class PostgreSql(AgentCheck):
"""Collects per-database, and optionally per-relation metrics, custom metrics"""
SOURCE_TYPE_NAME = 'postgresql'
SERVICE_CHECK_NAME = 'postgres.can_connect'
METADATA_TRANSFORMERS = {'version': transform_version}
def __init__(self, name, init_config, instances):
super(PostgreSql, self).__init__(name, init_config, instances)
self.db = None
self._version = None
self._is_aurora = None
# Deprecate custom_metrics in favor of custom_queries
if 'custom_metrics' in self.instance:
self.warning(
"DEPRECATION NOTICE: Please use the new custom_queries option "
"rather than the now deprecated custom_metrics"
)
self.config = PostgresConfig(self.instance)
self.metrics_cache = PostgresMetricsCache(self.config)
self.statement_metrics = PostgresStatementMetrics(self.config)
self._clean_state()
def _clean_state(self):
self._version = None
self._is_aurora = None
self.metrics_cache.clean_state()
def _get_replication_role(self):
cursor = self.db.cursor()
cursor.execute('SELECT pg_is_in_recovery();')
role = cursor.fetchone()[0]
# value fetched for role is of <type 'bool'>
return "standby" if role else "master"
@property
def version(self):
if self._version is None:
raw_version = get_raw_version(self.db)
self._version = parse_version(raw_version)
self.set_metadata('version', raw_version)
return self._version
@property
def is_aurora(self):
if self._is_aurora is None:
self._is_aurora = is_aurora(self.db)
return self._is_aurora
def _build_relations_config(self, yamlconfig):
"""Builds a dictionary from relations configuration while maintaining compatibility"""
config = {}
for element in yamlconfig:
if isinstance(element, str):
config[element] = {'relation_name': element, 'schemas': [ALL_SCHEMAS]}
elif isinstance(element, dict):
if not ('relation_name' in element or 'relation_regex' in element):
self.log.warning(
"Parameter 'relation_name' or 'relation_regex' is required for relation element %s", element
)
continue
if 'relation_name' in element and 'relation_regex' in element:
self.log.warning(
"Expecting only of parameters 'relation_name', 'relation_regex' for relation element %s",
element,
)
continue
schemas = element.get('schemas', [])
if not isinstance(schemas, list):
self.log.warning("Expected a list of schemas for %s", element)
continue
name = element.get('relation_name') or element['relation_regex']
config[name] = element.copy()
if len(schemas) == 0:
config[name]['schemas'] = [ALL_SCHEMAS]
else:
self.log.warning('Unhandled relations config type: %s', element)
return config
def _run_query_scope(self, cursor, scope, is_custom_metrics, relations_config, cols, descriptors, is_relations):
if scope is None:
return None
if scope == REPLICATION_METRICS or not self.version >= V9:
log_func = self.log.debug
else:
log_func = self.log.warning
results = None
try:
query = fmt.format(scope['query'], metrics_columns=", ".join(cols))
# if this is a relation-specific query, we need to list all relations last
if is_relations:
schema_field = get_schema_field(descriptors)
relations_filter = build_relations_filter(relations_config, schema_field)
self.log.debug("Running query: %s with relations matching: %s", query, relations_filter)
cursor.execute(query.format(relations=relations_filter))
else:
self.log.debug("Running query: %s", query)
cursor.execute(query.replace(r'%', r'%%'))
results = cursor.fetchall()
except psycopg2.errors.FeatureNotSupported as e:
