#
# SPDX-License-Identifier: AGPL-3.0
-from __future__ import print_function
-
import arvados
import collections
-import crunchstat_summary.chartjs
+import crunchstat_summary.dygraphs
import crunchstat_summary.reader
import datetime
import functools
# number of GiB. (Actual nodes tend to be sold in sizes like 8 GiB
# that have amounts like 7.5 GiB according to the kernel.)
AVAILABLE_RAM_RATIO = 0.95
-
+MB=2**20
# Workaround datetime.datetime.strptime() thread-safety bug by calling
# it once before starting threads. https://bugs.python.org/issue7980
datetime.datetime.strptime('1999-12-31_23:59:59', '%Y-%m-%d_%H:%M:%S')
+WEBCHART_CLASS = crunchstat_summary.dygraphs.DygraphsChart
+
+
class Task(object):
def __init__(self):
self.starttime = None
+ self.finishtime = None
self.series = collections.defaultdict(list)
class Summarizer(object):
- def __init__(self, logdata, label=None, skip_child_jobs=False):
+ def __init__(self, logdata, label=None, skip_child_jobs=False, uuid=None, **kwargs):
self._logdata = logdata
+ self.uuid = uuid
self.label = label
self.starttime = None
self.finishtime = None
def run(self):
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:
- seq = int(m.group('seq'))
- uuid = m.group('task_uuid')
- self.seq_to_uuid[seq] = uuid
- 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|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'))
- self.task_stats[task_id]['time'] = {'elapsed': elapsed}
- if elapsed > self.stats_max['time']['elapsed']:
- self.stats_max['time']['elapsed'] = elapsed
- continue
-
- m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) stderr Queued job (?P<uuid>\S+)$', line)
- if m:
- uuid = m.group('uuid')
- if self._skip_child_jobs:
- logger.warning('%s: omitting stats from child job %s'
- ' because --skip-child-jobs flag is on',
- self.label, uuid)
+ with self._logdata as logdata:
+ self._run(logdata)
+
+ def _run(self, logdata):
+ self.detected_crunch1 = False
+ for line in logdata:
+ if not self.detected_crunch1 and '-8i9sb-' in line:
+ self.detected_crunch1 = True
+
+ if self.detected_crunch1:
+ m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) job_task (?P<task_uuid>\S+)$', line)
+ if m:
+ seq = int(m.group('seq'))
+ uuid = m.group('task_uuid')
+ self.seq_to_uuid[seq] = uuid
+ logger.debug('%s: seq %d is task %s', self.label, seq, uuid)
continue
- logger.debug('%s: follow %s', self.label, uuid)
- child_summarizer = JobSummarizer(uuid)
- 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
- m = re.search(r'^(?P<timestamp>[^\s.]+)(\.\d+)? (?P<job_uuid>\S+) \d+ (?P<seq>\d+) stderr crunchstat: (?P<category>\S+) (?P<current>.*?)( -- interval (?P<interval>.*))?\n', line)
- if not m:
- continue
+ 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'))
+ self.task_stats[task_id]['time'] = {'elapsed': elapsed}
+ if elapsed > self.stats_max['time']['elapsed']:
+ self.stats_max['time']['elapsed'] = elapsed
+ continue
- try:
- if self.label is None:
- self.label = m.group('job_uuid')
- logger.debug('%s: using job uuid as label', self.label)
- if m.group('category').endswith(':'):
- # "stderr crunchstat: notice: ..."
