bytes
If not None, called as f(key), should return bytes. Default: None.
value_serializer (callable): used to convert user-supplied message
values to bytes. If not None, called as f(value), should return
bytes. Default: None.
acks (0, 1, 'all'): The number of acknowledgments the producer requires
the leader to have received before considering a request complete.
This controls the durability of records that are sent. The
following settings are common:
0: Producer will not wait for any acknowledgment from the server.
The message will immediately be added to the socket
buffer and considered sent. No guarantee can be made that the
server has received the record in this case, and the retries
configuration will not take effect (as the client won't
generally know of any failures). The offset given back for each
record will always be set to -1.
1: Wait for leader to write the record to its local log only.
Broker will respond without awaiting full acknowledgement from
all followers. In this case should the leader fail immediately
after acknowledging the record but before the followers have
replicated it then the record will be lost.
all: Wait for the full set of in-sync replicas to write the record.
This guarantees that the record will not be lost as long as at
least one in-sync replica remains alive. This is the strongest
available guarantee.
If unset, defaults to acks=1.
compression_type (str): The compression type for all data generated by
the producer. Valid values are 'gzip', 'snappy', 'lz4', or None.
Compression is of full batches of data, so the efficacy of batching
will also impact the compression ratio (more batching means better
compression). Default: None.
retries (int): Setting a value greater than zero will cause the client
to resend any record whose send fails with a potentially transient
error. Note that this retry is no different than if the client
resent the record upon receiving the error. Allowing retries
without setting max_in_flight_requests_per_connection to 1 will
potentially change the ordering of records because if two batches
are sent to a single partition, and the first fails and is retried
but the second succeeds, then the records in the second batch may
appear first.
Default: 0.
batch_size (int): Requests sent to brokers will contain multiple
batches, one for each partition with data available to be sent.
A small batch size will make batching less common and may reduce
throughput (a batch size of zero will disable batching entirely).
Default: 16384
linger_ms (int): The producer groups together any records that arrive
in between request transmissions into a single batched request.
Normally this occurs only under load when records arrive faster
than they can be sent out. However in some circumstances the client
may want to reduce the number of requests even under moderate load.
This setting accomplishes this by adding a small amount of
artificial delay; that is, rather than immediately sending out a
record the producer will wait for up to the given delay to allow
other records to be sent so that the sends can be batched together.
This can be thought of as analogous to Nagle's algorithm in TCP.
This setting gives the upper bound on the delay for batching: once
we get batch_size worth of records for a partition it will be sent
immediately regardless of this setting, however if we have fewer
than this many bytes accumulated for this partition we will
'linger' for the specified time waiting for more records to show
up. This setting defaults to 0 (i.e. no delay). Setting linger_ms=5
would have the effect of reducing the number of requests sent but
would add up to 5ms of latency to records sent in the absense of
load. Default: 0.
partitioner (callable): Callable used to determine which partition
each message is assigned to. Called (after key serialization):
partitioner(key_bytes, all_partitions, available_partitions).
The default partitioner implementation hashes each non-None key
using the same murmur2 algorithm as the java client so that
messages with the same key are assigned to the same partition.
When a key is None, the message is delivered to a random partition
(filtered to partitions with available leaders only, if possible).
buffer_memory (int): The total bytes of memory the producer should use
to buffer records waiting to be sent to the server. If records are
sent faster than they can be delivered to the server the producer
will block up to max_block_ms, raising an exception on timeout.
In the current implementation, this setting is an approximation.
Default: 33554432 (32MB)
connections_max_idle_ms: Close idle connections after the number of
milliseconds specified by this config. The broker closes idle
connections after connections.max.idle.ms, so this avoids hitting
unexpected socket disconnected errors on the client.
Default: 540000
max_block_ms (int): Number of milliseconds to block during
:meth:`~kafka.KafkaProducer.send` and
:meth:`~kafka.KafkaProducer.partitions_for`. These methods can be
blocked either because the buffer is full or metadata unavailable.
Blocking in the user-supplied serializers or partitioner will not be
counted against this timeout. Default: 60000.
max_request_size (int): The maximum size of a request. This is also
effectively a cap on the maximum record size. Note that the server
has its own cap on record size which may be different from this.
This setting will limit the number of record batches the producer
will send in a single request to avoid sending huge requests.
Default: 1048576.
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 30000.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). Java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). Java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
reconnect_backoff_max_ms (int): The maximum amount of time in
milliseconds to wait when reconnecting to a broker that has
repeatedly failed to connect. If provided, the backoff per host
will increase exponentially for each consecutive connection
failure, up to this maximum. To avoid connection storms, a
randomization factor of 0.2 will be applied to the backoff
resulting in a random range between 20% below and 20% above
the computed value. Default: 1000.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Note that if this setting is set to be greater
than 1 and there are failed sends, there is a risk of message
re-ordering due to retries (i.e., if retries are enabled).
Default: 5.
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL, SASL_PLAINTEXT, SASL_SSL.
Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
default: true.
ssl_cafile (str): optional filename of ca file to use in certificate
veriication. default: none.
ssl_certfile (str): optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. default: none.
ssl_keyfile (str): optional filename containing the client private key.
default: none.
ssl_password (str): optional password to be used when loading the
certificate chain. default: none.
ssl_crlfile (str): optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
default: none.
ssl_ciphers (str): optionally set the available ciphers for ssl
connections. It should be a string in the OpenSSL cipher list
format. If no cipher can be selected (because compile-time options
or other configuration forbids use of all the specified ciphers),
an ssl.SSLError will be raised. See ssl.SSLContext.set_ciphers
api_version (tuple): Specify which Kafka API version to use. If set to
None, the client will attempt to infer the broker version by probing
various APIs. Example: (0, 10, 2). Default: None
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to None.
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The maximum age in milliseconds of
samples used to compute metrics. Default: 30000
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
sasl_mechanism (str): Authentication mechanism when security_protocol
is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are:
PLAIN, GSSAPI, OAUTHBEARER.
sasl_plain_username (str): username for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_plain_password (str): password for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_kerberos_service_name (str): Service name to include in GSSAPI
sasl mechanism handshake. Default: 'kafka'
sasl_kerberos_domain_name (str): kerberos domain name to use in GSSAPI
sasl mechanism handshake. Default: one of bootstrap servers
sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider
instance. (See kafka.oauth.abstract). Default: None
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/0100/configuration.html#producerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': None,
'key_serializer': None,
'value_serializer': None,
'acks': 1,
'bootstrap_topics_filter': set(),
'compression_type': None,
'retries': 0,
'batch_size': 16384,
'linger_ms': 0,
'partitioner': DefaultPartitioner(),
'buffer_memory': 33554432,
'connections_max_idle_ms': 9 * 60 * 1000,
'max_block_ms': 60000,
'max_request_size': 1048576,
'metadata_max_age_ms': 300000,
'retry_backoff_ms': 100,
'request_timeout_ms': 30000,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'sock_chunk_bytes': 4096, # undocumented experimental option
'sock_chunk_buffer_count': 1000, # undocumented experimental option
'reconnect_backoff_ms': 50,
'reconnect_backoff_max_ms': 1000,
'max_in_flight_requests_per_connection': 5,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_crlfile': None,
'ssl_password': None,
'ssl_ciphers': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
'selector': selectors.DefaultSelector,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
'sasl_kerberos_service_name': 'kafka',
'sasl_kerberos_domain_name': None,
'sasl_oauth_token_provider': None
}
_COMPRESSORS = {
'gzip': (has_gzip, LegacyRecordBatchBuilder.CODEC_GZIP),
'snappy': (has_snappy, LegacyRecordBatchBuilder.CODEC_SNAPPY),
'lz4': (has_lz4, LegacyRecordBatchBuilder.CODEC_LZ4),
None: (lambda: True, LegacyRecordBatchBuilder.CODEC_NONE),
}
def __init__(self, **configs):
log.debug("Starting the Kafka producer") # trace
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % (configs,)
if self.config['client_id'] is None:
self.config['client_id'] = 'kafka-python-producer-%s' % \
(PRODUCER_CLIENT_ID_SEQUENCE.increment(),)
if self.config['acks'] == 'all':
self.config['acks'] = -1
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, deprecated.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
client = KafkaClient(metrics=self._metrics, metric_group_prefix='producer',
wakeup_timeout_ms=self.config['max_block_ms'],
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = client.config['api_version']
if self.config['compression_type'] == 'lz4':
assert self.config['api_version'] >= (0, 8, 2), 'LZ4 Requires >= Kafka 0.8.2 Brokers'
# Check compression_type for library support
ct = self.config['compression_type']
if ct not in self._COMPRESSORS:
raise ValueError("Not supported codec: {}".format(ct))
else:
checker, compression_attrs = self._COMPRESSORS[ct]
assert checker(), "Libraries for {} compression codec not found".format(ct)
self.config['compression_attrs'] = compression_attrs
message_version = self._max_usable_produce_magic()
self._accumulator = RecordAccumulator(message_version=message_version, metrics=self._metrics, **self.config)
self._metadata = client.cluster
guarantee_message_order = bool(self.config['max_in_flight_requests_per_connection'] == 1)
self._sender = Sender(client, self._metadata,
self._accumulator, self._metrics,
guarantee_message_order=guarantee_message_order,
**self.config)
self._sender.daemon = True
self._sender.start()
self._closed = False
self._cleanup = self._cleanup_factory()
atexit.register(self._cleanup)
log.debug("Kafka producer started")
def bootstrap_connected(self):
"""Return True if the bootstrap is connected."""
return self._sender.bootstrap_connected()
def _cleanup_factory(self):
"""Build a cleanup clojure that doesn't increase our ref count"""
_self = weakref.proxy(self)
def wrapper():
try:
_self.close(timeout=0)
except (ReferenceError, AttributeError):
pass
return wrapper
def _unregister_cleanup(self):
if getattr(self, '_cleanup', None):
if hasattr(atexit, 'unregister'):
atexit.unregister(self._cleanup) # pylint: disable=no-member
# py2 requires removing from private attribute...
else:
# ValueError on list.remove() if the exithandler no longer exists
# but that is fine here
try:
atexit._exithandlers.remove( # pylint: disable=no-member
(self._cleanup, (), {}))
except ValueError:
pass
self._cleanup = None
def __del__(self):
self.close(timeout=0)
def close(self, timeout=None):
"""Close this producer.
Arguments:
timeout (float, optional): timeout in seconds to wait for completion.
"""
# drop our atexit handler now to avoid leaks
self._unregister_cleanup()
if not hasattr(self, '_closed') or self._closed:
log.info('Kafka producer closed')
return
if timeout is None:
# threading.TIMEOUT_MAX is available in Python3.3+
timeout = getattr(threading, 'TIMEOUT_MAX', float('inf'))
if getattr(threading, 'TIMEOUT_MAX', False):
assert 0 <= timeout <= getattr(threading, 'TIMEOUT_MAX')
else:
assert timeout >= 0
log.info("Closing the Kafka producer with %s secs timeout.", timeout)
invoked_from_callback = bool(threading.current_thread() is self._sender)
if timeout > 0:
if invoked_from_callback:
log.warning("Overriding close timeout %s secs to 0 in order to"
" prevent useless blocking due to self-join. This"
" means you have incorrectly invoked close with a"
" non-zero timeout from the producer call-back.",
timeout)
else:
# Try to close gracefully.
if self._sender is not None:
self._sender.initiate_close()
self._sender.join(timeout)
if self._sender is not None and self._sender.is_alive():
log.info("Proceeding to force close the producer since pending"
" requests could not be completed within timeout %s.",
timeout)
self._sender.force_close()
# Only join the sender thread when not calling from callback.
if not invoked_from_callback:
self._sender.join()
self._metrics.close()
try:
self.config['key_serializer'].close()
except AttributeError:
pass
try:
self.config['value_serializer'].close()
except AttributeError:
pass
self._closed = True
log.debug("The Kafka producer has closed.")
def partitions_for(self, topic):
"""Returns set of all known partitions for the topic."""
max_wait = self.config['max_block_ms'] / 1000.0
return self._wait_on_metadata(topic, max_wait)
def _max_usable_produce_magic(self):
if self.config['api_version'] >= (0, 11):
return 2
elif self.config['api_version'] >= (0, 10):
return 1
else:
return 0
def _estimate_size_in_bytes(self, key, value, headers=[]):
magic = self._max_usable_produce_magic()
if magic == 2:
return DefaultRecordBatchBuilder.estimate_size_in_bytes(
key, value, headers)
else:
return LegacyRecordBatchBuilder.estimate_size_in_bytes(
magic, self.config['compression_type'], key, value)
def send(self, topic, value=None, key=None, headers=None, partition=None, timestamp_ms=None):
"""Publish a message to a topic.
Arguments:
topic (str): topic where the message will be published
value (optional): message value. Must be type bytes, or be
serializable to bytes via configured value_serializer. If value
is None, key is required and message acts as a 'delete'.
See kafka compaction documentation for more details:
https://kafka.apache.org/documentation.html#compaction
(compaction requires kafka >= 0.8.1)
partition (int, optional): optionally specify a partition. If not
set, the partition will be selected using the configured
'partitioner'.
key (optional): a key to associate with the message. Can be used to
determine which partition to send the message to. If partition
is None (and producer's partitioner config is left as default),
then messages with the same key will be delivered to the same
partition (but if key is None, partition is chosen randomly).
Must be type bytes, or be serializable to bytes via configured
key_serializer.
headers (optional): a list of header key value pairs. List items
are tuples of str key and bytes value.
timestamp_ms (int, optional): epoch milliseconds (from Jan 1 1970 UTC)
to use as the message timestamp. Defaults to current time.
