check_ceph_status

Initial commit
Benjamin Renard

Benjamin Renard commited on 2013-12-03 22:49:19
Showing 3 changed files, with 283 additions and 0 deletions.

.gitignore
View file @ dfb2f2e
... ... 
@@ -0,0 +1 @@
1 
+*~
README
View file @ dfb2f2e
... ... 
@@ -0,0 +1,49 @@
1 
+Nagios plugin to check Ceph cluster status
2 
+==========================================
3 
+
4 
+This plugin check ceph health, number of OSDs UP, number of MONs UP
5 
+and PGs states to determine Ceph cluster status.
6 
+
7 
+Usage
8 
+-----
9 
+
10 
+  Usage: check_ceph_status [options]
11 
+
12 
+  Options:
13 
+    -h, --help            show this help message and exit
14 
+    -d, --debug
15 
+    -b BIN, --bin=BIN     Ceph binary (default : /usr/bin/ceph)
16 
+    --conf=CONF           Ceph configuration file
17 
+    -m MON, --mon=MON     Ceph monitor address[:port]
18 
+    -i ID, --id=ID        Ceph client id
19 
+    -k KEYRING, --keyring=KEYRING
20 
+                          Ceph client keyring file
21 
+    -w WARNLOSTOSD, --warning-lost-osd=WARNLOSTOSD
22 
+                          Warning number of non-up OSDs (default : 1)
23 
+    -c CRITLOSTOSD, --critical-lost-osd=CRITLOSTOSD
24 
+                          Critical number of non-up OSDs (default : 2)
25 
+    -W WARNLOSTMON, --warning-lost-mon=WARNLOSTMON
26 
+                          Warning number of non-up MONs (default : 1)
27 
+    -C CRITLOSTMON, --critical-lost-mon=CRITLOSTMON
28 
+                          Critical number of non-up MONs (default : 2)
29 
+
30 
+Copyright
31 
+---------
32 
+
33 
+Copyright (c) 2013 Benjamin Renard <brenard@zionetrix.net>
34 
+
35 
+License
36 
+-------
37 
+
38 
+This program is free software; you can redistribute it and/or
39 
+modify it under the terms of the GNU General Public License version 2
40 
+as published by the Free Software Foundation.
41 
+
42 
+This program is distributed in the hope that it will be useful,
43 
+but WITHOUT ANY WARRANTY; without even the implied warranty of
44 
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
45 
+GNU General Public License for more details.
46 
+
47 
+You should have received a copy of the GNU General Public License
48 
+along with this program; if not, write to the Free Software
49 
+Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
check_ceph_status
View file @ dfb2f2e
... ... 
@@ -0,0 +1,233 @@
1 
+#!/usr/bin/python
2 
+#
3 
+# Nagios plugin to check Ceph cluster state
4 
+#
5 
+# This plugin check ceph health, number of OSDs UP, number of MONs UP
6 
+# and PGs states to determine Ceph cluster status.
7 
+#
8 
+#  Usage: check_ceph_status [options]
9 
+#
10 
+#  Options:
11 
+#    -h, --help            show this help message and exit
12 
+#    -d, --debug
13 
+#    -b BIN, --bin=BIN     Ceph binary (default : /usr/bin/ceph)
14 
+#    --conf=CONF           Ceph configuration file
15 
+#    -m MON, --mon=MON     Ceph monitor address[:port]
16 
+#    -i ID, --id=ID        Ceph client id
17 
+#    -k KEYRING, --keyring=KEYRING
18 
+#                          Ceph client keyring file
19 
+#    -w WARNLOSTOSD, --warning-lost-osd=WARNLOSTOSD
20 
+#                          Warning number of non-up OSDs (default : 1)
21 
+#    -c CRITLOSTOSD, --critical-lost-osd=CRITLOSTOSD
22 
+#                          Critical number of non-up OSDs (default : 2)
23 
+#    -W WARNLOSTMON, --warning-lost-mon=WARNLOSTMON
24 
+#                          Warning number of non-up MONs (default : 1)
25 
+#    -C CRITLOSTMON, --critical-lost-mon=CRITLOSTMON
26 
+#                          Critical number of non-up MONs (default : 2)
27 
+#
28 
+# Copyright (c) 2013 Benjamin Renard <brenard@zionetrix.net>
29 
+#
30 
+# This program is free software; you can redistribute it and/or
31 
+# modify it under the terms of the GNU General Public License version 2
32 
+# as published by the Free Software Foundation.
