1 CEPH-PROMETHEUS-ALERT-MIB DEFINITIONS ::= BEGIN
4 MODULE-IDENTITY, OBJECT-TYPE, NOTIFICATION-TYPE, enterprises, TimeTicks
10 ceph OBJECT IDENTIFIER ::= { enterprises 50495 }
11 prometheus OBJECT IDENTIFIER ::= { ceph 15 }
13 prometheusAlert MODULE-IDENTITY
14 LAST-UPDATED "201904010000Z" -- 1. Apr 2019
15 ORGANIZATION "The Ceph Project"
16 CONTACT-INFO "https://ceph.com"
17 DESCRIPTION "Prometheus Alert SNMP MIB"
18 REVISION "201904010000Z" -- 1. Apr 2019
19 DESCRIPTION "Initial version."
22 prometheusAlertObjects OBJECT IDENTIFIER ::= { prometheusAlert 1 }
23 prometheusAlertTraps OBJECT IDENTIFIER ::= { prometheusAlert 2 }
29 prometheusAlertNotificationAlertName OBJECT-TYPE
31 MAX-ACCESS accessible-for-notify
33 DESCRIPTION "The name of the Prometheus alert."
34 ::= { prometheusAlertObjects 1 }
36 prometheusAlertNotificationStatus OBJECT-TYPE
38 MAX-ACCESS accessible-for-notify
40 DESCRIPTION "The status of the Prometheus alert."
41 ::= { prometheusAlertObjects 2 }
43 prometheusAlertNotificationSeverity OBJECT-TYPE
45 MAX-ACCESS accessible-for-notify
47 DESCRIPTION "The severity of the Prometheus alert."
48 ::= { prometheusAlertObjects 3 }
50 prometheusAlertNotificationInstance OBJECT-TYPE
52 MAX-ACCESS accessible-for-notify
54 DESCRIPTION "Unique identifier for the Prometheus instance."
55 ::= { prometheusAlertObjects 4 }
57 prometheusAlertNotificationJob OBJECT-TYPE
59 MAX-ACCESS accessible-for-notify
61 DESCRIPTION "The name of the Prometheus job."
62 ::= { prometheusAlertObjects 5 }
64 prometheusAlertNotificationDescription OBJECT-TYPE
66 MAX-ACCESS accessible-for-notify
68 DESCRIPTION "The Prometheus alert description field."
69 ::= { prometheusAlertObjects 6 }
71 prometheusAlertNotificationLabels OBJECT-TYPE
73 MAX-ACCESS accessible-for-notify
75 DESCRIPTION "Additional Prometheus alert labels as JSON string."
76 ::= { prometheusAlertObjects 7 }
78 prometheusAlertNotificationTimestamp OBJECT-TYPE
80 MAX-ACCESS accessible-for-notify
82 DESCRIPTION "The time when the Prometheus alert occurred."
83 ::= { prometheusAlertObjects 8 }
85 prometheusAlertNotificationRawData OBJECT-TYPE
87 MAX-ACCESS accessible-for-notify
89 DESCRIPTION "The raw Prometheus alert as JSON string."
90 ::= { prometheusAlertObjects 9 }
96 prometheusAlertTrapDefault NOTIFICATION-TYPE
98 prometheusAlertNotificationAlertName,
99 prometheusAlertNotificationStatus,
100 prometheusAlertNotificationSeverity,
101 prometheusAlertNotificationInstance,
102 prometheusAlertNotificationJob,
103 prometheusAlertNotificationDescription,
104 prometheusAlertNotificationLabels,
105 prometheusAlertNotificationTimestamp,
106 prometheusAlertNotificationRawData
109 DESCRIPTION "Default trap."
110 ::= { prometheusAlertTraps 1 }
112 prometheusAlertClusterHealthTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 2 }
113 prometheusAlertMonTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 3 }
114 prometheusAlertOsdTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 4 }
115 prometheusAlertMdsTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 5 }
116 prometheusAlertMgrTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 6 }
117 prometheusAlertPgsTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 7 }
118 prometheusAlertNodesTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 8 }