# This happens for example when trying to get replication metrics from readers in Aurora. Let's ignore it.
log_func(e)
self.db.rollback()
self._is_aurora = None
except psycopg2.errors.UndefinedFunction as e:
log_func(e)
log_func(
"It seems the PG version has been incorrectly identified as %s. "
"A reattempt to identify the right version will happen on next agent run." % self._version
)
self._clean_state()
self.db.rollback()
except (psycopg2.ProgrammingError, psycopg2.errors.QueryCanceled) as e:
log_func("Not all metrics may be available: %s" % str(e))
self.db.rollback()
if not results:
return None
if is_custom_metrics and len(results) > MAX_CUSTOM_RESULTS:
self.warning(
"Query: %s returned more than %s results (%s). Truncating", query, MAX_CUSTOM_RESULTS, len(results)
)
results = results[:MAX_CUSTOM_RESULTS]
if is_relations and len(results) > self.config.max_relations:
self.warning(
"Query: %s returned more than %s results (%s). "
"Truncating. You can edit this limit by setting the `max_relations` config option",
query,
self.config.max_relations,
len(results),
)
results = results[: self.config.max_relations]
return results
def _query_scope(self, cursor, scope, instance_tags, is_custom_metrics, relations_config):
if scope is None:
return None
# build query
cols = list(scope['metrics']) # list of metrics to query, in some order
# we must remember that order to parse results
# A descriptor is the association of a Postgres column name (e.g. 'schemaname')
# to a tag name (e.g. 'schema').
descriptors = scope['descriptors']
is_relations = scope['relation'] and len(relations_config) > 0
results = self._run_query_scope(
cursor, scope, is_custom_metrics, relations_config, cols, descriptors, is_relations
)
if not results:
return None
# Parse and submit results.
num_results = 0
for row in results:
# A row contains descriptor values on the left (used for tagging), and
# metric values on the right (used as values for metrics).
# E.g.: (descriptor, descriptor, ..., value, value, value, value, ...)
expected_number_of_columns = len(descriptors) + len(cols)
if len(row) != expected_number_of_columns:
raise RuntimeError(
'Row does not contain enough values: '
'expected {} ({} descriptors + {} columns), got {}'.format(
expected_number_of_columns, len(descriptors), len(cols), len(row)
)
)
descriptor_values = row[: len(descriptors)]
column_values = row[len(descriptors) :]
# build a map of descriptors and their values
desc_map = {name: value for (_, name), value in zip(descriptors, descriptor_values)}
# Build tags.
# Add tags from the instance.
# Special-case the "db" tag, which overrides the one that is passed as instance_tag
# The reason is that pg_stat_database returns all databases regardless of the
# connection.
if not scope['relation'] and not scope.get('use_global_db_tag', False):
tags = [t for t in instance_tags if not t.startswith("db:")]
else:
tags = copy.copy(instance_tags)
# Add tags from descriptors.
tags += [("%s:%s" % (k, v)) for (k, v) in iteritems(desc_map)]
# Submit metrics to the Agent.
for column, value in zip(cols, column_values):
name, submit_metric = scope['metrics'][column]
submit_metric(self, name, value, tags=set(tags))
num_results += 1
return num_results
def _collect_stats(self, instance_tags):
"""Query pg_stat_* for various metrics
If relations is not an empty list, gather per-relation metrics
on top of that.
If custom_metrics is not an empty list, gather custom metrics defined in postgres.yaml
"""
db_instance_metrics = self.metrics_cache.get_instance_metrics(self.version)
bgw_instance_metrics = self.metrics_cache.get_bgw_metrics(self.version)
archiver_instance_metrics = self.metrics_cache.get_archiver_metrics(self.version)
metric_scope = [CONNECTION_METRICS]
if self.config.collect_function_metrics:
metric_scope.append(FUNCTION_METRICS)
if self.config.collect_count_metrics:
metric_scope.append(self.metrics_cache.get_count_metrics())
