X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/0f8dc3d824f03b82b8db9f61bbcc0592b62b998f..refs/heads/10472-crunchstat-summary-job-components:/tools/crunchstat-summary/crunchstat_summary/summarizer.py diff --git a/tools/crunchstat-summary/crunchstat_summary/summarizer.py b/tools/crunchstat-summary/crunchstat_summary/summarizer.py index 2ac12abcba..f1c97c282c 100644 --- a/tools/crunchstat-summary/crunchstat_summary/summarizer.py +++ b/tools/crunchstat-summary/crunchstat_summary/summarizer.py @@ -11,6 +11,7 @@ import math import re import sys import threading +import _strptime from arvados.api import OrderedJsonModel from crunchstat_summary import logger @@ -59,6 +60,21 @@ class Summarizer(object): logger.debug("%s: logdata %s", self.label, logdata) + def run_child(self, 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) + return + 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) + def run(self): logger.debug("%s: parsing logdata %s", self.label, self._logdata) for line in self._logdata: @@ -79,102 +95,99 @@ class Summarizer(object): self.stats_max['time']['elapsed'] = elapsed continue + # Old style job logs only - newer style uses job['components'] + uuid = None m = re.search(r'^\S+ \S+ \d+ (?P\d+) stderr Queued job (?P\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 = 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) + self.run_child(m.group('uuid')) continue m = re.search(r'^(?P[^\s.]+)(\.\d+)? (?P\S+) \d+ (?P\d+) stderr crunchstat: (?P\S+) (?P.*?)( -- interval (?P.*))?\n', line) if not m: continue - 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: ..." - continue - elif m.group('category') in ('error', 'caught'): - continue - elif m.group('category') == 'read': - # "stderr crunchstat: read /proc/1234/net/dev: ..." - # (crunchstat formatting fixed, but old logs still say this) - continue - task_id = self.seq_to_uuid[int(m.group('seq'))] - task = self.tasks[task_id] - - # Use the first and last crunchstat timestamps as - # approximations of starttime and finishtime. - 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) - task.finishtime = timestamp - - if not self.starttime: - self.starttime = timestamp - self.finishtime = timestamp - - this_interval_s = None - for group in ['current', 'interval']: - if not m.group(group): + 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: ..." continue - category = m.group('category') - words = m.group(group).split(' ') - stats = {} - 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 in "{}": {!r}'.format( - stat, line, 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 + elif m.group('category') in ('error', 'caught'): + continue + elif m.group('category') == 'read': + # "stderr crunchstat: read /proc/1234/net/dev: ..." + # (crunchstat formatting fixed, but old logs still say this) + continue + task_id = self.seq_to_uuid[int(m.group('seq'))] + task = self.tasks[task_id] + + # Use the first and last crunchstat timestamps as + # approximations of starttime and finishtime. + 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) + task.finishtime = timestamp + + if not self.starttime: + self.starttime = timestamp + 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 = {} + 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)) else: - stat = stat + '__rate' - val = val / this_interval_s - if stat in ['user+sys__rate', 'tx+rx__rate']: + if stat in ['rss']: task.series[category, stat].append( (timestamp - self.starttime, val)) - else: - if stat in ['rss']: - 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.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)) + logger.debug('%s: done parsing log', self.label) + + # Enabling this will roll up stats for all subjobs into the parent job + if False and 'components' in self.job: + for cname, component in self.job['components'].iteritems(): + self.run_child(component) self.job_tot = collections.defaultdict( functools.partial(collections.defaultdict, int)) @@ -401,8 +414,8 @@ class JobSummarizer(Summarizer): class PipelineSummarizer(object): def __init__(self, pipeline_instance_uuid, **kwargs): - arv = arvados.api('v1', model=OrderedJsonModel()) - instance = arv.pipeline_instances().get( + self.arv = arvados.api('v1', model=OrderedJsonModel()) + instance = self.arv.pipeline_instances().get( uuid=pipeline_instance_uuid).execute() self.summarizers = collections.OrderedDict() for cname, component in instance['components'].iteritems(): @@ -410,14 +423,20 @@ class PipelineSummarizer(object): 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.summarize_job(cname, component['job'], **kwargs) self.label = pipeline_instance_uuid + def summarize_job(self, cname, job, **kwargs): + uuid = job['uuid'] + logger.info("%s: job %s", cname, uuid) + summarizer = JobSummarizer(job, **kwargs) + summarizer.label = '{} {}'.format(cname, uuid) + self.summarizers[cname] = summarizer + if 'components' in job: + for cname, uuid in job['components'].iteritems(): + subjob = self.arv.jobs().get(uuid=uuid).execute() + self.summarize_job(cname, subjob, **kwargs) + def run(self): threads = [] for summarizer in self.summarizers.itervalues():