Returns:
FutureRecordMetadata: resolves to RecordMetadata
Raises:
KafkaTimeoutError: if unable to fetch topic metadata, or unable
to obtain memory buffer prior to configured max_block_ms
"""
assert value is not None or self.config['api_version'] >= (0, 8, 1), (
'Null messages require kafka >= 0.8.1')
assert not (value is None and key is None), 'Need at least one: key or value'
key_bytes = value_bytes = None
try:
self._wait_on_metadata(topic, self.config['max_block_ms'] / 1000.0)
key_bytes = self._serialize(
self.config['key_serializer'],
topic, key)
value_bytes = self._serialize(
self.config['value_serializer'],
topic, value)
assert type(key_bytes) in (bytes, bytearray, memoryview, type(None))
assert type(value_bytes) in (bytes, bytearray, memoryview, type(None))
partition = self._partition(topic, partition, key, value,
key_bytes, value_bytes)
if headers is None:
headers = []
assert type(headers) == list
assert all(type(item) == tuple and len(item) == 2 and type(item[0]) == str and type(item[1]) == bytes for item in headers)
message_size = self._estimate_size_in_bytes(key_bytes, value_bytes, headers)
self._ensure_valid_record_size(message_size)
tp = TopicPartition(topic, partition)
log.debug("Sending (key=%r value=%r headers=%r) to %s", key, value, headers, tp)
result = self._accumulator.append(tp, timestamp_ms,
key_bytes, value_bytes, headers,
self.config['max_block_ms'],
estimated_size=message_size)
future, batch_is_full, new_batch_created = result
if batch_is_full or new_batch_created:
log.debug("Waking up the sender since %s is either full or"
" getting a new batch", tp)
self._sender.wakeup()
return future
# handling exceptions and record the errors;
# for API exceptions return them in the future,
# for other exceptions raise directly
except Errors.BrokerResponseError as e:
log.debug("Exception occurred during message send: %s", e)
return FutureRecordMetadata(
FutureProduceResult(TopicPartition(topic, partition)),
-1, None, None,
len(key_bytes) if key_bytes is not None else -1,
len(value_bytes) if value_bytes is not None else -1,
sum(len(h_key.encode("utf-8")) + len(h_value) for h_key, h_value in headers) if headers else -1,
).failure(e)
def flush(self, timeout=None):
"""
Invoking this method makes all buffered records immediately available
to send (even if linger_ms is greater than 0) and blocks on the
completion of the requests associated with these records. The
post-condition of :meth:`~kafka.KafkaProducer.flush` is that any
previously sent record will have completed
(e.g. Future.is_done() == True). A request is considered completed when
either it is successfully acknowledged according to the 'acks'
configuration for the producer, or it results in an error.
Other threads can continue sending messages while one thread is blocked
waiting for a flush call to complete; however, no guarantee is made
about the completion of messages sent after the flush call begins.
Arguments:
timeout (float, optional): timeout in seconds to wait for completion.
Raises:
KafkaTimeoutError: failure to flush buffered records within the
provided timeout
"""
log.debug("Flushing accumulated records in producer.") # trace
self._accumulator.begin_flush()
self._sender.wakeup()
self._accumulator.await_flush_completion(timeout=timeout)
def _ensure_valid_record_size(self, size):
"""Validate that the record size isn't too large."""
if size > self.config['max_request_size']:
raise Errors.MessageSizeTooLargeError(
"The message is %d bytes when serialized which is larger than"
" the maximum request size you have configured with the"
" max_request_size configuration" % (size,))
if size > self.config['buffer_memory']:
raise Errors.MessageSizeTooLargeError(
"The message is %d bytes when serialized which is larger than"
" the total memory buffer you have configured with the"
" buffer_memory configuration." % (size,))
def _wait_on_metadata(self, topic, max_wait):
"""
Wait for cluster metadata including partitions for the given topic to
be available.
Arguments:
topic (str): topic we want metadata for
max_wait (float): maximum time in secs for waiting on the metadata
Returns:
set: partition ids for the topic
Raises:
KafkaTimeoutError: if partitions for topic were not obtained before
specified max_wait timeout
"""
# add topic to metadata topic list if it is not there already.
self._sender.add_topic(topic)
begin = time.time()
elapsed = 0.0
metadata_event = None
while True:
partitions = self._metadata.partitions_for_topic(topic)
if partitions is not None:
return partitions
if not metadata_event:
metadata_event = threading.Event()
log.debug("Requesting metadata update for topic %s", topic)
metadata_event.clear()
future = self._metadata.request_update()
future.add_both(lambda e, *args: e.set(), metadata_event)
self._sender.wakeup()
metadata_event.wait(max_wait - elapsed)
elapsed = time.time() - begin
if not metadata_event.is_set():
raise Errors.KafkaTimeoutError(
"Failed to update metadata after %.1f secs." % (max_wait,))
elif topic in self._metadata.unauthorized_topics:
raise Errors.TopicAuthorizationFailedError(topic)
else:
log.debug("_wait_on_metadata woke after %s secs.", elapsed)
def _serialize(self, f, topic, data):
if not f:
return data
if isinstance(f, Serializer):
return f.serialize(topic, data)
return f(data)
def _partition(self, topic, partition, key, value,
serialized_key, serialized_value):
if partition is not None:
assert partition >= 0
assert partition in self._metadata.partitions_for_topic(topic), 'Unrecognized partition'
return partition
all_partitions = sorted(self._metadata.partitions_for_topic(topic))
available = list(self._metadata.available_partitions_for_topic(topic))
return self.config['partitioner'](serialized_key,
all_partitions,
available)
def metrics(self, raw=False):
"""Get metrics on producer performance.
This is ported from the Java Producer, for details see:
https://kafka.apache.org/documentation/#producer_monitoring
Warning:
This is an unstable interface. It may change in future
releases without warning.
"""
if raw:
return self._metrics.metrics.copy()
metrics = {}
for k, v in six.iteritems(self._metrics.metrics.copy()):
if k.group not in metrics:
metrics[k.group] = {}
if k.name not in metrics[k.group]:
metrics[k.group][k.name] = {}
metrics[k.group][k.name] = v.value()
return metrics
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): called as
callback(offsets, response) response will be either an Exception
or a OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used.
Default: [RangePartitionAssignor, RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). The java client defaults to 131072.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). The java client defaults to 32768.
consumer_timeout_ms (int): number of millisecond to throw a timeout
exception to the consumer if no message is available for
consumption. Default: -1 (dont throw exception)
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
default: true.
ssl_cafile (str): optional filename of ca file to use in certificate
veriication. default: none.
ssl_certfile (str): optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. default: none.
ssl_keyfile (str): optional filename containing the client private key.
default: none.
ssl_crlfile (str): optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
default: none.
api_version (str): specify which kafka API version to use.
0.9 enables full group coordination features; 0.8.2 enables
kafka-storage offset commits; 0.8.1 enables zookeeper-storage
offset commits; 0.8.0 is what is left. If set to 'auto', will
attempt to infer the broker version by probing various APIs.
Default: auto
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to 'auto'
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The number of samples maintained to
compute metrics. Default: 30000
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/090/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers':
'localhost',
'client_id':
'kafka-python-' + __version__,
'group_id':
'kafka-python-default-group',
'key_deserializer':
None,
'value_deserializer':
None,
'fetch_max_wait_ms':
500,
'fetch_min_bytes':
1,
'max_partition_fetch_bytes':
1 * 1024 * 1024,
'request_timeout_ms':
40 * 1000,
'retry_backoff_ms':
100,
'reconnect_backoff_ms':
50,
'max_in_flight_requests_per_connection':
5,
'auto_offset_reset':
'latest',
'enable_auto_commit':
True,
'auto_commit_interval_ms':
5000,
'default_offset_commit_callback':
lambda offsets, response: True,
'check_crcs':
True,
'metadata_max_age_ms':
5 * 60 * 1000,
'partition_assignment_strategy':
(RangePartitionAssignor, RoundRobinPartitionAssignor),
'heartbeat_interval_ms':
3000,
'session_timeout_ms':
30000,
'send_buffer_bytes':
None,
'receive_buffer_bytes':
None,
'consumer_timeout_ms':
-1,
'security_protocol':
'PLAINTEXT',
'ssl_context':
None,
'ssl_check_hostname':
True,
'ssl_cafile':
None,
'ssl_certfile':
None,
'ssl_keyfile':
None,
'ssl_crlfile':
None,
'api_version':
'auto',
'api_version_auto_timeout_ms':
2000,
'connections_max_idle_ms':
9 * 60 * 1000, # not implemented yet
'metric_reporters': [],
'metrics_num_samples':
2,
'metrics_sample_window_ms':
30000,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(
samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [
reporter() for reporter in self.config['metric_reporters']
]
reporters.append(DictReporter('kafka.consumer'))
self._metrics = Metrics(metric_config, reporters)
metric_group_prefix = 'consumer'
# TODO _metrics likely needs to be passed to KafkaClient, etc.
self._client = KafkaClient(**self.config)
# Check Broker Version if not set explicitly
if self.config['api_version'] == 'auto':
self.config['api_version'] = self._client.check_version(
timeout=(self.config['api_version_auto_timeout_ms'] / 1000))
assert self.config['api_version'] in (
'0.10', '0.9', '0.8.2', '0.8.1',
'0.8.0'), 'Unrecognized api version'
# Convert api_version config to tuple for easy comparisons
self.config['api_version'] = tuple(
map(int, self.config['api_version'].split('.')))
self._subscription = SubscriptionState(
self.config['auto_offset_reset'])
self._fetcher = Fetcher(self._client, self._subscription,
self._metrics, metric_group_prefix,
**self.config)
self._coordinator = ConsumerCoordinator(
self._client,
self._subscription,
self._metrics,
metric_group_prefix,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
self._metrics.close()
self._client.close()
try:
self.config['key_deserializer'].close()
except AttributeError:
pass
try:
self.config['value_deserializer'].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
callback (callable, optional): called as callback(offsets, response)
with response as either an Exception or a OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(offsets,
callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = self._client.cluster
if self._client._metadata_refresh_in_progress and self._client._topics:
future = cluster.request_update()
self._client.poll(future=future)
stash = cluster.need_all_topic_metadata
cluster.need_all_topic_metadata = True
future = cluster.request_update()
self._client.poll(future=future)
cluster.need_all_topic_metadata = stash
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): milliseconds spent waiting in poll if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
"""
assert timeout_ms >= 0, 'Timeout must not be negative'
assert self._iterator is None, 'Incompatible with iterator interface'
# poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining)
if records:
# before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
self._fetcher.init_fetches()
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms):
"""
Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block
Returns:
dict: map of topic to list of records (may be empty)
"""
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config[
'api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(
self._subscription.missing_fetch_positions())
# init any new fetches (won't resend pending fetches)
records = self._fetcher.fetched_records()
# if data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
if records:
return records
self._fetcher.init_fetches()
self._client.poll(timeout_ms=timeout_ms, sleep=True)
return self._fetcher.fetched_records()
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(
partition), 'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions([partition])
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest available message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if not FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(
partition), 'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): partitions to pause
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def paused(self):
"""Get the partitions that were previously paused by a call to pause().
Returns:
set: {partition (TopicPartition), ...}
"""
return self._subscription.paused_partitions()
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): partitions to resume
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0; or if partition is not
currently assigned.
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(
), 'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(
), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp,
OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(
), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp,
OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_all_topic_metadata = True
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics([])
log.debug(
"Unsubscribed all topics or patterns and assigned partitions")
def _use_consumer_group(self):
"""Return True iff this consumer can/should join a broker-coordinated group."""
if self.config['api_version'] < (0, 9):
return False
elif self.config['group_id'] is None:
return False
elif not self._subscription.partitions_auto_assigned():
return False
return True
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if (self.config['api_version'] >= (0, 8, 1)
and self.config['group_id'] is not None):
# refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# then do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription(
) is not None, 'No topic subscription or manual partition assignment'
while time.time() < self._consumer_timeout:
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config[
'api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch offsets for any subscribed partitions that we arent tracking yet
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
poll_ms = 1000 * (self._consumer_timeout - time.time())
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(timeout_ms=poll_ms, sleep=True)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = self._next_timeout()
# Because the consumer client poll does not sleep unless blocking on
# network IO, we need to explicitly sleep when we know we are idle
# because we haven't been assigned any partitions to fetch / consume
if self._use_consumer_group() and not self.assignment():
sleep_time = max(timeout_at - time.time(), 0)
if sleep_time > 0 and not self._client.in_flight_request_count(
):
log.debug('No partitions assigned; sleeping for %s',
sleep_time)
time.sleep(sleep_time)
continue
# Short-circuit the fetch iterator if we are already timed out
# to avoid any unintentional interaction with fetcher setup
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# an else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# and we assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.init_fetches()
def _next_timeout(self):
timeout = min(self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time())
# Although the delayed_tasks timeout above should cover processing
# HeartbeatRequests, it is still possible that HeartbeatResponses
# are left unprocessed during a long _fetcher iteration without
# an intermediate poll(). And because tasks are responsible for
# rescheduling themselves, an unprocessed response will prevent
# the next heartbeat from being sent. This check should help
# avoid that.
if self._use_consumer_group():
heartbeat = time.time() + self._coordinator.heartbeat.ttl()
timeout = min(timeout, heartbeat)
return timeout
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config['consumer_timeout_ms'] >= 0:
self._consumer_timeout = time.time() + (
self.config['consumer_timeout_ms'] / 1000.0)
# old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError('deprecated -- initialize a new consumer')
def set_topic_partitions(self, *topics):
raise NotImplementedError('deprecated -- use subscribe() or assign()')
def fetch_messages(self):
raise NotImplementedError(
'deprecated -- use poll() or iterator interface')
def get_partition_offsets(self, topic, partition, request_time_ms,
max_num_offsets):
raise NotImplementedError(
'deprecated -- send an OffsetRequest with KafkaClient')
def offsets(self, group=None):
raise NotImplementedError('deprecated -- use committed(partition)')
def task_done(self, message):
raise NotImplementedError(
'deprecated -- commit offsets manually if needed')
def metrics():
return Metrics()
class KafkaAdmin(object):
"""An class for administering the kafka cluster.