# Do we need relation-specific metrics?
relations_config = {}
if self.config.relations:
metric_scope += [LOCK_METRICS, REL_METRICS, IDX_METRICS, SIZE_METRICS, STATIO_METRICS]
relations_config = self._build_relations_config(self.config.relations)
replication_metrics = self.metrics_cache.get_replication_metrics(self.version, self.is_aurora)
if replication_metrics:
replication_metrics_query = copy.deepcopy(REPLICATION_METRICS)
replication_metrics_query['metrics'] = replication_metrics
metric_scope.append(replication_metrics_query)
cursor = self.db.cursor()
results_len = self._query_scope(cursor, db_instance_metrics, instance_tags, False, relations_config)
if results_len is not None:
self.gauge("postgresql.db.count", results_len, tags=[t for t in instance_tags if not t.startswith("db:")])
self._query_scope(cursor, bgw_instance_metrics, instance_tags, False, relations_config)
self._query_scope(cursor, archiver_instance_metrics, instance_tags, False, relations_config)
if self.config.collect_activity_metrics:
activity_metrics = self.metrics_cache.get_activity_metrics(self.version)
self._query_scope(cursor, activity_metrics, instance_tags, False, relations_config)
for scope in list(metric_scope) + self.config.custom_metrics:
self._query_scope(cursor, scope, instance_tags, scope in self.config.custom_metrics, relations_config)
cursor.close()
def _connect(self):
"""Get and memoize connections to instances"""
if self.db and self.db.closed:
# Reset the connection object to retry to connect
self.db = None
if self.db:
if self.db.status != psycopg2.extensions.STATUS_READY:
# Some transaction went wrong and the connection is in an unhealthy state. Let's fix that
self.db.rollback()
else:
if self.config.host == 'localhost' and self.config.password == '':
# Use ident method
connection_string = "user=%s dbname=%s application_name=%s" % (
self.config.user,
self.config.dbname,
self.config.application_name,
)
if self.config.query_timeout:
connection_string += " options='-c statement_timeout=%s'" % self.config.query_timeout
self.db = psycopg2.connect(connection_string)
else:
args = {
'host': self.config.host,
'user': self.config.user,
'password': self.config.password,
'database': self.config.dbname,
'sslmode': self.config.ssl_mode,
'application_name': self.config.application_name,
}
if self.config.port:
args['port'] = self.config.port
if self.config.query_timeout:
args['options'] = '-c statement_timeout=%s' % self.config.query_timeout
self.db = psycopg2.connect(**args)
def _collect_custom_queries(self, tags):
"""
Given a list of custom_queries, execute each query and parse the result for metrics
"""
for custom_query in self.config.custom_queries:
metric_prefix = custom_query.get('metric_prefix')
if not metric_prefix:
self.log.error("custom query field `metric_prefix` is required")
continue
metric_prefix = metric_prefix.rstrip('.')
query = custom_query.get('query')
if not query:
self.log.error("custom query field `query` is required for metric_prefix `%s`", metric_prefix)
continue
columns = custom_query.get('columns')
if not columns:
self.log.error("custom query field `columns` is required for metric_prefix `%s`", metric_prefix)
continue
cursor = self.db.cursor()
with closing(cursor) as cursor:
try:
self.log.debug("Running query: %s", query)
cursor.execute(query)
except (psycopg2.ProgrammingError, psycopg2.errors.QueryCanceled) as e:
self.log.error("Error executing query for metric_prefix %s: %s", metric_prefix, str(e))
self.db.rollback()
continue
for row in cursor:
if not row:
self.log.debug("query result for metric_prefix %s: returned an empty result", metric_prefix)
continue
if len(columns) != len(row):
self.log.error(
"query result for metric_prefix %s: expected %s columns, got %s",
metric_prefix,
len(columns),
len(row),
)
continue
metric_info = []
query_tags = list(custom_query.get('tags', []))
query_tags.extend(tags)
for column, value in zip(columns, row):
# Columns can be ignored via configuration.
if not column:
continue
name = column.get('name')
if not name:
self.log.error("column field `name` is required for metric_prefix `%s`", metric_prefix)
break
column_type = column.get('type')
if not column_type:
self.log.error(
"column field `type` is required for column `%s` of metric_prefix `%s`",
name,
metric_prefix,
)
break
if column_type == 'tag':
query_tags.append('{}:{}'.format(name, value))
else:
if not hasattr(self, column_type):
self.log.error(
"invalid submission method `%s` for column `%s` of metric_prefix `%s`",
column_type,
name,
metric_prefix,
)
break
try:
metric_info.append(('{}.{}'.format(metric_prefix, name), float(value), column_type))
except (ValueError, TypeError):
self.log.error(
"non-numeric value `%s` for metric column `%s` of metric_prefix `%s`",
value,
name,
metric_prefix,
)
