| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package cockroachdb |
| 4 | |
| 5 | // Architecture Overview |
| 6 | // https://www.cockroachlabs.com/docs/stable/architecture/overview.html |
| 7 | |
| 8 | // Web Dashboards |
| 9 | // https://github.com/cockroachdb/cockroach/tree/master/pkg/ui/src/views/cluster/containers/nodeGraphs/dashboards |
| 10 | |
| 11 | // Process |
| 12 | const ( |
| 13 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/server/status/runtime.go |
| 14 | metricSysCPUUserNs = "sys_cpu_user_ns" |
| 15 | metricSysCPUSysNs = "sys_cpu_sys_ns" |
| 16 | metricSysCPUUserPercent = "sys_cpu_user_percent" |
| 17 | metricSysCPUSysPercent = "sys_cpu_sys_percent" |
| 18 | metricSysCPUCombinedPercentNormalized = "sys_cpu_combined_percent_normalized" |
| 19 | metricSysRSS = "sys_rss" |
| 20 | metricSysFDOpen = "sys_fd_open" |
| 21 | metricSysFDSoftLimit = "sys_fd_softlimit" |
| 22 | metricSysUptime = "sys_uptime" |
| 23 | ) |
| 24 | |
| 25 | // Host Disk/Network Cumulative |
| 26 | const ( |
| 27 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/server/status/runtime.go |
| 28 | metricSysHostDiskReadBytes = "sys_host_disk_read_bytes" |
| 29 | metricSysHostDiskWriteBytes = "sys_host_disk_write_bytes" |
| 30 | metricSysHostDiskReadCount = "sys_host_disk_read_count" |
| 31 | metricSysHostDiskWriteCount = "sys_host_disk_write_count" |
| 32 | metricSysHostDiskIOPSInProgress = "sys_host_disk_iopsinprogress" |
| 33 | metricSysHostNetSendBytes = "sys_host_net_send_bytes" |
| 34 | metricSysHostNetRecvBytes = "sys_host_net_recv_bytes" |
| 35 | metricSysHostNetSendPackets = "sys_host_net_send_packets" |
| 36 | metricSysHostNetRecvPackets = "sys_host_net_recv_packets" |
| 37 | ) |
| 38 | |
| 39 | // Liveness |
| 40 | const ( |
| 41 | //https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/node_liveness.go |
| 42 | metricLiveNodes = "liveness_livenodes" |
| 43 | metricHeartBeatSuccesses = "liveness_heartbeatsuccesses" |
| 44 | metricHeartBeatFailures = "liveness_heartbeatfailures" |
| 45 | ) |
| 46 | |
| 47 | // Capacity |
| 48 | const ( |
| 49 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 50 | metricCapacity = "capacity" |
| 51 | metricCapacityAvailable = "capacity_available" |
| 52 | metricCapacityUsed = "capacity_used" |
| 53 | //metricCapacityReserved = "capacity_reserved" |
| 54 | ) |
| 55 | |
| 56 | // SQL |
| 57 | const ( |
| 58 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/sql/pgwire/server.go |
| 59 | metricSQLConnections = "sql_conns" |
| 60 | metricSQLBytesIn = "sql_bytesin" |
| 61 | metricSQLBytesOut = "sql_bytesout" |
| 62 | |
| 63 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/sql/exec_util.go |
| 64 | // Started Statements |
| 65 | metricSQLQueryStartedCount = "sql_query_started_count" // Cumulative (Statements + Transaction Statements) |
| 66 | metricSQLSelectStartedCount = "sql_select_started_count" |
| 67 | metricSQLUpdateStartedCount = "sql_update_started_count" |
| 68 | metricSQLInsertStartedCount = "sql_insert_started_count" |
| 69 | metricSQLDeleteStartedCount = "sql_delete_started_count" |
| 70 | metricSQLSavepointStartedCount = "sql_savepoint_started_count" |