The KafkaAdmin class will negotiate for the latest version of each message protocol format supported
by both the kafka-python client library and the kafka broker. Usage of optional fields from protocol
versions that are not supported by the broker will result in UnsupportedVersionError exceptions.
Use of this class requires a minimum broker version >= 0.10.0.0.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
reconnect_backoff_max_ms (int): The maximum amount of time in
milliseconds to wait when reconnecting to a broker that has
repeatedly failed to connect. If provided, the backoff per host
will increase exponentially for each consecutive connection
failure, up to this maximum. To avoid connection storms, a
randomization factor of 0.2 will be applied to the backoff
resulting in a random range between 20% below and 20% above
the computed value. Default: 1000.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 30000.
connections_max_idle_ms: Close idle connections after the number of
milliseconds specified by this config. The broker closes idle
connections after connections.max.idle.ms, so this avoids hitting
unexpected socket disconnected errors on the client.
Default: 540000
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). Java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). Java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): Pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): Flag to configure whether SSL handshake
should verify that the certificate matches the broker's hostname.
Default: True.
ssl_cafile (str): Optional filename of CA file to use in certificate
veriication. Default: None.
ssl_certfile (str): Optional filename of file in PEM format containing
the client certificate, as well as any CA certificates needed to
establish the certificate's authenticity. Default: None.
ssl_keyfile (str): Optional filename containing the client private key.
Default: None.
ssl_password (str): Optional password to be used when loading the
certificate chain. Default: None.
ssl_crlfile (str): Optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
Default: None.
api_version (tuple): Specify which Kafka API version to use. If set
to None, KafkaClient will attempt to infer the broker version by
probing various APIs. Example: (0, 10, 2). Default: None
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version is None
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
metrics (kafka.metrics.Metrics): Optionally provide a metrics
instance for capturing network IO stats. Default: None.
metric_group_prefix (str): Prefix for metric names. Default: ''
sasl_mechanism (str): string picking sasl mechanism when security_protocol
is SASL_PLAINTEXT or SASL_SSL. Currently only PLAIN is supported.
Default: None
sasl_plain_username (str): username for sasl PLAIN authentication.
Default: None
sasl_plain_password (str): password for sasl PLAIN authentication.
Default: None
sasl_kerberos_service_name (str): Service name to include in GSSAPI
sasl mechanism handshake. Default: 'kafka'
"""
DEFAULT_CONFIG = {
# client configs
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'request_timeout_ms': 30000,
'connections_max_idle_ms': 9 * 60 * 1000,
'reconnect_backoff_ms': 50,
'reconnect_backoff_max_ms': 1000,
'max_in_flight_requests_per_connection': 5,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'sock_chunk_bytes': 4096, # undocumented experimental option
'sock_chunk_buffer_count': 1000, # undocumented experimental option
'retry_backoff_ms': 100,
'metadata_max_age_ms': 300000,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_password': None,
'ssl_crlfile': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'selector': selectors.DefaultSelector,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
'sasl_kerberos_service_name': 'kafka',
# metrics configs
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
}
def __init__(self, **configs):
log.debug("Starting Kafka administration interface")
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError("Unrecognized configs: %s" %
extra_configs)
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(
map(int, deprecated.split('.')))
log.warning(
'use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(
samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [
reporter() for reporter in self.config['metric_reporters']
]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug('Kafka administration interface started')
def close(self):
"""Close the administration connection to the kafka broker"""
if not hasattr(self, '_closed') or self._closed:
log.info('Kafka administration interface already closed')
return
self._metrics.close()
self._client.close()
self._closed = True
log.debug('Kafka administartion interface has closed')
def _matching_api_version(self, operation):
"""Find matching api version, the lesser of either the latest api version the library supports, or
the max version supported by the broker
:param operation: An operation array from kafka.protocol
:return: The max matching version number between client and broker
"""
version = min(
len(operation) - 1,
self._client.get_api_versions()[operation[0].API_KEY][1])
if version < self._client.get_api_versions()[operation[0].API_KEY][0]:
# max library version is less than min broker version. Not sure any brokers
# actually set a min version greater than 0 right now, tho. But maybe in the future?
raise UnsupportedVersionError(
"Could not find matching protocol version for {}".format(
operation.__name__))
return version
def _validate_timeout(self, timeout_ms):
"""Validate the timeout is set or use the configuration default
:param timeout_ms: The timeout provided by api call, in milliseconds
:return: The timeout to use for the operation
"""
return timeout_ms or self.config['request_timeout_ms']
def _refresh_controller_id(self):
"""Determine the kafka cluster controller
"""
response = self._send_request_to_node(self._client.least_loaded_node(),
MetadataRequest[1]([]))
self._controller_id = response.controller_id
version = self._client.check_version(self._controller_id)
if version < (0, 10, 0):
raise UnsupportedVersionError(
"Kafka Admin interface not supported for cluster controller version {} < 0.10.0.0"
.format(version))
def _send_request_to_node(self, node, request):
"""Send a kafka protocol message to a specific broker. Will block until the message result is received.
:param node: The broker id to which to send the message
:param request: The message to send
:return: The kafka protocol response for the message
:exception: The exception if the message could not be sent
"""
while not self._client.ready(node):
# connection to broker not ready, poll until it is or send will fail with NodeNotReadyError
self._client.poll()
future = self._client.send(node, request)
self._client.poll(future=future)
if future.succeeded():
return future.value
else:
raise future.exception # pylint: disable-msg=raising-bad-type
def _send(self, request):
"""Send a kafka protocol message to the cluster controller. Will block until the message result is received.
:param request: The message to send
:return The kafka protocol response for the message
:exception NodeNotReadyError: If the controller connection can't be established
"""
remaining_tries = 2
while remaining_tries > 0:
remaining_tries = remaining_tries - 1
try:
return self._send_request_to_node(self._controller_id, request)
except (NotControllerError, KafkaConnectionError) as e:
# controller changed? refresh it
self._refresh_controller_id()
raise NodeNotReadyError(self._controller_id)
@staticmethod
def _convert_new_topic_request(new_topic):
return (
new_topic.name, new_topic.num_partitions,
new_topic.replication_factor, [
(partition_id, replicas) for partition_id, replicas in
new_topic.replica_assignments.items()
],
[(config_key, config_value)
for config_key, config_value in new_topic.topic_configs.items()])
def create_topics(self, new_topics, timeout_ms=None, validate_only=None):
"""Create new topics in the cluster.
:param new_topics: Array of NewTopic objects
:param timeout_ms: Milliseconds to wait for new topics to be created before broker returns
:param validate_only: If True, don't actually create new topics. Not supported by all versions.
:return: Appropriate version of CreateTopicResponse class
"""
version = self._matching_api_version(CreateTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version == 0:
if validate_only:
raise UnsupportedVersionError(
"validate_only not supported on cluster version {}".format(
self.config['api_version']))
request = CreateTopicsRequest[version](create_topic_requests=[
self._convert_new_topic_request(new_topic)
for new_topic in new_topics
],
timeout=timeout_ms)
elif version <= 2:
validate_only = validate_only or False
request = CreateTopicsRequest[version](create_topic_requests=[
self._convert_new_topic_request(new_topic)
for new_topic in new_topics
],
timeout=timeout_ms,
validate_only=validate_only)
else:
raise UnsupportedVersionError(
"missing implementation of CreateTopics for library supported version {}"
.format(version))
return self._send(request)
def delete_topics(self, topics, timeout_ms=None):
"""Delete topics from the cluster
:param topics: Array of topic name strings
:param timeout_ms: Milliseconds to wait for topics to be deleted before broker returns
:return: Appropriate version of DeleteTopicsResponse class
"""
version = self._matching_api_version(DeleteTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version <= 1:
request = DeleteTopicsRequest[version](topics=topics,
timeout=timeout_ms)
else:
raise UnsupportedVersionError(
"missing implementation of DeleteTopics for library supported version {}"
.format(version))
return self._send(request)
# list topics functionality is in ClusterMetadata
# describe topics functionality is in ClusterMetadata
# describe cluster functionality is in ClusterMetadata
# describe_acls protocol not implemented
# create_acls protocol not implemented
# delete_acls protocol not implemented
@staticmethod
def _convert_describe_config_resource_request(config_resource):
return (config_resource.resource_type, config_resource.name, [
config_key
for config_key, config_value in config_resource.configs.items()
] if config_resource.configs else None)
def describe_configs(self, config_resources, include_synonyms=None):
"""Fetch configuration parameters for one or more kafka resources.
:param config_resources: An array of ConfigResource objects.
Any keys in ConfigResource.configs dict will be used to filter the result. The configs dict should be None
to get all values. An empty dict will get zero values (as per kafka protocol).
:param include_synonyms: If True, return synonyms in response. Not supported by all versions.
:return: Appropriate version of DescribeConfigsResponse class
"""
version = self._matching_api_version(DescribeConfigsRequest)
if version == 0:
if include_synonyms:
raise UnsupportedVersionError(
"include_synonyms not supported on cluster version {}".
format(self.config['api_version']))
request = DescribeConfigsRequest[version](resources=[
self._convert_describe_config_resource_request(config_resource)
for config_resource in config_resources
])
elif version <= 1:
include_synonyms = include_synonyms or False
request = DescribeConfigsRequest[version](
resources=[
self._convert_describe_config_resource_request(
config_resource)
for config_resource in config_resources
],
include_synonyms=include_synonyms)
else:
raise UnsupportedVersionError(
"missing implementation of DescribeConfigs for library supported version {}"
.format(version))
return self._send(request)
@staticmethod
def _convert_alter_config_resource_request(config_resource):
return (config_resource.resource_type, config_resource.name, [
(config_key, config_value)
for config_key, config_value in config_resource.configs.items()
])
def alter_configs(self, config_resources):
"""Alter configuration parameters of one or more kafka resources.
:param config_resources: An array of ConfigResource objects.
:return: Appropriate version of AlterConfigsResponse class
"""
version = self._matching_api_version(AlterConfigsRequest)
if version == 0:
request = AlterConfigsRequest[version](resources=[
self._convert_alter_config_resource_request(config_resource)
for config_resource in config_resources
])
else:
raise UnsupportedVersionError(
"missing implementation of AlterConfigs for library supported version {}"
.format(version))
return self._send(request)
# alter replica logs dir protocol not implemented
# describe log dirs protocol not implemented
@staticmethod
def _convert_create_partitions_request(topic_name, new_partitions):
return (topic_name, (new_partitions.total_count,
new_partitions.new_assignments))
def create_partitions(self,
topic_partitions,
timeout_ms=None,
validate_only=None):
"""Create additional partitions for an existing topic.
:param topic_partitions: A map of topic name strings to NewPartition objects
:param timeout_ms: Milliseconds to wait for new partitions to be created before broker returns
:param validate_only: If True, don't actually create new partitions.
:return: Appropriate version of CreatePartitionsResponse class
"""
version = self._matching_api_version(CreatePartitionsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
validate_only = validate_only or False
if version == 0:
request = CreatePartitionsRequest[version](
topic_partitions=[
self._convert_create_partitions_request(
topic_name, new_partitions)
for topic_name, new_partitions in topic_partitions.items()
],
timeout=timeout_ms,
validate_only=validate_only)
else:
raise UnsupportedVersionError(
"missing implementation of CreatePartitions for library supported version {}"
.format(version))
return self._send(request)
# delete records protocol not implemented
# create delegation token protocol not implemented
# renew delegation token protocol not implemented
# expire delegation_token protocol not implemented
# describe delegation_token protocol not implemented
def describe_consumer_groups(self, group_ids):
"""Describe a set of consumer groups.
:param group_ids: A list of consumer group id names
:return: Appropriate version of DescribeGroupsResponse class
"""
version = self._matching_api_version(DescribeGroupsRequest)
if version <= 1:
request = DescribeGroupsRequest[version](groups=group_ids)
else:
raise UnsupportedVersionError(
"missing implementation of DescribeGroups for library supported version {}"
.format(version))
return self._send(request)
def list_consumer_groups(self):
"""List all consumer groups known to the cluster.
:return: Appropriate version of ListGroupsResponse class
"""
version = self._matching_api_version(ListGroupsRequest)
if version <= 1:
request = ListGroupsRequest[version]()
else:
raise UnsupportedVersionError(
"missing implementation of ListGroups for library supported version {}"
.format(version))
return self._send(request)
def coordinator(client):
return ConsumerCoordinator(client, SubscriptionState(), Metrics())
def __init__(self, **configs):
log.debug("Starting the Kafka producer") # trace
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % (configs,)
if self.config['client_id'] is None:
self.config['client_id'] = 'kafka-python-producer-%s' % \
(PRODUCER_CLIENT_ID_SEQUENCE.increment(),)
if self.config['acks'] == 'all':
self.config['acks'] = -1
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, deprecated.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
client = KafkaClient(metrics=self._metrics, metric_group_prefix='producer',
wakeup_timeout_ms=self.config['max_block_ms'],
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = client.config['api_version']
if self.config['compression_type'] == 'lz4':
assert self.config['api_version'] >= (0, 8, 2), 'LZ4 Requires >= Kafka 0.8.2 Brokers'
# Check compression_type for library support
ct = self.config['compression_type']
if ct not in self._COMPRESSORS:
raise ValueError("Not supported codec: {}".format(ct))
else:
checker, compression_attrs = self._COMPRESSORS[ct]
assert checker(), "Libraries for {} compression codec not found".format(ct)
self.config['compression_attrs'] = compression_attrs
message_version = self._max_usable_produce_magic()
self._accumulator = RecordAccumulator(message_version=message_version, metrics=self._metrics, **self.config)
self._metadata = client.cluster
guarantee_message_order = bool(self.config['max_in_flight_requests_per_connection'] == 1)
self._sender = Sender(client, self._metadata,
self._accumulator, self._metrics,
guarantee_message_order=guarantee_message_order,
**self.config)
self._sender.daemon = True
self._sender.start()
self._closed = False
self._cleanup = self._cleanup_factory()
atexit.register(self._cleanup)
log.debug("Kafka producer started")
class KafkaProducer(object):
"""A Kafka client that publishes records to the Kafka cluster.