break
# Only submit metrics if there were absolutely no errors - all or nothing.
else:
for info in metric_info:
metric, value, method = info
getattr(self, method)(metric, value, tags=set(query_tags))
def _collect_per_statement_metrics(self, tags):
metrics = self.statement_metrics.collect_per_statement_metrics(self.db)
for metric_name, metric_value, metrics_tags in metrics:
self.count(metric_name, metric_value, tags=list(set(metrics_tags + tags)))
def check(self, _):
tags = copy.copy(self.config.tags)
# Collect metrics
try:
# Check version
self._connect()
if self.config.tag_replication_role:
tags.extend(["replication_role:{}".format(self._get_replication_role())])
self.log.debug("Running check against version %s", str(self.version))
self._collect_stats(tags)
self._collect_custom_queries(tags)
if self.config.deep_database_monitoring:
self._collect_per_statement_metrics(tags)
except Exception as e:
self.log.error("Unable to collect postgres metrics.")
self._clean_state()
self.db = None
message = u'Error establishing connection to postgres://{}:{}/{}, error is {}'.format(
self.config.host, self.config.port, self.config.dbname, str(e)
)
self.service_check(
self.SERVICE_CHECK_NAME, AgentCheck.CRITICAL, tags=self.config.service_check_tags, message=message
)
raise e
else:
message = u'Established connection to postgres://%s:%s/%s' % (
self.config.host,
self.config.port,
self.config.dbname,
)
self.service_check(
self.SERVICE_CHECK_NAME, AgentCheck.OK, tags=self.config.service_check_tags, message=message
)
try:
# commit to close the current query transaction
self.db.commit()
except Exception as e:
self.log.warning("Unable to commit: %s", e)
self._version = None # We don't want to cache versions between runs to capture minor updates for metadata
class PostgreSql(AgentCheck):
"""Collects per-database, and optionally per-relation metrics, custom metrics"""
SOURCE_TYPE_NAME = 'postgresql'
SERVICE_CHECK_NAME = 'postgres.can_connect'
METADATA_TRANSFORMERS = {'version': VersionUtils.transform_version}
def __init__(self, name, init_config, instances):
super(PostgreSql, self).__init__(name, init_config, instances)
self.db = None
self._resolved_hostname = None
self._agent_hostname = None
self._version = None
self._is_aurora = None
self._version_utils = VersionUtils()
# Deprecate custom_metrics in favor of custom_queries
if 'custom_metrics' in self.instance:
self.warning(
"DEPRECATION NOTICE: Please use the new custom_queries option "
"rather than the now deprecated custom_metrics")
self._config = PostgresConfig(self.instance)
self.pg_settings = {}
self._warnings_by_code = {}
self.metrics_cache = PostgresMetricsCache(self._config)
self.statement_metrics = PostgresStatementMetrics(
self, self._config, shutdown_callback=self._close_db_pool)
self.statement_samples = PostgresStatementSamples(
self, self._config, shutdown_callback=self._close_db_pool)
self._relations_manager = RelationsManager(self._config.relations)
self._clean_state()
self.check_initializations.append(
lambda: RelationsManager.validate_relations_config(self._config.