| 71 | metricSQLRestartSavepointStartedCount = "sql_restart_savepoint_started_count" |
| 72 | metricSQLRestartSavepointReleaseStartedCount = "sql_restart_savepoint_release_started_count" |
| 73 | metricSQLRestartSavepointRollbackStartedCount = "sql_restart_savepoint_rollback_started_count" |
| 74 | metricSQLDDLStartedCount = "sql_ddl_started_count" |
| 75 | metricSQLMiscStartedCount = "sql_misc_started_count" |
| 76 | // Started Transaction Statements |
| 77 | metricSQLTXNBeginStartedCount = "sql_txn_begin_started_count" |
| 78 | metricSQLTXNCommitStartedCount = "sql_txn_commit_started_count" |
| 79 | metricSQLTXNRollbackStartedCount = "sql_txn_rollback_started_count" |
| 80 | |
| 81 | // Executed Statements |
| 82 | metricSQLQueryCount = "sql_query_count" // Cumulative (Statements + Transaction Statements) |
| 83 | metricSQLSelectCount = "sql_select_count" |
| 84 | metricSQLUpdateCount = "sql_update_count" |
| 85 | metricSQLInsertCount = "sql_insert_count" |
| 86 | metricSQLDeleteCount = "sql_delete_count" |
| 87 | metricSQLSavepointCount = "sql_savepoint_count" |
| 88 | metricSQLRestartSavepointCount = "sql_restart_savepoint_count" |
| 89 | metricSQLRestartSavepointReleaseCount = "sql_restart_savepoint_release_count" |
| 90 | metricSQLRestartSavepointRollbackCount = "sql_restart_savepoint_rollback_count" |
| 91 | metricSQLDDLCount = "sql_ddl_count" |
| 92 | metricSQLMiscCount = "sql_misc_count" |
| 93 | // Executed Transaction statements |
| 94 | metricSQLTXNBeginCount = "sql_txn_begin_count" |
| 95 | metricSQLTXNCommitCount = "sql_txn_commit_count" |
| 96 | metricSQLTXNRollbackCount = "sql_txn_rollback_count" |
| 97 | |
| 98 | // Statements Resulted In An Error |
| 99 | metricSQLFailureCount = "sql_failure_count" |
| 100 | // Transaction Resulted In Abort Errors |
| 101 | metricSQLTXNAbortCount = "sql_txn_abort_count" |
| 102 | |
| 103 | // Distributed SQL |
| 104 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/sql/execinfra/metrics.go |
| 105 | metricSQLDistSQLQueriesActive = "sql_distsql_queries_active" |
| 106 | metricSQLDistSQLFlowsActive = "sql_distsql_flows_active" |
| 107 | metricSQLDistSQLFlowsQueued = "sql_distsql_flows_queued" |
| 108 | ) |
| 109 | |
| 110 | // Storage |
| 111 | const ( |
| 112 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 113 | metricLiveBytes = "livebytes" |
| 114 | metricSysBytes = "sysbytes" |
| 115 | metricKeyBytes = "keybytes" |
| 116 | metricValBytes = "valbytes" |
| 117 | metricKeyCount = "keycount" |
| 118 | metricValCount = "valcount" |
| 119 | ) |
| 120 | |
| 121 | // KV Transactions |
| 122 | const ( |
| 123 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/kv/txn_metrics.go |
| 124 | metricTxnCommits = "txn_commits" |
| 125 | metricTxnCommits1PC = "txn_commits1PC" |
| 126 | metricTxnAborts = "txn_aborts" |
| 127 | metricTxnRestartsWriteTooOld = "txn_restarts_writetooold" |
| 128 | metricTxnRestartsWriteTooOldMulti = "txn_restarts_writetoooldmulti" |
| 129 | metricTxnRestartsSerializable = "txn_restarts_serializable" |
| 130 | metricTxnRestartsPossibleReplay = "txn_restarts_possiblereplay" |
| 131 | metricTxnRestartsAsyncWriteFailure = "txn_restarts_asyncwritefailure" |
