"collection {} has {} files; need exactly one".format(
collection_id, len(filenames)))
self._reader = collection.open(filenames[0])
+ self._label = "{}/{}".format(collection_id, filenames[0])
+
+ def __str__(self):
+ return self._label
def __iter__(self):
return iter(self._reader)
def __init__(self, job_uuid):
logger.debug('load stderr events for job %s', job_uuid)
- self._filters = [
- ['object_uuid', '=', job_uuid],
- ['event_type', '=', 'stderr']]
- self._label = job_uuid
+ self.job_uuid = job_uuid
+
+ def __str__(self):
+ return self.job_uuid
def _get_all_pages(self):
got = 0
last_id = 0
- while True:
- page = arvados.api().logs().index(
- limit=1000,
- order=['id asc'],
- filters=self._filters + [['id','>',str(last_id)]],
- ).execute(num_retries=2)
- got += len(page['items'])
- logger.debug(
- '%s: received %d of %d log events',
- self._label, got,
- got + page['items_available'] - len(page['items']))
- for i in page['items']:
- for line in i['properties']['text'].split('\n'):
- self._queue.put(line+'\n')
- last_id = i['id']
- if (len(page['items']) == 0 or
- len(page['items']) >= page['items_available']):
- break
- self._queue.put(self.EOF)
+ filters = [
+ ['object_uuid', '=', self.job_uuid],
+ ['event_type', '=', 'stderr']]
+ try:
+ while True:
+ page = arvados.api().logs().index(
+ limit=1000,
+ order=['id asc'],
+ filters=filters + [['id','>',str(last_id)]],
+ select=['id', 'properties'],
+ ).execute(num_retries=2)
+ got += len(page['items'])
+ logger.debug(
+ '%s: received %d of %d log events',
+ self.job_uuid, got,
+ got + page['items_available'] - len(page['items']))
+ for i in page['items']:
+ for line in i['properties']['text'].split('\n'):
+ self._queue.put(line+'\n')
+ last_id = i['id']
+ if (len(page['items']) == 0 or
+ len(page['items']) >= page['items_available']):
+ break
+ finally:
+ self._queue.put(self.EOF)
def __iter__(self):
self._queue = Queue.Queue()
def next(self):
line = self._queue.get()
if line is self.EOF:
+ self._thread.join()
raise StopIteration
return line
import math
import re
import sys
+import threading
from arvados.api import OrderedJsonModel
from crunchstat_summary import logger
# stats_max: {category: {stat: val}}
self.stats_max = collections.defaultdict(
- functools.partial(collections.defaultdict,
- lambda: float('-Inf')))
+ functools.partial(collections.defaultdict, lambda: 0))
# task_stats: {task_id: {category: {stat: val}}}
self.task_stats = collections.defaultdict(
functools.partial(collections.defaultdict, dict))
# constructor will overwrite this with something useful.
self.existing_constraints = {}
- logger.debug("%s: logdata %s", self.label, repr(logdata))
+ logger.debug("%s: logdata %s", self.label, logdata)
def run(self):
- logger.debug("%s: parsing log data", self.label)
+ logger.debug("%s: parsing logdata %s", self.label, self._logdata)
for line in self._logdata:
m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) job_task (?P<task_uuid>\S+)$', line)
if m:
logger.debug('%s: seq %d is task %s', self.label, seq, uuid)
continue
- m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) success in (?P<elapsed>\d+) seconds', line)
+ m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) (success in|failure \(#., permanent\) after) (?P<elapsed>\d+) seconds', line)
if m:
task_id = self.seq_to_uuid[int(m.group('seq'))]
elapsed = int(m.group('elapsed'))
child_summarizer.stats_max = self.stats_max
child_summarizer.task_stats = self.task_stats
child_summarizer.tasks = self.tasks
+ child_summarizer.starttime = self.starttime
child_summarizer.run()
logger.debug('%s: done %s', self.label, uuid)
continue
val = val / this_interval_s
if stat in ['user+sys__rate', 'tx+rx__rate']:
task.series[category, stat].append(
- (timestamp - task.starttime, val))
+ (timestamp - self.starttime, val))
else:
if stat in ['rss']:
task.series[category, stat].append(
- (timestamp - task.starttime, val))
+ (timestamp - self.starttime, val))
self.task_stats[task_id][category][stat] = val
if val > self.stats_max[category][stat]:
self.stats_max[category][stat] = val
return label
def text_report(self):
+ if not self.tasks:
+ return "(no report generated)\n"
return "\n".join(itertools.chain(
self._text_report_gen(),
self._recommend_gen())) + "\n"
lambda x: x * 100),
('Overall CPU usage: {}%',
self.job_tot['cpu']['user+sys'] /
- self.job_tot['time']['elapsed'],
+ self.job_tot['time']['elapsed']
+ if self.job_tot['time']['elapsed'] > 0 else 0,
lambda x: x * 100),
('Max memory used by a single task: {}GB',
self.stats_max['mem']['rss'],
lambda x: x / 1e9),
('Max network traffic in a single task: {}GB',
- self.stats_max['net:eth0']['tx+rx'],
+ self.stats_max['net:eth0']['tx+rx'] +
+ self.stats_max['net:keep0']['tx+rx'],
lambda x: x / 1e9),
('Max network speed in a single interval: {}MB/s',
- self.stats_max['net:eth0']['tx+rx__rate'],
- lambda x: x / 1e6)):
+ self.stats_max['net:eth0']['tx+rx__rate'] +
+ self.stats_max['net:keep0']['tx+rx__rate'],
+ lambda x: x / 1e6),
+ ('Keep cache miss rate {}%',
+ (float(self.job_tot['keepcache']['miss']) /
+ float(self.job_tot['keepcalls']['get']))
+ if self.job_tot['keepcalls']['get'] > 0 else 0,
+ lambda x: x * 100.0),
+ ('Keep cache utilization {}%',
+ (float(self.job_tot['blkio:0:0']['read']) /
+ float(self.job_tot['net:keep0']['rx']))
+ if self.job_tot['net:keep0']['rx'] > 0 else 0,
+ lambda x: x * 100.0)):
format_string, val, transform = args
if val == float('-Inf'):
continue
def _recommend_gen(self):
return itertools.chain(
self._recommend_cpu(),
- self._recommend_ram())
+ self._recommend_ram(),
+ self._recommend_keep_cache())
def _recommend_cpu(self):
"""Recommend asking for 4 cores if max CPU usage was 333%"""
if cpu_max_rate == float('-Inf'):
logger.warning('%s: no CPU usage data', self.label)
return
- used_cores = int(math.ceil(cpu_max_rate))
+ used_cores = max(1, int(math.ceil(cpu_max_rate)))
asked_cores = self.existing_constraints.get('min_cores_per_node')
if asked_cores is None or used_cores < asked_cores:
yield (
int(used_mib),
int(math.ceil(nearlygibs(used_mib))*AVAILABLE_RAM_RATIO*1024))
+ def _recommend_keep_cache(self):
+ """Recommend increasing keep cache if miss rate is above 0.2%"""
+ if self.job_tot['keepcalls']['get'] == 0:
+ return
+ miss_rate = float(self.job_tot['keepcache']['miss']) / float(self.job_tot['keepcalls']['get']) * 100.0
+ asked_mib = self.existing_constraints.get('keep_cache_mb_per_task', 256)
+
+ if miss_rate > 0.2:
+ yield (
+ '#!! {} Keep cache miss rate was {:.2f}% -- '
+ 'try runtime_constraints "keep_cache_mb_per_task":{}'
+ ).format(
+ self.label,
+ miss_rate,
+ asked_mib*2)
+
+
def _format(self, val):
"""Return a string representation of a stat.
self.job = arv.jobs().get(uuid=job).execute()
else:
self.job = job
- if self.job['log']:
- rdr = crunchstat_summary.reader.CollectionReader(self.job['log'])
- label = self.job['uuid']
- else:
+ rdr = None
+ if self.job.get('log'):
+ try:
+ rdr = crunchstat_summary.reader.CollectionReader(self.job['log'])
+ except arvados.errors.NotFoundError as e:
+ logger.warning("Trying event logs after failing to read "
+ "log collection %s: %s", self.job['log'], e)
+ else:
+ label = self.job['uuid']
+ if rdr is None:
rdr = crunchstat_summary.reader.LiveLogReader(self.job['uuid'])
label = self.job['uuid'] + ' (partial)'
super(JobSummarizer, self).__init__(rdr, **kwargs)
if 'job' not in component:
logger.warning(
"%s: skipping component with no job assigned", cname)
- elif component['job'].get('log') is None:
- logger.warning(
- "%s: skipping job %s with no log available",
- cname, component['job'].get('uuid'))
else:
logger.info(
- "%s: logdata %s", cname, component['job']['log'])
+ "%s: job %s", cname, component['job']['uuid'])
summarizer = JobSummarizer(component['job'], **kwargs)
- summarizer.label = cname
+ summarizer.label = '{} {}'.format(
+ cname, component['job']['uuid'])
self.summarizers[cname] = summarizer
self.label = pipeline_instance_uuid
def run(self):
+ threads = []
for summarizer in self.summarizers.itervalues():
- summarizer.run()
+ t = threading.Thread(target=summarizer.run)
+ t.daemon = True
+ t.start()
+ threads.append(t)
+ for t in threads:
+ t.join()
def text_report(self):
txt = ''
projects / arvados.git / commitdiff