The producer is thread safe and sharing a single producer instance across
threads will generally be faster than having multiple instances.
The producer consists of a pool of buffer space that holds records that
haven't yet been transmitted to the server as well as a background I/O
thread that is responsible for turning these records into requests and
transmitting them to the cluster.
:meth:`~kafka.KafkaProducer.send` is asynchronous. When called it adds the
record to a buffer of pending record sends and immediately returns. This
allows the producer to batch together individual records for efficiency.
The 'acks' config controls the criteria under which requests are considered
complete. The "all" setting will result in blocking on the full commit of
the record, the slowest but most durable setting.
If the request fails, the producer can automatically retry, unless
'retries' is configured to 0. Enabling retries also opens up the
possibility of duplicates (see the documentation on message
delivery semantics for details:
https://kafka.apache.org/documentation.html#semantics
).
The producer maintains buffers of unsent records for each partition. These
buffers are of a size specified by the 'batch_size' config. Making this
larger can result in more batching, but requires more memory (since we will
generally have one of these buffers for each active partition).
By default a buffer is available to send immediately even if there is
additional unused space in the buffer. However if you want to reduce the
number of requests you can set 'linger_ms' to something greater than 0.
This will instruct the producer to wait up to that number of milliseconds
before sending a request in hope that more records will arrive to fill up
the same batch. This is analogous to Nagle's algorithm in TCP. Note that
records that arrive close together in time will generally batch together
even with linger_ms=0 so under heavy load batching will occur regardless of
the linger configuration; however setting this to something larger than 0
can lead to fewer, more efficient requests when not under maximal load at
the cost of a small amount of latency.
The buffer_memory controls the total amount of memory available to the
producer for buffering. If records are sent faster than they can be
transmitted to the server then this buffer space will be exhausted. When
the buffer space is exhausted additional send calls will block.
The key_serializer and value_serializer instruct how to turn the key and
value objects the user provides into bytes.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the producer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client.
Default: 'kafka-python-producer-#' (appended with a unique number
per instance)
key_serializer (callable): used to convert user-supplied keys to bytes
If not None, called as f(key), should return bytes. Default: None.
value_serializer (callable): used to convert user-supplied message
values to bytes. If not None, called as f(value), should return
bytes. Default: None.
acks (0, 1, 'all'): The number of acknowledgments the producer requires
the leader to have received before considering a request complete.
This controls the durability of records that are sent. The
following settings are common:
0: Producer will not wait for any acknowledgment from the server.
The message will immediately be added to the socket
buffer and considered sent. No guarantee can be made that the
server has received the record in this case, and the retries
configuration will not take effect (as the client won't
generally know of any failures). The offset given back for each
record will always be set to -1.
1: Wait for leader to write the record to its local log only.
Broker will respond without awaiting full acknowledgement from
all followers. In this case should the leader fail immediately
after acknowledging the record but before the followers have
replicated it then the record will be lost.
all: Wait for the full set of in-sync replicas to write the record.
This guarantees that the record will not be lost as long as at
least one in-sync replica remains alive. This is the strongest
available guarantee.
If unset, defaults to acks=1.
compression_type (str): The compression type for all data generated by
the producer. Valid values are 'gzip', 'snappy', 'lz4', or None.
Compression is of full batches of data, so the efficacy of batching
will also impact the compression ratio (more batching means better
compression). Default: None.
retries (int): Setting a value greater than zero will cause the client
to resend any record whose send fails with a potentially transient
error. Note that this retry is no different than if the client
resent the record upon receiving the error. Allowing retries
without setting max_in_flight_requests_per_connection to 1 will
potentially change the ordering of records because if two batches
are sent to a single partition, and the first fails and is retried
but the second succeeds, then the records in the second batch may
appear first.
Default: 0.
batch_size (int): Requests sent to brokers will contain multiple
batches, one for each partition with data available to be sent.
A small batch size will make batching less common and may reduce
throughput (a batch size of zero will disable batching entirely).
Default: 16384
linger_ms (int): The producer groups together any records that arrive
in between request transmissions into a single batched request.
Normally this occurs only under load when records arrive faster
than they can be sent out. However in some circumstances the client
may want to reduce the number of requests even under moderate load.
This setting accomplishes this by adding a small amount of
artificial delay; that is, rather than immediately sending out a
record the producer will wait for up to the given delay to allow
other records to be sent so that the sends can be batched together.
This can be thought of as analogous to Nagle's algorithm in TCP.
This setting gives the upper bound on the delay for batching: once
we get batch_size worth of records for a partition it will be sent
immediately regardless of this setting, however if we have fewer
than this many bytes accumulated for this partition we will
'linger' for the specified time waiting for more records to show
up. This setting defaults to 0 (i.e. no delay). Setting linger_ms=5
would have the effect of reducing the number of requests sent but
would add up to 5ms of latency to records sent in the absense of
load. Default: 0.
partitioner (callable): Callable used to determine which partition
each message is assigned to. Called (after key serialization):
partitioner(key_bytes, all_partitions, available_partitions).
The default partitioner implementation hashes each non-None key
using the same murmur2 algorithm as the java client so that
messages with the same key are assigned to the same partition.
When a key is None, the message is delivered to a random partition
(filtered to partitions with available leaders only, if possible).
buffer_memory (int): The total bytes of memory the producer should use
to buffer records waiting to be sent to the server. If records are
sent faster than they can be delivered to the server the producer
will block up to max_block_ms, raising an exception on timeout.
In the current implementation, this setting is an approximation.
Default: 33554432 (32MB)
connections_max_idle_ms: Close idle connections after the number of
milliseconds specified by this config. The broker closes idle
connections after connections.max.idle.ms, so this avoids hitting
unexpected socket disconnected errors on the client.
Default: 540000
max_block_ms (int): Number of milliseconds to block during
:meth:`~kafka.KafkaProducer.send` and
:meth:`~kafka.KafkaProducer.partitions_for`. These methods can be
blocked either because the buffer is full or metadata unavailable.
Blocking in the user-supplied serializers or partitioner will not be
counted against this timeout. Default: 60000.
max_request_size (int): The maximum size of a request. This is also
effectively a cap on the maximum record size. Note that the server
has its own cap on record size which may be different from this.
This setting will limit the number of record batches the producer
will send in a single request to avoid sending huge requests.
Default: 1048576.
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 30000.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). Java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). Java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
reconnect_backoff_max_ms (int): The maximum amount of time in
milliseconds to wait when reconnecting to a broker that has
repeatedly failed to connect. If provided, the backoff per host
will increase exponentially for each consecutive connection
failure, up to this maximum. To avoid connection storms, a
randomization factor of 0.2 will be applied to the backoff
resulting in a random range between 20% below and 20% above
the computed value. Default: 1000.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Note that if this setting is set to be greater
than 1 and there are failed sends, there is a risk of message
re-ordering due to retries (i.e., if retries are enabled).
Default: 5.
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL, SASL_PLAINTEXT, SASL_SSL.
Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
default: true.
ssl_cafile (str): optional filename of ca file to use in certificate
veriication. default: none.
ssl_certfile (str): optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. default: none.
ssl_keyfile (str): optional filename containing the client private key.
default: none.
ssl_password (str): optional password to be used when loading the
certificate chain. default: none.
ssl_crlfile (str): optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
default: none.
ssl_ciphers (str): optionally set the available ciphers for ssl
connections. It should be a string in the OpenSSL cipher list
format. If no cipher can be selected (because compile-time options
or other configuration forbids use of all the specified ciphers),
an ssl.SSLError will be raised. See ssl.SSLContext.set_ciphers
api_version (tuple): Specify which Kafka API version to use. If set to
None, the client will attempt to infer the broker version by probing
various APIs. Example: (0, 10, 2). Default: None
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to 'auto'
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The maximum age in milliseconds of
samples used to compute metrics. Default: 30000
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
sasl_mechanism (str): Authentication mechanism when security_protocol
is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are:
PLAIN, GSSAPI, OAUTHBEARER.
sasl_plain_username (str): username for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_plain_password (str): password for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_kerberos_service_name (str): Service name to include in GSSAPI
sasl mechanism handshake. Default: 'kafka'
sasl_kerberos_domain_name (str): kerberos domain name to use in GSSAPI
sasl mechanism handshake. Default: one of bootstrap servers
sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider
instance. (See kafka.oauth.abstract). Default: None
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/0100/configuration.html#producerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': None,
'key_serializer': None,
'value_serializer': None,
'acks': 1,
'bootstrap_topics_filter': set(),
'compression_type': None,
'retries': 0,
'batch_size': 16384,
'linger_ms': 0,
'partitioner': DefaultPartitioner(),
'buffer_memory': 33554432,
'connections_max_idle_ms': 9 * 60 * 1000,
'max_block_ms': 60000,
'max_request_size': 1048576,
'metadata_max_age_ms': 300000,
'retry_backoff_ms': 100,
'request_timeout_ms': 30000,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'sock_chunk_bytes': 4096, # undocumented experimental option
'sock_chunk_buffer_count': 1000, # undocumented experimental option
'reconnect_backoff_ms': 50,
'reconnect_backoff_max_ms': 1000,
'max_in_flight_requests_per_connection': 5,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_crlfile': None,
'ssl_password': None,
'ssl_ciphers': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
'selector': selectors.DefaultSelector,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
'sasl_kerberos_service_name': 'kafka',
'sasl_kerberos_domain_name': None,
'sasl_oauth_token_provider': None
}
_COMPRESSORS = {
'gzip': (has_gzip, LegacyRecordBatchBuilder.CODEC_GZIP),
'snappy': (has_snappy, LegacyRecordBatchBuilder.CODEC_SNAPPY),
'lz4': (has_lz4, LegacyRecordBatchBuilder.CODEC_LZ4),
None: (lambda: True, LegacyRecordBatchBuilder.CODEC_NONE),
}
def __init__(self, **configs):
log.debug("Starting the Kafka producer") # trace
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % (configs,)
if self.config['client_id'] is None:
self.config['client_id'] = 'kafka-python-producer-%s' % \
(PRODUCER_CLIENT_ID_SEQUENCE.increment(),)
if self.config['acks'] == 'all':
self.config['acks'] = -1
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
deprecated = self.config['api_version']
if deprecated == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, deprecated.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), deprecated)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
client = KafkaClient(metrics=self._metrics, metric_group_prefix='producer',
wakeup_timeout_ms=self.config['max_block_ms'],
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = client.config['api_version']
if self.config['compression_type'] == 'lz4':
assert self.config['api_version'] >= (0, 8, 2), 'LZ4 Requires >= Kafka 0.8.2 Brokers'
# Check compression_type for library support
ct = self.config['compression_type']
if ct not in self._COMPRESSORS:
raise ValueError("Not supported codec: {}".format(ct))
else:
checker, compression_attrs = self._COMPRESSORS[ct]
assert checker(), "Libraries for {} compression codec not found".format(ct)
self.config['compression_attrs'] = compression_attrs
message_version = self._max_usable_produce_magic()
self._accumulator = RecordAccumulator(message_version=message_version, metrics=self._metrics, **self.config)
self._metadata = client.cluster
guarantee_message_order = bool(self.config['max_in_flight_requests_per_connection'] == 1)
self._sender = Sender(client, self._metadata,
self._accumulator, self._metrics,
guarantee_message_order=guarantee_message_order,
**self.config)
self._sender.daemon = True
self._sender.start()
self._closed = False
self._cleanup = self._cleanup_factory()
atexit.register(self._cleanup)
log.debug("Kafka producer started")
def _cleanup_factory(self):
"""Build a cleanup clojure that doesn't increase our ref count"""
_self = weakref.proxy(self)
def wrapper():
try:
_self.close(timeout=0)
except (ReferenceError, AttributeError):
pass
return wrapper
def _unregister_cleanup(self):
if getattr(self, '_cleanup', None):
if hasattr(atexit, 'unregister'):
atexit.unregister(self._cleanup) # pylint: disable=no-member
# py2 requires removing from private attribute...
else:
# ValueError on list.remove() if the exithandler no longer exists
# but that is fine here
try:
atexit._exithandlers.remove( # pylint: disable=no-member
(self._cleanup, (), {}))
except ValueError:
pass
self._cleanup = None
def __del__(self):
self.close(timeout=0)
def close(self, timeout=None):
"""Close this producer.
Arguments:
timeout (float, optional): timeout in seconds to wait for completion.