| 132 | metricTxnRestartsReadWithInUncertainty = "txn_restarts_readwithinuncertainty" |
| 133 | metricTxnRestartsTxnAborted = "txn_restarts_txnaborted" |
| 134 | metricTxnRestartsTxnPush = "txn_restarts_txnpush" |
| 135 | metricTxnRestartsUnknown = "txn_restarts_unknown" |
| 136 | ) |
| 137 | |
| 138 | // Ranges |
| 139 | const ( |
| 140 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 141 | metricRanges = "ranges" |
| 142 | metricRangesUnavailable = "ranges_unavailable" |
| 143 | metricRangesUnderReplicated = "ranges_underreplicated" |
| 144 | metricRangesOverReplicated = "ranges_overreplicated" |
| 145 | // Range Events Metrics |
| 146 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 147 | metricRangeSplits = "range_splits" |
| 148 | metricRangeAdds = "range_adds" |
| 149 | metricRangeRemoves = "range_removes" |
| 150 | metricRangeMerges = "range_merges" |
| 151 | metricRangeSnapshotsGenerated = "range_snapshots_generated" |
| 152 | metricRangeSnapshotsPreemptiveApplied = "range_snapshots_preemptive_applied" |
| 153 | metricRangeSnapshotsLearnerApplied = "range_snapshots_learner_applied" |
| 154 | metricRangeSnapshotsNormalApplied = "range_snapshots_normal_applied" |
| 155 | ) |
| 156 | |
| 157 | // RocksDB |
| 158 | const ( |
| 159 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 160 | metricRocksDBReadAmplification = "rocksdb_read_amplification" |
| 161 | metricRocksDBNumSSTables = "rocksdb_num_sstables" |
| 162 | metricRocksDBBlockCacheUsage = "rocksdb_block_cache_usage" |
| 163 | metricRocksDBBlockCacheHits = "rocksdb_block_cache_hits" |
| 164 | metricRocksDBBlockCacheMisses = "rocksdb_block_cache_misses" |
| 165 | metricRocksDBCompactions = "rocksdb_compactions" |
| 166 | metricRocksDBFlushes = "rocksdb_flushes" |
| 167 | ) |
| 168 | |
| 169 | // Replication |
| 170 | const ( |
| 171 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 172 | metricReplicas = "replicas" |
| 173 | // metricReplicasReserved = "replicas_reserved" |
| 174 | metricReplicasLeaders = "replicas_leaders" |
| 175 | metricReplicasLeadersNotLeaseholders = "replicas_leaders_not_leaseholders" |
| 176 | metricReplicasLeaseholders = "replicas_leaseholders" |
| 177 | metricReplicasQuiescent = "replicas_quiescent" |
| 178 | ) |
| 179 | |
| 180 | // Queues |
| 181 | const ( |
| 182 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 183 | metricQueueGCProcessFailure = "queue_gc_process_failure" |
| 184 | metricQueueReplicaGCProcessFailure = "queue_replicagc_process_failure" |
| 185 | metricQueueReplicateProcessFailure = "queue_replicate_process_failure" |
| 186 | metricQueueSplitProcessFailure = "queue_split_process_failure" |
| 187 | metricQueueConsistencyProcessFailure = "queue_consistency_process_failure" |
| 188 | metricQueueRaftLogProcessFailure = "queue_raftlog_process_failure" |
| 189 | metricQueueRaftSnapshotProcessFailure = "queue_raftsnapshot_process_failure" |
| 190 | metricQueueTSMaintenanceProcessFailure = "queue_tsmaintenance_process_failure" |
| 191 | ) |
| 192 | |
| 193 | // Rebalancing |
| 194 | const ( |
| 195 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 196 | metricRebalancingQueriesPerSecond = "rebalancing_queriespersecond" |
