| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package cassandra |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | |
| 8 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 9 | ) |
| 10 | |
| 11 | const ( |
| 12 | prioClientRequestsRate = collectorapi.Priority + iota |
| 13 | |
| 14 | prioClientRequestReadLatency |
| 15 | prioClientRequestWriteLatency |
| 16 | prioClientRequestsLatency |
| 17 | |
| 18 | prioKeyCacheHitRatio |
| 19 | prioRowCacheHitRatio |
| 20 | prioKeyCacheHitRate |
| 21 | prioRowCacheHitRate |
| 22 | prioKeyCacheUtilization |
| 23 | prioRowCacheUtilization |
| 24 | prioKeyCacheSize |
| 25 | prioRowCacheSize |
| 26 | |
| 27 | prioStorageLiveDiskSpaceUsed |
| 28 | |
| 29 | prioCompactionCompletedTasksRate |
| 30 | prioCompactionPendingTasksCount |
| 31 | prioCompactionBytesCompactedRate |
| 32 | |
| 33 | prioThreadPoolActiveTasksCount |
| 34 | prioThreadPoolPendingTasksCount |
| 35 | prioThreadPoolBlockedTasksCount |
| 36 | prioThreadPoolBlockedTasksRate |
| 37 | |
| 38 | prioJVMMemoryUsed |
| 39 | prioJVMGCCount |
| 40 | prioJVMGCTime |
| 41 | |
| 42 | prioDroppedMessagesRate |
| 43 | prioRequestsTimeoutsRate |
| 44 | prioRequestsUnavailablesRate |
| 45 | prioRequestsFailuresRate |
| 46 | prioStorageExceptionsRate |
| 47 | ) |
| 48 | |
| 49 | var baseCharts = collectorapi.Charts{ |
| 50 | chartClientRequestsRate.Copy(), |
| 51 | |
| 52 | chartClientRequestsLatency.Copy(), |
| 53 | chartClientRequestReadLatencyHistogram.Copy(), |
| 54 | chartClientRequestWriteLatencyHistogram.Copy(), |
| 55 | |
| 56 | chartKeyCacheHitRatio.Copy(), |
| 57 | chartRowCacheHitRatio.Copy(), |
| 58 | chartKeyCacheHitRate.Copy(), |
| 59 | chartRowCacheHitRate.Copy(), |
| 60 | chartKeyCacheUtilization.Copy(), |
| 61 | chartRowCacheUtilization.Copy(), |
| 62 | chartKeyCacheSize.Copy(), |
| 63 | chartRowCacheSize.Copy(), |
| 64 | |
| 65 | chartStorageLiveDiskSpaceUsed.Copy(), |
| 66 | |
| 67 | chartCompactionCompletedTasksRate.Copy(), |
| 68 | chartCompactionPendingTasksCount.Copy(), |
| 69 | chartCompactionBytesCompactedRate.Copy(), |
| 70 | |
| 71 | chartJVMMemoryUsed.Copy(), |
| 72 | chartJVMGCRate.Copy(), |
| 73 | chartJVMGCTime.Copy(), |
| 74 | |
| 75 | chartDroppedMessagesRate.Copy(), |
| 76 | chartClientRequestTimeoutsRate.Copy(), |
| 77 | chartClientRequestUnavailablesRate.Copy(), |
| 78 | chartClientRequestFailuresRate.Copy(), |
| 79 | chartStorageExceptionsRate.Copy(), |
| 80 | } |
| 81 | |
| 82 | var ( |
| 83 | chartClientRequestsRate = collectorapi.Chart{ |
| 84 | ID: "client_requests_rate", |
| 85 | Title: "Client requests rate", |
| 86 | Units: "requests/s", |
| 87 | Fam: "throughput", |
| 88 | Ctx: "cassandra.client_requests_rate", |
| 89 | Priority: prioClientRequestsRate, |
| 90 | Dims: collectorapi.Dims{ |
