X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/9a7e2a24f5f3d261e554ac3815b7e2a4c2e24503..a7631a1ccb6e2a6925d00a06562e171c4ce4ea2f:/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 bf905a3946..463c552c4f 100644 --- a/tools/crunchstat-summary/crunchstat_summary/summarizer.py +++ b/tools/crunchstat-summary/crunchstat_summary/summarizer.py @@ -105,7 +105,7 @@ class Summarizer(object): self.label, uuid) continue logger.debug('%s: follow %s', self.label, uuid) - child_summarizer = ProcessSummarizer(uuid) + child_summarizer = NewSummarizer(uuid) child_summarizer.stats_max = self.stats_max child_summarizer.task_stats = self.task_stats child_summarizer.tasks = self.tasks @@ -129,13 +129,14 @@ class Summarizer(object): try: self.label = m.group('job_uuid') except IndexError: - self.label = 'container' - if m.group('category').endswith(':'): + self.label = 'label #1' + category = m.group('category') + if category.endswith(':'): # "stderr crunchstat: notice: ..." continue - elif m.group('category') in ('error', 'caught'): + elif category in ('error', 'caught'): continue - elif m.group('category') in ('read', 'open', 'cgroup', 'CID', 'Running'): + 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 @@ -167,9 +168,9 @@ class Summarizer(object): if task.finishtime is None or timestamp > task.finishtime: task.finishtime = timestamp - if self.starttime is None or timestamp < task.starttime: + if self.starttime is None or timestamp < self.starttime: self.starttime = timestamp - if self.finishtime is None or timestamp < task.finishtime: + if self.finishtime is None or timestamp > self.finishtime: self.finishtime = timestamp if (not self.detected_crunch1) and task.starttime is not None and task.finishtime is not None: @@ -207,24 +208,25 @@ class Summarizer(object): 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.items(): - if group == 'interval': - if stat == 'seconds': - this_interval_s = val - continue - elif not (this_interval_s > 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)) - continue - else: + 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', 'tx+rx__rate']: + 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)) else: - if stat in ['rss']: + if stat in ['rss','used','total']: task.series[category, stat].append( (timestamp - self.starttime, val)) self.task_stats[task_id][category][stat] = val @@ -314,7 +316,13 @@ class Summarizer(object): (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 @@ -327,7 +335,9 @@ class Summarizer(object): 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%""" @@ -354,7 +364,7 @@ class Summarizer(object): constraint_key, int(used_cores)) - # FIXME: This needs to be updated to account for current nodemanager algorithms + # 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. @@ -430,6 +440,21 @@ class Summarizer(object): 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): """Return a string representation of a stat. @@ -445,7 +470,7 @@ class Summarizer(object): elif self.detected_crunch1: return JobSummarizer.runtime_constraint_mem_unit else: - return ContainerSummarizer.runtime_constraint_mem_unit + return ContainerRequestSummarizer.runtime_constraint_mem_unit def _map_runtime_constraint(self, key): if hasattr(self, 'map_runtime_constraint'): @@ -476,12 +501,15 @@ def NewSummarizer(process_or_uuid, **kwargs): if '-dz642-' in uuid: if process is None: - process = arv.containers().get(uuid=uuid).execute() - klass = ContainerTreeSummarizer + # 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 = ContainerTreeSummarizer + klass = ContainerRequestTreeSummarizer elif '-8i9sb-' in uuid: if process is None: process = arv.jobs().get(uuid=uuid).execute() @@ -498,21 +526,25 @@ def NewSummarizer(process_or_uuid, **kwargs): class ProcessSummarizer(Summarizer): - """Process is a job, pipeline, container, or container request.""" + """Process is a job, pipeline, or container request.""" def __init__(self, process, label=None, **kwargs): rdr = None self.process = process if label is None: label = self.process.get('name', self.process['uuid']) - if self.process.get('log'): + # 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.process['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.process['log'], e) if rdr is None: - rdr = crunchstat_summary.reader.LiveLogReader(self.process['uuid']) + 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', {}) @@ -527,7 +559,7 @@ class JobSummarizer(ProcessSummarizer): } -class ContainerSummarizer(ProcessSummarizer): +class ContainerRequestSummarizer(ProcessSummarizer): runtime_constraint_mem_unit = 1 @@ -628,7 +660,7 @@ class PipelineSummarizer(MultiSummarizer): **kwargs) -class ContainerTreeSummarizer(MultiSummarizer): +class ContainerRequestTreeSummarizer(MultiSummarizer): def __init__(self, root, skip_child_jobs=False, **kwargs): arv = arvados.api('v1', model=OrderedJsonModel()) @@ -641,10 +673,8 @@ class ContainerTreeSummarizer(MultiSummarizer): current = todo.popleft() label = current['name'] sort_key = current['created_at'] - if current['uuid'].find('-xvhdp-') > 0: - current = arv.containers().get(uuid=current['container_uuid']).execute() - summer = ContainerSummarizer(current, label=label, **kwargs) + summer = ContainerRequestSummarizer(current, label=label, **kwargs) summer.sort_key = sort_key children[current['uuid']] = summer @@ -653,7 +683,7 @@ class ContainerTreeSummarizer(MultiSummarizer): child_crs = arv.container_requests().index( order=['uuid asc'], filters=page_filters+[ - ['requesting_container_uuid', '=', current['uuid']]], + ['requesting_container_uuid', '=', current['container_uuid']]], ).execute() if not child_crs['items']: break @@ -671,7 +701,7 @@ class ContainerTreeSummarizer(MultiSummarizer): sorted_children = collections.OrderedDict() for uuid in sorted(list(children.keys()), key=lambda uuid: children[uuid].sort_key): sorted_children[uuid] = children[uuid] - super(ContainerTreeSummarizer, self).__init__( + super(ContainerRequestTreeSummarizer, self).__init__( children=sorted_children, label=root['name'], **kwargs)