| 197 | metricRebalancingWritesPerSecond = "rebalancing_writespersecond" |
| 198 | ) |
| 199 | |
| 200 | // Slow Requests |
| 201 | const ( |
| 202 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/metrics.go |
| 203 | metricRequestsSlowLease = "requests_slow_lease" |
| 204 | metricRequestsSlowLatch = "requests_slow_latch" |
| 205 | metricRequestsSlowRaft = "requests_slow_raft" |
| 206 | ) |
| 207 | |
| 208 | // Time Series |
| 209 | const ( |
| 210 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/ts/metrics.go |
| 211 | metricTimeSeriesWriteSamples = "timeseries_write_samples" |
| 212 | metricTimeSeriesWriteErrors = "timeseries_write_errors" |
| 213 | metricTimeSeriesWriteBytes = "timeseries_write_bytes" |
| 214 | ) |
| 215 | |
| 216 | // Go/Cgo |
| 217 | const ( |
| 218 | // https://github.com/cockroachdb/cockroach/blob/master/pkg/server/status/runtime.go |
| 219 | metricSysGoAllocBytes = "sys_go_allocbytes" |
| 220 | metricSysCGoAllocBytes = "sys_cgo_allocbytes" |
| 221 | metricSysCGoCalls = "sys_cgocalls" |
| 222 | metricSysGoroutines = "sys_goroutines" |
| 223 | metricSysGCCount = "sys_gc_count" |
| 224 | metricSysGCPauseNs = "sys_gc_pause_ns" |
| 225 | ) |
| 226 | |
| 227 | const ( |
| 228 | // Calculated Metrics |
| 229 | metricCapacityUsable = "capacity_usable" |
| 230 | metricCapacityUnusable = "capacity_unusable" |
| 231 | metricCapacityUsedPercentage = "capacity_used_percent" |
| 232 | metricCapacityUsableUsedPercentage = "capacity_usable_used_percent" |
| 233 | metricRocksDBBlockCacheHitRate = "rocksdb_block_cache_hit_rate" |
| 234 | metricReplicasActive = "replicas_active" |
| 235 | ) |
| 236 | |
| 237 | var metrics = []string{ |
| 238 | metricSysCPUUserNs, |
| 239 | metricSysCPUSysNs, |
| 240 | metricSysCPUUserPercent, |
| 241 | metricSysCPUSysPercent, |
| 242 | metricSysCPUCombinedPercentNormalized, |
| 243 | metricSysRSS, |
| 244 | metricSysFDOpen, |
| 245 | metricSysFDSoftLimit, |
| 246 | metricSysUptime, |
| 247 | |
| 248 | metricSysHostDiskReadBytes, |
| 249 | metricSysHostDiskWriteBytes, |
| 250 | metricSysHostDiskReadCount, |
| 251 | metricSysHostDiskWriteCount, |
| 252 | metricSysHostDiskIOPSInProgress, |
| 253 | metricSysHostNetSendBytes, |
| 254 | metricSysHostNetRecvBytes, |
| 255 | metricSysHostNetSendPackets, |
| 256 | metricSysHostNetRecvPackets, |
| 257 | |
| 258 | metricLiveNodes, |
| 259 | metricHeartBeatSuccesses, |
| 260 | metricHeartBeatFailures, |
| 261 | |
| 262 | metricCapacity, |
| 263 | metricCapacityAvailable, |
| 264 | metricCapacityUsed, |
| 265 | |
| 266 | metricSQLConnections, |
| 267 | metricSQLBytesIn, |
| 268 | metricSQLBytesOut, |
| 269 | metricSQLQueryStartedCount, |
| 270 | metricSQLSelectStartedCount, |
| 271 | metricSQLUpdateStartedCount, |
| 272 | metricSQLInsertStartedCount, |
| 273 | metricSQLDeleteStartedCount, |
| 274 | metricSQLSavepointStartedCount, |
| 275 | metricSQLRestartSavepointStartedCount, |
| 276 | metricSQLRestartSavepointReleaseStartedCount, |
| 277 | metricSQLRestartSavepointRollbackStartedCount, |
| 278 | metricSQLDDLStartedCount, |
| 279 | metricSQLMiscStartedCount, |
| 280 | metricSQLTXNBeginStartedCount, |
| 281 | metricSQLTXNCommitStartedCount, |
