[checkmk-commits] agent_aws: Default resolution of 5 min requires an aysnchronous execution

Module: check_mk Branch: master Commit: 816f61c737dc9ddcf56e12b4c68e54597836bb5b URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=816f61c737dc9d… Author: Simon Betz &lt;si(a)mathias-kettner.de&gt; Date: Tue Dec 11 13:50:12 2018 +0100 agent_aws: Default resolution of 5 min requires an aysnchronous execution Change-Id: Id24213ae2bf13b1be06450b6b707ef010d46ab02 --- agents/special/agent_aws | 97 +++++++++++++++++++++++++++---------- checks/agent_aws | 1 + cmk/paths.py | 2 +- tests/unit/checks/test_agent_aws.py | 2 +- 4 files changed, 75 insertions(+), 27 deletions(-) diff --git a/agents/special/agent_aws b/agents/special/agent_aws index 471f2b9..07e2862 100755 --- a/agents/special/agent_aws +++ b/agents/special/agent_aws @@ -34,11 +34,12 @@ import json import logging import sys import time +import os from typing import NamedTuple, Any - -import boto3 -import botocore - +from pathlib2 import Path +import boto3 # type: ignore +import botocore # type: ignore +from cmk.paths import tmp_dir import cmk.password_store # .--helpers-------------------------------------------------------------. @@ -99,6 +100,8 @@ AWSSectionResult = NamedTuple("Result", [ ("content", Any), ]) +AWSStateFilesPath = Path(os.environ['OMD_ROOT']) / tmp_dir / "agents/agent_aws" + class AWSSection(object): __metaclass__ = abc.ABCMeta @@ -121,6 +124,10 @@ class AWSSection(object): """ pass + @property + def period(self): + return 2 * self.interval + def run(self): if self._client is None: msg = "%s: No client initialized" % self.__class__.__name__ @@ -223,7 +230,8 @@ class EC2(AWSSection): class AWSSections(object): - def __init__(self, session): + def __init__(self, session, hostname): + self._hostname = hostname self._session = session self._sections = [] self._results = {} @@ -262,8 +270,12 @@ class AWSSections(object): logging.info(e) self._exceptions.append(e) - def run(self): + def run(self, use_cache=True): + now = time.time() for section in self._sections: + if use_cache and self._section_cache_is_recent_enough(now, section): + continue + try: result = section.run() except Exception as e: @@ -273,38 +285,70 @@ class AWSSections(object): self._results.setdefault((section.name, section.interval), result) if not self._results: - logging.info("No results") + logging.info("No results or cached data") return - self._write_results() + self._write_exceptions() + self._write_section_results(now) + + def _section_cache_is_recent_enough(self, now, section): + state_dir = AWSStateFilesPath / self._hostname + state_dir.mkdir(parents=True, exist_ok=True) + state_file = state_dir / Path("%s-%s" % (self._session.region_name, section.name)) + if not state_file.exists(): + logging.info("New state file of section '%s'", section.name) + state_file.touch() + return False - def _write_results(self): + try: + age = now - state_file.stat().st_mtime + except OSError as e: + if e.errno == 2: # No such file or directory + logging.info("Cannot calculate age; new state file of section '%s'", section.name) + state_file.touch() + return False + else: + raise + + if age >= section.interval: + logging.info("State file of section '%s' is outdated: %s >= %s", section.name, age, + section.interval) + state_file.touch() + return False + return True + + def _write_exceptions(self): sys.stdout.write("<<<aws_exceptions>>>\n") if self._exceptions: - for e in self._exceptions: - sys.stdout.write("%s\n" % e.message) + out = "\n".join([e.message for e in self._exceptions]) else: - sys.stdout.write("No exceptions\n") + out = "No exceptions" + sys.stdout.write("%s\n" % out) - now = int(time.time()) + def _write_section_results(self, now): for (section_name, section_interval), result in self._results.iteritems(): if not result: - msg = "No results of %s" % section_name - logging.info(msg) + logging.info("No results of %s", section_name) continue + if not isinstance(result, list): msg = "Section result must be formatted as a list of 'AWSSectionResult's" logging.info(msg) raise AWSSectionError(msg) - section_header = "<<<aws_%s:cached(%s,%s)>>>\n" % (section_name, now, section_interval) - for row in result: - if row.piggyback_hostname: - sys.stdout.write("<<<<%s>>>>\n" % row.piggyback_hostname) - sys.stdout.write(section_header) - sys.stdout.write("%s\n" % json.dumps(row.content, default=_datetime_converter)) - if row.piggyback_hostname: - sys.stdout.write("<<<<>>>>\n") + self._write_section_result(section_name, int(now + section_interval + 60), result) + + def _write_section_result(self, section_name, until, result): + section_header = "<<<aws_%s:persist(%s)>>>\n" % (section_name, until) + for row in result: + write_piggyback_header = row.piggyback_hostname\ + and row.piggyback_hostname != self._hostname + if write_piggyback_header: + sys.stdout.write("<<<<%s>>>>\n" % row.piggyback_hostname) + sys.stdout.write(section_header) + sys.stdout.write("%s\n" % json.dumps(row.content, default=_datetime_converter)) + if write_piggyback_header: + sys.stdout.write("<<<<>>>>\n") #. @@ -326,6 +370,9 @@ def parse_arguments(argv): parser.add_argument( "--aws-secret-access-key", required=True, help="The secret key for your AWS account.") parser.add_argument("--region", required=True, help="The region to use.") + parser.add_argument( + "--no-cache", action="store_true", help="Execute all sections, do not rely on cached data.") + parser.add_argument("hostname") return parser.parse_args(argv) @@ -353,8 +400,8 @@ def main(args=None): region_name=args.region) try: - aws_sections = AWSSections(s) - aws_sections.run() + aws_sections = AWSSections(s, args.hostname) + aws_sections.run(use_cache=not args.no_cache) except Exception as e: if args.debug: raise diff --git a/checks/agent_aws b/checks/agent_aws index f08a435..0393d3b 100644 --- a/checks/agent_aws +++ b/checks/agent_aws @@ -33,6 +33,7 @@ def agent_aws_arguments(params, hostname, ipaddress): passwordstore_get_cmdline("%s", params["--aws-secret-access-key"]), "--region", params["--region"], + hostname, ] diff --git a/cmk/paths.py b/cmk/paths.py index dd702e5..6431fc0 100644 --- a/cmk/paths.py +++ b/cmk/paths.py @@ -51,7 +51,7 @@ tcp_cache_dir = None data_source_cache_dir = None snmp_scan_cache_dir = None include_cache_dir = None -tmp_dir = None +tmp_dir = '' logwatch_dir = None nagios_objects_file = None nagios_command_pipe_path = None diff --git a/tests/unit/checks/test_agent_aws.py b/tests/unit/checks/test_agent_aws.py index bba4178..0e65b5c 100644 --- a/tests/unit/checks/test_agent_aws.py +++ b/tests/unit/checks/test_agent_aws.py @@ -6,7 +6,7 @@ import pytest "--aws-access-key-id": "user", "--aws-secret-access-key": "d1ng", "--region": 'region-eu', - }, ['--aws-access-key-id', 'user', '--aws-secret-access-key', 'd1ng', '--region', 'region-eu']), + }, ['--aws-access-key-id', 'user', '--aws-secret-access-key', 'd1ng', '--region', 'region-eu', 'host']), ]) def test_aws(check_manager, params, result): agent = check_manager.get_special_agent("agent_aws")