check_ceph.py

#!/usr/bin/python

# Examples:
# osd status, warn at 2 missing, crit at 3: ./check_ceph.py -C ceph.conf --id icinga -k ceph.client.icinga.keyring --osd -w 2 -c 3
# general health statis: /check_ceph.py -C ceph.conf --id icinga -k ceph.client.icinga.keyring --health
# pg status, does not take warning or critical arguments yet. Only warns on PGs not in an active+clean state which means some PGs are not in an optimal state. ./check_ceph.py -C ceph.conf --id icinga -k ceph.client.icinga.keyring --pg
# extra performance metrics (iops, read/write bytes/sec): ./check_ceph.py -C ceph.conf --id icinga -k ceph.client.icinga.keyring --perf
# disk space, if run with --pool you only alert on that pool. when run without --pool the thresholds are for every pool. warning and ciritcal are the max avail fields from `ceph df`: ./check_ceph.py -C ceph.conf --id icinga -k ceph.client.icinga.keyring --df -w 100 -c 50
#
#
import sys
import argparse
import json
import subprocess

# ceph osd stat
# ceph mon stat
# ceph pg stat
# ceph health statua
# ceph mon_status
# ceph quorum status


def checkHealth(args):

    ceph_health_json = subprocess.check_output(
        ["ceph --id {0} -c {1} -k {2} --format json health".format(args.id, args.conf, args.keyring)], shell=True)
    ceph_health_dict = json.loads(ceph_health_json)

    if ceph_health_dict['status'] == 'HEALTH_ERR':
        try:
            print "%s: %s" % (ceph_health_dict['overall_status'], ceph_health_dict['summary'][0]['summary'])
        except KeyError:
            print "%s: %s" % (ceph_health_dict['status'], ceph_health_dict['checks'].keys()[0])
        sys.exit(2)
    elif ceph_health_dict['status'] == 'HEALTH_WARN':
        try:
            print "%s: %s" % (ceph_health_dict['overall_status'], ceph_health_dict['summary'][0]['summary'])
        except KeyError:
            print "%s: %s" % (ceph_health_dict['status'], ceph_health_dict['checks'].keys()[0])
        sys.exit(1)
    elif ceph_health_dict['status'] == 'HEALTH_OK':
        print "%s" % (ceph_health_dict['status'])
        sys.exit(0)


def checkOSD(args):
    if args.warning:
        WARN = float(args.warning)
    if args.critical:
        CRIT = float(args.critical)
    osd_stat_json = subprocess.check_output(
        ["ceph --id {0} -c {1} -k {2} --format json osd stat".format(args.id, args.conf, args.keyring)], shell=True)
    osd_stat_dict = json.loads(osd_stat_json)
    try:
        osd_not_up = osd_stat_dict['num_osds'] - osd_stat_dict['num_up_osds']
    except KeyError:
        osd_stat_dict = osd_stat_dict['osdmap']
        osd_not_up = osd_stat_dict['num_osds'] - osd_stat_dict['num_up_osds']
    
    osd_not_in = osd_stat_dict['num_osds'] - osd_stat_dict['num_in_osds']
    perf_string = "num_osds={0} num_up_osds={1} num_in_osds={2}".format(
        osd_stat_dict['num_osds'], osd_stat_dict['num_up_osds'], osd_stat_dict['num_in_osds'])

# Build in logic to handle the full and near full keys that are returned in the json
    if (osd_not_up >= WARN and osd_not_up < CRIT) or (osd_not_in >= WARN and osd_not_in < CRIT):
        print "WARNING: ALL OSDs are not up and in. {0} OSDS. {1} up, {2} in|{3}".format(osd_stat_dict['num_osds'], osd_stat_dict['num_up_osds'], osd_stat_dict['num_in_osds'], perf_string)
        sys.exit(1)
    elif (osd_not_up >= CRIT) or (osd_not_in >= CRIT):
        print "CRITICAL: ALL OSDs are not up and in. {0} OSDS. {1} up, {2} in|{3}".format(osd_stat_dict['num_osds'], osd_stat_dict['num_up_osds'], osd_stat_dict['num_in_osds'], perf_string)
        sys.exit(2)
    elif (osd_stat_dict['num_osds'] == osd_stat_dict['num_in_osds']) and (osd_stat_dict['num_osds'] == osd_stat_dict['num_up_osds']):
        print "ALL OSDs are up and in. {0} OSDS. {1} up, {2} in|{3}".format(osd_stat_dict['num_osds'], osd_stat_dict['num_up_osds'], osd_stat_dict['num_in_osds'], perf_string)
        sys.exit(0)
    else:
        print "Script shouldn't reach this point. Thar be bugs!"
        sys.exit(3)