| 91 | {ID: "client_request_latency_reads", Name: "read", Algo: collectorapi.Incremental}, |
| 92 | {ID: "client_request_latency_writes", Name: "write", Algo: collectorapi.Incremental, Mul: -1}, |
| 93 | }, |
| 94 | } |
| 95 | ) |
| 96 | |
| 97 | var ( |
| 98 | chartClientRequestReadLatencyHistogram = collectorapi.Chart{ |
| 99 | ID: "client_request_read_latency_histogram", |
| 100 | Title: "Client request read latency histogram", |
| 101 | Units: "seconds", |
| 102 | Fam: "latency", |
| 103 | Ctx: "cassandra.client_request_read_latency_histogram", |
| 104 | Priority: prioClientRequestReadLatency, |
| 105 | Dims: collectorapi.Dims{ |
| 106 | {ID: "client_request_read_latency_p50", Name: "p50", Div: 1e6}, |
| 107 | {ID: "client_request_read_latency_p75", Name: "p75", Div: 1e6}, |
| 108 | {ID: "client_request_read_latency_p95", Name: "p95", Div: 1e6}, |
| 109 | {ID: "client_request_read_latency_p98", Name: "p98", Div: 1e6}, |
| 110 | {ID: "client_request_read_latency_p99", Name: "p99", Div: 1e6}, |
| 111 | {ID: "client_request_read_latency_p999", Name: "p999", Div: 1e6}, |
| 112 | }, |
| 113 | } |
| 114 | chartClientRequestWriteLatencyHistogram = collectorapi.Chart{ |
| 115 | ID: "client_request_write_latency_histogram", |
| 116 | Title: "Client request write latency histogram", |
| 117 | Units: "seconds", |
| 118 | Fam: "latency", |
| 119 | Ctx: "cassandra.client_request_write_latency_histogram", |
| 120 | Priority: prioClientRequestWriteLatency, |
| 121 | Dims: collectorapi.Dims{ |
| 122 | {ID: "client_request_write_latency_p50", Name: "p50", Div: 1e6}, |
| 123 | {ID: "client_request_write_latency_p75", Name: "p75", Div: 1e6}, |
| 124 | {ID: "client_request_write_latency_p95", Name: "p95", Div: 1e6}, |
| 125 | {ID: "client_request_write_latency_p98", Name: "p98", Div: 1e6}, |
| 126 | {ID: "client_request_write_latency_p99", Name: "p99", Div: 1e6}, |
| 127 | {ID: "client_request_write_latency_p999", Name: "p999", Div: 1e6}, |
| 128 | }, |
| 129 | } |
| 130 | chartClientRequestsLatency = collectorapi.Chart{ |
| 131 | ID: "client_requests_latency", |
| 132 | Title: "Client requests total latency", |
| 133 | Units: "seconds", |
| 134 | Fam: "latency", |
| 135 | Ctx: "cassandra.client_requests_latency", |
| 136 | Priority: prioClientRequestsLatency, |
| 137 | Dims: collectorapi.Dims{ |
| 138 | {ID: "client_request_total_latency_reads", Name: "read", Algo: collectorapi.Incremental, Div: 1e6}, |
| 139 | {ID: "client_request_total_latency_writes", Name: "write", Algo: collectorapi.Incremental, Div: 1e6}, |
| 140 | }, |
| 141 | } |
| 142 | ) |
| 143 | |
| 144 | var ( |
| 145 | chartKeyCacheHitRatio = collectorapi.Chart{ |
| 146 | ID: "key_cache_hit_ratio", |
| 147 | Title: "Key cache hit ratio", |
| 148 | Units: "percentage", |
