def optimalWriteOrder():
"Generates metrics with the most cached values first and applies a soft rate limit on new metrics"
global lastCreateInterval
global createCount
metrics = MetricCache.counts()
t = time.time()
#metrics.sort(key=lambda item: item[1], reverse=True) # by queue size, descending
log.msg("Sorted %d cache queues in %.6f seconds" % (len(metrics), time.time() - t))
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
for metric, queueSize in metrics:
#Handling special characters in metric names
metric_sanit = list(metric)
for i in range(0,len(metric_sanit),1):
if metric_sanit[i] < '\x20' or metric_sanit[i] > '\x7e':
metric_sanit[i] = '_'
metric_sanit = "".join(metric_sanit)
try:
dbFilePath = getFilesystemPath(metric_sanit)
dbFileExists = exists(dbFilePath)
except:
log.err()
log.msg("dbFilePath: %s" % (dbFilePath))
continue
if not dbFileExists:
createCount += 1
now = time.time()
if now - lastCreateInterval >= 60:
lastCreateInterval = now
createCount = 1
elif createCount >= settings.MAX_CREATES_PER_MINUTE:
# dropping queued up datapoints for new metrics prevents filling up the entire cache
# when a bunch of new metrics are received.
try:
MetricCache.pop(metric)
except KeyError:
pass
continue
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" % metric)
continue # we simply move on to the next metric when this race condition occurs
yield (metric_sanit, datapoints, dbFilePath, dbFileExists)
def optimalWriteOrder():
"Generates metrics with the most cached values first and applies a soft rate limit on new metrics"
global lastCreateInterval
global createCount
metrics = MetricCache.counts()
t = time.time()
metrics.sort(key=lambda item: item[1], reverse=True) # by queue size, descending
log.msg("Sorted %d cache queues in %.6f seconds" % (len(metrics), time.time() - t))
for metric, queueSize in metrics:
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
# Let our persister do its own check, and ignore the metric if needed.
if not persister.pre_get_datapoints_check(metric):
continue
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" % metric)
continue # we simply move on to the next metric when this race condition occurs
dbInfo = persister.get_dbinfo(metric)
dbIdentifier = dbInfo[0]
dbExists = dbInfo[1]
yield (metric, datapoints, dbIdentifier, dbExists)
def _flush(prefix=None):
""" Write/create whisped files at maximal speed """
assert(prefix==None or hasattr(prefix, 'startswith'))
log.msg("flush started (prefix: %s)" % prefix)
started = time.time()
metrics = MetricCache.counts()
updates = 0
write_lock.acquire()
try:
for metric, queueSize in metrics:
if prefix and not metric.startswith(prefix):
continue
dbFilePath = getFilesystemPath(metric)
dbFileExists = exists(dbFilePath)
try:
datapoints = MetricCache.pop(metric)
except KeyError:
continue
if not createWhisperFile(metric, dbFilePath, dbFileExists):
continue
if not writeWhisperFile(dbFilePath, datapoints):
continue
updates += 1
finally:
write_lock.release()
log.msg('flush finished (updates: %d, time: %.5f sec)' % (updates, time.time()-started))
return updates
def optimalWriteOrder():
"Generates metrics with the most cached values first and applies a soft rate limit on new metrics"
global lastCreateInterval
global createCount
metrics = [ (metric, len(datapoints)) for metric,datapoints in MetricCache.items() ]
t = time.time()
metrics.sort(key=lambda item: item[1], reverse=True) # by queue size, descending
log.msg("Sorted %d cache queues in %.6f seconds" % (len(metrics), time.time() - t))
for metric, queueSize in metrics:
dbFilePath = getFilesystemPath(metric)
dbFileExists = exists(dbFilePath)
if not dbFileExists:
createCount += 1
now = time.time()
if now - lastCreateInterval >= 60:
lastCreateInterval = now
createCount = 1
elif createCount >= settings.MAX_CREATES_PER_MINUTE:
continue
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" % metric)
continue # we simply move on to the next metric when this race condition occurs
yield (metric, datapoints, dbFilePath, dbFileExists)
def getMetrics(self):
metrics = MetricCache.counts()
for metric, queueSize in metrics:
datapoints = MetricCache.pop(metric)
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
yield (metric, datapoints)
def test_write_strategy_sorted(self):
"""Create a metric cache, insert metrics, ensure sorted writes"""
self.assertEqual("sorted", MetricCache.method)
now = time.time()
datapoint1 = (now - 10, float(1))
datapoint2 = (now, float(2))
MetricCache.store("d.e.f", datapoint1)
MetricCache.store("a.b.c", datapoint1)
MetricCache.store("a.b.c", datapoint2)
(m, d) = MetricCache.pop()
self.assertEqual(("a.b.c", deque([datapoint1, datapoint2])), (m, d))
(m, d) = MetricCache.pop()
self.assertEqual(("d.e.f", deque([datapoint1])), (m, d))
self.assertEqual(0, MetricCache.size)
def optimalWriteOrder():
"""Generates metrics with the most cached values first and applies a soft
rate limit on new metrics"""
global lastCreateInterval
global createCount
metrics = MetricCache.counts()
t = time.time()
metrics.sort(key=lambda item: item[1],
reverse=True) # by queue size, descending
log.debug("Sorted %d cache queues in %.6f seconds" %
(len(metrics), time.time() - t))
for metric, queueSize in metrics:
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
dbFilePath = getFilesystemPath(metric)
dbFileExists = exists(dbFilePath)
if not dbFileExists:
createCount += 1
now = time.time()
if now - lastCreateInterval >= 60:
lastCreateInterval = now
createCount = 1
elif createCount >= settings.MAX_CREATES_PER_MINUTE:
# dropping queued up datapoints for new metrics prevents filling up the entire cache
# when a bunch of new metrics are received.
try:
MetricCache.pop(metric)
except KeyError:
pass
continue
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" %
metric)
continue # we simply move on to the next metric when this race condition occurs
yield (metric, datapoints, dbFilePath, dbFileExists)
def optimalWriteOrder():
"""Generates metrics with the most cached values first and applies a soft
rate limit on new metrics"""
global lastCreateInterval
global createCount
metrics = MetricCache.counts()
time_ = time.time()
metrics.sort(key=lambda item: item[1], reverse=True) # by queue size, descending
log.debug("Sorted %d cache queues in %.6f seconds" % (len(metrics),
time.time() - time_))
for metric, queueSize in metrics:
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
dbFileExists = APP_DB.exists(metric)
if not dbFileExists:
createCount += 1
now = time.time()
if now - lastCreateInterval >= 60:
lastCreateInterval = now
createCount = 1
elif createCount >= settings.MAX_CREATES_PER_MINUTE:
# dropping queued up datapoints for new metrics prevents filling up the entire cache
# when a bunch of new metrics are received.
try:
MetricCache.pop(metric)
except KeyError:
pass
continue
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" % metric)
continue # we simply move on to the next metric when this race condition occurs
yield (metric, datapoints, dbFileExists)
def pre_retrieve_metric_check(self, metric):
dbinfo = self.get_dbinfo(metric)
dbFilePath = dbInfo[0]
dbFileExists = dbInfo[1]
if not dbFileExists:
createCount += 1
now = time.time()
if now - lastCreateInterval >= 60:
lastCreateInterval = now
createCount = 1
elif createCount >= settings.MAX_CREATES_PER_MINUTE:
# dropping queued up datapoints for new metrics prevents filling up the entire cache
# when a bunch of new metrics are received.
try:
MetricCache.pop(metric)
except KeyError:
pass
return False
return True
def optimalWriteOrder():
log.msg("Entered optimalWriteOrder")
metrics = MetricCache.counts()
t = time.time()
metrics.sort(key=lambda item: item[1], reverse=True) # by queue size, descending
log.msg("Sorted %d cache queues in %.6f seconds" % (len(metrics), time.time() - t))
for metric, queueSize in metrics:
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
try: # metrics can momentarily disappear from the MetricCache due to the implementation of MetricCache.store()
datapoints = MetricCache.pop(metric)
except KeyError:
log.msg("MetricCache contention, skipping %s update for now" % metric)
continue # we simply move on to the next metric when this race condition occurs
yield (metric, datapoints)
def optimalWriteOrder():
"""Generates metrics with the most cached values first and applies a soft
rate limit on new metrics"""
while MetricCache:
(metric, datapoints) = MetricCache.pop()
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
dbFilePath = getFilesystemPath(metric)
dbFileExists = exists(dbFilePath)
if not dbFileExists and CREATE_BUCKET:
# If our tokenbucket has enough tokens available to create a new metric
# file then yield the metric data to complete that operation. Otherwise
# we'll just drop the metric on the ground and move on to the next
# metric.
# XXX This behavior should probably be configurable to no tdrop metrics
# when rate limitng unless our cache is too big or some other legit
# reason.
if CREATE_BUCKET.drain(1):
yield (metric, datapoints, dbFilePath, dbFileExists)
continue
yield (metric, datapoints, dbFilePath, dbFileExists)
def optimalWriteOrder():
"""Generates metrics with the most cached values first and applies a soft
rate limit on new metrics"""
while MetricCache:
(metric, datapoints) = MetricCache.pop()
if state.cacheTooFull and MetricCache.size < CACHE_SIZE_LOW_WATERMARK:
events.cacheSpaceAvailable()
dbFilePath = getFilesystemPath(metric)
dbFileExists = exists(dbFilePath)
if not dbFileExists and CREATE_BUCKET:
# If our tokenbucket has enough tokens available to create a new metric
# file then yield the metric data to complete that operation. Otherwise
# we'll just drop the metric on the ground and move on to the next
# metric.
# XXX This behavior should probably be configurable to no tdrop metrics
# when rate limitng unless our cache is too big or some other legit
# reason.
if CREATE_BUCKET.drain(1):
yield (metric, datapoints, dbFilePath, dbFileExists)
continue
yield (metric, datapoints, dbFilePath, dbFileExists)