+ m = re.search(r'^\S+ \S+ \d+ (?P<seq>\d+) stderr Queued job (?P<uuid>\S+)$', line)
+ if m:
+ uuid = m.group('uuid')
+ if self._skip_child_jobs:
+ logger.warning('%s: omitting stats from child job %s'
+ ' because --skip-child-jobs flag is on',
+ self.label, uuid)
+ continue
+ logger.debug('%s: follow %s', self.label, uuid)
+ child_summarizer = NewSummarizer(uuid)
+ child_summarizer.stats_max = self.stats_max
+ child_summarizer.task_stats = self.task_stats
+ child_summarizer.tasks = self.tasks
+ child_summarizer.starttime = self.starttime
+ child_summarizer.run()
+ logger.debug('%s: done %s', self.label, uuid)
continue
- elif m.group('category') in ('error', 'caught'):
+
+ # 2017-12-02_17:15:08 e51c5-8i9sb-mfp68stkxnqdd6m 63676 0 stderr crunchstat: keepcalls 0 put 2576 get -- interval 10.0000 seconds 0 put 2576 get
+ m = re.search(r'^(?P<timestamp>[^\s.]+)(\.\d+)? (?P<job_uuid>\S+) \d+ (?P<seq>\d+) stderr (?P<crunchstat>crunchstat: )(?P<category>\S+) (?P<current>.*?)( -- interval (?P<interval>.*))?\n$', line)
+ if not m:
continue
- elif m.group('category') == 'read':
- # "stderr crunchstat: read /proc/1234/net/dev: ..."
- # (crunchstat formatting fixed, but old logs still say this)
+ else:
+ # crunch2
+ # 2017-12-01T16:56:24.723509200Z crunchstat: keepcalls 0 put 3 get -- interval 10.0000 seconds 0 put 3 get
+ m = re.search(r'^(?P<timestamp>\S+) (?P<crunchstat>crunchstat: )?(?P<category>\S+) (?P<current>.*?)( -- interval (?P<interval>.*))?\n$', line)
+ if not m:
continue
+
+ if self.label is None:
+ try:
+ self.label = m.group('job_uuid')
+ except IndexError:
+ self.label = 'label #1'
+ category = m.group('category')
+ if category.endswith(':'):
+ # "stderr crunchstat: notice: ..."
+ continue
+ elif category in ('error', 'caught'):
+ continue
+ elif category in ('read', 'open', 'cgroup', 'CID', 'Running'):
+ # "stderr crunchstat: read /proc/1234/net/dev: ..."
+ # (old logs are less careful with unprefixed error messages)
+ continue
+
+ if self.detected_crunch1:
task_id = self.seq_to_uuid[int(m.group('seq'))]
- task = self.tasks[task_id]
+ else:
+ task_id = 'container'
+ task = self.tasks[task_id]
- # Use the first and last crunchstat timestamps as
- # approximations of starttime and finishtime.
+ # Use the first and last crunchstat timestamps as
+ # approximations of starttime and finishtime.
+ timestamp = m.group('timestamp')
+ if timestamp[10:11] == '_':
timestamp = datetime.datetime.strptime(
- m.group('timestamp'), '%Y-%m-%d_%H:%M:%S')
- if not task.starttime:
- task.starttime = timestamp
- logger.debug('%s: task %s starttime %s',
- self.label, task_id, timestamp)
+ timestamp, '%Y-%m-%d_%H:%M:%S')
+ elif timestamp[10:11] == 'T':
+ timestamp = datetime.datetime.strptime(
+ timestamp[:19], '%Y-%m-%dT%H:%M:%S')
+ else:
+ raise ValueError("Cannot parse timestamp {!r}".format(
+ timestamp))
+
+ if task.starttime is None:
+ logger.debug('%s: task %s starttime %s',
+ self.label, task_id, timestamp)
+ if task.starttime is None or timestamp < task.starttime:
+ task.starttime = timestamp
+ if task.finishtime is None or timestamp > task.finishtime:
task.finishtime = timestamp
- if not self.starttime:
- self.starttime = timestamp
+ if self.starttime is None or timestamp < self.starttime:
+ self.starttime = timestamp
+ if self.finishtime is None or timestamp > self.finishtime:
self.finishtime = timestamp
- this_interval_s = None
- for group in ['current', 'interval']:
- if not m.group(group):
- continue
- category = m.group('category')
- words = m.group(group).split(' ')
- stats = {}
+ if (not self.detected_crunch1) and task.starttime is not None and task.finishtime is not None:
+ elapsed = (task.finishtime - task.starttime).seconds
+ self.task_stats[task_id]['time'] = {'elapsed': elapsed}
+ if elapsed > self.stats_max['time']['elapsed']:
+ self.stats_max['time']['elapsed'] = elapsed
+
+ this_interval_s = None
+ for group in ['current', 'interval']:
+ if not m.group(group):
+ continue
+ category = m.group('category')
+ words = m.group(group).split(' ')
+ stats = {}
+ try:
for val, stat in zip(words[::2], words[1::2]):
- try:
- if '.' in val:
- stats[stat] = float(val)
- else:
- stats[stat] = int(val)
- except ValueError as e:
- raise ValueError(
- 'Error parsing {} stat: {!r}'.format(
- stat, e))
- if 'user' in stats or 'sys' in stats:
- stats['user+sys'] = stats.get('user', 0) + stats.get('sys', 0)
- if 'tx' in stats or 'rx' in stats:
- stats['tx+rx'] = stats.get('tx', 0) + stats.get('rx', 0)
- for stat, val in stats.iteritems():
- if group == 'interval':
- if stat == 'seconds':
- this_interval_s = val
- continue
- elif not (this_interval_s > 0):
- logger.error(
- "BUG? interval stat given with duration {!r}".