| 149 | Fam: "cache", |
| 150 | Ctx: "cassandra.key_cache_hit_ratio", |
| 151 | Priority: prioKeyCacheHitRatio, |
| 152 | Dims: collectorapi.Dims{ |
| 153 | {ID: "key_cache_hit_ratio", Name: "hit_ratio", Div: 1000}, |
| 154 | }, |
| 155 | } |
| 156 | chartKeyCacheHitRate = collectorapi.Chart{ |
| 157 | ID: "key_cache_hit_rate", |
| 158 | Title: "Key cache hit rate", |
| 159 | Units: "events/s", |
| 160 | Fam: "cache", |
| 161 | Ctx: "cassandra.key_cache_hit_rate", |
| 162 | Priority: prioKeyCacheHitRate, |
| 163 | Type: collectorapi.Stacked, |
| 164 | Dims: collectorapi.Dims{ |
| 165 | {ID: "key_cache_hits", Name: "hits", Algo: collectorapi.Incremental}, |
| 166 | {ID: "key_cache_misses", Name: "misses", Algo: collectorapi.Incremental}, |
| 167 | }, |
| 168 | } |
| 169 | chartKeyCacheUtilization = collectorapi.Chart{ |
| 170 | ID: "key_cache_utilization", |
| 171 | Title: "Key cache utilization", |
| 172 | Units: "percentage", |
| 173 | Fam: "cache", |
| 174 | Ctx: "cassandra.key_cache_utilization", |
| 175 | Priority: prioKeyCacheUtilization, |
| 176 | Dims: collectorapi.Dims{ |
| 177 | {ID: "key_cache_utilization", Name: "used", Div: 1000}, |
| 178 | }, |
| 179 | } |
| 180 | chartKeyCacheSize = collectorapi.Chart{ |
| 181 | ID: "key_cache_size", |
| 182 | Title: "Key cache size", |
| 183 | Units: "bytes", |
| 184 | Fam: "cache", |
| 185 | Ctx: "cassandra.key_cache_size", |
| 186 | Priority: prioKeyCacheSize, |
| 187 | Dims: collectorapi.Dims{ |
| 188 | {ID: "key_cache_size", Name: "size"}, |
| 189 | }, |
| 190 | } |
| 191 | |
| 192 | chartRowCacheHitRatio = collectorapi.Chart{ |
| 193 | ID: "row_cache_hit_ratio", |
| 194 | Title: "Row cache hit ratio", |
| 195 | Units: "percentage", |
| 196 | Fam: "cache", |
| 197 | Ctx: "cassandra.row_cache_hit_ratio", |
| 198 | Priority: prioRowCacheHitRatio, |
| 199 | Dims: collectorapi.Dims{ |
| 200 | {ID: "row_cache_hit_ratio", Name: "hit_ratio", Div: 1000}, |
| 201 | }, |
| 202 | } |
| 203 | chartRowCacheHitRate = collectorapi.Chart{ |
| 204 | ID: "row_cache_hit_rate", |
| 205 | Title: "Row cache hit rate", |
| 206 | Units: "events/s", |
| 207 | Fam: "cache", |
| 208 | Ctx: "cassandra.row_cache_hit_rate", |
| 209 | Priority: prioRowCacheHitRate, |
| 210 | Type: collectorapi.Stacked, |
| 211 | Dims: collectorapi.Dims{ |
| 212 | {ID: "row_cache_hits", Name: "hits", Algo: collectorapi.Incremental}, |
| 213 | {ID: "row_cache_misses", Name: "misses", Algo: collectorapi.Incremental}, |
| 214 | }, |
| 215 | } |
| 216 | chartRowCacheUtilization = collectorapi.Chart{ |
| 217 | ID: "row_cache_utilization", |
| 218 | Title: "Row cache utilization", |
| 219 | Units: "percentage", |
| 220 | Fam: "cache", |
| 221 | Ctx: "cassandra.row_cache_utilization", |
| 222 | Priority: prioRowCacheUtilization, |
