ceph/monitoring/snmp/CEPH-PROMETHEUS-ALERT-MIB.txt

   1 CEPH-PROMETHEUS-ALERT-MIB DEFINITIONS ::= BEGIN
   2
   3 IMPORTS
   4     MODULE-IDENTITY, OBJECT-TYPE, NOTIFICATION-TYPE, enterprises, TimeTicks
   5         FROM SNMPv2-SMI
   6     DisplayString
   7         FROM SNMPv2-TC
   8 ;
   9
  10 ceph       OBJECT IDENTIFIER ::= { enterprises 50495 }
  11 prometheus OBJECT IDENTIFIER ::= { ceph 15 }
  12
  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."
  20     ::= { prometheus 1 }
  21
  22 prometheusAlertObjects OBJECT IDENTIFIER ::= { prometheusAlert 1 }
  23 prometheusAlertTraps   OBJECT IDENTIFIER ::= { prometheusAlert 2 }
  24
  25 --
  26 -- Objects
  27 --
  28
  29 prometheusAlertNotificationAlertName OBJECT-TYPE
  30     SYNTAX      DisplayString
  31     MAX-ACCESS  accessible-for-notify
  32     STATUS      current
  33     DESCRIPTION "The name of the Prometheus alert."
  34 ::= { prometheusAlertObjects 1 }
  35
  36 prometheusAlertNotificationStatus OBJECT-TYPE
  37     SYNTAX      DisplayString
  38     MAX-ACCESS  accessible-for-notify
  39     STATUS      current
  40     DESCRIPTION "The status of the Prometheus alert."
  41 ::= { prometheusAlertObjects 2 }
  42
  43 prometheusAlertNotificationSeverity OBJECT-TYPE
  44     SYNTAX      DisplayString
  45     MAX-ACCESS  accessible-for-notify
  46     STATUS      current
  47     DESCRIPTION "The severity of the Prometheus alert."
  48 ::= { prometheusAlertObjects 3 }
  49
  50 prometheusAlertNotificationInstance OBJECT-TYPE
  51     SYNTAX      DisplayString
  52     MAX-ACCESS  accessible-for-notify
  53     STATUS      current
  54     DESCRIPTION "Unique identifier for the Prometheus instance."
  55 ::= { prometheusAlertObjects 4 }
  56
  57 prometheusAlertNotificationJob OBJECT-TYPE
  58     SYNTAX      DisplayString
  59     MAX-ACCESS  accessible-for-notify
  60     STATUS      current
  61     DESCRIPTION "The name of the Prometheus job."
  62 ::= { prometheusAlertObjects 5 }
  63
  64 prometheusAlertNotificationDescription OBJECT-TYPE
  65     SYNTAX      DisplayString
  66     MAX-ACCESS  accessible-for-notify
  67     STATUS      current
  68     DESCRIPTION "The Prometheus alert description field."
  69 ::= { prometheusAlertObjects 6 }
  70
  71 prometheusAlertNotificationLabels OBJECT-TYPE
  72     SYNTAX      DisplayString
  73     MAX-ACCESS  accessible-for-notify
  74     STATUS      current
  75     DESCRIPTION "Additional Prometheus alert labels as JSON string."
  76 ::= { prometheusAlertObjects 7 }
  77
  78 prometheusAlertNotificationTimestamp OBJECT-TYPE
  79     SYNTAX      TimeTicks
  80     MAX-ACCESS  accessible-for-notify
  81     STATUS      current
  82     DESCRIPTION "The time when the Prometheus alert occurred."
  83 ::= { prometheusAlertObjects 8 }
  84
  85 prometheusAlertNotificationRawData OBJECT-TYPE
  86     SYNTAX      DisplayString
  87     MAX-ACCESS  accessible-for-notify
  88     STATUS      current
  89     DESCRIPTION "The raw Prometheus alert as JSON string."
  90 ::= { prometheusAlertObjects 9 }
  91
  92 --
  93 -- Traps
  94 --
  95
  96 prometheusAlertTrapDefault NOTIFICATION-TYPE
  97     OBJECTS     {
  98                     prometheusAlertNotificationAlertName,
  99                     prometheusAlertNotificationStatus,
 100                     prometheusAlertNotificationSeverity,
 101                     prometheusAlertNotificationInstance,
 102                     prometheusAlertNotificationJob,
 103                     prometheusAlertNotificationDescription,
 104                     prometheusAlertNotificationLabels,
 105                     prometheusAlertNotificationTimestamp,
 106                     prometheusAlertNotificationRawData
 107                 }
 108     STATUS      current
 109     DESCRIPTION "Default trap."
 110 ::= { prometheusAlertTraps 1 }
 111
 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 }
 120
 121 prometheusAlertClusterHealthTrapHealthError NOTIFICATION-TYPE
 122     OBJECTS     {
 123                     prometheusAlertNotificationAlertName,
 124                     prometheusAlertNotificationStatus,
 125                     prometheusAlertNotificationSeverity,