def checkMON(args):
    if args.warning:
        WARN = float(args.warning)
    if args.critical:
        CRIT = float(args.critical)
    # not written yet, more important things


def checkPG(args):
    pg_stat_json = subprocess.check_output(
        ["ceph --id {0} -c {1} -k {2} --format json pg stat".format(args.id, args.conf, args.keyring)], shell=True)
    pg_stat_dict = json.loads(pg_stat_json)
    # cheap fix for nautilus change in json output
    if 'num_pgs' in pg_stat_dict.keys():
        # pre nautilus json format
        pg_summary = pg_stat_dict
    elif 'pg_summary' in pg_stat_dict.keys():
        # nautilus json format
        pg_summary = pg_stat_dict['pg_summary']
    num_pgs = pg_summary['num_pgs']
    active_pgs = 0
    perf_string = ""
    for x in pg_summary['num_pg_by_state']:
        if "active+clean" in x['name']:
            active_pgs += x['num']
        perf_string += "%s=%s " % (x['name'], x['num'])
# Maybe build in a percentage based threshold for users who want to have thresholds like that
    if active_pgs < num_pgs:
        print "WARNING: All PGs are not active+clean: {0} PGs Total, {1}|{1}".format(num_pgs, perf_string)
        sys.exit(1)
    elif active_pgs == num_pgs:
        print "All PGs are active+clean: {0} PGs Total, {1}|{1}".format(num_pgs, perf_string)
        sys.exit(0)
    else:
        print "Script shouldn't reach this point. Thar be bugs!"
        sys.exit(3)


def checkPerf(args):
    pg_stat_json = subprocess.check_output(
        ["ceph --id {0} -c {1} -k {2} --format json pg stat".format(args.id, args.conf, args.keyring)], shell=True)
    pg_stat_dict = json.loads(pg_stat_json)
    if 'read_bytes_sec' not in pg_stat_dict:
        pg_stat_dict['read_bytes_sec'] = 0
    if 'write_bytes_sec' not in pg_stat_dict:
        pg_stat_dict['write_bytes_sec'] = 0
    if 'io_sec' not in pg_stat_dict:
        pg_stat_dict['io_sec'] = 0
    perf_string = "read_bytes_sec={0} write_bytes_sec={1} io_sec={2}".format(
        pg_stat_dict['read_bytes_sec'], pg_stat_dict['write_bytes_sec'], pg_stat_dict['io_sec'])
    print "Healthy: Additional perf stats for cluster {0}|{0}".format(perf_string)
    sys.exit(0)


def checkDF(args):
    if args.warning:
        WARN = float(args.warning)
    if args.critical:
        CRIT = float(args.critical)
    if args.byte:
        if args.byte == "T":
            byte_divisor = 1024**4
            perf_metric = "TB"
        elif args.byte == "G":
            byte_divisor = 1024**3
            perf_metric = "GB"
        elif args.byte == "P":
            byte_divisor = 1024**5
            perf_metric = "PB"
    else:
        byte_divisor = 1024**4
        perf_metric = "TB"

    ceph_df_json = subprocess.check_output(
        ["ceph --id {0} -c {1} -k {2} --format json df".format(args.id, args.conf, args.keyring)], shell=True)
    ceph_df_dict = json.loads(ceph_df_json)
    # get global stats
    global_bytes, global_used_bytes, global_avail_bytes = ceph_df_dict['stats'][
        'total_bytes'], ceph_df_dict['stats']['total_used_bytes'], ceph_df_dict['stats']['total_avail_bytes']
    global_total = global_bytes/byte_divisor
    global_used = global_used_bytes/byte_divisor
    global_avail = global_avail_bytes/byte_divisor

    # get all pool stats
    pool_stats = {}
    for pool in ceph_df_dict['pools']:
        pool_stats[pool['name']] = {'bytes_used': pool['stats']['bytes_used']/byte_divisor,
                                    'max_avail': pool['stats']['max_avail']/byte_divisor, 'objects': pool['stats']['objects']}

    perf_string = "global_total_bytes={0}{3} global_used_bytes={1}{3} global_avail_bytes={2}{3} ".format(
        global_bytes/byte_divisor, global_used_bytes/byte_divisor, global_avail_bytes/byte_divisor, perf_metric)
    for item in pool_stats.keys():
        perf_string += "{0}_bytes_used={1}{2} {0}_max_avail={3}{2} {0}_objects={4} ".format(
            item, pool_stats[item]['bytes_used'], perf_metric, pool_stats[item]['max_avail'], pool_stats[item]['objects'])