- format(this_interval_s))
- continue
- else:
- stat = stat + '__rate'
- val = val / this_interval_s
- if stat in ['user+sys__rate', 'tx+rx__rate']:
- task.series[category, stat].append(
- (timestamp - self.starttime, val))
+ if '.' in val:
+ stats[stat] = float(val)
else:
- if stat in ['rss']:
+ stats[stat] = int(val)
+ except ValueError as e:
+ # If the line doesn't start with 'crunchstat:' we
+ # might have mistaken an error message for a
+ # structured crunchstat line.
+ if m.group("crunchstat") is None or m.group("category") == "crunchstat":
+ logger.warning("%s: log contains message\n %s", self.label, line)
+ else:
+ logger.warning(
+ '%s: Error parsing value %r (stat %r, category %r): %r',
+ self.label, val, stat, category, e)
+ logger.warning('%s', line)
+ continue
+ if 'user' in stats or 'sys' in stats:
+ stats['user+sys'] = stats.get('user', 0) + stats.get('sys', 0)
+ if 'tx' in stats or 'rx' in stats:
+ stats['tx+rx'] = stats.get('tx', 0) + stats.get('rx', 0)
+ if group == 'interval':
+ if 'seconds' in stats:
+ this_interval_s = stats.get('seconds',0)
+ del stats['seconds']
+ if this_interval_s <= 0:
+ logger.error(
+ "BUG? interval stat given with duration {!r}".
+ format(this_interval_s))
+ else:
+ logger.error('BUG? interval stat missing duration')
+ for stat, val in stats.items():
+ if group == 'interval' and this_interval_s:
+ stat = stat + '__rate'
+ val = val / this_interval_s
+ if stat in ['user+sys__rate', 'user__rate', 'sys__rate', 'tx+rx__rate', 'rx__rate', 'tx__rate']:
task.series[category, stat].append(
(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
- except Exception as e:
- logger.info('Skipping malformed line: {}Error was: {}\n'.format(line, e))
+ else:
+ if stat in ['rss','used','total']:
+ task.series[category, stat].append(
+ (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
logger.debug('%s: done parsing', self.label)
self.job_tot = collections.defaultdict(
functools.partial(collections.defaultdict, int))
- for task_id, task_stat in self.task_stats.iteritems():
- for category, stat_last in task_stat.iteritems():
- for stat, val in stat_last.iteritems():
+ for task_id, task_stat in self.task_stats.items():
+ for category, stat_last in task_stat.items():
+ for stat, val in stat_last.items():
if stat in ['cpus', 'cache', 'swap', 'rss']:
# meaningless stats like 16 cpu cores x 5 tasks = 80
continue
def long_label(self):
label = self.label
+ if hasattr(self, 'process') and self.process['uuid'] not in label:
+ label = '{} ({})'.format(label, self.process['uuid'])
if self.finishtime:
label += ' -- elapsed time '
s = (self.finishtime - self.starttime).total_seconds()
self._recommend_gen())) + "\n"
def html_report(self):
- return crunchstat_summary.chartjs.ChartJS(self.label, [self]).html()
+ return WEBCHART_CLASS(self.label, [self]).html()
def _text_report_gen(self):
yield "\t".join(['category', 'metric', 'task_max', 'task_max_rate', 'job_total'])
- for category, stat_max in sorted(self.stats_max.iteritems()):
- for stat, val in sorted(stat_max.iteritems()):
+ for category, stat_max in sorted(self.stats_max.items()):
+ for stat, val in sorted(stat_max.items()):
if stat.endswith('__rate'):
continue
max_rate = self._format(stat_max.get(stat+'__rate', '-'))
self.stats_max['cpu']['user+sys__rate'],
lambda x: x * 100),
('Overall CPU usage: {}%',
