def get_processes(gs, docker_host, container_id):
"""Gets the list of all processes in the 'docker_host' and 'container_id'.
If the container is not found, returns an empty list of processes.
Args:
gs: global state.
docker_host: the Docker host running the container.
container_id: the container running the processes.
Returns:
list of wrapped process objects.
Each element in the list is the result of
utilities.wrap_object(process, 'Process', ...)
Raises:
CollectorError in case of failure to fetch data from Docker.
Other exceptions may be raised due to exectution errors.
"""
processes_label = '%s/%s' % (docker_host, container_id)
processes, timestamp_secs = gs.get_processes_cache().lookup(
processes_label)
if timestamp_secs is not None:
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) cache hit',
docker_host, container_id)
return processes
container = get_one_container(gs, docker_host, container_id)
if container is not None:
assert utilities.is_wrapped_object(container, 'Container')
container_short_hex_id = utilities.object_to_hex_id(container['properties'])
assert utilities.valid_string(container_short_hex_id)
else:
# Parent container not found. Container might have crashed while we were
# looking for it.
return []
# NOTE: there is no trailing /json in this URL - this looks like a bug in the
# Docker API
url = ('http://{docker_host}:{port}/containers/{container_id}/top?'
'ps_args=aux'.format(docker_host=docker_host,
port=gs.get_docker_port(),
container_id=container_id))
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
# The typical value of 'container_id' is:
# k8s_php-redis.b317029a_guestbook-controller-ls6k1.default.api_f991d53e-b949-11e4-8246-42010af0c3dd_8dcdfec8
# Use just the tail of the container ID after the last '_' sign.
fname = '{host}-processes-{id}'.format(
host=docker_host.split('.')[0], id=container_id.split('_')[-1])
try:
# TODO(vasbala): what should we do in cases where the container is gone
# (and replaced by a different one)?
result = fetch_data(gs, url, fname, expect_missing=True)
except ValueError:
# this container does not exist anymore
return []
except collector_error.CollectorError:
raise
except:
msg = 'fetching %s failed with exception %s' % (url, sys.exc_info()[0])
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
if not isinstance(utilities.get_attribute(result, ['Titles']),
types.ListType):
invalid_processes(gs, url)
if not isinstance(utilities.get_attribute(result, ['Processes']),
types.ListType):
invalid_processes(gs, url)
pstats = result['Titles']
processes = []
now = time.time()
for pvalues in result['Processes']:
process = {}
if not isinstance(pvalues, types.ListType):
invalid_processes(gs, url)
if len(pstats) != len(pvalues):
invalid_processes(gs, url)
for pstat, pvalue in zip(pstats, pvalues):
process[pstat] = pvalue
# Prefix with container Id to ensure uniqueness across the whole graph.
process_id = '%s/%s' % (container_short_hex_id, process['PID'])
processes.append(utilities.wrap_object(
process, 'Process', process_id, now, label=process['PID']))
ret_value = gs.get_processes_cache().update(
processes_label, processes, now)
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) returns %d processes',
docker_host, container_id, len(processes))
return ret_value
def get_image(gs, docker_host, container):
"""Gets the information of the given image in the given host.
Args:
gs: global state.
docker_host: Docker host name. Must not be empty.
container: the container which runs the image.
Returns:
If image was found, returns the wrapped image object, which is the result of
utilities.wrap_object(image, 'Image', ...)
If the image was not found, returns None.
Raises:
CollectorError: in case of failure to fetch data from Docker.
ValueError: in case the container does not contain a valid image ID.
Other exceptions may be raised due to exectution errors.
"""
assert utilities.is_wrapped_object(container, 'Container')
# The 'image_id' should be a long hexadecimal string.
image_id = utilities.get_attribute(container, ['properties', 'Image'])
if not utilities.valid_hex_id(image_id):
msg = 'missing or invalid image ID in container ID=%s' % container['id']
gs.logger_error(msg)
raise ValueError(msg)
# The 'image_name' should be a symbolic name (not a hexadecimal string).
image_name = utilities.get_attribute(
container, ['properties', 'Config', 'Image'])
if ((not utilities.valid_string(image_name)) or
utilities.valid_hex_id(image_name)):
msg = 'missing or invalid image name in container ID=%s' % container['id']
gs.logger_error(msg)
raise ValueError(msg)
cache_key = '%s|%s' % (docker_host, image_id)
image, timestamp_secs = gs.get_images_cache().lookup(cache_key)
if timestamp_secs is not None:
gs.logger_info('get_image(docker_host=%s, image_id=%s) cache hit',
docker_host, image_id)