# if pool is defined alert on that. if pool is not defined alert on the max_avail of all pools if any cross threshold
    if args.pool in pool_stats.keys():
        #        print pool_stats[args.pool]
        # add in percentage later
        if (pool_stats[args.pool]['max_avail'] < WARN) and (pool_stats[args.pool]['max_avail'] > CRIT):
            print "WARNING: Ceph pool {0} has {1}{2} availbale|{3}".format(args.pool, pool_stats[args.pool]['max_avail'], perf_metric, perf_string)
            sys.exit(1)
        elif pool_stats[args.pool]['max_avail'] < CRIT:
            print "CRITICAL: Ceph pool {0} has {1}{2} availbale|{3}".format(args.pool, pool_stats[args.pool]['max_avail'], perf_metric, perf_string)
            sys.exit(2)
        elif pool_stats[args.pool]['max_avail'] > WARN:
            print "Healthy: Ceph pool {0} has {1}{2} availbale|{3}".format(args.pool, pool_stats[args.pool]['max_avail'], perf_metric, perf_string)
            sys.exit(0)
        else:
            print "Script shouldn't reach this point. Thar be bugs!"
            sys.exit(3)

    else:
        # Alerts based on all pools. If any pool is crossing the threshold we alert on it
        warn_list = []
        crit_list = []

        for key in pool_stats.keys():
            if (pool_stats[key]['max_avail'] < WARN) and (pool_stats[key]['max_avail'] > CRIT):
                warn_list.append("%s:%s%s" % (
                    key, pool_stats[key]['max_avail'], perf_metric))
            elif pool_stats[key]['max_avail'] < CRIT:
                crit_list.append("%s:%s%s" % (
                    key, pool_stats[key]['max_avail'], perf_metric))

        if (len(warn_list) > 0) and (len(crit_list) == 0):
            print "WARNING: Ceph pool(s) low on free space. {0}|{1}".format(warn_list, perf_string)
            sys.exit(1)
        elif len(crit_list) > 0:
            print "CRITICAL: Ceph pool(s) critically low on free space. Critial:{0} Warning:{1}|{2}".format(crit_list, warn_list, perf_string)
            sys.exit(2)
        elif (len(warn_list) == 0) and (len(crit_list) == 0):
            print "Healthy: All ceph pools are within free space thresholds|{0}".format(perf_string)
        else:
            print "Script shouldn't reach this point. Thar be bugs!"
            sys.exit(3)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description='Runs health checks against a ceph cluster. This is designed to run on the monitoring server using the ceph client software. Supply a ceph.conf, keyring, and user to access the cluster.')
    parser.add_argument(
        '-C', '--conf', help='ceph.conf file, defaults to /etc/ceph/ceph.conf.')
    parser.add_argument('-id', '--id', help='Ceph authx user', required=True)
    parser.add_argument(
        '-k', '--keyring', help='Path to ceph keyring if not in /etc/ceph/client.\$id.keyring')
    parser.add_argument(
        '--health', help='Get general health status. ex. HEALTH_OK, HEALTH_WARN', action="store_true")
    parser.add_argument(
        '-o', '--osd', help='OSD status. Thresholds are in number of OSDs missing', action="store_true")
    parser.add_argument(
        '-m', '--mon', help='MON status. Thesholds are in number of mons missing')
    parser.add_argument(
        '-p', '--pg', help='PG status. No thresholds due to the large number of pg states.', action="store_true")
    parser.add_argument(
        '--perf', help='collects additional ceph performance statistics', action='store_true')
    parser.add_argument('--df', help='Disk/cluster usage. Reports global and all pools unless --pool is used. Warning and critical are number of -b free to the pools. This is not Raw Free, but Max Avail to the pools based on rep or k,m settings. If you do not define a pool the threshold is run agains all the pools in the cluster.', action="store_true")
    parser.add_argument(
        '-b', '--byte', help="Format to use for displaying DF data. G=Gigabyte, T=Terabyte. Use with the --df option. Defults to TB")
    parser.add_argument('--pool', help='Pool. Use with df')
    parser.add_argument('--objects', help='Object counts based on pool')
    parser.add_argument(
        '-w', '--warning', help='Warning threshold. See specific checks for value types')
    parser.add_argument(
        '-c', '--critical', help='Critical threshold. See specific checks for value types')

    args = parser.parse_args()

    if args.health:
        checkHealth(args)
    elif args.osd:
        checkOSD(args)
    elif args.pg:
        checkPG(args)
    elif args.df:
        checkDF(args)
    elif args.perf:
        checkPerf(args)