- self.job_tot['cpu']['user+sys'] /
+ float(self.job_tot['cpu']['user+sys']) /
self.job_tot['time']['elapsed']
if self.job_tot['time']['elapsed'] > 0 else 0,
lambda x: x * 100),
(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)):
+ lambda x: x * 100.0),
+ ('Temp disk utilization {}%',
+ (float(self.job_tot['statfs']['used']) /
+ float(self.job_tot['statfs']['total']))
+ if self.job_tot['statfs']['total'] > 0 else 0,
+ lambda x: x * 100.0),
+ ):
format_string, val, transform = args
if val == float('-Inf'):
continue
yield "# "+format_string.format(self._format(val))
def _recommend_gen(self):
+ # TODO recommend fixing job granularity if elapsed time is too short
return itertools.chain(
self._recommend_cpu(),
self._recommend_ram(),
- self._recommend_keep_cache())
+ self._recommend_keep_cache(),
+ self._recommend_temp_disk(),
+ )
def _recommend_cpu(self):
"""Recommend asking for 4 cores if max CPU usage was 333%"""
+ constraint_key = self._map_runtime_constraint('vcpus')
cpu_max_rate = self.stats_max['cpu']['user+sys__rate']
- if cpu_max_rate == float('-Inf'):
+ if cpu_max_rate == float('-Inf') or cpu_max_rate == 0.0:
logger.warning('%s: no CPU usage data', self.label)
return
+ # TODO Don't necessarily want to recommend on isolated max peak
+ # take average CPU usage into account as well or % time at max
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:
+ asked_cores = self.existing_constraints.get(constraint_key)
+ if asked_cores is None:
+ asked_cores = 1
+ # TODO: This should be more nuanced in cases where max >> avg
+ if used_cores < asked_cores:
yield (
'#!! {} max CPU usage was {}% -- '
- 'try runtime_constraints "min_cores_per_node":{}'
+ 'try reducing runtime_constraints to "{}":{}'
).format(
self.label,
- int(math.ceil(cpu_max_rate*100)),
+ math.ceil(cpu_max_rate*100),
+ constraint_key,
int(used_cores))
+ # FIXME: This needs to be updated to account for current a-d-c algorithms
def _recommend_ram(self):
"""Recommend an economical RAM constraint for this job.
the memory we want -- even if that happens to be 8192 MiB.
"""
+ constraint_key = self._map_runtime_constraint('ram')
used_bytes = self.stats_max['mem']['rss']
if used_bytes == float('-Inf'):
logger.warning('%s: no memory usage data', self.label)
return
- used_mib = math.ceil(float(used_bytes) / 1048576)
- asked_mib = self.existing_constraints.get('min_ram_mb_per_node')
+ used_mib = math.ceil(float(used_bytes) / MB)
+ asked_mib = self.existing_constraints.get(constraint_key)
nearlygibs = lambda mebibytes: mebibytes/AVAILABLE_RAM_RATIO/1024
- if asked_mib is None or (
- math.ceil(nearlygibs(used_mib)) < nearlygibs(asked_mib)):
+ if used_mib > 0 and (asked_mib is None or (
+ math.ceil(nearlygibs(used_mib)) < nearlygibs(asked_mib))):
yield (
'#!! {} max RSS was {} MiB -- '
- 'try runtime_constraints "min_ram_mb_per_node":{}'
+ 'try reducing runtime_constraints to "{}":{}'
).format(
self.label,
int(used_mib),
- int(math.ceil(nearlygibs(used_mib))*AVAILABLE_RAM_RATIO*1024))
+ constraint_key,
+ int(math.ceil(nearlygibs(used_mib))*AVAILABLE_RAM_RATIO*1024*(MB)/self._runtime_constraint_mem_unit()))
def _recommend_keep_cache(self):
"""Recommend increasing keep cache if utilization < 80%"""
+ constraint_key = self._map_runtime_constraint('keep_cache_ram')
if self.job_tot['net:keep0']['rx'] == 0:
return
