def collectMetricStatistics(self, autostack, metric):
""" Get a sequence of min/max statistics for a given metric from the
Autostack's instances
:param autostack: an autostack object
:type autostack: TODO
:param metric: a metric object linked to the given autostack
:type metric: TODO
:returns: a possibly empty, unordered sequence of InstanceMetricData
objects, each containing a single MetricRecord object in its `records`
attribute.
"""
self._updateInstanceCache((autostack, ))
executionStart = time.time()
# Initialize the result
result = tuple()
# Determine the maximum time range for gathering individual statistics
timeSlice = self._getMetricStatisticsTimeSlice(
period=metric.poll_interval)
# Create tasks for concurrent execution
tasks = self._createMetricStatisticsCollectionTasks(
autostack=autostack,
metric=metric,
stats=self._STATS_OF_INTEREST.keys(),
timeSlice=timeSlice,
instanceCache=self._instanceCache)
# Execute the tasks, if any, via worker pool
numFailedTasks = 0
sampleInfo = None
if tasks:
# TODO: evaluate performance with
# chunksize=max(1, len(tasks)//self._WORKER_PROCESS_POOL_SIZE)
taskResults = self._processPool.map(_collectMetrics,
tasks,
chunksize=1)
assert len(taskResults) == len(tasks), (len(taskResults),
len(tasks))
result, numFailedTasks, sampleInfo = (
self._processMetricStatisticsResults(taskResults=taskResults,
log=self._log))
self._log.info(
"Completed collection of stats for metric=<%s> of autostack=<%s>: "
"numTasks=%d; numFailedTasks=%s; sampleCounts=%s; numSlices=%s; "
"duration=%ss", getMetricLogPrefix(metric),
getAutostackLogPrefix(autostack), len(tasks), numFailedTasks,
sampleInfo, len(result),
time.time() - executionStart)
return result
def collectMetricStatistics(self, autostack, metric):
""" Get a sequence of min/max statistics for a given metric from the
Autostack's instances
:param autostack: an autostack object
:type autostack: TODO
:param metric: a metric object linked to the given autostack
:type metric: TODO
:returns: a possibly empty, unordered sequence of InstanceMetricData
objects, each containing a single MetricRecord object in its `records`
attribute.
"""
self._updateInstanceCache((autostack,))
executionStart = time.time()
# Initialize the result
result = tuple()
# Determine the maximum time range for gathering individual statistics
timeSlice = self._getMetricStatisticsTimeSlice(period=metric.poll_interval)
# Create tasks for concurrent execution
tasks = self._createMetricStatisticsCollectionTasks(
autostack=autostack,
metric=metric,
stats=self._STATS_OF_INTEREST.keys(),
timeSlice=timeSlice,
instanceCache=self._instanceCache)
# Execute the tasks, if any, via worker pool
numFailedTasks = 0
sampleInfo = None
if tasks:
# TODO: evaluate performance with
# chunksize=max(1, len(tasks)//self._WORKER_PROCESS_POOL_SIZE)
taskResults = self._processPool.map(_collectMetrics, tasks, chunksize=1)
assert len(taskResults) == len(tasks), (len(taskResults), len(tasks))
result, numFailedTasks, sampleInfo = (
self._processMetricStatisticsResults(
taskResults=taskResults,
log=self._log))
self._log.info(
"Completed collection of stats for metric=<%s> of autostack=<%s>: "
"numTasks=%d; numFailedTasks=%s; sampleCounts=%s; numSlices=%s; "
"duration=%ss",
getMetricLogPrefix(metric), getAutostackLogPrefix(autostack), len(tasks),
numFailedTasks, sampleInfo, len(result), time.time() - executionStart)
return result
def _processAutostackMetricRequests(self, engine, requests, modelSwapper):
""" Execute autostack metric requests, aggregate and stream
collected metric data
:param engine: SQLAlchemy engine object
:type engine: sqlalchemy.engine.Engine
:param requests: sequence of AutostackMetricRequest objects
:param modelSwapper: Model Swapper
"""
# Start collecting requested metric data
collectionIter = self._metricGetter.collectMetricData(requests)
# Aggregate each collection and dispatch to app MetricStreamer
for metricCollection in collectionIter:
request = requests[metricCollection.refID]
metricObj = request.metric
data = None
if metricCollection.slices:
aggregationFn = getAggregationFn(metricObj)
if aggregationFn:
data = aggregate(metricCollection.slices, aggregationFn=aggregationFn)
else:
data = aggregate(metricCollection.slices)
try:
with engine.connect() as conn:
repository.retryOnTransientErrors(repository.setMetricLastTimestamp)(
conn, metricObj.uid, metricCollection.nextMetricTime
)
except ObjectNotFoundError:
self._log.warning(
"Processing autostack data collection results for " "unknown model=%s (model deleted?)",
metricObj.uid,
)
continue
if data:
try:
self.metricStreamer.streamMetricData(data, metricID=metricObj.uid, modelSwapper=modelSwapper)
except ObjectNotFoundError:
# We expect that the model exists but in the odd case that it has
# already been deleted we don't want to crash the process.
self._log.info("Metric not found when adding data. metric=%s" % metricObj.uid)
self._log.debug(
"{TAG:APP.AGG.DATA.PUB} Published numItems=%d for metric=%s;"
"timeRange=[%sZ-%sZ]; headTS=%sZ; tailTS=%sZ",
len(data),
getMetricLogPrefix(metricObj),
metricCollection.timeRange.start.isoformat(),
metricCollection.timeRange.end.isoformat(),
data[0][0].isoformat(),
data[-1][0].isoformat(),
)
else:
self._log.info(
"{TAG:APP.AGG.DATA.NONE} No data for metric=%s;" "timeRange=[%sZ-%sZ]",
getMetricLogPrefix(metricObj),
metricCollection.timeRange.start.isoformat(),
metricCollection.timeRange.end.isoformat(),
)
def _processAutostackMetricRequests(self, engine, requests, modelSwapper):
""" Execute autostack metric requests, aggregate and stream
collected metric data
:param engine: SQLAlchemy engine object
:type engine: sqlalchemy.engine.Engine
:param requests: sequence of AutostackMetricRequest objects
:param modelSwapper: Model Swapper
"""
# Start collecting requested metric data
collectionIter = self._metricGetter.collectMetricData(requests)
# Aggregate each collection and dispatch to app MetricStreamer
for metricCollection in collectionIter:
request = requests[metricCollection.refID]
metricObj = request.metric
data = None
if metricCollection.slices:
aggregationFn = getAggregationFn(metricObj)
if aggregationFn:
data = aggregate(metricCollection.slices,
aggregationFn=aggregationFn)
else:
data = aggregate(metricCollection.slices)
try:
with engine.connect() as conn:
repository.retryOnTransientErrors(repository.setMetricLastTimestamp)(
conn, metricObj.uid, metricCollection.nextMetricTime)
except ObjectNotFoundError:
self._log.warning("Processing autostack data collection results for "
"unknown model=%s (model deleted?)", metricObj.uid)
continue
if data:
try:
self.metricStreamer.streamMetricData(data,
metricID=metricObj.uid,
modelSwapper=modelSwapper)
except ObjectNotFoundError:
# We expect that the model exists but in the odd case that it has
# already been deleted we don't want to crash the process.
self._log.info("Metric not found when adding data. metric=%s" %
metricObj.uid)
self._log.debug(
"{TAG:APP.AGG.DATA.PUB} Published numItems=%d for metric=%s;"
"timeRange=[%sZ-%sZ]; headTS=%sZ; tailTS=%sZ",
len(data), getMetricLogPrefix(metricObj),
metricCollection.timeRange.start.isoformat(),
metricCollection.timeRange.end.isoformat(),
data[0][0].isoformat(), data[-1][0].isoformat())
else:
self._log.info(
"{TAG:APP.AGG.DATA.NONE} No data for metric=%s;"
"timeRange=[%sZ-%sZ]", getMetricLogPrefix(metricObj),
metricCollection.timeRange.start.isoformat(),
metricCollection.timeRange.end.isoformat())