 126                     prometheusAlertNotificationInstance,
 127                     prometheusAlertNotificationJob,
 128                     prometheusAlertNotificationDescription,
 129                     prometheusAlertNotificationLabels,
 130                     prometheusAlertNotificationTimestamp,
 131                     prometheusAlertNotificationRawData
 132                 }
 133     STATUS      current
 134     DESCRIPTION "Ceph in health_error state for more than 5m."
 135 ::= { prometheusAlertClusterHealthTraps 1 }
 136
 137 prometheusAlertClusterHealthTrapHealthWarn NOTIFICATION-TYPE
 138     OBJECTS     {
 139                     prometheusAlertNotificationAlertName,
 140                     prometheusAlertNotificationStatus,
 141                     prometheusAlertNotificationSeverity,
 142                     prometheusAlertNotificationInstance,
 143                     prometheusAlertNotificationJob,
 144                     prometheusAlertNotificationDescription,
 145                     prometheusAlertNotificationLabels,
 146                     prometheusAlertNotificationTimestamp,
 147                     prometheusAlertNotificationRawData
 148                 }
 149     STATUS      current
 150     DESCRIPTION "Ceph in health_warn for more than 15m."
 151 ::= { prometheusAlertClusterHealthTraps 2 }
 152
 153 prometheusAlertMonTrapLowMonitorQuorumCount NOTIFICATION-TYPE
 154     OBJECTS     {
 155                     prometheusAlertNotificationAlertName,
 156                     prometheusAlertNotificationStatus,
 157                     prometheusAlertNotificationSeverity,
 158                     prometheusAlertNotificationInstance,
 159                     prometheusAlertNotificationJob,
 160                     prometheusAlertNotificationDescription,
 161                     prometheusAlertNotificationLabels,
 162                     prometheusAlertNotificationTimestamp,
 163                     prometheusAlertNotificationRawData
 164                 }
 165     STATUS      current
 166     DESCRIPTION "Monitor count in quorum is low."
 167 ::= { prometheusAlertMonTraps 1 }
 168
 169 prometheusAlertOsdTrap10PercentOsdsDown NOTIFICATION-TYPE
 170     OBJECTS     {
 171                     prometheusAlertNotificationAlertName,
 172                     prometheusAlertNotificationStatus,
 173                     prometheusAlertNotificationSeverity,
 174                     prometheusAlertNotificationInstance,
 175                     prometheusAlertNotificationJob,
 176                     prometheusAlertNotificationDescription,
 177                     prometheusAlertNotificationLabels,
 178                     prometheusAlertNotificationTimestamp,
 179                     prometheusAlertNotificationRawData
 180                 }
 181     STATUS      current
 182     DESCRIPTION "More than 10% of OSDs are down."
 183 ::= { prometheusAlertOsdTraps 1 }
 184
 185 prometheusAlertOsdTrapOsdDown NOTIFICATION-TYPE
 186     OBJECTS     {
 187                     prometheusAlertNotificationAlertName,
 188                     prometheusAlertNotificationStatus,
 189                     prometheusAlertNotificationSeverity,
 190                     prometheusAlertNotificationInstance,
 191                     prometheusAlertNotificationJob,
 192                     prometheusAlertNotificationDescription,
 193                     prometheusAlertNotificationLabels,
 194                     prometheusAlertNotificationTimestamp,
 195                     prometheusAlertNotificationRawData
 196                 }
 197     STATUS      current
 198     DESCRIPTION "One or more OSDs down for more than 15 minutes."
 199 ::= { prometheusAlertOsdTraps 2 }
 200
 201 prometheusAlertOsdTrapOsdsNearFull NOTIFICATION-TYPE
 202     OBJECTS     {
 203                     prometheusAlertNotificationAlertName,
 204                     prometheusAlertNotificationStatus,
 205                     prometheusAlertNotificationSeverity,
 206                     prometheusAlertNotificationInstance,
 207                     prometheusAlertNotificationJob,
 208                     prometheusAlertNotificationDescription,