utilization = (float(self.job_tot['blkio:0:0']['read']) /
float(self.job_tot['net:keep0']['rx']))
- asked_mib = self.existing_constraints.get('keep_cache_mb_per_task', 256)
+ # FIXME: the default on this get won't work correctly
+ asked_cache = self.existing_constraints.get(constraint_key, 256) * self._runtime_constraint_mem_unit()
if utilization < 0.8:
yield (
'#!! {} Keep cache utilization was {:.2f}% -- '
- 'try runtime_constraints "keep_cache_mb_per_task":{} (or more)'
+ 'try doubling runtime_constraints to "{}":{} (or more)'
).format(
self.label,
utilization * 100.0,
- asked_mib*2)
+ constraint_key,
+ math.ceil(asked_cache * 2 / self._runtime_constraint_mem_unit()))
+
+
+ def _recommend_temp_disk(self):
+ """Recommend decreasing temp disk if utilization < 50%"""
+ total = float(self.job_tot['statfs']['total'])
+ utilization = (float(self.job_tot['statfs']['used']) / total) if total > 0 else 0.0
+
+ if utilization < 50.8 and total > 0:
+ yield (
+ '#!! {} max temp disk utilization was {:.0f}% of {:.0f} MiB -- '
+ 'consider reducing "tmpdirMin" and/or "outdirMin"'
+ ).format(
+ self.label,
+ utilization * 100.0,
+ total / MB)
def _format(self, val):
else:
return '{}'.format(val)
+ def _runtime_constraint_mem_unit(self):
+ if hasattr(self, 'runtime_constraint_mem_unit'):
+ return self.runtime_constraint_mem_unit
+ elif self.detected_crunch1:
+ return JobSummarizer.runtime_constraint_mem_unit
+ else:
+ return ContainerRequestSummarizer.runtime_constraint_mem_unit
+
+ def _map_runtime_constraint(self, key):
+ if hasattr(self, 'map_runtime_constraint'):
+ return self.map_runtime_constraint[key]
+ elif self.detected_crunch1:
+ return JobSummarizer.map_runtime_constraint[key]
+ else:
+ return key
+
class CollectionSummarizer(Summarizer):
def __init__(self, collection_id, **kwargs):
self.label = collection_id
-class JobSummarizer(Summarizer):
- def __init__(self, job, **kwargs):
- arv = arvados.api('v1')
- if isinstance(job, basestring):
- self.job = arv.jobs().get(uuid=job).execute()
- else:
- self.job = job
+def NewSummarizer(process_or_uuid, **kwargs):
+ """Construct with the appropriate subclass for this uuid/object."""
+
+ if isinstance(process_or_uuid, dict):
+ process = process_or_uuid
+ uuid = process['uuid']
+ else:
+ uuid = process_or_uuid
+ process = None
+ arv = arvados.api('v1', model=OrderedJsonModel())
+
+ if '-dz642-' in uuid:
+ if process is None:
+ # Get the associated CR. Doesn't matter which since they all have the same logs
+ crs = arv.container_requests().list(filters=[['container_uuid','=',uuid]],limit=1).execute()['items']
+ if len(crs) > 0:
+ process = crs[0]
+ klass = ContainerRequestTreeSummarizer
+ elif '-xvhdp-' in uuid:
+ if process is None:
+ process = arv.container_requests().get(uuid=uuid).execute()
+ klass = ContainerRequestTreeSummarizer
+ elif '-8i9sb-' in uuid:
+ if process is None:
+ process = arv.jobs().get(uuid=uuid).execute()
+ klass = JobTreeSummarizer
+ elif '-d1hrv-' in uuid:
+ if process is None:
+ process = arv.pipeline_instances().get(uuid=uuid).execute()
+ klass = PipelineSummarizer
+ elif '-4zz18-' in uuid:
+ return CollectionSummarizer(collection_id=uuid)
+ else:
+ raise ArgumentError("Unrecognized uuid %s", uuid)
+ return klass(process, uuid=uuid, **kwargs)
+
+
+class ProcessSummarizer(Summarizer):
+ """Process is a job, pipeline, or container request."""