119 prometheusAlertPoolsTraps OBJECT IDENTIFIER ::= { prometheusAlertTraps 9 }
121 prometheusAlertClusterHealthTrapHealthError NOTIFICATION-TYPE
123 prometheusAlertNotificationAlertName,
124 prometheusAlertNotificationStatus,
125 prometheusAlertNotificationSeverity,
126 prometheusAlertNotificationInstance,
127 prometheusAlertNotificationJob,
128 prometheusAlertNotificationDescription,
129 prometheusAlertNotificationLabels,
130 prometheusAlertNotificationTimestamp,
131 prometheusAlertNotificationRawData
134 DESCRIPTION "Ceph in health_error state for more than 5m."
135 ::= { prometheusAlertClusterHealthTraps 1 }
137 prometheusAlertClusterHealthTrapHealthWarn NOTIFICATION-TYPE
139 prometheusAlertNotificationAlertName,
140 prometheusAlertNotificationStatus,
141 prometheusAlertNotificationSeverity,
142 prometheusAlertNotificationInstance,
143 prometheusAlertNotificationJob,
144 prometheusAlertNotificationDescription,
145 prometheusAlertNotificationLabels,
146 prometheusAlertNotificationTimestamp,
147 prometheusAlertNotificationRawData
150 DESCRIPTION "Ceph in health_warn for more than 15m."
151 ::= { prometheusAlertClusterHealthTraps 2 }
153 prometheusAlertMonTrapLowMonitorQuorumCount NOTIFICATION-TYPE
155 prometheusAlertNotificationAlertName,
156 prometheusAlertNotificationStatus,
157 prometheusAlertNotificationSeverity,
158 prometheusAlertNotificationInstance,
159 prometheusAlertNotificationJob,
160 prometheusAlertNotificationDescription,
161 prometheusAlertNotificationLabels,
162 prometheusAlertNotificationTimestamp,
163 prometheusAlertNotificationRawData
166 DESCRIPTION "Monitor count in quorum is low."
167 ::= { prometheusAlertMonTraps 1 }
169 prometheusAlertOsdTrap10PercentOsdsDown NOTIFICATION-TYPE
171 prometheusAlertNotificationAlertName,
172 prometheusAlertNotificationStatus,
173 prometheusAlertNotificationSeverity,
174 prometheusAlertNotificationInstance,
175 prometheusAlertNotificationJob,
176 prometheusAlertNotificationDescription,
177 prometheusAlertNotificationLabels,
178 prometheusAlertNotificationTimestamp,
179 prometheusAlertNotificationRawData
182 DESCRIPTION "More than 10% of OSDs are down."
183 ::= { prometheusAlertOsdTraps 1 }
185 prometheusAlertOsdTrapOsdDown NOTIFICATION-TYPE
187 prometheusAlertNotificationAlertName,
188 prometheusAlertNotificationStatus,
189 prometheusAlertNotificationSeverity,
190 prometheusAlertNotificationInstance,
191 prometheusAlertNotificationJob,
192 prometheusAlertNotificationDescription,
193 prometheusAlertNotificationLabels,
194 prometheusAlertNotificationTimestamp,
195 prometheusAlertNotificationRawData
198 DESCRIPTION "One or more OSDs down for more than 15 minutes."
199 ::= { prometheusAlertOsdTraps 2 }
201 prometheusAlertOsdTrapOsdsNearFull NOTIFICATION-TYPE
203 prometheusAlertNotificationAlertName,
204 prometheusAlertNotificationStatus,
205 prometheusAlertNotificationSeverity,
206 prometheusAlertNotificationInstance,
207 prometheusAlertNotificationJob,
208 prometheusAlertNotificationDescription,
209 prometheusAlertNotificationLabels,
210 prometheusAlertNotificationTimestamp,
211 prometheusAlertNotificationRawData
214 DESCRIPTION "An OSD is dangerously full, over 80%."
215 ::= { prometheusAlertOsdTraps 3 }
217 prometheusAlertOsdTrapFlapOsd NOTIFICATION-TYPE
219 prometheusAlertNotificationAlertName,
220 prometheusAlertNotificationStatus,
221 prometheusAlertNotificationSeverity,
222 prometheusAlertNotificationInstance,
223 prometheusAlertNotificationJob,
224 prometheusAlertNotificationDescription,
225 prometheusAlertNotificationLabels,
226 prometheusAlertNotificationTimestamp,
227 prometheusAlertNotificationRawData
230 DESCRIPTION "An OSD was marked down at back up at least once a minute for 5 minutes."
231 ::= { prometheusAlertOsdTraps 4 }
233 prometheusAlertOsdTrapHighPgCountDeviation NOTIFICATION-TYPE
235 prometheusAlertNotificationAlertName,
236 prometheusAlertNotificationStatus,
237 prometheusAlertNotificationSeverity,
238 prometheusAlertNotificationInstance,
239 prometheusAlertNotificationJob,
240 prometheusAlertNotificationDescription,
241 prometheusAlertNotificationLabels,
242 prometheusAlertNotificationTimestamp,
243 prometheusAlertNotificationRawData
246 DESCRIPTION "An OSD deviates by more then 30% from average PG count."
247 ::= { prometheusAlertOsdTraps 5 }
249 prometheusAlertPgsTrapPgsInactive NOTIFICATION-TYPE