| 223 | Dims: collectorapi.Dims{ |
| 224 | {ID: "row_cache_utilization", Name: "used", Div: 1000}, |
| 225 | }, |
| 226 | } |
| 227 | chartRowCacheSize = collectorapi.Chart{ |
| 228 | ID: "row_cache_size", |
| 229 | Title: "Row cache size", |
| 230 | Units: "bytes", |
| 231 | Fam: "cache", |
| 232 | Ctx: "cassandra.row_cache_size", |
| 233 | Priority: prioRowCacheSize, |
| 234 | Dims: collectorapi.Dims{ |
| 235 | {ID: "row_cache_size", Name: "size"}, |
| 236 | }, |
| 237 | } |
| 238 | ) |
| 239 | |
| 240 | var ( |
| 241 | chartStorageLiveDiskSpaceUsed = collectorapi.Chart{ |
| 242 | ID: "storage_live_disk_space_used", |
| 243 | Title: "Disk space used by live data", |
| 244 | Units: "bytes", |
| 245 | Fam: "disk usage", |
| 246 | Ctx: "cassandra.storage_live_disk_space_used", |
| 247 | Priority: prioStorageLiveDiskSpaceUsed, |
| 248 | Dims: collectorapi.Dims{ |
| 249 | {ID: "storage_load", Name: "used"}, |
| 250 | }, |
| 251 | } |
| 252 | ) |
| 253 | |
| 254 | var ( |
| 255 | chartCompactionCompletedTasksRate = collectorapi.Chart{ |
| 256 | ID: "compaction_completed_tasks_rate", |
| 257 | Title: "Completed compactions rate", |
| 258 | Units: "tasks/s", |
| 259 | Fam: "compaction", |
| 260 | Ctx: "cassandra.compaction_completed_tasks_rate", |
| 261 | Priority: prioCompactionCompletedTasksRate, |
| 262 | Dims: collectorapi.Dims{ |
| 263 | {ID: "compaction_completed_tasks", Name: "completed", Algo: collectorapi.Incremental}, |
| 264 | }, |
| 265 | } |
| 266 | chartCompactionPendingTasksCount = collectorapi.Chart{ |
| 267 | ID: "compaction_pending_tasks_count", |
| 268 | Title: "Pending compactions", |
| 269 | Units: "tasks", |
| 270 | Fam: "compaction", |
| 271 | Ctx: "cassandra.compaction_pending_tasks_count", |
| 272 | Priority: prioCompactionPendingTasksCount, |
| 273 | Dims: collectorapi.Dims{ |
| 274 | {ID: "compaction_pending_tasks", Name: "pending"}, |
| 275 | }, |
| 276 | } |
| 277 | chartCompactionBytesCompactedRate = collectorapi.Chart{ |
| 278 | ID: "compaction_compacted_rate", |
| 279 | Title: "Compaction data rate", |
| 280 | Units: "bytes/s", |
| 281 | Fam: "compaction", |
| 282 | Ctx: "cassandra.compaction_compacted_rate", |
| 283 | Priority: prioCompactionBytesCompactedRate, |
| 284 | Dims: collectorapi.Dims{ |
| 285 | {ID: "compaction_bytes_compacted", Name: "compacted", Algo: collectorapi.Incremental}, |
| 286 | }, |
| 287 | } |
| 288 | ) |
| 289 | |
| 290 | var ( |
| 291 | chartsTmplThreadPool = collectorapi.Charts{ |
| 292 | chartTmplThreadPoolActiveTasksCount.Copy(), |
| 293 | chartTmplThreadPoolPendingTasksCount.Copy(), |
| 294 | chartTmplThreadPoolBlockedTasksCount.Copy(), |
| 295 | chartTmplThreadPoolBlockedTasksRate.Copy(), |
| 296 | } |
| 297 | |
| 298 | chartTmplThreadPoolActiveTasksCount = collectorapi.Chart{ |