 209                     prometheusAlertNotificationLabels,
 210                     prometheusAlertNotificationTimestamp,
 211                     prometheusAlertNotificationRawData
 212                 }
 213     STATUS      current
 214     DESCRIPTION "An OSD is dangerously full, over 80%."
 215 ::= { prometheusAlertOsdTraps 3 }
 216
 217 prometheusAlertOsdTrapFlapOsd NOTIFICATION-TYPE
 218     OBJECTS     {
 219                     prometheusAlertNotificationAlertName,
 220                     prometheusAlertNotificationStatus,
 221                     prometheusAlertNotificationSeverity,
 222                     prometheusAlertNotificationInstance,
 223                     prometheusAlertNotificationJob,
 224                     prometheusAlertNotificationDescription,
 225                     prometheusAlertNotificationLabels,
 226                     prometheusAlertNotificationTimestamp,
 227                     prometheusAlertNotificationRawData
 228                 }
 229     STATUS      current
 230     DESCRIPTION "An OSD was marked down at back up at least once a minute for 5 minutes."
 231 ::= { prometheusAlertOsdTraps 4 }
 232
 233 prometheusAlertOsdTrapHighPgCountDeviation NOTIFICATION-TYPE
 234     OBJECTS     {
 235                     prometheusAlertNotificationAlertName,
 236                     prometheusAlertNotificationStatus,
 237                     prometheusAlertNotificationSeverity,
 238                     prometheusAlertNotificationInstance,
 239                     prometheusAlertNotificationJob,
 240                     prometheusAlertNotificationDescription,
 241                     prometheusAlertNotificationLabels,
 242                     prometheusAlertNotificationTimestamp,
 243                     prometheusAlertNotificationRawData
 244                 }
 245     STATUS      current
 246     DESCRIPTION "An OSD deviates by more then 30% from average PG count."
 247 ::= { prometheusAlertOsdTraps 5 }
 248
 249 prometheusAlertPgsTrapPgsInactive NOTIFICATION-TYPE
 250     OBJECTS     {
 251                     prometheusAlertNotificationAlertName,
 252                     prometheusAlertNotificationStatus,
 253                     prometheusAlertNotificationSeverity,
 254                     prometheusAlertNotificationInstance,
 255                     prometheusAlertNotificationJob,
 256                     prometheusAlertNotificationDescription,
 257                     prometheusAlertNotificationLabels,
 258                     prometheusAlertNotificationTimestamp,
 259                     prometheusAlertNotificationRawData
 260                 }
 261     STATUS      current
 262     DESCRIPTION "One or more PGs are inactive for more than 5 minutes."
 263 ::= { prometheusAlertPgsTraps 1 }
 264
 265 prometheusAlertPgsTrapPgsUnclean NOTIFICATION-TYPE
 266     OBJECTS     {
 267                     prometheusAlertNotificationAlertName,
 268                     prometheusAlertNotificationStatus,
 269                     prometheusAlertNotificationSeverity,
 270                     prometheusAlertNotificationInstance,
 271                     prometheusAlertNotificationJob,
 272                     prometheusAlertNotificationDescription,
 273                     prometheusAlertNotificationLabels,
 274                     prometheusAlertNotificationTimestamp,
 275                     prometheusAlertNotificationRawData
 276                 }
 277     STATUS      current
 278     DESCRIPTION "One or more PGs are not clean for more than 15 minutes."
 279 ::= { prometheusAlertPgsTraps 2 }
 280
 281 prometheusAlertNodesTrapRootVolumeFull NOTIFICATION-TYPE
 282     OBJECTS     {
 283                     prometheusAlertNotificationAlertName,
 284                     prometheusAlertNotificationStatus,
 285                     prometheusAlertNotificationSeverity,
 286                     prometheusAlertNotificationInstance,
 287                     prometheusAlertNotificationJob,
 288                     prometheusAlertNotificationDescription,
 289                     prometheusAlertNotificationLabels,
 290                     prometheusAlertNotificationTimestamp,