251 prometheusAlertNotificationAlertName,
252 prometheusAlertNotificationStatus,
253 prometheusAlertNotificationSeverity,
254 prometheusAlertNotificationInstance,
255 prometheusAlertNotificationJob,
256 prometheusAlertNotificationDescription,
257 prometheusAlertNotificationLabels,
258 prometheusAlertNotificationTimestamp,
259 prometheusAlertNotificationRawData
262 DESCRIPTION "One or more PGs are inactive for more than 5 minutes."
263 ::= { prometheusAlertPgsTraps 1 }
265 prometheusAlertPgsTrapPgsUnclean NOTIFICATION-TYPE
267 prometheusAlertNotificationAlertName,
268 prometheusAlertNotificationStatus,
269 prometheusAlertNotificationSeverity,
270 prometheusAlertNotificationInstance,
271 prometheusAlertNotificationJob,
272 prometheusAlertNotificationDescription,
273 prometheusAlertNotificationLabels,
274 prometheusAlertNotificationTimestamp,
275 prometheusAlertNotificationRawData
278 DESCRIPTION "One or more PGs are not clean for more than 15 minutes."
279 ::= { prometheusAlertPgsTraps 2 }
281 prometheusAlertNodesTrapRootVolumeFull NOTIFICATION-TYPE
283 prometheusAlertNotificationAlertName,
284 prometheusAlertNotificationStatus,
285 prometheusAlertNotificationSeverity,
286 prometheusAlertNotificationInstance,
287 prometheusAlertNotificationJob,
288 prometheusAlertNotificationDescription,
289 prometheusAlertNotificationLabels,
290 prometheusAlertNotificationTimestamp,
291 prometheusAlertNotificationRawData
294 DESCRIPTION "Root volume (OSD and MON store) is dangerously full (< 5% free)."
295 ::= { prometheusAlertNodesTraps 1 }
297 prometheusAlertNodesTrapNetworkPacketsDropped NOTIFICATION-TYPE
299 prometheusAlertNotificationAlertName,
300 prometheusAlertNotificationStatus,
301 prometheusAlertNotificationSeverity,
302 prometheusAlertNotificationInstance,
303 prometheusAlertNotificationJob,
304 prometheusAlertNotificationDescription,
305 prometheusAlertNotificationLabels,
306 prometheusAlertNotificationTimestamp,
307 prometheusAlertNotificationRawData
310 DESCRIPTION "A node experiences packet drop > 1 packet/s on an interface."
311 ::= { prometheusAlertNodesTraps 2 }
313 prometheusAlertNodesTrapNetworkPacketErrors NOTIFICATION-TYPE
315 prometheusAlertNotificationAlertName,
316 prometheusAlertNotificationStatus,
317 prometheusAlertNotificationSeverity,
318 prometheusAlertNotificationInstance,
319 prometheusAlertNotificationJob,
320 prometheusAlertNotificationDescription,
321 prometheusAlertNotificationLabels,
322 prometheusAlertNotificationTimestamp,
323 prometheusAlertNotificationRawData
326 DESCRIPTION "A node experiences packet errors > 1 packet/s on an interface."
327 ::= { prometheusAlertNodesTraps 3 }
329 prometheusAlertNodesTrapStorageFilling NOTIFICATION-TYPE
331 prometheusAlertNotificationAlertName,
332 prometheusAlertNotificationStatus,
333 prometheusAlertNotificationSeverity,
334 prometheusAlertNotificationInstance,
335 prometheusAlertNotificationJob,
336 prometheusAlertNotificationDescription,
337 prometheusAlertNotificationLabels,
338 prometheusAlertNotificationTimestamp,
339 prometheusAlertNotificationRawData
342 DESCRIPTION "A mountpoint will be full in less then 5 days assuming the average fillup rate of the past 48 hours."
343 ::= { prometheusAlertNodesTraps 4 }
345 prometheusAlertPoolsTrapPoolFull NOTIFICATION-TYPE
347 prometheusAlertNotificationAlertName,
348 prometheusAlertNotificationStatus,
349 prometheusAlertNotificationSeverity,
350 prometheusAlertNotificationInstance,
351 prometheusAlertNotificationJob,
352 prometheusAlertNotificationDescription,
353 prometheusAlertNotificationLabels,
354 prometheusAlertNotificationTimestamp,
355 prometheusAlertNotificationRawData
358 DESCRIPTION "A pool is at 90% capacity or over."
359 ::= { prometheusAlertPoolsTraps 1 }
361 prometheusAlertPoolsTrapPoolFillingUp NOTIFICATION-TYPE
363 prometheusAlertNotificationAlertName,
364 prometheusAlertNotificationStatus,
365 prometheusAlertNotificationSeverity,
366 prometheusAlertNotificationInstance,
367 prometheusAlertNotificationJob,
368 prometheusAlertNotificationDescription,
369 prometheusAlertNotificationLabels,
370 prometheusAlertNotificationTimestamp,
371 prometheusAlertNotificationRawData
374 DESCRIPTION "A pool will be full in less then 5 days assuming the average fillup rate of the past 48 hours."
375 ::= { prometheusAlertPoolsTraps 2 }