"""
# drop our atexit handler now to avoid leaks
self._unregister_cleanup()
if not hasattr(self, '_closed') or self._closed:
log.info('Kafka producer closed')
return
if timeout is None:
# threading.TIMEOUT_MAX is available in Python3.3+
timeout = getattr(threading, 'TIMEOUT_MAX', float('inf'))
if getattr(threading, 'TIMEOUT_MAX', False):
assert 0 <= timeout <= getattr(threading, 'TIMEOUT_MAX')
else:
assert timeout >= 0
log.info("Closing the Kafka producer with %s secs timeout.", timeout)
#first_exception = AtomicReference() # this will keep track of the first encountered exception
invoked_from_callback = bool(threading.current_thread() is self._sender)
if timeout > 0:
if invoked_from_callback:
log.warning("Overriding close timeout %s secs to 0 in order to"
" prevent useless blocking due to self-join. This"
" means you have incorrectly invoked close with a"
" non-zero timeout from the producer call-back.",
timeout)
else:
# Try to close gracefully.
if self._sender is not None:
self._sender.initiate_close()
self._sender.join(timeout)
if self._sender is not None and self._sender.is_alive():
log.info("Proceeding to force close the producer since pending"
" requests could not be completed within timeout %s.",
timeout)
self._sender.force_close()
# Only join the sender thread when not calling from callback.
if not invoked_from_callback:
self._sender.join()
self._metrics.close()
try:
self.config['key_serializer'].close()
except AttributeError:
pass
try:
self.config['value_serializer'].close()
except AttributeError:
pass
self._closed = True
log.debug("The Kafka producer has closed.")
def partitions_for(self, topic):
"""Returns set of all known partitions for the topic."""
max_wait = self.config['max_block_ms'] / 1000.0
return self._wait_on_metadata(topic, max_wait)
def _max_usable_produce_magic(self):
if self.config['api_version'] >= (0, 11):
return 2
elif self.config['api_version'] >= (0, 10):
return 1
else:
return 0
def _estimate_size_in_bytes(self, key, value, headers=[]):
magic = self._max_usable_produce_magic()
if magic == 2:
return DefaultRecordBatchBuilder.estimate_size_in_bytes(
key, value, headers)
else:
return LegacyRecordBatchBuilder.estimate_size_in_bytes(
magic, self.config['compression_type'], key, value)
def send(self, topic, value=None, key=None, headers=None, partition=None, timestamp_ms=None):
"""Publish a message to a topic.
Arguments:
topic (str): topic where the message will be published
value (optional): message value. Must be type bytes, or be
serializable to bytes via configured value_serializer. If value
is None, key is required and message acts as a 'delete'.
See kafka compaction documentation for more details:
https://kafka.apache.org/documentation.html#compaction
(compaction requires kafka >= 0.8.1)
partition (int, optional): optionally specify a partition. If not
set, the partition will be selected using the configured
'partitioner'.
key (optional): a key to associate with the message. Can be used to
determine which partition to send the message to. If partition
is None (and producer's partitioner config is left as default),
then messages with the same key will be delivered to the same
partition (but if key is None, partition is chosen randomly).
Must be type bytes, or be serializable to bytes via configured
key_serializer.
headers (optional): a list of header key value pairs. List items
are tuples of str key and bytes value.
timestamp_ms (int, optional): epoch milliseconds (from Jan 1 1970 UTC)
to use as the message timestamp. Defaults to current time.
Returns:
FutureRecordMetadata: resolves to RecordMetadata
Raises:
KafkaTimeoutError: if unable to fetch topic metadata, or unable
to obtain memory buffer prior to configured max_block_ms
"""
assert value is not None or self.config['api_version'] >= (0, 8, 1), (
'Null messages require kafka >= 0.8.1')
assert not (value is None and key is None), 'Need at least one: key or value'
key_bytes = value_bytes = None
try:
self._wait_on_metadata(topic, self.config['max_block_ms'] / 1000.0)
key_bytes = self._serialize(
self.config['key_serializer'],
topic, key)
value_bytes = self._serialize(
self.config['value_serializer'],
topic, value)
assert type(key_bytes) in (bytes, bytearray, memoryview, type(None))
assert type(value_bytes) in (bytes, bytearray, memoryview, type(None))
partition = self._partition(topic, partition, key, value,
key_bytes, value_bytes)
if headers is None:
headers = []
assert type(headers) == list
assert all(type(item) == tuple and len(item) == 2 and type(item[0]) == str and type(item[1]) == bytes for item in headers)
message_size = self._estimate_size_in_bytes(key_bytes, value_bytes, headers)
self._ensure_valid_record_size(message_size)
tp = TopicPartition(topic, partition)
log.debug("Sending (key=%r value=%r headers=%r) to %s", key, value, headers, tp)
result = self._accumulator.append(tp, timestamp_ms,
key_bytes, value_bytes, headers,
self.config['max_block_ms'],
estimated_size=message_size)
future, batch_is_full, new_batch_created = result
if batch_is_full or new_batch_created:
log.debug("Waking up the sender since %s is either full or"
" getting a new batch", tp)
self._sender.wakeup()
return future
# handling exceptions and record the errors;
# for API exceptions return them in the future,
# for other exceptions raise directly
except Errors.BrokerResponseError as e:
log.debug("Exception occurred during message send: %s", e)
return FutureRecordMetadata(
FutureProduceResult(TopicPartition(topic, partition)),
-1, None, None,
len(key_bytes) if key_bytes is not None else -1,
len(value_bytes) if value_bytes is not None else -1,
sum(len(h_key.encode("utf-8")) + len(h_value) for h_key, h_value in headers) if headers else -1,
).failure(e)
def flush(self, timeout=None):
"""
Invoking this method makes all buffered records immediately available
to send (even if linger_ms is greater than 0) and blocks on the
completion of the requests associated with these records. The
post-condition of :meth:`~kafka.KafkaProducer.flush` is that any
previously sent record will have completed
(e.g. Future.is_done() == True). A request is considered completed when
either it is successfully acknowledged according to the 'acks'
configuration for the producer, or it results in an error.
Other threads can continue sending messages while one thread is blocked
waiting for a flush call to complete; however, no guarantee is made
about the completion of messages sent after the flush call begins.
Arguments:
timeout (float, optional): timeout in seconds to wait for completion.
Raises:
KafkaTimeoutError: failure to flush buffered records within the
provided timeout
"""
log.debug("Flushing accumulated records in producer.") # trace
self._accumulator.begin_flush()
self._sender.wakeup()
self._accumulator.await_flush_completion(timeout=timeout)
def _ensure_valid_record_size(self, size):
"""Validate that the record size isn't too large."""
if size > self.config['max_request_size']:
raise Errors.MessageSizeTooLargeError(
"The message is %d bytes when serialized which is larger than"
" the maximum request size you have configured with the"
" max_request_size configuration" % (size,))
if size > self.config['buffer_memory']:
raise Errors.MessageSizeTooLargeError(
"The message is %d bytes when serialized which is larger than"
" the total memory buffer you have configured with the"
" buffer_memory configuration." % (size,))
def _wait_on_metadata(self, topic, max_wait):
"""
Wait for cluster metadata including partitions for the given topic to
be available.
Arguments:
topic (str): topic we want metadata for
max_wait (float): maximum time in secs for waiting on the metadata
Returns:
set: partition ids for the topic
Raises:
KafkaTimeoutError: if partitions for topic were not obtained before
specified max_wait timeout
"""
# add topic to metadata topic list if it is not there already.
self._sender.add_topic(topic)
begin = time.time()
elapsed = 0.0
metadata_event = None
while True:
partitions = self._metadata.partitions_for_topic(topic)
if partitions is not None:
return partitions
if not metadata_event:
metadata_event = threading.Event()
log.debug("Requesting metadata update for topic %s", topic)
metadata_event.clear()
future = self._metadata.request_update()
future.add_both(lambda e, *args: e.set(), metadata_event)
self._sender.wakeup()
metadata_event.wait(max_wait - elapsed)
elapsed = time.time() - begin
if not metadata_event.is_set():
raise Errors.KafkaTimeoutError(
"Failed to update metadata after %.1f secs." % (max_wait,))
elif topic in self._metadata.unauthorized_topics:
raise Errors.TopicAuthorizationFailedError(topic)
else:
log.debug("_wait_on_metadata woke after %s secs.", elapsed)
def _serialize(self, f, topic, data):
if not f:
return data
if isinstance(f, Serializer):
return f.serialize(topic, data)
return f(data)
def _partition(self, topic, partition, key, value,
serialized_key, serialized_value):
if partition is not None:
assert partition >= 0
assert partition in self._metadata.partitions_for_topic(topic), 'Unrecognized partition'
return partition
all_partitions = sorted(self._metadata.partitions_for_topic(topic))
available = list(self._metadata.available_partitions_for_topic(topic))
return self.config['partitioner'](serialized_key,
all_partitions,
available)
def metrics(self, raw=False):
"""Get metrics on producer performance.
This is ported from the Java Producer, for details see:
https://kafka.apache.org/documentation/#producer_monitoring
Warning:
This is an unstable interface. It may change in future
releases without warning.
"""
if raw:
return self._metrics.metrics.copy()
metrics = {}
for k, v in six.iteritems(self._metrics.metrics.copy()):
if k.group not in metrics:
metrics[k.group] = {}
if k.name not in metrics[k.group]:
metrics[k.group][k.name] = {}
metrics[k.group][k.name] = v.value()
return metrics
def coordinator(conn):
return ConsumerCoordinator(KafkaClient(), SubscriptionState(), Metrics(),
'consumer')
def metrics(request, config, reporter):
metrics = Metrics(config, [reporter], enable_expiration=True)
request.addfinalizer(lambda: metrics.close())
return metrics
class KafkaAdminClient(object):
"""A class for administering the Kafka cluster.
Warning:
This is an unstable interface that was recently added and is subject to
change without warning. In particular, many methods currently return
raw protocol tuples. In future releases, we plan to make these into
nicer, more pythonic objects. Unfortunately, this will likely break
those interfaces.
The KafkaAdminClient class will negotiate for the latest version of each message
protocol format supported by both the kafka-python client library and the
Kafka broker. Usage of optional fields from protocol versions that are not
supported by the broker will result in IncompatibleBrokerVersion exceptions.
Use of this class requires a minimum broker version >= 0.10.0.0.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
reconnect_backoff_max_ms (int): The maximum amount of time in
milliseconds to wait when reconnecting to a broker that has
repeatedly failed to connect. If provided, the backoff per host
will increase exponentially for each consecutive connection
failure, up to this maximum. To avoid connection storms, a
randomization factor of 0.2 will be applied to the backoff
resulting in a random range between 20% below and 20% above
the computed value. Default: 1000.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 30000.
connections_max_idle_ms: Close idle connections after the number of
milliseconds specified by this config. The broker closes idle
connections after connections.max.idle.ms, so this avoids hitting
unexpected socket disconnected errors on the client.
Default: 540000
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). Java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). Java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): Pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): Flag to configure whether SSL handshake
should verify that the certificate matches the broker's hostname.
Default: True.
ssl_cafile (str): Optional filename of CA file to use in certificate
veriication. Default: None.
ssl_certfile (str): Optional filename of file in PEM format containing
the client certificate, as well as any CA certificates needed to
establish the certificate's authenticity. Default: None.
ssl_keyfile (str): Optional filename containing the client private key.
Default: None.
ssl_password (str): Optional password to be used when loading the
certificate chain. Default: None.
ssl_crlfile (str): Optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
Default: None.
api_version (tuple): Specify which Kafka API version to use. If set
to None, KafkaClient will attempt to infer the broker version by
probing various APIs. Example: (0, 10, 2). Default: None
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version is None
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
metrics (kafka.metrics.Metrics): Optionally provide a metrics
instance for capturing network IO stats. Default: None.
metric_group_prefix (str): Prefix for metric names. Default: ''
sasl_mechanism (str): Authentication mechanism when security_protocol
is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are:
PLAIN, GSSAPI, OAUTHBEARER.
sasl_plain_username (str): username for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_plain_password (str): password for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_kerberos_service_name (str): Service name to include in GSSAPI
sasl mechanism handshake. Default: 'kafka'
sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider
instance. (See kafka.oauth.abstract). Default: None
"""
DEFAULT_CONFIG = {
# client configs
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'request_timeout_ms': 30000,
'connections_max_idle_ms': 9 * 60 * 1000,
'reconnect_backoff_ms': 50,
'reconnect_backoff_max_ms': 1000,
'max_in_flight_requests_per_connection': 5,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'sock_chunk_bytes': 4096, # undocumented experimental option
'sock_chunk_buffer_count': 1000, # undocumented experimental option
'retry_backoff_ms': 100,
'metadata_max_age_ms': 300000,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_password': None,
'ssl_crlfile': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'selector': selectors.DefaultSelector,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
'sasl_kerberos_service_name': 'kafka',
'sasl_oauth_token_provider': None,
# metrics configs
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
}
def __init__(self, **configs):
log.debug("Starting KafkaAdminClient with configuration: %s", configs)
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError(
"Unrecognized configs: {}".format(extra_configs))
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(
samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [
reporter() for reporter in self.config['metric_reporters']
]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug("KafkaAdminClient started.")
def close(self):
"""Close the KafkaAdminClient connection to the Kafka broker."""
if not hasattr(self, '_closed') or self._closed:
log.info("KafkaAdminClient already closed.")
return
self._metrics.close()
self._client.close()
self._closed = True
log.debug("KafkaAdminClient is now closed.")
def _matching_api_version(self, operation):
"""Find the latest version of the protocol operation supported by both
this library and the broker.
This resolves to the lesser of either the latest api version this
library supports, or the max version supported by the broker.
:param operation: A list of protocol operation versions from kafka.protocol.
:return: The max matching version number between client and broker.
"""
version = min(
len(operation) - 1,
self._client.get_api_versions()[operation[0].API_KEY][1])
if version < self._client.get_api_versions()[operation[0].API_KEY][0]:
# max library version is less than min broker version. Currently,
# no Kafka versions specify a min msg version. Maybe in the future?
raise IncompatibleBrokerVersion(
"No version of the '{}' Kafka protocol is supported by both the client and broker."
.format(operation.__name__))
return version
def _validate_timeout(self, timeout_ms):
"""Validate the timeout is set or use the configuration default.
:param timeout_ms: The timeout provided by api call, in milliseconds.
:return: The timeout to use for the operation.
"""
return timeout_ms or self.config['request_timeout_ms']
def _refresh_controller_id(self):
"""Determine the Kafka cluster controller."""
version = self._matching_api_version(MetadataRequest)
if 1 <= version <= 6:
request = MetadataRequest[version]()
future = self._send_request_to_node(
self._client.least_loaded_node(), request)
self._wait_for_futures([future])
response = future.value
controller_id = response.controller_id
# verify the controller is new enough to support our requests
controller_version = self._client.check_version(controller_id)
if controller_version < (0, 10, 0):
raise IncompatibleBrokerVersion(
"The controller appears to be running Kafka {}. KafkaAdminClient requires brokers >= 0.10.0.0."