+
+ def __init__(self, process, label=None, **kwargs):
rdr = None
- if self.job.get('log'):
+ self.process = process
+ if label is None:
+ label = self.process.get('name', self.process['uuid'])
+ # Pre-Arvados v1.4 everything is in 'log'
+ # For 1.4+ containers have no logs and container_requests have them in 'log_uuid', not 'log'
+ log_collection = self.process.get('log', self.process.get('log_uuid'))
+ if log_collection and self.process.get('state') != 'Uncommitted': # arvados.util.CR_UNCOMMITTED:
try:
- rdr = crunchstat_summary.reader.CollectionReader(self.job['log'])
+ rdr = crunchstat_summary.reader.CollectionReader(log_collection)
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']
+ "log collection %s: %s", self.process['log'], e)
if rdr is None:
- rdr = crunchstat_summary.reader.LiveLogReader(self.job['uuid'])
- label = self.job['uuid'] + ' (partial)'
- super(JobSummarizer, self).__init__(rdr, **kwargs)
- self.label = label
- self.existing_constraints = self.job.get('runtime_constraints', {})
+ uuid = self.process.get('container_uuid', self.process.get('uuid'))
+ rdr = crunchstat_summary.reader.LiveLogReader(uuid)
+ label = label + ' (partial)'
+ super(ProcessSummarizer, self).__init__(rdr, label=label, **kwargs)
+ self.existing_constraints = self.process.get('runtime_constraints', {})
-class PipelineSummarizer(object):
- def __init__(self, pipeline_instance_uuid, **kwargs):
- arv = arvados.api('v1', model=OrderedJsonModel())
- instance = arv.pipeline_instances().get(
- uuid=pipeline_instance_uuid).execute()
- self.summarizers = collections.OrderedDict()
- for cname, component in instance['components'].iteritems():
- if 'job' not in component:
- logger.warning(
- "%s: skipping component with no job assigned", cname)
- else:
- logger.info(
- "%s: job %s", cname, component['job']['uuid'])
- summarizer = JobSummarizer(component['job'], **kwargs)
- summarizer.label = '{} {}'.format(
- cname, component['job']['uuid'])
- self.summarizers[cname] = summarizer
- self.label = pipeline_instance_uuid
+class JobSummarizer(ProcessSummarizer):
+ runtime_constraint_mem_unit = MB
+ map_runtime_constraint = {
+ 'keep_cache_ram': 'keep_cache_mb_per_task',
+ 'ram': 'min_ram_mb_per_node',
+ 'vcpus': 'min_cores_per_node',
+ }
+
+
+class ContainerRequestSummarizer(ProcessSummarizer):
+ runtime_constraint_mem_unit = 1
+
+
+class MultiSummarizer(object):
+ def __init__(self, children={}, label=None, threads=1, **kwargs):
+ self.throttle = threading.Semaphore(threads)
+ self.children = children
+ self.label = label
+
+ def run_and_release(self, target, *args, **kwargs):
+ try:
+ return target(*args, **kwargs)
+ finally:
+ self.throttle.release()
def run(self):
threads = []
- for summarizer in self.summarizers.itervalues():
- t = threading.Thread(target=summarizer.run)
+ for child in self.children.values():
+ self.throttle.acquire()
+ t = threading.Thread(target=self.run_and_release, args=(child.run, ))
t.daemon = True
t.start()
threads.append(t)
def text_report(self):
txt = ''
- for cname, summarizer in self.summarizers.iteritems():
- txt += '### Summary for {} ({})\n'.format(
- cname, summarizer.job['uuid'])
- txt += summarizer.text_report()
+ d = self._descendants()
+ for child in d.values():
+ if len(d) > 1:
+ txt += '### Summary for {} ({})\n'.format(
+ child.label, child.process['uuid'])
+ txt += child.text_report()
txt += '\n'
return txt
+ def _descendants(self):
+ """Dict of self and all descendants.
+
+ Nodes with nothing of their own to report (like
+ MultiSummarizers) are omitted.