return image
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
# The typical value of 'image_name' is:
# brendanburns/php-redis
# We convert embedded '/' and ':' characters to '-' to avoid interference with
# the directory structure or file system.
url = 'http://{docker_host}:{port}/images/{image_id}/json'.format(
docker_host=docker_host, port=gs.get_docker_port(), image_id=image_id)
fname = '{host}-image-{id}'.format(
host=docker_host.split('.')[0],
id=image_name.replace('/', '-').replace(':', '-'))
try:
image = fetch_data(gs, url, fname, expect_missing=True)
except ValueError:
# image not found.
msg = 'image not found for image_id: %s' % image_id
gs.logger_info(msg)
return None
except collector_error.CollectorError:
raise
except:
msg = 'fetching %s failed with exception %s' % (url, sys.exc_info()[0])
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
now = time.time()
# compute the two labels of the image.
# The first is a 12-digit hexadecimal number shown by "docker images".
# The second is the symbolic name of the image.
full_hex_label = image.get('Id')
if not (isinstance(full_hex_label, types.StringTypes) and full_hex_label):
msg = 'Image id=%s has an invalid "Id" attribute value' % image_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
short_hex_label = utilities.object_to_hex_id(image)
if short_hex_label is None:
msg = 'Could not compute short hex ID of image %s' % image_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
wrapped_image = utilities.wrap_object(
image, 'Image', full_hex_label, now,
label=short_hex_label, alt_label=image_name)
ret_value = gs.get_images_cache().update(cache_key, wrapped_image, now)
gs.logger_info('get_image(docker_host=%s, image_id=%s, image_name=%s)',
docker_host, image_id, image_name)
return ret_value
def get_containers(gs, docker_host):
"""Gets the list of all containers in 'docker_host'.
Args:
gs: global state.
docker_host: the Docker host running the containers.
Returns:
list of wrapped container objects.
Each element in the list is the result of
utilities.wrap_object(container, 'Container', ...)
Raises:
CollectorError: in case of failure to fetch data from Docker.
Other exceptions may be raised due to exectution errors.
"""
containers, timestamp = gs.get_containers_cache().lookup(docker_host)
if timestamp is not None:
gs.logger_info(
'get_containers(docker_host=%s) cache hit returns '
'%d containers', docker_host, len(containers))
return containers
url = 'http://{docker_host}:{port}/containers/json'.format(
docker_host=docker_host, port=gs.get_docker_port())
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
fname = '{host}-containers'.format(host=docker_host.split('.')[0])
try:
containers_list = fetch_data(gs, url, fname)
except collector_error.CollectorError:
raise
except:
msg = ('fetching %s or %s failed with exception %s' %
(url, fname, sys.exc_info()[0]))
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
if not isinstance(containers_list, types.ListType):
msg = 'invalid response from fetching %s' % url
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
containers = []
timestamps = []
for container_info in containers_list:
# NOTE: container 'Name' is stable across container re-starts whereas
# container 'Id' is not.
# This may be because Kubernertes assigns the Name while Docker assigns
# the Id (?)
# The container Name is the only element of the array 'Names' -
# why is Names an array here?
# skip the leading / in the Name
if not (isinstance(container_info.get('Names'), types.ListType) and
container_info['Names'] and
utilities.valid_string(container_info['Names'][0]) and
container_info['Names'][0][0] == '/'):
msg = 'invalid containers data format. docker_host=%s' % docker_host
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
container_id = container_info['Names'][0][1:]
container, ts = _inspect_container(gs, docker_host, container_id)
if container is None:
continue
if not utilities.valid_string(container.get('Name')):
msg = ('missing or invalid Name attribute in container %s' %
container_id)
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
if container['Name'] != ('/' + container_id):
msg = ('container %s\'s Name attribute is "%s"; expecting "%s"' %
(container_id, container['Name'], '/' + container_id))
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
short_hex_id = utilities.object_to_hex_id(container)
if short_hex_id is None:
msg = 'Could not compute short hex ID of container %s' % container_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
wrapped_container = utilities.wrap_object(
container, 'Container', container_id, ts, label=short_hex_id)
containers.append(wrapped_container)
timestamps.append(ts)
# Modify the container's label after the wrapped container was added
# to the containers list.
# Compute the container's short name to create a better container label:
# short_container_name/short_hex_id.
# For example: "cassandra/d85b599c17d8".
parent_pod_id = utilities.get_parent_pod_id(wrapped_container)
if parent_pod_id is None:
continue
parent_pod = kubernetes.get_one_pod(gs, docker_host, parent_pod_id)
if parent_pod is None:
continue
short_container_name = utilities.get_short_container_name(
wrapped_container, parent_pod)
if not utilities.valid_string(short_container_name):
continue
wrapped_container['annotations']['label'] = (short_container_name + '/' +
short_hex_id)
ret_value = gs.get_containers_cache().update(
docker_host, containers,
min(timestamps) if timestamps else time.time())
gs.logger_info(
'get_containers(docker_host=%s) returns %d containers',
docker_host, len(containers))
return ret_value
def get_processes(gs, docker_host, container_id):
"""Gets the list of all processes in the 'docker_host' and 'container_id'.