.format(controller_version))
self._controller_id = controller_id
else:
raise UnrecognizedBrokerVersion(
"Kafka Admin interface cannot determine the controller using MetadataRequest_v{}."
.format(version))
def _find_group_coordinator_id(self, group_id):
"""Find the broker node_id of the coordinator of the given group.
Sends a FindCoordinatorRequest message to the cluster. Will block until
the FindCoordinatorResponse is received. Any errors are immediately
raised.
:param group_id: The consumer group ID. This is typically the group
name as a string.
:return: The node_id of the broker that is the coordinator.
"""
# Note: Java may change how this is implemented in KAFKA-6791.
#
# TODO add support for dynamically picking version of
# GroupCoordinatorRequest which was renamed to FindCoordinatorRequest.
# When I experimented with this, GroupCoordinatorResponse_v1 didn't
# match GroupCoordinatorResponse_v0 and I couldn't figure out why.
gc_request = GroupCoordinatorRequest[0](group_id)
future = self._send_request_to_node(self._client.least_loaded_node(),
gc_request)
self._wait_for_futures([future])
gc_response = future.value
# use the extra error checking in add_group_coordinator() rather than
# immediately returning the group coordinator.
success = self._client.cluster.add_group_coordinator(
group_id, gc_response)
if not success:
error_type = Errors.for_code(gc_response.error_code)
assert error_type is not Errors.NoError
# Note: When error_type.retriable, Java will retry... see
# KafkaAdminClient's handleFindCoordinatorError method
raise error_type(
"Could not identify group coordinator for group_id '{}' from response '{}'."
.format(group_id, gc_response))
group_coordinator = self._client.cluster.coordinator_for_group(
group_id)
# will be None if the coordinator was never populated, which should never happen here
assert group_coordinator is not None
# will be -1 if add_group_coordinator() failed... but by this point the
# error should have been raised.
assert group_coordinator != -1
return group_coordinator
def _send_request_to_node(self, node_id, request):
"""Send a Kafka protocol message to a specific broker.
Returns a future that may be polled for status and results.
:param node_id: The broker id to which to send the message.
:param request: The message to send.
:return: A future object that may be polled for status and results.
:exception: The exception if the message could not be sent.
"""
while not self._client.ready(node_id):
# poll until the connection to broker is ready, otherwise send()
# will fail with NodeNotReadyError
self._client.poll()
return self._client.send(node_id, request)
def _send_request_to_controller(self, request):
"""Send a Kafka protocol message to the cluster controller.
Will block until the message result is received.
:param request: The message to send.
:return: The Kafka protocol response for the message.
"""
tries = 2 # in case our cached self._controller_id is outdated
while tries:
tries -= 1
future = self._send_request_to_node(self._controller_id, request)
self._wait_for_futures([future])
response = future.value
# In Java, the error fieldname is inconsistent:
# - CreateTopicsResponse / CreatePartitionsResponse uses topic_errors
# - DeleteTopicsResponse uses topic_error_codes
# So this is a little brittle in that it assumes all responses have
# one of these attributes and that they always unpack into
# (topic, error_code) tuples.
topic_error_tuples = (response.topic_errors if hasattr(
response, 'topic_errors') else response.topic_error_codes)
# Also small py2/py3 compatibility -- py3 can ignore extra values
# during unpack via: for x, y, *rest in list_of_values. py2 cannot.
# So for now we have to map across the list and explicitly drop any
# extra values (usually the error_message)
for topic, error_code in map(lambda e: e[:2], topic_error_tuples):
error_type = Errors.for_code(error_code)
if tries and error_type is NotControllerError:
# No need to inspect the rest of the errors for
# non-retriable errors because NotControllerError should
# either be thrown for all errors or no errors.
self._refresh_controller_id()
break
elif error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'.".format(
request, response))
else:
return response
raise RuntimeError(
"This should never happen, please file a bug with full stacktrace if encountered"
)
@staticmethod
def _convert_new_topic_request(new_topic):
return (
new_topic.name, new_topic.num_partitions,
new_topic.replication_factor, [
(partition_id, replicas) for partition_id, replicas in
new_topic.replica_assignments.items()
],
[(config_key, config_value)
for config_key, config_value in new_topic.topic_configs.items()])
def create_topics(self, new_topics, timeout_ms=None, validate_only=False):
"""Create new topics in the cluster.
:param new_topics: A list of NewTopic objects.
:param timeout_ms: Milliseconds to wait for new topics to be created
before the broker returns.
:param validate_only: If True, don't actually create new topics.
Not supported by all versions. Default: False
:return: Appropriate version of CreateTopicResponse class.
"""
version = self._matching_api_version(CreateTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version == 0:
if validate_only:
raise IncompatibleBrokerVersion(
"validate_only requires CreateTopicsRequest >= v1, which is not supported by Kafka {}."
.format(self.config['api_version']))
request = CreateTopicsRequest[version](create_topic_requests=[
self._convert_new_topic_request(new_topic)
for new_topic in new_topics
],
timeout=timeout_ms)
elif version <= 2:
request = CreateTopicsRequest[version](create_topic_requests=[
self._convert_new_topic_request(new_topic)
for new_topic in new_topics
],
timeout=timeout_ms,
validate_only=validate_only)
else:
raise NotImplementedError(
"Support for CreateTopics v{} has not yet been added to KafkaAdminClient."
.format(version))
# TODO convert structs to a more pythonic interface
# TODO raise exceptions if errors
return self._send_request_to_controller(request)
def delete_topics(self, topics, timeout_ms=None):
"""Delete topics from the cluster.
:param topics: A list of topic name strings.
:param timeout_ms: Milliseconds to wait for topics to be deleted
before the broker returns.
:return: Appropriate version of DeleteTopicsResponse class.
"""
version = self._matching_api_version(DeleteTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version <= 1:
request = DeleteTopicsRequest[version](topics=topics,
timeout=timeout_ms)
response = self._send_request_to_controller(request)
else:
raise NotImplementedError(
"Support for DeleteTopics v{} has not yet been added to KafkaAdminClient."
.format(version))
return response
# list topics functionality is in ClusterMetadata
# Note: if implemented here, send the request to the least_loaded_node()
# describe topics functionality is in ClusterMetadata
# Note: if implemented here, send the request to the controller
# describe cluster functionality is in ClusterMetadata
# Note: if implemented here, send the request to the least_loaded_node()
# describe_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
# create_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
# delete_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
@staticmethod
def _convert_describe_config_resource_request(config_resource):
return (config_resource.resource_type, config_resource.name, [
config_key
for config_key, config_value in config_resource.configs.items()
] if config_resource.configs else None)
def describe_configs(self, config_resources, include_synonyms=False):
"""Fetch configuration parameters for one or more Kafka resources.
:param config_resources: An list of ConfigResource objects.
Any keys in ConfigResource.configs dict will be used to filter the
result. Setting the configs dict to None will get all values. An
empty dict will get zero values (as per Kafka protocol).
:param include_synonyms: If True, return synonyms in response. Not
supported by all versions. Default: False.
:return: Appropriate version of DescribeConfigsResponse class.
"""
version = self._matching_api_version(DescribeConfigsRequest)
if version == 0:
if include_synonyms:
raise IncompatibleBrokerVersion(
"include_synonyms requires DescribeConfigsRequest >= v1, which is not supported by Kafka {}."
.format(self.config['api_version']))
request = DescribeConfigsRequest[version](resources=[
self._convert_describe_config_resource_request(config_resource)
for config_resource in config_resources
])
elif version == 1:
request = DescribeConfigsRequest[version](
resources=[
self._convert_describe_config_resource_request(
config_resource)
for config_resource in config_resources
],
include_synonyms=include_synonyms)
else:
raise NotImplementedError(
"Support for DescribeConfigs v{} has not yet been added to KafkaAdminClient."
.format(version))
future = self._send_request_to_node(self._client.least_loaded_node(),
request)
self._wait_for_futures([future])
return future.value
@staticmethod
def _convert_alter_config_resource_request(config_resource):
return (config_resource.resource_type, config_resource.name, [
(config_key, config_value)
for config_key, config_value in config_resource.configs.items()
])
def alter_configs(self, config_resources):
"""Alter configuration parameters of one or more Kafka resources.
Warning:
This is currently broken for BROKER resources because those must be
sent to that specific broker, versus this always picks the
least-loaded node. See the comment in the source code for details.
We would happily accept a PR fixing this.
:param config_resources: A list of ConfigResource objects.
:return: Appropriate version of AlterConfigsResponse class.
"""
version = self._matching_api_version(AlterConfigsRequest)
if version == 0:
request = AlterConfigsRequest[version](resources=[
self._convert_alter_config_resource_request(config_resource)
for config_resource in config_resources
])
else:
raise NotImplementedError(
"Support for AlterConfigs v{} has not yet been added to KafkaAdminClient."
.format(version))
# TODO the Java client has the note:
# // We must make a separate AlterConfigs request for every BROKER resource we want to alter
# // and send the request to that specific broker. Other resources are grouped together into
# // a single request that may be sent to any broker.
#
# So this is currently broken as it always sends to the least_loaded_node()
future = self._send_request_to_node(self._client.least_loaded_node(),
request)
self._wait_for_futures([future])
return future.value
# alter replica logs dir protocol not yet implemented
# Note: have to lookup the broker with the replica assignment and send the request to that broker
# describe log dirs protocol not yet implemented
# Note: have to lookup the broker with the replica assignment and send the request to that broker
@staticmethod
def _convert_create_partitions_request(topic_name, new_partitions):
return (topic_name, (new_partitions.total_count,
new_partitions.new_assignments))
def create_partitions(self,
topic_partitions,
timeout_ms=None,
validate_only=False):
"""Create additional partitions for an existing topic.
:param topic_partitions: A map of topic name strings to NewPartition objects.
:param timeout_ms: Milliseconds to wait for new partitions to be
created before the broker returns.
:param validate_only: If True, don't actually create new partitions.
Default: False
:return: Appropriate version of CreatePartitionsResponse class.
"""
version = self._matching_api_version(CreatePartitionsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version == 0:
request = CreatePartitionsRequest[version](
topic_partitions=[
self._convert_create_partitions_request(
topic_name, new_partitions)
for topic_name, new_partitions in topic_partitions.items()
],
timeout=timeout_ms,
validate_only=validate_only)
else:
raise NotImplementedError(
"Support for CreatePartitions v{} has not yet been added to KafkaAdminClient."
.format(version))
return self._send_request_to_controller(request)
# delete records protocol not yet implemented
# Note: send the request to the partition leaders
# create delegation token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# renew delegation token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# expire delegation_token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# describe delegation_token protocol not yet implemented
# Note: send the request to the least_loaded_node()
def describe_consumer_groups(self, group_ids, group_coordinator_id=None):
"""Describe a set of consumer groups.
Any errors are immediately raised.
:param group_ids: A list of consumer group IDs. These are typically the
group names as strings.
:param group_coordinator_id: The node_id of the groups' coordinator
broker. If set to None, it will query the cluster for each group to
find that group's coordinator. Explicitly specifying this can be
useful for avoiding extra network round trips if you already know
the group coordinator. This is only useful when all the group_ids
have the same coordinator, otherwise it will error. Default: None.
:return: A list of group descriptions. For now the group descriptions
are the raw results from the DescribeGroupsResponse. Long-term, we
plan to change this to return namedtuples as well as decoding the
partition assignments.
"""
group_descriptions = []
futures = []
version = self._matching_api_version(DescribeGroupsRequest)
for group_id in group_ids:
if group_coordinator_id is not None:
this_groups_coordinator_id = group_coordinator_id
else:
this_groups_coordinator_id = self._find_group_coordinator_id(
group_id)
if version <= 1:
# Note: KAFKA-6788 A potential optimization is to group the
# request per coordinator and send one request with a list of
# all consumer groups. Java still hasn't implemented this
# because the error checking is hard to get right when some
# groups error and others don't.
request = DescribeGroupsRequest[version](groups=(group_id, ))
futures.append(
self._send_request_to_node(this_groups_coordinator_id,
request))
else:
raise NotImplementedError(
"Support for DescribeGroups v{} has not yet been added to KafkaAdminClient."
.format(version))
self._wait_for_futures(futures)
for future in futures:
response = future.value
assert len(response.groups) == 1
# TODO need to implement converting the response tuple into
# a more accessible interface like a namedtuple and then stop
# hardcoding tuple indices here. Several Java examples,
# including KafkaAdminClient.java
group_description = response.groups[0]
error_code = group_description[0]
error_type = Errors.for_code(error_code)
# Java has the note: KAFKA-6789, we can retry based on the error code
if error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'.".format(
request, response))
# TODO Java checks the group protocol type, and if consumer
# (ConsumerProtocol.PROTOCOL_TYPE) or empty string, it decodes
# the members' partition assignments... that hasn't yet been
# implemented here so just return the raw struct results
group_descriptions.append(group_description)
return group_descriptions
def list_consumer_groups(self, broker_ids=None):
"""List all consumer groups known to the cluster.
This returns a list of Consumer Group tuples. The tuples are
composed of the consumer group name and the consumer group protocol
type.
Only consumer groups that store their offsets in Kafka are returned.
The protocol type will be an empty string for groups created using
Kafka < 0.9 APIs because, although they store their offsets in Kafka,
they don't use Kafka for group coordination. For groups created using
Kafka >= 0.9, the protocol type will typically be "consumer".
As soon as any error is encountered, it is immediately raised.
:param broker_ids: A list of broker node_ids to query for consumer
groups. If set to None, will query all brokers in the cluster.
Explicitly specifying broker(s) can be useful for determining which
consumer groups are coordinated by those broker(s). Default: None
:return list: List of tuples of Consumer Groups.
