src/ceph/qa/tasks/check_counter.py

   1
   2 import logging
   3 import json
   4
   5 from teuthology.task import Task
   6 from teuthology import misc
   7 import ceph_manager
   8
   9 log = logging.getLogger(__name__)
  10
  11
  12 class CheckCounter(Task):
  13     """
  14     Use this task to validate that some daemon perf counters were
  15     incremented by the nested tasks.
  16
  17     Config:
  18      'cluster_name': optional, specify which cluster
  19      'target': dictionary of daemon type to list of performance counters.
  20      'dry_run': just log the value of the counters, don't fail if they
  21                 aren't nonzero.
  22
  23     Success condition is that for all of the named counters, at least
  24     one of the daemons of that type has the counter nonzero.
  25
  26     Example to check cephfs dirfrag splits are happening:
  27     - install:
  28     - ceph:
  29     - ceph-fuse:
  30     - check-counter:
  31         counters:
  32             mds:
  33                 - "mds.dir_split"
  34     - workunit: ...
  35     """
  36
  37     def start(self):
  38         log.info("START")
  39
  40     def end(self):
  41         cluster_name = self.config.get('cluster_name', None)
  42         dry_run = self.config.get('dry_run', False)
  43         targets = self.config.get('counters', {})
  44
  45         if cluster_name is None:
  46             cluster_name = self.ctx.managers.keys()[0]
  47
  48         for daemon_type, counters in targets.items():
  49             # List of 'a', 'b', 'c'...
  50             daemon_ids = list(misc.all_roles_of_type(self.ctx.cluster, daemon_type))
  51             daemons = dict([(daemon_id,
  52                              self.ctx.daemons.get_daemon(daemon_type, daemon_id))
  53                             for daemon_id in daemon_ids])
  54
  55             seen = set()
  56
  57             for daemon_id, daemon in daemons.items():
  58                 if not daemon.running():
  59                     log.info("Ignoring daemon {0}, it isn't running".format(daemon_id))
  60                     continue
  61                 else:
  62                     log.debug("Getting stats from {0}".format(daemon_id))
  63
  64                 manager = self.ctx.managers[cluster_name]
  65                 proc = manager.admin_socket(daemon_type, daemon_id, ["perf", "dump"])
  66                 response_data = proc.stdout.getvalue().strip()
  67                 if response_data:
  68                     perf_dump = json.loads(response_data)
  69                 else:
  70                     log.warning("No admin socket response from {0}, skipping".format(daemon_id))
  71                     continue
  72
  73                 for counter in counters:
  74                     subsys, counter_id = counter.split(".")
  75                     if subsys not in perf_dump or counter_id not in perf_dump[subsys]:
  76                         log.warning("Counter '{0}' not found on daemon {1}.{2}".format(
  77                             counter, daemon_type, daemon_id))
  78                         continue
  79                     value = perf_dump[subsys][counter_id]
  80
  81                     log.info("Daemon {0}.{1} {2}={3}".format(
  82                         daemon_type, daemon_id, counter, value
  83                     ))
  84
  85                     if value > 0:
  86                         seen.add(counter)
  87
  88             if not dry_run:
  89                 unseen = set(counters) - set(seen)
  90                 if unseen:
  91                     raise RuntimeError("The following counters failed to be set "
  92                                        "on {0} daemons: {1}".format(
  93                         daemon_type, unseen
  94                     ))
  95
  96 task = CheckCounter