33 
+#
34 
+# This program is distributed in the hope that it will be useful,
35 
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
36 
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
37 
+# GNU General Public License for more details.
38 
+#
39 
+# You should have received a copy of the GNU General Public License
40 
+# along with this program; if not, write to the Free Software
41 
+# Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
42 
+#
43 
+
44 
+import sys,os,json,subprocess,re
45 
+from optparse import OptionParser
46 
+
47 
+# default ceph values
48 
+CEPH_COMMAND = '/usr/bin/ceph'
49 
+WARN_LOST_OSD = 1
50 
+CRIT_LOST_OSD = 2
51 
+WARN_LOST_MON = 1
52 
+CRIT_LOST_MON = 2
53 
+
54 
+# nagios exit code
55 
+STATUS = {
56 
+	'OK': 0,
57 
+	'WARNING': 1,
58 
+	'CRITICAL': 2,
59 
+	'UNKNOWN': 3
60 
+}
61 
+
62 
+parser = OptionParser()
63 
+parser.add_option('-d',
64 
+                  '--debug',
65 
+                  action="store_true",
66 
+                  dest="debug",
67 
+                  default=False)
68 
+
69 
+parser.add_option('-b',
70 
+                  '--bin',
71 
+                  action="store",
72 
+                  dest="bin",
73 
+                  help="Ceph binary (default : %s)" % CEPH_COMMAND,
74 
+                  type='string',
75 
+                  default=CEPH_COMMAND)
76 
+
77 
+parser.add_option('--conf',
78 
+                  action="store",
79 
+                  dest="conf",
80 
+                  help="Ceph configuration file",
81 
+                  type='string',
82 
+                  default=None)
83 
+
84 
+parser.add_option('-m',
85 
+                  '--mon',
86 
+                  action="store",
87 
+                  dest="mon",
88 
+                  help="Ceph monitor address[:port]",
89 
+                  type='string',
90 
+                  default=None)
91 
+
92 
+parser.add_option('-i',
93 
+                  '--id',
94 
+                  action="store",
95 
+                  dest="id",
96 
+                  help="Ceph client id",
97 
+                  type='string',
98 
+                  default=None)
99 
+
100 
+parser.add_option('-k',
101 
+                  '--keyring',
102 
+                  action="store",
103 
+                  dest="keyring",
104 
+                  help="Ceph client keyring file",
105 
+                  type='string',
106 
+                  default=None)
107 
+
108 
+parser.add_option('-w',
109 
+                  '--warning-lost-osd',
110 
+                  action="store",
111 
+                  dest="warnlostosd",
112 
+                  help="Warning number of non-up OSDs (default : %s)" % WARN_LOST_OSD,
113 
+                  type='int',
114 
+                  default=WARN_LOST_OSD)
115 
+
116 
+parser.add_option('-c',
117 
+                  '--critical-lost-osd',
118 
+                  action="store",
119 
+                  dest="critlostosd",
120 
+                  help="Critical number of non-up OSDs (default : %s)" % CRIT_LOST_OSD,
121 
+                  type='int',
122 
+                  default=CRIT_LOST_OSD)
123 
+
124 
+parser.add_option('-W',
125 
+                  '--warning-lost-mon',
126 
+                  action="store",
127 
+                  dest="warnlostmon",
128 
+                  help="Warning number of non-up MONs (default : %s)" % WARN_LOST_MON,
129 
+                  type='int',
130 
+                  default=WARN_LOST_MON)
131 
+
132 
+parser.add_option('-C',
133 
+                  '--critical-lost-mon',
134 
+                  action="store",
135 
+                  dest="critlostmon",
136 
