From c5489cb1052e68d1d0db71224ed09f1499ee776f Mon Sep 17 00:00:00 2001 From: Peter Amstutz Date: Thu, 22 Feb 2024 11:30:53 -0500 Subject: [PATCH] 19744: Don't warn about missing metrics when the elapsed time is short Arvados-DCO-1.1-Signed-off-by: Peter Amstutz --- .../crunchstat_summary/summarizer.py | 43 ++++++++++--------- 1 file changed, 23 insertions(+), 20 deletions(-) diff --git a/tools/crunchstat-summary/crunchstat_summary/summarizer.py b/tools/crunchstat-summary/crunchstat_summary/summarizer.py index a67b72f89c..371dc20cd3 100644 --- a/tools/crunchstat-summary/crunchstat_summary/summarizer.py +++ b/tools/crunchstat-summary/crunchstat_summary/summarizer.py @@ -251,26 +251,29 @@ class Summarizer(object): self.job_tot[category][stat] += val logger.debug('%s: done totals', self.label) - missing_category = { - 'cpu': 'CPU', - 'mem': 'memory', - 'net:': 'network I/O', - 'statfs': 'storage space', - } - for task_stat in self.task_stats.values(): - for category in task_stat.keys(): - for checkcat in missing_category: - if checkcat.endswith(':'): - if category.startswith(checkcat): - missing_category.pop(checkcat) - break - else: - if category == checkcat: - missing_category.pop(checkcat) - break - for catlabel in missing_category.values(): - logger.warning('%s: %s stats are missing -- possible cluster configuration issue', - self.label, catlabel) + if self.stats_max['time'].get('elapsed', 0) > 20: + # needs to have executed for at least 20 seconds or we may + # not have collected any metrics and these warnings are duds. + missing_category = { + 'cpu': 'CPU', + 'mem': 'memory', + 'net:': 'network I/O', + 'statfs': 'storage space', + } + for task_stat in self.task_stats.values(): + for category in task_stat.keys(): + for checkcat in missing_category: + if checkcat.endswith(':'): + if category.startswith(checkcat): + missing_category.pop(checkcat) + break + else: + if category == checkcat: + missing_category.pop(checkcat) + break + for catlabel in missing_category.values(): + logger.warning('%s: %s stats are missing -- possible cluster configuration issue', + self.label, catlabel) def long_label(self): label = self.label -- 2.30.2