| 282 | metricSQLTXNRollbackStartedCount, |
| 283 | metricSQLQueryCount, |
| 284 | metricSQLSelectCount, |
| 285 | metricSQLUpdateCount, |
| 286 | metricSQLInsertCount, |
| 287 | metricSQLDeleteCount, |
| 288 | metricSQLSavepointCount, |
| 289 | metricSQLRestartSavepointCount, |
| 290 | metricSQLRestartSavepointReleaseCount, |
| 291 | metricSQLRestartSavepointRollbackCount, |
| 292 | metricSQLDDLCount, |
| 293 | metricSQLMiscCount, |
| 294 | metricSQLTXNBeginCount, |
| 295 | metricSQLTXNCommitCount, |
| 296 | metricSQLTXNRollbackCount, |
| 297 | metricSQLFailureCount, |
| 298 | metricSQLTXNAbortCount, |
| 299 | metricSQLDistSQLQueriesActive, |
| 300 | metricSQLDistSQLFlowsActive, |
| 301 | metricSQLDistSQLFlowsQueued, |
| 302 | |
| 303 | metricLiveBytes, |
| 304 | metricSysBytes, |
| 305 | metricKeyBytes, |
| 306 | metricValBytes, |
| 307 | metricKeyCount, |
| 308 | metricValCount, |
| 309 | |
| 310 | metricTxnCommits, |
| 311 | metricTxnCommits1PC, |
| 312 | metricTxnAborts, |
| 313 | metricTxnRestartsWriteTooOld, |
| 314 | metricTxnRestartsWriteTooOldMulti, |
| 315 | metricTxnRestartsSerializable, |
| 316 | metricTxnRestartsPossibleReplay, |
| 317 | metricTxnRestartsAsyncWriteFailure, |
| 318 | metricTxnRestartsReadWithInUncertainty, |
| 319 | metricTxnRestartsTxnAborted, |
| 320 | metricTxnRestartsTxnPush, |
| 321 | metricTxnRestartsUnknown, |
| 322 | |
| 323 | metricRanges, |
| 324 | metricRangesUnavailable, |
| 325 | metricRangesUnderReplicated, |
| 326 | metricRangesOverReplicated, |
| 327 | metricRangeSplits, |
| 328 | metricRangeAdds, |
| 329 | metricRangeRemoves, |
| 330 | metricRangeMerges, |
| 331 | metricRangeSnapshotsGenerated, |
| 332 | metricRangeSnapshotsPreemptiveApplied, |
| 333 | metricRangeSnapshotsLearnerApplied, |
| 334 | metricRangeSnapshotsNormalApplied, |
| 335 | |
| 336 | metricRocksDBReadAmplification, |
| 337 | metricRocksDBNumSSTables, |
| 338 | metricRocksDBBlockCacheUsage, |
| 339 | metricRocksDBBlockCacheHits, |
| 340 | metricRocksDBBlockCacheMisses, |
| 341 | metricRocksDBCompactions, |
| 342 | metricRocksDBFlushes, |
| 343 | |
| 344 | metricReplicas, |
| 345 | metricReplicasLeaders, |
| 346 | metricReplicasLeadersNotLeaseholders, |
| 347 | metricReplicasLeaseholders, |
| 348 | metricReplicasQuiescent, |
| 349 | |
| 350 | metricQueueGCProcessFailure, |
| 351 | metricQueueReplicaGCProcessFailure, |
| 352 | metricQueueReplicateProcessFailure, |
| 353 | metricQueueSplitProcessFailure, |
| 354 | metricQueueConsistencyProcessFailure, |
| 355 | metricQueueRaftLogProcessFailure, |
| 356 | metricQueueRaftSnapshotProcessFailure, |
| 357 | metricQueueTSMaintenanceProcessFailure, |
| 358 | |
| 359 | metricRebalancingQueriesPerSecond, |
| 360 | metricRebalancingWritesPerSecond, |
| 361 | |
| 362 | metricTimeSeriesWriteSamples, |
| 363 | metricTimeSeriesWriteErrors, |
| 364 | metricTimeSeriesWriteBytes, |
| 365 | |
| 366 | metricRequestsSlowLease, |
| 367 | metricRequestsSlowLatch, |
| 368 | metricRequestsSlowRaft, |
| 369 | |
| 370 | metricSysGoAllocBytes, |
| 371 | metricSysCGoAllocBytes, |
| 372 | metricSysCGoCalls, |
| 373 | metricSysGoroutines, |
| 374 | metricSysGCCount, |
| 375 | metricSysGCPauseNs, |
| 376 | } |