4yandex

custom-monitoring-checks-helper.py

#!/usr/bin/env python

import os, sys, commands
import argparse
import ConfigParser

version = '0.1.1'
default_monitoring_checks_cache_path = '/var/cache/monitoring'

def createParser ():
    parser = argparse.ArgumentParser(
        prog = 'custom-monitoring-checks-helper',
        description = '''Helper Python script to handle custom monitoring checks.
                         Checks are defined with config files (one for each).
                         Each check could be forced to run and display it's actual status,
                         or all checks latest statuses could be displayed from cache.
                         If there are no cached data for one of the checks, this check
                         would be also forced to run in order to retrieve it's status.''',
        epilog = '''(c) Antonio Kless 2014.
                    This software is available under BSD license.'''
    )

    parser.add_argument('-a', '--all', action='store_true', default=False, help='show all checks latest statuses (show cached status, or cause a new check run if no cached data found)')
    parser.add_argument('-f', '--force-check', help="force check run and show it's actual status", metavar='CHECK_NAME')
    parser.add_argument('-s', '--show', help='show config details for the check specified', metavar='CHECK_NAME')
    parser.add_argument('-d', '--config-file-scan-dir', default='./custom-monitoring-checks.d', help='specify a directory where config files should be looked for (default - ./custom-monitoring-checks.d)', metavar='DIR')
    parser.add_argument('-V', '--version', action='version', version='%(prog)s {}'.format(version), help='show version and exit')

    return parser

def config_file_scan_dir (dirname):
    if (os.path.exists(dirname)):
        files_found = os.listdir(dirname)
        if (len(files_found)) < 1:
            sys.stderr.write("No config files found in " + dirname + " - don't know what to do, aborting.\n")
            sys.exit(1)
        checks = {}
        for config_file in files_found:
            check = config_file_parse(config_file)
            if check:
                checks.update(check)

        if (len(checks)) < 1:
            sys.stderr.write("No valid config files found, no checks loaded, exiting.\n")
            sys.exit(0)

        print('Now ' + format(len(checks)) + ' check(s) loaded.')

        return checks
    else:
        sys.stderr.write("Directory " + dirname + " does not exist or not accessable, aborting.\n")
        sys.exit(1)

def config_file_parse (filename):
    print('Found config file: ' + filename)
    config = ConfigParser.ConfigParser()
    config.readfp(open(args.config_file_scan_dir + '/' + filename))
    #TODO: config values validation needed
    try:
        parsed_config = {
                         config.get('check', 'name'): {
                                                       'path': config.get('check', 'path'),
                                                       'description': config.get('check', 'description'),
                                                       'cronstring': config.get('check', 'cronstring'),
                                                       'cache': config.get('check', 'cache')
                                                      }
                        }
        print(' ...successfully parsed config for check "' + config.get('check', 'name') + '"')
        return parsed_config
    except BaseException:
        print(' ...not a valid config file, skipping')

def read_check_status_from_cache (filename, checkname):
    #FIXME: f.read() is danger to use as it may causes memory overflow

    if (os.path.exists(filename)):
        try:
            f = open(filename, 'r')
        except IOError:
            pass
        else:
            content = f.read()
            f.close()
            if (len(content) > 1):
                return content

    backupfilename = default_monitoring_checks_cache_path + '/' + checkname + '.last-run'

    if (os.path.exists(backupfilename)):
        try:
            f = open(backupfilename, 'r')
        except IOError:
            pass
        else:
            content = f.read()
            f.close()
            if (len(content) > 1):
                return content

    return False


if __name__ == '__main__':
    parser = createParser()

    if (len(sys.argv)) < 2:
        sys.stderr.write("No arguments passed - don't know what to do, aborting.\n")
        print("\n")
        parser.print_help()
        sys.exit(1)

    args = parser.parse_args()

checks = config_file_scan_dir(args.config_file_scan_dir)

print("\n")

if (args.show):
    if (args.show in checks):
        print('Config details for check "' + args.show + '":')
        print(' command to execute or path to executable file: ' + checks[args.show]['path'])
        print(' check description: ' + checks[args.show]['description'])
        print(' scheduling string for cron job: ' + checks[args.show]['cronstring'])
        print(' latest check status should be stored to: ' + checks[args.show]['cache'])
    else:
        sys.stderr.write("No such check found - " + args.show + ", aborting.\n")
        sys.exit(1)

elif (args.force_check):
    if (args.force_check in checks):
        print('Force run for check "' + args.force_check + '":')
        print(commands.getoutput(checks[args.force_check]['path']))
    else:
        sys.stderr.write("No such check found - " + args.force_check + ", aborting.\n")
        sys.exit(1)

elif (args.all):
    for check in checks:
        print('Latest value for check "' + check + '":')

        data_from_cache = read_check_status_from_cache(checks[check]['cache'], check)

        if (data_from_cache):
            print ' ...cached data available:'
            print data_from_cache
        else:
            print ' ...cached data unavailable, force check run:'
            print(commands.getoutput(checks[check]['path']))

else:
    sys.stderr.write("Not enough arguments passed - don't know what to do, aborting.\n")
    print("\n")
    parser.print_help()
    sys.exit(1)

#TODO: when processing --force_check or --all, it's better to format check's status output to some human-friendly view

some_check.conf

[check]
name        = new_os_updates_available
path        = /opt/scripts/monitoring/new-os-updates-available.sh
description = Returns 1, if there are some new OS updates available to install, and 0 instead
cronstring  = */30 * * * *
cache       = /any/custom/path/to/cache/new-os-updates-available.last-run