:exception GroupCoordinatorNotAvailableError: The coordinator is not
available, so cannot process requests.
:exception GroupLoadInProgressError: The coordinator is loading and
hence can't process requests.
"""
# While we return a list, internally use a set to prevent duplicates
# because if a group coordinator fails after being queried, and its
# consumer groups move to new brokers that haven't yet been queried,
# then the same group could be returned by multiple brokers.
consumer_groups = set()
futures = []
if broker_ids is None:
broker_ids = [
broker.nodeId for broker in self._client.cluster.brokers()
]
version = self._matching_api_version(ListGroupsRequest)
if version <= 2:
request = ListGroupsRequest[version]()
for broker_id in broker_ids:
futures.append(self._send_request_to_node(broker_id, request))
self._wait_for_futures(futures)
for future in futures:
response = future.value
error_type = Errors.for_code(response.error_code)
if error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'.".format(
request, response))
consumer_groups.update(response.groups)
else:
raise NotImplementedError(
"Support for ListGroups v{} has not yet been added to KafkaAdminClient."
.format(version))
return list(consumer_groups)
def list_consumer_group_offsets(self,
group_id,
group_coordinator_id=None,
partitions=None):
"""Fetch Consumer Group Offsets.
Note:
This does not verify that the group_id or partitions actually exist
in the cluster.
As soon as any error is encountered, it is immediately raised.
:param group_id: The consumer group id name for which to fetch offsets.
:param group_coordinator_id: The node_id of the group's coordinator
broker. If set to None, will query the cluster to find the group
coordinator. Explicitly specifying this can be useful to prevent
that extra network round trip if you already know the group
coordinator. Default: None.
:param partitions: A list of TopicPartitions for which to fetch
offsets. On brokers >= 0.10.2, this can be set to None to fetch all
known offsets for the consumer group. Default: None.
:return dictionary: A dictionary with TopicPartition keys and
OffsetAndMetada values. Partitions that are not specified and for
which the group_id does not have a recorded offset are omitted. An
offset value of `-1` indicates the group_id has no offset for that
TopicPartition. A `-1` can only happen for partitions that are
explicitly specified.
"""
group_offsets_listing = {}
if group_coordinator_id is None:
group_coordinator_id = self._find_group_coordinator_id(group_id)
version = self._matching_api_version(OffsetFetchRequest)
if version <= 3:
if partitions is None:
if version <= 1:
raise ValueError(
"""OffsetFetchRequest_v{} requires specifying the
partitions for which to fetch offsets. Omitting the
partitions is only supported on brokers >= 0.10.2.
For details, see KIP-88.""".format(version))
topics_partitions = None
else:
# transform from [TopicPartition("t1", 1), TopicPartition("t1", 2)] to [("t1", [1, 2])]
topics_partitions_dict = defaultdict(set)
for topic, partition in partitions:
topics_partitions_dict[topic].add(partition)
topics_partitions = list(six.iteritems(topics_partitions_dict))
request = OffsetFetchRequest[version](group_id, topics_partitions)
future = self._send_request_to_node(group_coordinator_id, request)
self._wait_for_futures([future])
response = future.value
if version > 1: # OffsetFetchResponse_v1 lacks a top-level error_code
error_type = Errors.for_code(response.error_code)
if error_type is not Errors.NoError:
# optionally we could retry if error_type.retriable
raise error_type(
"Request '{}' failed with response '{}'.".format(
request, response))
# transform response into a dictionary with TopicPartition keys and
# OffsetAndMetada values--this is what the Java AdminClient returns
for topic, partitions in response.topics:
for partition, offset, metadata, error_code in partitions:
error_type = Errors.for_code(error_code)
if error_type is not Errors.NoError:
raise error_type(
"Unable to fetch offsets for group_id {}, topic {}, partition {}"
.format(group_id, topic, partition))
group_offsets_listing[TopicPartition(
topic,
partition)] = OffsetAndMetadata(offset, metadata)
else:
raise NotImplementedError(
"Support for OffsetFetch v{} has not yet been added to KafkaAdminClient."
.format(version))
return group_offsets_listing
# delete groups protocol not yet implemented
# Note: send the request to the group's coordinator.
def _wait_for_futures(self, futures):
while not all(future.succeeded() for future in futures):
for future in futures:
self._client.poll(future=future)
if future.failed():
raise future.exception # pylint: disable-msg=raising-bad-type
class KafkaAdminClient(object):
"""A class for administering the Kafka cluster.
Warning:
This is an unstable interface that was recently added and is subject to
change without warning. In particular, many methods currently return
raw protocol tuples. In future releases, we plan to make these into
nicer, more pythonic objects. Unfortunately, this will likely break
those interfaces.
The KafkaAdminClient class will negotiate for the latest version of each message
protocol format supported by both the kafka-python client library and the
Kafka broker. Usage of optional fields from protocol versions that are not
supported by the broker will result in IncompatibleBrokerVersion exceptions.
Use of this class requires a minimum broker version >= 0.10.0.0.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
reconnect_backoff_max_ms (int): The maximum amount of time in
milliseconds to wait when reconnecting to a broker that has
repeatedly failed to connect. If provided, the backoff per host
will increase exponentially for each consecutive connection
failure, up to this maximum. To avoid connection storms, a
randomization factor of 0.2 will be applied to the backoff
resulting in a random range between 20% below and 20% above
the computed value. Default: 1000.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 30000.
connections_max_idle_ms: Close idle connections after the number of
milliseconds specified by this config. The broker closes idle
connections after connections.max.idle.ms, so this avoids hitting
unexpected socket disconnected errors on the client.
Default: 540000
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). Java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). Java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): Pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): Flag to configure whether SSL handshake
should verify that the certificate matches the broker's hostname.
Default: True.
ssl_cafile (str): Optional filename of CA file to use in certificate
veriication. Default: None.
ssl_certfile (str): Optional filename of file in PEM format containing
the client certificate, as well as any CA certificates needed to
establish the certificate's authenticity. Default: None.
ssl_keyfile (str): Optional filename containing the client private key.
Default: None.
ssl_password (str): Optional password to be used when loading the
certificate chain. Default: None.
ssl_crlfile (str): Optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
Default: None.
api_version (tuple): Specify which Kafka API version to use. If set
to None, KafkaClient will attempt to infer the broker version by
probing various APIs. Example: (0, 10, 2). Default: None
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version is None
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
metrics (kafka.metrics.Metrics): Optionally provide a metrics
instance for capturing network IO stats. Default: None.
metric_group_prefix (str): Prefix for metric names. Default: ''
sasl_mechanism (str): Authentication mechanism when security_protocol
is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are:
PLAIN, GSSAPI, OAUTHBEARER.
sasl_plain_username (str): username for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_plain_password (str): password for sasl PLAIN authentication.
Required if sasl_mechanism is PLAIN.
sasl_kerberos_service_name (str): Service name to include in GSSAPI
sasl mechanism handshake. Default: 'kafka'
sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider
instance. (See kafka.oauth.abstract). Default: None
"""
DEFAULT_CONFIG = {
# client configs
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'request_timeout_ms': 30000,
'connections_max_idle_ms': 9 * 60 * 1000,
'reconnect_backoff_ms': 50,
'reconnect_backoff_max_ms': 1000,
'max_in_flight_requests_per_connection': 5,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'sock_chunk_bytes': 4096, # undocumented experimental option
'sock_chunk_buffer_count': 1000, # undocumented experimental option
'retry_backoff_ms': 100,
'metadata_max_age_ms': 300000,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_password': None,
'ssl_crlfile': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'selector': selectors.DefaultSelector,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
'sasl_kerberos_service_name': 'kafka',
'sasl_oauth_token_provider': None,
# metrics configs
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
}
def __init__(self, **configs):
log.debug("Starting KafkaAdminClient with configuration: %s", configs)
extra_configs = set(configs).difference(self.DEFAULT_CONFIG)
if extra_configs:
raise KafkaConfigurationError("Unrecognized configs: {}".format(extra_configs))
self.config = copy.copy(self.DEFAULT_CONFIG)
self.config.update(configs)
# Configure metrics
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
self._client = KafkaClient(metrics=self._metrics,
metric_group_prefix='admin',
**self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._closed = False
self._refresh_controller_id()
log.debug("KafkaAdminClient started.")
def close(self):
"""Close the KafkaAdminClient connection to the Kafka broker."""
if not hasattr(self, '_closed') or self._closed:
log.info("KafkaAdminClient already closed.")
return
self._metrics.close()
self._client.close()
self._closed = True
log.debug("KafkaAdminClient is now closed.")
def _matching_api_version(self, operation):
"""Find the latest version of the protocol operation supported by both
this library and the broker.
This resolves to the lesser of either the latest api version this
library supports, or the max version supported by the broker.
:param operation: A list of protocol operation versions from kafka.protocol.
:return: The max matching version number between client and broker.
"""
version = min(len(operation) - 1,
self._client.get_api_versions()[operation[0].API_KEY][1])
if version < self._client.get_api_versions()[operation[0].API_KEY][0]:
# max library version is less than min broker version. Currently,
# no Kafka versions specify a min msg version. Maybe in the future?
raise IncompatibleBrokerVersion(
"No version of the '{}' Kafka protocol is supported by both the client and broker."
.format(operation.__name__))
return version
def _validate_timeout(self, timeout_ms):
"""Validate the timeout is set or use the configuration default.
:param timeout_ms: The timeout provided by api call, in milliseconds.
:return: The timeout to use for the operation.
"""
return timeout_ms or self.config['request_timeout_ms']
def _refresh_controller_id(self):
"""Determine the Kafka cluster controller."""
version = self._matching_api_version(MetadataRequest)
if 1 <= version <= 6:
request = MetadataRequest[version]()
response = self._send_request_to_node(self._client.least_loaded_node(), request)
controller_id = response.controller_id
# verify the controller is new enough to support our requests
controller_version = self._client.check_version(controller_id)
if controller_version < (0, 10, 0):
raise IncompatibleBrokerVersion(
"The controller appears to be running Kafka {}. KafkaAdminClient requires brokers >= 0.10.0.0."
.format(controller_version))
self._controller_id = controller_id
else:
raise UnrecognizedBrokerVersion(
"Kafka Admin interface cannot determine the controller using MetadataRequest_v{}."
.format(version))
def _find_group_coordinator_id(self, group_id):
"""Find the broker node_id of the coordinator of the given group.
Sends a FindCoordinatorRequest message to the cluster. Will block until
the FindCoordinatorResponse is received. Any errors are immediately
raised.
:param group_id: The consumer group ID. This is typically the group
name as a string.
:return: The node_id of the broker that is the coordinator.
"""
# Note: Java may change how this is implemented in KAFKA-6791.
#
# TODO add support for dynamically picking version of
# GroupCoordinatorRequest which was renamed to FindCoordinatorRequest.
# When I experimented with this, GroupCoordinatorResponse_v1 didn't
# match GroupCoordinatorResponse_v0 and I couldn't figure out why.
gc_request = GroupCoordinatorRequest[0](group_id)
gc_response = self._send_request_to_node(self._client.least_loaded_node(), gc_request)
# use the extra error checking in add_group_coordinator() rather than
# immediately returning the group coordinator.
success = self._client.cluster.add_group_coordinator(group_id, gc_response)
if not success:
error_type = Errors.for_code(gc_response.error_code)
assert error_type is not Errors.NoError
# Note: When error_type.retriable, Java will retry... see
# KafkaAdminClient's handleFindCoordinatorError method
raise error_type(
"Could not identify group coordinator for group_id '{}' from response '{}'."
.format(group_id, gc_response))
group_coordinator = self._client.cluster.coordinator_for_group(group_id)
# will be None if the coordinator was never populated, which should never happen here
assert group_coordinator is not None
# will be -1 if add_group_coordinator() failed... but by this point the
# error should have been raised.
assert group_coordinator != -1
return group_coordinator
def _send_request_to_node(self, node_id, request):
"""Send a Kafka protocol message to a specific broker.
Will block until the message result is received.
:param node_id: The broker id to which to send the message.
:param request: The message to send.
:return: The Kafka protocol response for the message.
:exception: The exception if the message could not be sent.
"""
while not self._client.ready(node_id):
# poll until the connection to broker is ready, otherwise send()
# will fail with NodeNotReadyError
self._client.poll()
future = self._client.send(node_id, request)
self._client.poll(future=future)
if future.succeeded():
return future.value
else:
raise future.exception # pylint: disable-msg=raising-bad-type
def _send_request_to_controller(self, request):
"""Send a Kafka protocol message to the cluster controller.
Will block until the message result is received.
:param request: The message to send.
:return: The Kafka protocol response for the message.
"""
tries = 2 # in case our cached self._controller_id is outdated
while tries:
tries -= 1
response = self._send_request_to_node(self._controller_id, request)
# In Java, the error fieldname is inconsistent:
# - CreateTopicsResponse / CreatePartitionsResponse uses topic_errors
# - DeleteTopicsResponse uses topic_error_codes
# So this is a little brittle in that it assumes all responses have
# one of these attributes and that they always unpack into
# (topic, error_code) tuples.
topic_error_tuples = (response.topic_errors if hasattr(response, 'topic_errors')
else response.topic_error_codes)
# Also small py2/py3 compatibility -- py3 can ignore extra values
# during unpack via: for x, y, *rest in list_of_values. py2 cannot.
# So for now we have to map across the list and explicitly drop any
# extra values (usually the error_message)
for topic, error_code in map(lambda e: e[:2], topic_error_tuples):
error_type = Errors.for_code(error_code)
if tries and error_type is NotControllerError:
# No need to inspect the rest of the errors for
# non-retriable errors because NotControllerError should
# either be thrown for all errors or no errors.
self._refresh_controller_id()
break
elif error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'."