+                  help="Critical number of non-up MONs (default : %s)" % CRIT_LOST_MON,
137 
+                  type='int',
138 
+                  default=CRIT_LOST_MON)
139 
+
140 
+(options, args) = parser.parse_args()
141 
+
142 
+ # validate args
143 
+if not os.path.exists(options.bin):
144 
+    print "ERROR: ceph executable '%s' doesn't exist" % options.bin
145 
+    sys.exit(STATUS['UNKNOWN'])
146 
+
147 
+if options.conf and not os.path.exists(options.conf):
148 
+    print "ERROR: ceph conf file '%s' doesn't exist" % options.conf
149 
+    sys.exit(STATUS['UNKNOWN'])
150 
+
151 
+if options.keyring and not os.path.exists(options.keyring):
152 
+    print "ERROR: keyring file '%s' doesn't exist" % options.keyring
153 
+    sys.exit(STATUS['UNKNOWN'])
154 
+
155 
+# build command
156 
+ceph_cmd = [options.bin]
157 
+if options.mon:
158 
+    ceph_cmd.append('-m')
159 
+    ceph_cmd.append(options.mon)
160 
+if options.conf:
161 
+    ceph_cmd.append('-c')
162 
+    ceph_cmd.append(options.conf)
163 
+if options.id:
164 
+    ceph_cmd.append('--id')
165 
+    ceph_cmd.append(options.id)
166 
+if options.keyring:
167 
+    ceph_cmd.append('--keyring')
168 
+    ceph_cmd.append(options.keyring)
169 
+ceph_cmd.append('status')
170 
+ceph_cmd.append('--format=json')
171 
+
172 
+# exec command
173 
+p = subprocess.Popen(ceph_cmd,stdout=subprocess.PIPE,stderr=subprocess.PIPE)
174 
+output, err = p.communicate()
175 
+
176 
+if output:
177 
+	data=json.loads(output)
178 
+
179 
+	status='OK'
180 
+
181 
+	health=data['health']['overall_status']
182 
+	if health=='HEALTH_WARN':
183 
+		status='WARNING'
184 
+	elif health=='HEALTH_CRIT':
185 
+		status='CRITICAL'
186 
+
187 
+	total_mon=len(data['monmap']['mons'])
188 
+	total_mon_up=len(data['health']['timechecks']['mons'])
189 
+
190 
+	num_lost_mon=total_mon-total_mon_up
191 
+	if num_lost_mon==0:
192 
+		monstate="(MONs UP : %s/%s)" % (total_mon_up,total_mon)
193 
+	else:
194 
+		monstate="%s MONs down (MONs UP : %s/%s)" % (num_lost_mon,total_mon_up,total_mon)
195 
+		if num_lost_mon >= options.critlostmon:
196 
+			status='CRITICAL'
197 
+		elif num_lost_mon >= options.warnlostmon and status!='CRITICAL':
198 
+			status='WARNING'
199 
+
200 
+	total_osd=data['osdmap']['osdmap']['num_osds']
201 
+	total_osd_up=data['osdmap']['osdmap']['num_up_osds']
202 
+
203 
+	num_lost_osd=total_osd-total_osd_up
204 
+
205 
+	if num_lost_osd>=options.critlostosd:
206 
+		status='CRITICAL'
207 
+	elif num_lost_osd>=options.warnlostosd and status!='CRITICAL':
208 
+		status='WARNING'
209 
+
210 
+	total_pg=data['pgmap']['num_pgs']
211 
+	pgstate=""
212 
+	for st in data['pgmap']['pgs_by_state']:
213 
+		if re.search('(down|inconsistent|imcomplete|stale)',st['state_name'],re.IGNORECASE):
214 
+			status='CRITICAL'
215 
+			pgstate="%s / %s PGs %s" % (pgstate,st['count'],st['state_name'])
216 
+		elif re.search('(replay|degraded|repair|recovering|backfill)',st['state_name'],re.IGNORECASE):
217 
+			if status!='CRITICAL':
218 
+				status="WARNING"
219 
+			pgstate="%s / %s PGs %s" % (pgstate,st['count'],st['state_name'])
220 
+		elif st['state_name']=="active+clean":
221 
+			pgstate="%s / %s/%s PGs active+clean" % (pgstate,st['count'],total_pg)
222 
+
223 
+	msg="%s : %s%s %s" % (status,health,pgstate,monstate)
224 
+
225 
+
226 
+	if num_lost_osd==0:
227 
+		print "%s (OSDs UP : %s/%s)" % (msg,total_osd_up,total_osd)
228 
+	else:
229 
+		print "%s / %s OSDs down (OSDs UP : %s/%s)" % (msg,num_lost_osd,total_osd_up,total_osd)
230 
+	sys.exit(STATUS[status])
231 
+else:
232 
+	print "UNKNOWN : fail to execute ceph status command"
233 
+	sys.exit(STATUS['UNKNOWN'])
0 234