 291                     prometheusAlertNotificationRawData
 292                 }
 293     STATUS      current
 294     DESCRIPTION "Root volume (OSD and MON store) is dangerously full (< 5% free)."
 295 ::= { prometheusAlertNodesTraps 1 }
 296
 297 prometheusAlertNodesTrapNetworkPacketsDropped NOTIFICATION-TYPE
 298     OBJECTS     {
 299                     prometheusAlertNotificationAlertName,
 300                     prometheusAlertNotificationStatus,
 301                     prometheusAlertNotificationSeverity,
 302                     prometheusAlertNotificationInstance,
 303                     prometheusAlertNotificationJob,
 304                     prometheusAlertNotificationDescription,
 305                     prometheusAlertNotificationLabels,
 306                     prometheusAlertNotificationTimestamp,
 307                     prometheusAlertNotificationRawData
 308                 }
 309     STATUS      current
 310     DESCRIPTION "A node experiences packet drop > 1 packet/s on an interface."
 311 ::= { prometheusAlertNodesTraps 2 }
 312
 313 prometheusAlertNodesTrapNetworkPacketErrors NOTIFICATION-TYPE
 314     OBJECTS     {
 315                     prometheusAlertNotificationAlertName,
 316                     prometheusAlertNotificationStatus,
 317                     prometheusAlertNotificationSeverity,
 318                     prometheusAlertNotificationInstance,
 319                     prometheusAlertNotificationJob,
 320                     prometheusAlertNotificationDescription,
 321                     prometheusAlertNotificationLabels,
 322                     prometheusAlertNotificationTimestamp,
 323                     prometheusAlertNotificationRawData
 324                 }
 325     STATUS      current
 326     DESCRIPTION "A node experiences packet errors > 1 packet/s on an interface."
 327 ::= { prometheusAlertNodesTraps 3 }
 328
 329 prometheusAlertNodesTrapStorageFilling NOTIFICATION-TYPE
 330     OBJECTS     {
 331                     prometheusAlertNotificationAlertName,
 332                     prometheusAlertNotificationStatus,
 333                     prometheusAlertNotificationSeverity,
 334                     prometheusAlertNotificationInstance,
 335                     prometheusAlertNotificationJob,
 336                     prometheusAlertNotificationDescription,
 337                     prometheusAlertNotificationLabels,
 338                     prometheusAlertNotificationTimestamp,
 339                     prometheusAlertNotificationRawData
 340                 }
 341     STATUS      current
 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 }
 344
 345 prometheusAlertPoolsTrapPoolFull NOTIFICATION-TYPE
 346     OBJECTS     {
 347                     prometheusAlertNotificationAlertName,
 348                     prometheusAlertNotificationStatus,
 349                     prometheusAlertNotificationSeverity,
 350                     prometheusAlertNotificationInstance,
 351                     prometheusAlertNotificationJob,
 352                     prometheusAlertNotificationDescription,
 353                     prometheusAlertNotificationLabels,
 354                     prometheusAlertNotificationTimestamp,
 355                     prometheusAlertNotificationRawData
 356                 }
 357     STATUS      current
 358     DESCRIPTION "A pool is at 90% capacity or over."
 359 ::= { prometheusAlertPoolsTraps 1 }
 360
 361 prometheusAlertPoolsTrapPoolFillingUp NOTIFICATION-TYPE
 362     OBJECTS     {
 363                     prometheusAlertNotificationAlertName,
 364                     prometheusAlertNotificationStatus,
 365                     prometheusAlertNotificationSeverity,
 366                     prometheusAlertNotificationInstance,
 367                     prometheusAlertNotificationJob,
 368                     prometheusAlertNotificationDescription,
 369                     prometheusAlertNotificationLabels,
 370                     prometheusAlertNotificationTimestamp,
 371                     prometheusAlertNotificationRawData
 372                 }
 373     STATUS      current
 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 }
 376
 377 END