def get_events(self, type=None, mlapp_node=None, agent=None, start_time=None, end_time=None, is_alert=True):
"""
Fetch events according to a time window and other filters.
:param mlapp_node: Set the mlapp_node to select events from. If None, events will be fetched for the entire ION.
:param agent: (optional) Either an agent ID or an agent object. If specified, the query will filter by agent.
:param start_time: (optional) Filter by start time, this should be a datetime object.
:param end_time: (optional) Filter by end time, this should be a datetime object.
:param is_alert: (optional) Filter by is_alert - between alerts and events. Default is True. If None is provided
all events are fetched.
:return: A dataframe containing information about the events obtained.
:raises: MLOpsException
:Example:
>>> from parallelm.mlops import mlops as pm
>>> from datetime import datetime, timedelta
>>> node_agent_list = pm.get_agents("0")
>>> now = datetime.utcnow()
>>> hour_ago = (now - timedelta(hours=1))
>>> df = pm.get_events("training_comp", node_agent_list[0].id, hour_ago, now)
The above example assumes the MLApp this code is running as part of a node called "my_training".
1. The code obtains the list of agents which are used by the "my_training" component.
2. It computes the timestamp of now.
3. It computes the timestamp of an hour ago.
4. It calls the get_events method asking for all the events generated by the "0" node of the first
agent during the past hour.
"""
self._verify_mlops_is_ready()
self._verify_time_window(start_time, end_time, allow_none=True)
start_time_ms_timestamp = time_to_str_timestamp_milli(start_time) if start_time is not None else None
end_time_ms_timestamp = time_to_str_timestamp_milli(end_time) if end_time is not None else None
ion = self._mlops_ctx.ion()
# Translate the mlapp_node to its pipeline_instance_id
pipeline_inst_id = None
if mlapp_node is not None:
mlapp_node = self.get_node(mlapp_node)
if mlapp_node is None:
raise MLOpsException("No such {} node: [{}]".format(Constants.ION_LITERAL,
mlapp_node))
pipeline_inst_id = mlapp_node.pipeline_instance_id
# Translate the agent to its host name
agent_host = None
if agent is not None:
if isinstance(agent, Agent):
agent_host = agent.hostname
elif isinstance(agent, six.string_types):
agent_obj = self._mlops_ctx.get_agent_by_id(agent)
if agent_obj is None:
raise MLOpsException("Agent ID [{}] does not " +
"exists in this {}".format(agent, Constants.ION_LITERAL))
agent_host = agent_obj.host
if self._api_test_mode:
self._logger.info("API testing mode - returning without performing call")
return
ef = EventFilter()
ef.ion_name = None
ef.ion_inst_id = ion.id
ef.pipeline_inst_id = pipeline_inst_id
ef.agent_host = agent_host
ef.time_window_start = start_time_ms_timestamp
ef.time_window_end = end_time_ms_timestamp
ef.is_alert = is_alert
events_df = self._event_broker.get_events(ef)
return events_df
def test_datetime_timestamps():
ts = time_to_str_timestamp_milli(datetime.now())
assert ts is not None
def get_stats(self, name, mlapp_node, agent, start_time, end_time, no_of_lines=-1):
"""
Get statistics from MLOps.
This call provides a way for python code uploaded to MLOps to obtain access to
MLApp statistics. The get_stats() call allows access to statistics from each of the MLApp components; these may be
filtered by the agent. The start_time and end_time can be used to define a time window for the returned stats.
:param name: The name of the statistic - this is the name used in the stats method
:param mlapp_node: The name of the MLApp node
:param agent: ID of the agent which collected the statistics
:param start_time: A datetime object representing the start of the time window
:param end_time: A datetime object representing the end of the time window
:param no_of_lines: DEPRECATED
:return: A dataframe representing the statistics collected over time
:raises: MLOPsException
:Example:
>>> from parallelm.mlops import mlops as pm
>>> from datetime import datetime, timedelta
>>> node_agent_list = pm.get_agents("0")
>>> now = datetime.utcnow()
>>> hour_ago = (now - timedelta(hours=1))
>>> df = pm.get_stats("myValue", "0", node_agent_list[0].id, hour_ago, now)
The above example assumes the MLApp this code is running as part of has a component called "my_training".
1. The code obtains the list of agents which are used by the "my_training" component.
2. It computes the timestamp of now.
3. It computes the timestamp of an hour ago.
4. It calls the get_stats method asking for the "myValue" statistic from the first agent.
The return value is a dataframe object.
"""
self._verify_mlops_is_ready()
ion = self._mlops_ctx.ion()
self._verify_get_stats_args(name, mlapp_node, agent)
self._verify_time_window(start_time, end_time)
start_time_ms_timestamp = time_to_str_timestamp_milli(start_time)
end_time_ms_timestamp = time_to_str_timestamp_milli(end_time)
self._logger.info("Calling get_stat with start_time: {} end_time: {}".format(
start_time_ms_timestamp, end_time_ms_timestamp))
if self._api_test_mode:
self._logger.info("API testing mode - returning without performing call")
return
# A valid system should exist past this point
agents_per_node = self._assemble_nodes_and_agents(ion, mlapp_node, agent)
self._logger.info("agents_per_node:" + str(agents_per_node))
df_helper = DataFrameHelper()
df_list = []
for node_name in agents_per_node.keys():
node_obj = self.get_node(node_name)
self._logger.info("{} component: {}".format(Constants.ION_LITERAL, node_name))
for agent_obj in agents_per_node[node_name]:
self._logger.info("stat: {} {} node: {} agent: {}".format(name,
Constants.ION_LITERAL,
node_obj.name,
agent_obj.id))
json = self._mlops_ctx.rest_helper().get_stat(stat_name=name,
ion_id=ion.id,
workflow_node_id=node_obj.id,
agent_id=agent_obj.id,
pipeline_id=
node_obj.pipeline_agent_set[
agent_obj.id],
start_time=start_time_ms_timestamp,
end_time=end_time_ms_timestamp)
# TODO: next set of patches will convert the JSON into a dataframe with the correct format
self._logger.info(pprint.pformat(json))
df = df_helper.create_data_frame(json, no_of_lines)
df_helper.add_col_with_same_value(df, DataframeColNames.ION_NODE, node_name)
df_helper.add_col_with_same_value(df, DataframeColNames.AGENT, agent_obj.id)
df_list.append(df)
final_df = pd.concat(df_list, ignore_index=True)
return final_df