check_openshift_pod_status_count

#!/usr/bin/python3

import argparse
import collections
import logging
import nagiosplugin
import re

from vshn_npo import constants
from vshn_npo import nagiosutils
from vshn_npo import oc_client
from vshn_npo import utils


def _find_column_pos(line, name):
  """Determine start and end of column.

  """
  m = re.search(r"\b({})\s+(\w)".format(re.escape(name)), line)
  if not m:
    msg = "Position of column \"{}\" not found".format(name)
    raise nagiosplugin.CheckError(msg)

  return (m.start(1), m.start(2))


def _extract_values(lines):
  (ready_start, ready_end) = _find_column_pos(lines[0], "READY")
  (status_start, status_end) = _find_column_pos(lines[0], "STATUS")

  for i in lines[1:]:
    ready = i[ready_start:ready_end].rstrip()
    status = i[status_start:status_end].rstrip()

    ready_parts = ready.split("/", 1)

    yield (int(ready_parts[0]), int(ready_parts[1]), status)


_STATUS_CAT_COMPLETED = "Completed"
_STATUS_CAT_CONTAINERCREATING = "ContainerCreating"
_STATUS_CAT_CRASHLOOPBACKOFF = "CrashLoopBackOff"
_STATUS_CAT_DEADLINEEXCEEDED = "DeadlineExceeded"
_STATUS_CAT_ERROR = "Error"
_STATUS_CAT_IMAGEPULLBACKOFF = "ImagePullBackOff"
_STATUS_CAT_INIT_ERROR = "InitError"
_STATUS_CAT_RUNNING = "Running"
_STATUS_CAT_TERMINATING = "Terminating"
_STATUS_CAT_UNRECOGNIZED = "Unrecognized"


# Status categories for performance data
_STATUS_CATEGORY_NAMES = frozenset([
  _STATUS_CAT_COMPLETED,
  _STATUS_CAT_CONTAINERCREATING,
  _STATUS_CAT_CRASHLOOPBACKOFF,
  _STATUS_CAT_DEADLINEEXCEEDED,
  _STATUS_CAT_ERROR,
  _STATUS_CAT_IMAGEPULLBACKOFF,
  _STATUS_CAT_INIT_ERROR,
  _STATUS_CAT_RUNNING,
  _STATUS_CAT_TERMINATING,
  _STATUS_CAT_UNRECOGNIZED,
  ])


# Status names without special handling
_STATUS_CAT_MAP = \
    dict((i.lower(), i) for i in [
      _STATUS_CAT_COMPLETED,
      _STATUS_CAT_CONTAINERCREATING,
      _STATUS_CAT_CRASHLOOPBACKOFF,
      _STATUS_CAT_DEADLINEEXCEEDED,
      _STATUS_CAT_ERROR,
      _STATUS_CAT_IMAGEPULLBACKOFF,
      _STATUS_CAT_RUNNING,
      _STATUS_CAT_TERMINATING,
      ])


def _categorize_status(status):
  lc_status = status.lower()

  if lc_status.startswith("init:"):
    return _STATUS_CAT_INIT_ERROR

  if lc_status in _STATUS_CAT_MAP:
    return _STATUS_CAT_MAP[lc_status]

  logging.info("Status \"%s\" not recognized", status)

  return _STATUS_CAT_UNRECOGNIZED



class PodStatusCount(nagiosplugin.Resource):
  def __init__(self, client):
    self._client = client

  def probe(self):
    output = self._client.capture_output([
      "get", "pod", "--all-namespaces", "--show-all",
      ])

    status_counter = collections.Counter()
    ready_containers = 0
    total_containers = 0

    for (ready_count, total_count, status) in _extract_values(output.splitlines()):
      cat = _categorize_status(status)
      assert cat in _STATUS_CATEGORY_NAMES
      status_counter[cat] += 1
      ready_containers += ready_count
      total_containers += total_count

    yield nagiosplugin.Metric("count-ready-containers", ready_containers, min=0, context="default")
    yield nagiosplugin.Metric("count-total-containers", total_containers, min=0, context="default")

    for i in sorted(_STATUS_CATEGORY_NAMES):
      yield nagiosplugin.Metric("status-{}".format(i), status_counter[i], min=0, context="default")


@nagiosplugin.guarded
def main():
  parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
  utils.add_verbose_argument(parser)
  parser.add_argument("--oc", type=str, default=constants.DEFAULT_OC_BINARY,
                      help="Path to OpenShift client binary")
  parser.add_argument("--config", type=str, default=None,
                      help="Configuration file with login credentials")
  args = parser.parse_args()

  utils.setup_basic_logging(args.verbose)

  client = oc_client.Client(args.oc, args.config)

  checks = [
      PodStatusCount(client),
      nagiosutils.FullSummary(),
      ]

  nagiosplugin.Check(*checks).main(verbose=args.verbose, timeout=None)


if __name__ == "__main__":
  main()

# vim: set sw=2 sts=2 et :