relations))
# map[dbname -> psycopg connection]
self._db_pool = {}
self._db_pool_lock = threading.Lock()
def cancel(self):
self.statement_samples.cancel()
self.statement_metrics.cancel()
def _clean_state(self):
self.log.debug("Cleaning state")
self._version = None
self._is_aurora = None
self.metrics_cache.clean_state()
def _get_debug_tags(self):
return ['agent_hostname:{}'.format(self.agent_hostname)]
def _get_service_check_tags(self):
service_check_tags = []
service_check_tags.extend(self._config.tags)
return list(service_check_tags)
def _get_replication_role(self):
cursor = self.db.cursor()
cursor.execute('SELECT pg_is_in_recovery();')
role = cursor.fetchone()[0]
# value fetched for role is of <type 'bool'>
return "standby" if role else "master"
def _collect_wal_metrics(self, instance_tags):
wal_file_age = self._get_wal_file_age()
if wal_file_age is not None:
self.gauge(
"postgresql.wal_age",
wal_file_age,
tags=[t for t in instance_tags if not t.startswith("db:")],
hostname=self.resolved_hostname,
)
def _get_wal_dir(self):
if self.version >= V10:
wal_dir = "pg_wal"
else:
wal_dir = "pg_xlog"
wal_log_dir = os.path.join(self._config.data_directory, wal_dir)
return wal_log_dir
def _get_wal_file_age(self):
wal_log_dir = self._get_wal_dir()
if not os.path.isdir(wal_log_dir):
self.log.warning(
"Cannot access WAL log directory: %s. Ensure that you are "
"running the agent on your local postgres database.",
wal_log_dir,
)
return None
all_dir_contents = os.listdir(wal_log_dir)
all_files = [
f for f in all_dir_contents
if os.path.isfile(os.path.join(wal_log_dir, f))
]
# files extensions that are not valid WAL files
exluded_file_exts = [".backup", ".history"]
all_wal_files = [
os.path.join(wal_log_dir, file_name) for file_name in all_files
if not any(
[ext for ext in exluded_file_exts if file_name.endswith(ext)])
]
if len(all_wal_files) < 1:
self.log.warning("No WAL files found in directory: %s.",
wal_log_dir)
return None
oldest_file = min(all_wal_files, key=os.path.getctime)
now = time()
oldest_file_age = now - os.path.getctime(oldest_file)
return oldest_file_age
@property
def version(self):
if self._version is None:
raw_version = self._version_utils.get_raw_version(self.db)
self._version = self._version_utils.parse_version(raw_version)
self.set_metadata('version', raw_version)
return self._version
@property
def is_aurora(self):
if self._is_aurora is None:
self._is_aurora = self._version_utils.is_aurora(self.db)
return self._is_aurora
@property
def resolved_hostname(self):
# type: () -> str
if self._resolved_hostname is None:
if self._config.reported_hostname:
self._resolved_hostname = self._config.reported_hostname
elif self._config.dbm_enabled or self.disable_generic_tags:
self._resolved_hostname = self.resolve_db_host()
else:
self._resolved_hostname = self.agent_hostname
return self._resolved_hostname
@property
def agent_hostname(self):
# type: () -> str
if self._agent_hostname is None:
self._agent_hostname = datadog_agent.get_hostname()
return self._agent_hostname
def resolve_db_host(self):
return agent_host_resolver(self._config.host)
def _run_query_scope(self, cursor, scope, is_custom_metrics, cols,
descriptors):
if scope is None:
return None
if scope == REPLICATION_METRICS or not self.version >= V9:
log_func = self.log.debug
else:
log_func = self.log.warning
results = None
is_relations = scope.get(
'relation') and self._relations_manager.has_relations
try:
query = fmt.format(scope['query'], metrics_columns=", ".join(cols))
# if this is a relation-specific query, we need to list all relations last
if is_relations:
schema_field = get_schema_field(descriptors)
formatted_query = self._relations_manager.filter_relation_query(
query, schema_field)
cursor.execute(formatted_query)
else:
self.log.debug("Running query: %s", str(query))
cursor.execute(query.replace(r'%', r'%%'))
results = cursor.fetchall()
except psycopg2.errors.FeatureNotSupported as e:
# This happens for example when trying to get replication metrics from readers in Aurora. Let's ignore it.
log_func(e)
self.db.rollback()
self.log.debug("Disabling replication metrics")
self._is_aurora = False
self.metrics_cache.replication_metrics = {}
except psycopg2.errors.UndefinedFunction as e:
log_func(e)
log_func(
"It seems the PG version has been incorrectly identified as %s. "
"A reattempt to identify the right version will happen on next agent run."
% self._version)
self._clean_state()
self.db.rollback()
except (psycopg2.ProgrammingError, psycopg2.errors.QueryCanceled) as e:
log_func("Not all metrics may be available: %s" % str(e))
self.db.rollback()
if not results:
return None
if is_custom_metrics and len(results) > MAX_CUSTOM_RESULTS:
self.warning(
"Query: %s returned more than %s results (%s). Truncating",
query, MAX_CUSTOM_RESULTS, len(results))
results = results[:MAX_CUSTOM_RESULTS]
if is_relations and len(results) > self._config.max_relations:
self.warning(
"Query: %s returned more than %s results (%s). "
"Truncating. You can edit this limit by setting the `max_relations` config option",
query,
self._config.max_relations,
len(results),
)
results = results[:self._config.max_relations]
return results
def _query_scope(self, cursor, scope, instance_tags, is_custom_metrics):
if scope is None:
return None
# build query
cols = list(
scope['metrics']) # list of metrics to query, in some order
# we must remember that order to parse results
# A descriptor is the association of a Postgres column name (e.g. 'schemaname')
# to a tag name (e.g. 'schema').
descriptors = scope['descriptors']
results = self._run_query_scope(cursor, scope, is_custom_metrics, cols,
descriptors)
if not results:
return None
# Parse and submit results.
num_results = 0
for row in results:
# A row contains descriptor values on the left (used for tagging), and
# metric values on the right (used as values for metrics).
# E.g.: (descriptor, descriptor, ..., value, value, value, value, ...)
expected_number_of_columns = len(descriptors) + len(cols)
if len(row) != expected_number_of_columns:
raise RuntimeError(
'Row does not contain enough values: '
'expected {} ({} descriptors + {} columns), got {}'.format(
expected_number_of_columns, len(descriptors),
len(cols), len(row)))
descriptor_values = row[:len(descriptors)]
column_values = row[len(descriptors):]
# build a map of descriptors and their values
desc_map = {
name: value
for (_, name), value in zip(descriptors, descriptor_values)
}
# Build tags.
# Add tags from the instance.
# Special-case the "db" tag, which overrides the one that is passed as instance_tag
# The reason is that pg_stat_database returns all databases regardless of the
# connection.
if not scope['relation'] and not scope.get('use_global_db_tag',
False):
tags = [t for t in instance_tags if not t.startswith("db:")]
else:
tags = copy.copy(instance_tags)
# Add tags from descriptors.
tags += [("%s:%s" % (k, v)) for (k, v) in iteritems(desc_map)]
# Submit metrics to the Agent.
for column, value in zip(cols, column_values):
name, submit_metric = scope['metrics'][column]
submit_metric(self,
name,
value,
tags=set(tags),
hostname=self.resolved_hostname)
num_results += 1
return num_results
def _collect_stats(self, instance_tags):
"""Query pg_stat_* for various metrics
If relations is not an empty list, gather per-relation metrics
on top of that.
If custom_metrics is not an empty list, gather custom metrics defined in postgres.yaml
"""
db_instance_metrics = self.metrics_cache.get_instance_metrics(
self.version)
bgw_instance_metrics = self.metrics_cache.get_bgw_metrics(self.version)
archiver_instance_metrics = self.metrics_cache.get_archiver_metrics(
self.version)
metric_scope = [CONNECTION_METRICS]
if self._config.collect_function_metrics:
metric_scope.append(FUNCTION_METRICS)
if self._config.collect_count_metrics:
metric_scope.append(self.metrics_cache.get_count_metrics())
# Do we need relation-specific metrics?
if self._config.relations:
metric_scope.extend(RELATION_METRICS)
if self._config.collect_bloat_metrics:
metric_scope.extend([INDEX_BLOAT, TABLE_BLOAT])
replication_metrics = self.metrics_cache.get_replication_metrics(
self.version, self.is_aurora)
if replication_metrics:
replication_metrics_query = copy.deepcopy(REPLICATION_METRICS)
replication_metrics_query['metrics'] = replication_metrics
metric_scope.append(replication_metrics_query)
replication_stats_metrics = self.metrics_cache.get_replication_stats_metrics(
self.version)
if replication_stats_metrics:
metric_scope.append(replication_stats_metrics)
cursor = self.db.cursor()
results_len = self._query_scope(cursor, db_instance_metrics,
instance_tags, False)
if results_len is not None:
self.gauge(
"postgresql.db.count",
results_len,
tags=[t for t in instance_tags if not t.startswith("db:")],
hostname=self.resolved_hostname,
)
self._query_scope(cursor, bgw_instance_metrics, instance_tags, False)
self._query_scope(cursor, archiver_instance_metrics, instance_tags,
False)
if self._config.collect_activity_metrics:
activity_metrics = self.metrics_cache.get_activity_metrics(
self.version)
self._query_scope(cursor, activity_metrics, instance_tags, False)
for scope in list(metric_scope) + self._config.custom_metrics:
self._query_scope(cursor, scope, instance_tags, scope
in self._config.custom_metrics)
cursor.close()
def _new_connection(self, dbname):
if self._config.host == 'localhost' and self._config.password == '':
# Use ident method
connection_string = "user=%s dbname=%s application_name=%s" % (
self._config.user,
dbname,
self._config.application_name,
)
if self._config.query_timeout:
connection_string += " options='-c statement_timeout=%s'" % self._config.query_timeout
conn = psycopg2.connect(connection_string)
else:
args = {
'host': self._config.host,
'user': self._config.user,
'password': self._config.password,
'database': dbname,
'sslmode': self._config.ssl_mode,
'application_name': self._config.application_name,
}
if self._config.port:
args['port'] = self._config.port
if self._config.query_timeout:
args[
'options'] = '-c statement_timeout=%s' % self._config.query_timeout
if self._config.ssl_cert:
args['sslcert'] = self._config.ssl_cert
if self._config.ssl_root_cert:
args['sslrootcert'] = self._config.ssl_root_cert
if self._config.ssl_key:
args['sslkey'] = self._config.ssl_key
if self._config.ssl_password:
args['sslpassword'] = self._config.ssl_password
conn = psycopg2.connect(**args)
# Autocommit is enabled by default for safety for all new connections (to prevent long-lived transactions).
conn.set_session(autocommit=True)
return conn
def _connect(self):
"""Get and memoize connections to instances"""
if self.db and self.db.closed:
# Reset the connection object to retry to connect
self.db = None
if self.db:
if self.db.status != psycopg2.extensions.STATUS_READY:
# Some transaction went wrong and the connection is in an unhealthy state. Let's fix that
self.db.rollback()
else:
self.db = self._new_connection(self._config.dbname)
# Reload pg_settings on a new connection to the main db
def _load_pg_settings(self, db):
try:
with db.cursor(
cursor_factory=psycopg2.extras.DictCursor) as cursor:
self.log.debug("Running query [%s]", PG_SETTINGS_QUERY)
cursor.execute(
PG_SETTINGS_QUERY,
("pg_stat_statements.max", "track_activity_query_size"),
)
rows = cursor.fetchall()
self.pg_settings.clear()
for setting in rows:
name, val = setting
self.pg_settings[name] = val
except (psycopg2.DatabaseError, psycopg2.OperationalError) as err:
self.log.warning("Failed to query for pg_settings: %s", repr(err))
self.count(
"dd.postgres.error",
1,
tags=self._config.tags + ["error:load-pg-settings"] +
self._get_debug_tags(),
hostname=self.resolved_hostname,
)
def _get_db(self, dbname):
"""
Returns a memoized psycopg2 connection to `dbname` with autocommit
Threadsafe as long as no transactions are used
:param dbname:
:return: a psycopg2 connection
"""
# TODO: migrate the rest of this check to use a connection from this pool
with self._db_pool_lock:
db = self._db_pool.get(dbname)
if not db or db.closed:
self.log.debug("initializing connection to dbname=%s", dbname)
db = self._new_connection(dbname)
db.set_session(autocommit=True)
self._db_pool[dbname] = db
if self._config.dbname == dbname:
# reload settings for the main DB only once every time the connection is reestablished
self._load_pg_settings(db)
if db.status != psycopg2.extensions.STATUS_READY:
# Some transaction went wrong and the connection is in an unhealthy state. Let's fix that
db.rollback()
return db
def _close_db_pool(self):
# TODO: add automatic aging out of connections after some time
with self._db_pool_lock:
for dbname, db in self._db_pool.items():
if db and not db.closed:
try:
db.close()
except Exception:
self._log.exception(
"failed to close DB connection for db=%s", dbname)
self._db_pool[dbname] = None
def _collect_custom_queries(self, tags):
"""
Given a list of custom_queries, execute each query and parse the result for metrics
"""
for custom_query in self._config.custom_queries:
metric_prefix = custom_query.get('metric_prefix')
if not metric_prefix:
self.log.error(
"custom query field `metric_prefix` is required")
continue
metric_prefix = metric_prefix.rstrip('.')