| 299 | ID: "thread_pool_%s_active_tasks_count", |
| 300 | Title: "Active tasks", |
| 301 | Units: "tasks", |
| 302 | Fam: "thread pools", |
| 303 | Ctx: "cassandra.thread_pool_active_tasks_count", |
| 304 | Priority: prioThreadPoolActiveTasksCount, |
| 305 | Dims: collectorapi.Dims{ |
| 306 | {ID: "thread_pool_%s_active_tasks", Name: "active"}, |
| 307 | }, |
| 308 | } |
| 309 | chartTmplThreadPoolPendingTasksCount = collectorapi.Chart{ |
| 310 | ID: "thread_pool_%s_pending_tasks_count", |
| 311 | Title: "Pending tasks", |
| 312 | Units: "tasks", |
| 313 | Fam: "thread pools", |
| 314 | Ctx: "cassandra.thread_pool_pending_tasks_count", |
| 315 | Priority: prioThreadPoolPendingTasksCount, |
| 316 | Dims: collectorapi.Dims{ |
| 317 | {ID: "thread_pool_%s_pending_tasks", Name: "pending"}, |
| 318 | }, |
| 319 | } |
| 320 | chartTmplThreadPoolBlockedTasksCount = collectorapi.Chart{ |
| 321 | ID: "thread_pool_%s_blocked_tasks_count", |
| 322 | Title: "Blocked tasks", |
| 323 | Units: "tasks", |
| 324 | Fam: "thread pools", |
| 325 | Ctx: "cassandra.thread_pool_blocked_tasks_count", |
| 326 | Priority: prioThreadPoolBlockedTasksCount, |
| 327 | Dims: collectorapi.Dims{ |
| 328 | {ID: "thread_pool_%s_blocked_tasks", Name: "blocked"}, |
| 329 | }, |
| 330 | } |
| 331 | chartTmplThreadPoolBlockedTasksRate = collectorapi.Chart{ |
| 332 | ID: "thread_pool_%s_blocked_tasks_rate", |
| 333 | Title: "Blocked tasks rate", |
| 334 | Units: "tasks/s", |
| 335 | Fam: "thread pools", |
| 336 | Ctx: "cassandra.thread_pool_blocked_tasks_rate", |
| 337 | Priority: prioThreadPoolBlockedTasksRate, |
| 338 | Dims: collectorapi.Dims{ |
| 339 | {ID: "thread_pool_%s_total_blocked_tasks", Name: "blocked", Algo: collectorapi.Incremental}, |
| 340 | }, |
| 341 | } |
| 342 | ) |
| 343 | |
| 344 | var ( |
| 345 | chartJVMMemoryUsed = collectorapi.Chart{ |
| 346 | ID: "jvm_memory_used", |
| 347 | Title: "Memory used", |
| 348 | Units: "bytes", |
| 349 | Fam: "jvm runtime", |
| 350 | Ctx: "cassandra.jvm_memory_used", |
| 351 | Priority: prioJVMMemoryUsed, |
| 352 | Type: collectorapi.Stacked, |
| 353 | Dims: collectorapi.Dims{ |
| 354 | {ID: "jvm_memory_heap_used", Name: "heap"}, |
| 355 | {ID: "jvm_memory_nonheap_used", Name: "nonheap"}, |
| 356 | }, |
| 357 | } |
| 358 | chartJVMGCRate = collectorapi.Chart{ |
| 359 | ID: "jvm_gc_rate", |
| 360 | Title: "Garbage collections rate", |
| 361 | Units: "gc/s", |
| 362 | Fam: "jvm runtime", |
| 363 | Ctx: "cassandra.jvm_gc_rate", |
| 364 | Priority: prioJVMGCCount, |
| 365 | Dims: collectorapi.Dims{ |
| 366 | {ID: "jvm_gc_parnew_count", Name: "parnew", Algo: collectorapi.Incremental}, |
| 367 | {ID: "jvm_gc_cms_count", Name: "cms", Algo: collectorapi.Incremental}, |
| 368 | }, |
| 369 | } |
| 370 | chartJVMGCTime = collectorapi.Chart{ |