If the container is not found, returns an empty list of processes.
Args:
gs: global state.
docker_host: the Docker host running the container.
container_id: the container running the processes.
Returns:
list of wrapped process objects.
Each element in the list is the result of
utilities.wrap_object(process, 'Process', ...)
Raises:
CollectorError in case of failure to fetch data from Docker.
Other exceptions may be raised due to exectution errors.
"""
processes_label = '%s/%s' % (docker_host, container_id)
processes, timestamp_secs = gs.get_processes_cache().lookup(
processes_label)
if timestamp_secs is not None:
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) cache hit',
docker_host, container_id)
return processes
container = get_one_container(gs, docker_host, container_id)
if container is not None:
assert utilities.is_wrapped_object(container, 'Container')
container_short_hex_id = utilities.object_to_hex_id(container['properties'])
assert utilities.valid_string(container_short_hex_id)
else:
# Parent container not found. Container might have crashed while we were
# looking for it.
return []
container_name = utilities.get_container_name(container)
if not utilities.valid_string(container_name):
msg = 'Invalid container "Name" attribute in container %s' % container_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
# NOTE: there is no trailing /json in this URL - this looks like a bug in the
# Docker API
# Note that the {container_id} in the URL must be the internal container
# name in container['properties']['Name'][1:]
# and not the container name in container['id'] which may contain an extra
# suffix.
url = ('http://{docker_host}:{port}/containers/{container_name}/top?'
'ps_args=aux'.format(docker_host=docker_host,
port=gs.get_docker_port(),
container_name=container_name))
fname = utilities.container_id_to_fname(
docker_host, 'processes', container_name)
try:
# TODO(vasbala): what should we do in cases where the container is gone
# (and replaced by a different one)?
result = fetch_data(gs, url, fname, expect_missing=True)
except ValueError:
# this container does not exist anymore
return []
except collector_error.CollectorError:
raise
except:
msg = 'fetching %s failed with exception %s' % (url, sys.exc_info()[0])
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
if not isinstance(utilities.get_attribute(result, ['Titles']),
types.ListType):
invalid_processes(gs, url)
if not isinstance(utilities.get_attribute(result, ['Processes']),
types.ListType):
invalid_processes(gs, url)
pstats = result['Titles']
processes = []
now = time.time()
for pvalues in result['Processes']:
process = {}
if not isinstance(pvalues, types.ListType):
invalid_processes(gs, url)
if len(pstats) != len(pvalues):
invalid_processes(gs, url)
for pstat, pvalue in zip(pstats, pvalues):
process[pstat] = pvalue
# Prefix with container Id to ensure uniqueness across the whole graph.
process_id = '%s/%s' % (container_short_hex_id, process['PID'])
processes.append(utilities.wrap_object(
process, 'Process', process_id, now, label=process['PID']))
ret_value = gs.get_processes_cache().update(
processes_label, processes, now)
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) returns %d processes',
docker_host, container_id, len(processes))
return ret_value
def get_image(gs, docker_host, container):
"""Gets the information of the given image in the given host.
Args:
gs: global state.
docker_host: Docker host name. Must not be empty.
container: the container which runs the image.
Returns:
If image was found, returns the wrapped image object, which is the result of
utilities.wrap_object(image, 'Image', ...)
If the image was not found, returns None.
Raises:
CollectorError: in case of failure to fetch data from Docker.
ValueError: in case the container does not contain a valid image ID.
Other exceptions may be raised due to exectution errors.