query = custom_query.get('query')
if not query:
self.log.error(
"custom query field `query` is required for metric_prefix `%s`",
metric_prefix)
continue
columns = custom_query.get('columns')
if not columns:
self.log.error(
"custom query field `columns` is required for metric_prefix `%s`",
metric_prefix)
continue
cursor = self.db.cursor()
with closing(cursor) as cursor:
try:
self.log.debug("Running query: %s", query)
cursor.execute(query)
except (psycopg2.ProgrammingError,
psycopg2.errors.QueryCanceled) as e:
self.log.error(
"Error executing query for metric_prefix %s: %s",
metric_prefix, str(e))
self.db.rollback()
continue
for row in cursor:
if not row:
self.log.debug(
"query result for metric_prefix %s: returned an empty result",
metric_prefix)
continue
if len(columns) != len(row):
self.log.error(
"query result for metric_prefix %s: expected %s columns, got %s",
metric_prefix,
len(columns),
len(row),
)
continue
metric_info = []
query_tags = list(custom_query.get('tags', []))
query_tags.extend(tags)
for column, value in zip(columns, row):
# Columns can be ignored via configuration.
if not column:
continue
name = column.get('name')
if not name:
self.log.error(
"column field `name` is required for metric_prefix `%s`",
metric_prefix)
break
column_type = column.get('type')
if not column_type:
self.log.error(
"column field `type` is required for column `%s` of metric_prefix `%s`",
name,
metric_prefix,
)
break
if column_type == 'tag':
query_tags.append('{}:{}'.format(name, value))
else:
if not hasattr(self, column_type):
self.log.error(
"invalid submission method `%s` for column `%s` of metric_prefix `%s`",
column_type,
name,
metric_prefix,
)
break
try:
metric_info.append(
('{}.{}'.format(metric_prefix, name),
float(value), column_type))
except (ValueError, TypeError):
self.log.error(
"non-numeric value `%s` for metric column `%s` of metric_prefix `%s`",
value,
name,
metric_prefix,
)
break
# Only submit metrics if there were absolutely no errors - all or nothing.
else:
for info in metric_info:
metric, value, method = info
getattr(self,
method)(metric,
value,
tags=set(query_tags),
hostname=self.resolved_hostname)
def record_warning(self, code, message):
# type: (DatabaseConfigurationError, str) -> None
self._warnings_by_code[code] = message
def _report_warnings(self):
messages = self._warnings_by_code.values()
# Reset the warnings for the next check run
self._warnings_by_code = {}
for warning in messages:
self.warning(warning)
def check(self, _):
tags = copy.copy(self._config.tags)
# Collect metrics
try:
# Check version
self._connect()
if self._config.tag_replication_role:
tags.extend([
"replication_role:{}".format(self._get_replication_role())
])
self.log.debug("Running check against version %s: is_aurora: %s",
str(self.version), str(self.is_aurora))
self._collect_stats(tags)
self._collect_custom_queries(tags)
if self._config.dbm_enabled:
self.statement_metrics.run_job_loop(tags)
self.statement_samples.run_job_loop(tags)
if self._config.collect_wal_metrics:
self._collect_wal_metrics(tags)
except Exception as e:
self.log.exception("Unable to collect postgres metrics.")
self._clean_state()
self.db = None
message = u'Error establishing connection to postgres://{}:{}/{}, error is {}'.format(
self._config.host, self._config.port, self._config.dbname,
str(e))
self.service_check(
self.SERVICE_CHECK_NAME,
AgentCheck.CRITICAL,
tags=self._get_service_check_tags(),
message=message,
hostname=self.resolved_hostname,
)
raise e
else:
self.service_check(
self.SERVICE_CHECK_NAME,
AgentCheck.OK,
tags=self._get_service_check_tags(),
hostname=self.resolved_hostname,
)
try:
# commit to close the current query transaction
self.db.commit()
except Exception as e:
self.log.warning("Unable to commit: %s", e)
self._version = None # We don't want to cache versions between runs to capture minor updates for metadata
finally:
# Add the warnings saved during the execution of the check
self._report_warnings()