+ """
+ d = collections.OrderedDict()
+ for key, child in self.children.items():
+ if isinstance(child, Summarizer):
+ d[key] = child
+ if isinstance(child, MultiSummarizer):
+ d.update(child._descendants())
+ return d
+
def html_report(self):
- return crunchstat_summary.chartjs.ChartJS(
- self.label, self.summarizers.itervalues()).html()
+ return WEBCHART_CLASS(self.label, iter(self._descendants().values())).html()
+
+
+class JobTreeSummarizer(MultiSummarizer):
+ """Summarizes a job and all children listed in its components field."""
+ def __init__(self, job, label=None, **kwargs):
+ arv = arvados.api('v1', model=OrderedJsonModel())
+ label = label or job.get('name', job['uuid'])
+ children = collections.OrderedDict()
+ children[job['uuid']] = JobSummarizer(job, label=label, **kwargs)
+ if job.get('components', None):
+ preloaded = {}
+ for j in arv.jobs().index(
+ limit=len(job['components']),
+ filters=[['uuid','in',list(job['components'].values())]]).execute()['items']:
+ preloaded[j['uuid']] = j
+ for cname in sorted(job['components'].keys()):
+ child_uuid = job['components'][cname]
+ j = (preloaded.get(child_uuid) or
+ arv.jobs().get(uuid=child_uuid).execute())
+ children[child_uuid] = JobTreeSummarizer(job=j, label=cname, **kwargs)
+
+ super(JobTreeSummarizer, self).__init__(
+ children=children,
+ label=label,
+ **kwargs)
+
+
+class PipelineSummarizer(MultiSummarizer):
+ def __init__(self, instance, **kwargs):
+ children = collections.OrderedDict()
+ for cname, component in instance['components'].items():
+ if 'job' not in component:
+ logger.warning(
+ "%s: skipping component with no job assigned", cname)
+ else:
+ logger.info(
+ "%s: job %s", cname, component['job']['uuid'])
+ summarizer = JobTreeSummarizer(component['job'], label=cname, **kwargs)
+ summarizer.label = '{} {}'.format(
+ cname, component['job']['uuid'])
+ children[cname] = summarizer
+ super(PipelineSummarizer, self).__init__(
+ children=children,
+ label=instance['uuid'],
+ **kwargs)
+
+
+class ContainerRequestTreeSummarizer(MultiSummarizer):
+ def __init__(self, root, skip_child_jobs=False, **kwargs):
+ arv = arvados.api('v1', model=OrderedJsonModel())
+
+ label = kwargs.pop('label', None) or root.get('name') or root['uuid']
+ root['name'] = label
+
+ children = collections.OrderedDict()
+ todo = collections.deque((root, ))
+ while len(todo) > 0:
+ current = todo.popleft()
+ label = current['name']
+ sort_key = current['created_at']
+
+ summer = ContainerRequestSummarizer(current, label=label, **kwargs)
+ summer.sort_key = sort_key
+ children[current['uuid']] = summer
+
+ page_filters = []
+ while True:
+ child_crs = arv.container_requests().index(
+ order=['uuid asc'],
+ filters=page_filters+[
+ ['requesting_container_uuid', '=', current['container_uuid']]],
+ ).execute()
+ if not child_crs['items']:
+ break
+ elif skip_child_jobs:
+ logger.warning('%s: omitting stats from %d child containers'
+ ' because --skip-child-jobs flag is on',
+ label, child_crs['items_available'])
+ break
+ page_filters = [['uuid', '>', child_crs['items'][-1]['uuid']]]
+ for cr in child_crs['items']:
+ if cr['container_uuid']:
+ logger.debug('%s: container req %s', current['uuid'], cr['uuid'])
+ cr['name'] = cr.get('name') or cr['uuid']
+ todo.append(cr)
+ sorted_children = collections.OrderedDict()
+ for uuid in sorted(list(children.keys()), key=lambda uuid: children[uuid].sort_key):
+ sorted_children[uuid] = children[uuid]
+ super(ContainerRequestTreeSummarizer, self).__init__(
+ children=sorted_children,
+ label=root['name'],
+ **kwargs)
projects / arvados.git / blobdiff