"""
assert utilities.is_wrapped_object(container, 'Container')
# The 'image_id' should be a long hexadecimal string.
image_id = utilities.get_attribute(container, ['properties', 'Image'])
if not utilities.valid_hex_id(image_id):
msg = 'missing or invalid image ID in container ID=%s' % container['id']
gs.logger_error(msg)
raise ValueError(msg)
# The 'image_name' should be a symbolic name (not a hexadecimal string).
image_name = utilities.get_attribute(container,
['properties', 'Config', 'Image'])
if ((not utilities.valid_string(image_name))
or utilities.valid_hex_id(image_name)):
msg = 'missing or invalid image name in container ID=%s' % container[
'id']
gs.logger_error(msg)
raise ValueError(msg)
cache_key = '%s|%s' % (docker_host, image_id)
image, timestamp_secs = gs.get_images_cache().lookup(cache_key)
if timestamp_secs is not None:
gs.logger_info('get_image(docker_host=%s, image_id=%s) cache hit',
docker_host, image_id)
return image
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
# The typical value of 'image_name' is:
# brendanburns/php-redis
# We convert embedded '/' and ':' characters to '-' to avoid interference with
# the directory structure or file system.
url = 'http://{docker_host}:{port}/images/{image_id}/json'.format(
docker_host=docker_host, port=gs.get_docker_port(), image_id=image_id)
fname = '{host}-image-{id}'.format(host=docker_host.split('.')[0],
id=image_name.replace('/', '-').replace(
':', '-'))
try:
image = fetch_data(gs, url, fname, expect_missing=True)
except ValueError:
# image not found.
msg = 'image not found for image_id: %s' % image_id
gs.logger_info(msg)
return None
except collector_error.CollectorError:
raise
except:
msg = 'fetching %s failed with exception %s' % (url, sys.exc_info()[0])
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
now = time.time()
# compute the two labels of the image.
# The first is a 12-digit hexadecimal number shown by "docker images".
# The second is the symbolic name of the image.
full_hex_label = image.get('Id')
if not (isinstance(full_hex_label, types.StringTypes) and full_hex_label):
msg = 'Image id=%s has an invalid "Id" attribute value' % image_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
short_hex_label = utilities.object_to_hex_id(image)
if short_hex_label is None:
msg = 'Could not compute short hex ID of image %s' % image_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
wrapped_image = utilities.wrap_object(image,
'Image',
full_hex_label,
now,
label=short_hex_label,
alt_label=image_name)
ret_value = gs.get_images_cache().update(cache_key, wrapped_image, now)
gs.logger_info('get_image(docker_host=%s, image_id=%s, image_name=%s)',
docker_host, image_id, image_name)
return ret_value
def get_processes(gs, docker_host, container_id):
"""Gets the list of all processes in the 'docker_host' and 'container_id'.
If the container is not found, returns an empty list of processes.
Args:
gs: global state.
docker_host: the Docker host running the container.
container_id: the container running the processes.
Returns:
list of wrapped process objects.
Each element in the list is the result of
utilities.wrap_object(process, 'Process', ...)
Raises:
CollectorError in case of failure to fetch data from Docker.
Other exceptions may be raised due to exectution errors.
"""
processes_label = '%s/%s' % (docker_host, container_id)
processes, timestamp_secs = gs.get_processes_cache().lookup(
processes_label)
if timestamp_secs is not None:
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) cache hit',
docker_host, container_id)
return processes
container = get_one_container(gs, docker_host, container_id)
if container is not None:
assert utilities.is_wrapped_object(container, 'Container')
container_short_hex_id = utilities.object_to_hex_id(
container['properties'])
assert utilities.valid_string(container_short_hex_id)
else:
# Parent container not found. Container might have crashed while we were
# looking for it.
return []
# NOTE: there is no trailing /json in this URL - this looks like a bug in the
# Docker API
url = ('http://{docker_host}:{port}/containers/{container_id}/top?'
'ps_args=aux'.format(docker_host=docker_host,
port=gs.get_docker_port(),
container_id=container_id))
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
# The typical value of 'container_id' is:
# k8s_php-redis.b317029a_guestbook-controller-ls6k1.default.api_f991d53e-b949-11e4-8246-42010af0c3dd_8dcdfec8
# Use just the tail of the container ID after the last '_' sign.
fname = '{host}-processes-{id}'.format(host=docker_host.split('.')[0],
id=container_id.split('_')[-1])
try:
# TODO(vasbala): what should we do in cases where the container is gone
# (and replaced by a different one)?
result = fetch_data(gs, url, fname, expect_missing=True)
except ValueError:
# this container does not exist anymore
return []
except collector_error.CollectorError:
raise
except:
msg = 'fetching %s failed with exception %s' % (url, sys.exc_info()[0])
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
if not isinstance(utilities.get_attribute(result, ['Titles']),
types.ListType):
invalid_processes(gs, url)
if not isinstance(utilities.get_attribute(result, ['Processes']),
types.ListType):
invalid_processes(gs, url)
pstats = result['Titles']
processes = []
now = time.time()
for pvalues in result['Processes']:
process = {}
if not isinstance(pvalues, types.ListType):
invalid_processes(gs, url)
if len(pstats) != len(pvalues):
invalid_processes(gs, url)
for pstat, pvalue in zip(pstats, pvalues):
process[pstat] = pvalue
# Prefix with container Id to ensure uniqueness across the whole graph.
process_id = '%s/%s' % (container_short_hex_id, process['PID'])
processes.append(
utilities.wrap_object(process,
'Process',
process_id,
now,
label=process['PID']))
ret_value = gs.get_processes_cache().update(processes_label, processes,
now)
gs.logger_info(
'get_processes(docker_host=%s, container_id=%s) returns %d processes',
docker_host, container_id, len(processes))
return ret_value
def get_containers(gs, docker_host):
"""Gets the list of all containers in 'docker_host'.
Args:
gs: global state.
docker_host: the Docker host running the containers.
Returns:
list of wrapped container objects.
Each element in the list is the result of
utilities.wrap_object(container, 'Container', ...)
Raises:
CollectorError: in case of failure to fetch data from Docker.
Other exceptions may be raised due to exectution errors.
"""
containers, timestamp = gs.get_containers_cache().lookup(docker_host)
if timestamp is not None:
gs.logger_info(
'get_containers(docker_host=%s) cache hit returns '
'%d containers', docker_host, len(containers))
return containers
url = 'http://{docker_host}:{port}/containers/json'.format(
docker_host=docker_host, port=gs.get_docker_port())
# A typical value of 'docker_host' is:
# k8s-guestbook-node-3.c.rising-apricot-840.internal
# Use only the first period-seperated element for the test file name.
fname = '{host}-containers'.format(host=docker_host.split('.')[0])
try:
containers_list = fetch_data(gs, url, fname)
except collector_error.CollectorError:
raise
except:
msg = ('fetching %s or %s failed with exception %s' %
(url, fname, sys.exc_info()[0]))
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
if not isinstance(containers_list, types.ListType):
msg = 'invalid response from fetching %s' % url
gs.logger_exception(msg)
raise collector_error.CollectorError(msg)
containers = []
timestamps = []
for container_info in containers_list:
# NOTE: container 'Name' is stable across container re-starts whereas
# container 'Id' is not.
# This may be because Kubernertes assigns the Name while Docker assigns
# the Id (?)
# The container Name is the only element of the array 'Names' -
# why is Names an array here?
# skip the leading / in the Name
if not (isinstance(container_info.get('Names'), types.ListType)
and container_info['Names']
and utilities.valid_string(container_info['Names'][0])
and container_info['Names'][0][0] == '/'):
msg = 'invalid containers data format. docker_host=%s' % docker_host
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
container_id = container_info['Names'][0][1:]
container, ts = _inspect_container(gs, docker_host, container_id)
if container is None:
continue
if not utilities.valid_string(container.get('Name')):
msg = ('missing or invalid Name attribute in container %s' %
container_id)
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
if container['Name'] != ('/' + container_id):
msg = ('container %s\'s Name attribute is "%s"; expecting "%s"' %
(container_id, container['Name'], '/' + container_id))
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
short_hex_id = utilities.object_to_hex_id(container)
if short_hex_id is None:
msg = 'Could not compute short hex ID of container %s' % container_id
gs.logger_error(msg)
raise collector_error.CollectorError(msg)
wrapped_container = utilities.wrap_object(container,
'Container',
container_id,
ts,
label=short_hex_id)
containers.append(wrapped_container)
timestamps.append(ts)
# Modify the container's label after the wrapped container was added
# to the containers list.
# Compute the container's short name to create a better container label:
# short_container_name/short_hex_id.
# For example: "cassandra/d85b599c17d8".
parent_pod_id = utilities.get_parent_pod_id(wrapped_container)
if parent_pod_id is None:
continue
parent_pod = kubernetes.get_one_pod(gs, docker_host, parent_pod_id)
if parent_pod is None:
continue
short_container_name = utilities.get_short_container_name(
wrapped_container, parent_pod)
if not utilities.valid_string(short_container_name):
continue
wrapped_container['annotations']['label'] = (short_container_name +
'/' + short_hex_id)
ret_value = gs.get_containers_cache().update(
docker_host, containers,
min(timestamps) if timestamps else time.time())
gs.logger_info('get_containers(docker_host=%s) returns %d containers',
docker_host, len(containers))
return ret_value