.format(request, response))
else:
return response
raise RuntimeError("This should never happen, please file a bug with full stacktrace if encountered")
@staticmethod
def _convert_new_topic_request(new_topic):
return (
new_topic.name,
new_topic.num_partitions,
new_topic.replication_factor,
[
(partition_id, replicas) for partition_id, replicas in new_topic.replica_assignments.items()
],
[
(config_key, config_value) for config_key, config_value in new_topic.topic_configs.items()
]
)
def create_topics(self, new_topics, timeout_ms=None, validate_only=False):
"""Create new topics in the cluster.
:param new_topics: A list of NewTopic objects.
:param timeout_ms: Milliseconds to wait for new topics to be created
before the broker returns.
:param validate_only: If True, don't actually create new topics.
Not supported by all versions. Default: False
:return: Appropriate version of CreateTopicResponse class.
"""
version = self._matching_api_version(CreateTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version == 0:
if validate_only:
raise IncompatibleBrokerVersion(
"validate_only requires CreateTopicsRequest >= v1, which is not supported by Kafka {}."
.format(self.config['api_version']))
request = CreateTopicsRequest[version](
create_topic_requests=[self._convert_new_topic_request(new_topic) for new_topic in new_topics],
timeout=timeout_ms
)
elif version <= 2:
request = CreateTopicsRequest[version](
create_topic_requests=[self._convert_new_topic_request(new_topic) for new_topic in new_topics],
timeout=timeout_ms,
validate_only=validate_only
)
else:
raise NotImplementedError(
"Support for CreateTopics v{} has not yet been added to KafkaAdminClient."
.format(version))
# TODO convert structs to a more pythonic interface
# TODO raise exceptions if errors
return self._send_request_to_controller(request)
def delete_topics(self, topics, timeout_ms=None):
"""Delete topics from the cluster.
:param topics: A list of topic name strings.
:param timeout_ms: Milliseconds to wait for topics to be deleted
before the broker returns.
:return: Appropriate version of DeleteTopicsResponse class.
"""
version = self._matching_api_version(DeleteTopicsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version <= 1:
request = DeleteTopicsRequest[version](
topics=topics,
timeout=timeout_ms
)
response = self._send_request_to_controller(request)
else:
raise NotImplementedError(
"Support for DeleteTopics v{} has not yet been added to KafkaAdminClient."
.format(version))
return response
# list topics functionality is in ClusterMetadata
# Note: if implemented here, send the request to the least_loaded_node()
# describe topics functionality is in ClusterMetadata
# Note: if implemented here, send the request to the controller
# describe cluster functionality is in ClusterMetadata
# Note: if implemented here, send the request to the least_loaded_node()
# describe_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
# create_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
# delete_acls protocol not yet implemented
# Note: send the request to the least_loaded_node()
@staticmethod
def _convert_describe_config_resource_request(config_resource):
return (
config_resource.resource_type,
config_resource.name,
[
config_key for config_key, config_value in config_resource.configs.items()
] if config_resource.configs else None
)
def describe_configs(self, config_resources, include_synonyms=False):
"""Fetch configuration parameters for one or more Kafka resources.
:param config_resources: An list of ConfigResource objects.
Any keys in ConfigResource.configs dict will be used to filter the
result. Setting the configs dict to None will get all values. An
empty dict will get zero values (as per Kafka protocol).
:param include_synonyms: If True, return synonyms in response. Not
supported by all versions. Default: False.
:return: Appropriate version of DescribeConfigsResponse class.
"""
version = self._matching_api_version(DescribeConfigsRequest)
if version == 0:
if include_synonyms:
raise IncompatibleBrokerVersion(
"include_synonyms requires DescribeConfigsRequest >= v1, which is not supported by Kafka {}."
.format(self.config['api_version']))
request = DescribeConfigsRequest[version](
resources=[self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources]
)
elif version == 1:
request = DescribeConfigsRequest[version](
resources=[self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources],
include_synonyms=include_synonyms
)
else:
raise NotImplementedError(
"Support for DescribeConfigs v{} has not yet been added to KafkaAdminClient."
.format(version))
return self._send_request_to_node(self._client.least_loaded_node(), request)
@staticmethod
def _convert_alter_config_resource_request(config_resource):
return (
config_resource.resource_type,
config_resource.name,
[
(config_key, config_value) for config_key, config_value in config_resource.configs.items()
]
)
def alter_configs(self, config_resources):
"""Alter configuration parameters of one or more Kafka resources.
Warning:
This is currently broken for BROKER resources because those must be
sent to that specific broker, versus this always picks the
least-loaded node. See the comment in the source code for details.
We would happily accept a PR fixing this.
:param config_resources: A list of ConfigResource objects.
:return: Appropriate version of AlterConfigsResponse class.
"""
version = self._matching_api_version(AlterConfigsRequest)
if version == 0:
request = AlterConfigsRequest[version](
resources=[self._convert_alter_config_resource_request(config_resource) for config_resource in config_resources]
)
else:
raise NotImplementedError(
"Support for AlterConfigs v{} has not yet been added to KafkaAdminClient."
.format(version))
# TODO the Java client has the note:
# // We must make a separate AlterConfigs request for every BROKER resource we want to alter
# // and send the request to that specific broker. Other resources are grouped together into
# // a single request that may be sent to any broker.
#
# So this is currently broken as it always sends to the least_loaded_node()
return self._send_request_to_node(self._client.least_loaded_node(), request)
# alter replica logs dir protocol not yet implemented
# Note: have to lookup the broker with the replica assignment and send the request to that broker
# describe log dirs protocol not yet implemented
# Note: have to lookup the broker with the replica assignment and send the request to that broker
@staticmethod
def _convert_create_partitions_request(topic_name, new_partitions):
return (
topic_name,
(
new_partitions.total_count,
new_partitions.new_assignments
)
)
def create_partitions(self, topic_partitions, timeout_ms=None, validate_only=False):
"""Create additional partitions for an existing topic.
:param topic_partitions: A map of topic name strings to NewPartition objects.
:param timeout_ms: Milliseconds to wait for new partitions to be
created before the broker returns.
:param validate_only: If True, don't actually create new partitions.
Default: False
:return: Appropriate version of CreatePartitionsResponse class.
"""
version = self._matching_api_version(CreatePartitionsRequest)
timeout_ms = self._validate_timeout(timeout_ms)
if version == 0:
request = CreatePartitionsRequest[version](
topic_partitions=[self._convert_create_partitions_request(topic_name, new_partitions) for topic_name, new_partitions in topic_partitions.items()],
timeout=timeout_ms,
validate_only=validate_only
)
else:
raise NotImplementedError(
"Support for CreatePartitions v{} has not yet been added to KafkaAdminClient."
.format(version))
return self._send_request_to_controller(request)
# delete records protocol not yet implemented
# Note: send the request to the partition leaders
# create delegation token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# renew delegation token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# expire delegation_token protocol not yet implemented
# Note: send the request to the least_loaded_node()
# describe delegation_token protocol not yet implemented
# Note: send the request to the least_loaded_node()
def describe_consumer_groups(self, group_ids, group_coordinator_id=None):
"""Describe a set of consumer groups.
Any errors are immediately raised.
:param group_ids: A list of consumer group IDs. These are typically the
group names as strings.
:param group_coordinator_id: The node_id of the groups' coordinator
broker. If set to None, it will query the cluster for each group to
find that group's coordinator. Explicitly specifying this can be
useful for avoiding extra network round trips if you already know
the group coordinator. This is only useful when all the group_ids
have the same coordinator, otherwise it will error. Default: None.
:return: A list of group descriptions. For now the group descriptions
are the raw results from the DescribeGroupsResponse. Long-term, we
plan to change this to return namedtuples as well as decoding the
partition assignments.
"""
group_descriptions = []
version = self._matching_api_version(DescribeGroupsRequest)
for group_id in group_ids:
if group_coordinator_id is not None:
this_groups_coordinator_id = group_coordinator_id
else:
this_groups_coordinator_id = self._find_group_coordinator_id(group_id)
if version <= 1:
# Note: KAFKA-6788 A potential optimization is to group the
# request per coordinator and send one request with a list of
# all consumer groups. Java still hasn't implemented this
# because the error checking is hard to get right when some
# groups error and others don't.
request = DescribeGroupsRequest[version](groups=(group_id,))
response = self._send_request_to_node(this_groups_coordinator_id, request)
assert len(response.groups) == 1
# TODO need to implement converting the response tuple into
# a more accessible interface like a namedtuple and then stop
# hardcoding tuple indices here. Several Java examples,
# including KafkaAdminClient.java
group_description = response.groups[0]
error_code = group_description[0]
error_type = Errors.for_code(error_code)
# Java has the note: KAFKA-6789, we can retry based on the error code
if error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'."
.format(request, response))
# TODO Java checks the group protocol type, and if consumer
# (ConsumerProtocol.PROTOCOL_TYPE) or empty string, it decodes
# the members' partition assignments... that hasn't yet been
# implemented here so just return the raw struct results
group_descriptions.append(group_description)
else:
raise NotImplementedError(
"Support for DescribeGroups v{} has not yet been added to KafkaAdminClient."
.format(version))
return group_descriptions
def list_consumer_groups(self, broker_ids=None):
"""List all consumer groups known to the cluster.
This returns a list of Consumer Group tuples. The tuples are
composed of the consumer group name and the consumer group protocol
type.
Only consumer groups that store their offsets in Kafka are returned.
The protocol type will be an empty string for groups created using
Kafka < 0.9 APIs because, although they store their offsets in Kafka,
they don't use Kafka for group coordination. For groups created using
Kafka >= 0.9, the protocol type will typically be "consumer".
As soon as any error is encountered, it is immediately raised.
:param broker_ids: A list of broker node_ids to query for consumer
groups. If set to None, will query all brokers in the cluster.
Explicitly specifying broker(s) can be useful for determining which
consumer groups are coordinated by those broker(s). Default: None
:return list: List of tuples of Consumer Groups.
:exception GroupCoordinatorNotAvailableError: The coordinator is not
available, so cannot process requests.
:exception GroupLoadInProgressError: The coordinator is loading and
hence can't process requests.
"""
# While we return a list, internally use a set to prevent duplicates
# because if a group coordinator fails after being queried, and its
# consumer groups move to new brokers that haven't yet been queried,
# then the same group could be returned by multiple brokers.
consumer_groups = set()
if broker_ids is None:
broker_ids = [broker.nodeId for broker in self._client.cluster.brokers()]
version = self._matching_api_version(ListGroupsRequest)
if version <= 2:
request = ListGroupsRequest[version]()
for broker_id in broker_ids:
response = self._send_request_to_node(broker_id, request)
error_type = Errors.for_code(response.error_code)
if error_type is not Errors.NoError:
raise error_type(
"Request '{}' failed with response '{}'."
.format(request, response))
consumer_groups.update(response.groups)
else:
raise NotImplementedError(
"Support for ListGroups v{} has not yet been added to KafkaAdminClient."
.format(version))
return list(consumer_groups)
def list_consumer_group_offsets(self, group_id, group_coordinator_id=None,
partitions=None):
"""Fetch Consumer Group Offsets.
Note:
This does not verify that the group_id or partitions actually exist
in the cluster.
As soon as any error is encountered, it is immediately raised.
:param group_id: The consumer group id name for which to fetch offsets.
:param group_coordinator_id: The node_id of the group's coordinator
broker. If set to None, will query the cluster to find the group
coordinator. Explicitly specifying this can be useful to prevent
that extra network round trip if you already know the group
coordinator. Default: None.
:param partitions: A list of TopicPartitions for which to fetch
offsets. On brokers >= 0.10.2, this can be set to None to fetch all
known offsets for the consumer group. Default: None.
:return dictionary: A dictionary with TopicPartition keys and
OffsetAndMetada values. Partitions that are not specified and for
which the group_id does not have a recorded offset are omitted. An
offset value of `-1` indicates the group_id has no offset for that
TopicPartition. A `-1` can only happen for partitions that are
explicitly specified.
"""
group_offsets_listing = {}
if group_coordinator_id is None:
group_coordinator_id = self._find_group_coordinator_id(group_id)
version = self._matching_api_version(OffsetFetchRequest)
if version <= 3:
if partitions is None:
if version <= 1:
raise ValueError(
"""OffsetFetchRequest_v{} requires specifying the
partitions for which to fetch offsets. Omitting the
partitions is only supported on brokers >= 0.10.2.
For details, see KIP-88.""".format(version))
topics_partitions = None
else:
# transform from [TopicPartition("t1", 1), TopicPartition("t1", 2)] to [("t1", [1, 2])]
topics_partitions_dict = defaultdict(set)
for topic, partition in partitions:
topics_partitions_dict[topic].add(partition)
topics_partitions = list(six.iteritems(topics_partitions_dict))
request = OffsetFetchRequest[version](group_id, topics_partitions)
response = self._send_request_to_node(group_coordinator_id, request)
if version > 1: # OffsetFetchResponse_v1 lacks a top-level error_code
error_type = Errors.for_code(response.error_code)
if error_type is not Errors.NoError:
# optionally we could retry if error_type.retriable
raise error_type(
"Request '{}' failed with response '{}'."
.format(request, response))
# transform response into a dictionary with TopicPartition keys and
# OffsetAndMetada values--this is what the Java AdminClient returns
for topic, partitions in response.topics:
for partition, offset, metadata, error_code in partitions:
error_type = Errors.for_code(error_code)
if error_type is not Errors.NoError:
raise error_type(
"Unable to fetch offsets for group_id {}, topic {}, partition {}"
.format(group_id, topic, partition))
group_offsets_listing[TopicPartition(topic, partition)] = OffsetAndMetadata(offset, metadata)
else:
raise NotImplementedError(
"Support for OffsetFetch v{} has not yet been added to KafkaAdminClient."
.format(version))
return group_offsets_listing
def metrics(request, config, reporter):
metrics = Metrics(config, [reporter], enable_expiration=True)
request.addfinalizer(lambda: metrics.close())
return metrics