| 371 | ID: "jvm_gc_time", |
| 372 | Title: "Garbage collection time", |
| 373 | Units: "seconds", |
| 374 | Fam: "jvm runtime", |
| 375 | Ctx: "cassandra.jvm_gc_time", |
| 376 | Priority: prioJVMGCTime, |
| 377 | Dims: collectorapi.Dims{ |
| 378 | {ID: "jvm_gc_parnew_time", Name: "parnew", Algo: collectorapi.Incremental, Div: 1e9}, |
| 379 | {ID: "jvm_gc_cms_time", Name: "cms", Algo: collectorapi.Incremental, Div: 1e9}, |
| 380 | }, |
| 381 | } |
| 382 | ) |
| 383 | |
| 384 | var ( |
| 385 | chartDroppedMessagesRate = collectorapi.Chart{ |
| 386 | ID: "dropped_messages_rate", |
| 387 | Title: "Dropped messages rate", |
| 388 | Units: "messages/s", |
| 389 | Fam: "errors", |
| 390 | Ctx: "cassandra.dropped_messages_rate", |
| 391 | Priority: prioDroppedMessagesRate, |
| 392 | Dims: collectorapi.Dims{ |
| 393 | {ID: "dropped_messages", Name: "dropped"}, |
| 394 | }, |
| 395 | } |
| 396 | chartClientRequestTimeoutsRate = collectorapi.Chart{ |
| 397 | ID: "client_requests_timeouts_rate", |
| 398 | Title: "Client requests timeouts rate", |
| 399 | Units: "timeouts/s", |
| 400 | Fam: "errors", |
| 401 | Ctx: "cassandra.client_requests_timeouts_rate", |
| 402 | Priority: prioRequestsTimeoutsRate, |
| 403 | Dims: collectorapi.Dims{ |
| 404 | {ID: "client_request_timeouts_reads", Name: "read", Algo: collectorapi.Incremental}, |
| 405 | {ID: "client_request_timeouts_writes", Name: "write", Algo: collectorapi.Incremental, Mul: -1}, |
| 406 | }, |
| 407 | } |
| 408 | chartClientRequestUnavailablesRate = collectorapi.Chart{ |
| 409 | ID: "client_requests_unavailables_rate", |
| 410 | Title: "Client requests unavailable exceptions rate", |
| 411 | Units: "exceptions/s", |
| 412 | Fam: "errors", |
| 413 | Ctx: "cassandra.client_requests_unavailables_rate", |
| 414 | Priority: prioRequestsUnavailablesRate, |
| 415 | Dims: collectorapi.Dims{ |
| 416 | {ID: "client_request_unavailables_reads", Name: "read", Algo: collectorapi.Incremental}, |
| 417 | {ID: "client_request_unavailables_writes", Name: "write", Algo: collectorapi.Incremental, Mul: -1}, |
| 418 | }, |
| 419 | } |
| 420 | chartClientRequestFailuresRate = collectorapi.Chart{ |
| 421 | ID: "client_requests_failures_rate", |
| 422 | Title: "Client requests failures rate", |
| 423 | Units: "failures/s", |
| 424 | Fam: "errors", |
| 425 | Ctx: "cassandra.client_requests_failures_rate", |
| 426 | Priority: prioRequestsFailuresRate, |
| 427 | Dims: collectorapi.Dims{ |
| 428 | {ID: "client_request_failures_reads", Name: "read", Algo: collectorapi.Incremental}, |
| 429 | {ID: "client_request_failures_writes", Name: "write", Algo: collectorapi.Incremental, Mul: -1}, |
| 430 | }, |
| 431 | } |
| 432 | chartStorageExceptionsRate = collectorapi.Chart{ |
| 433 | ID: "storage_exceptions_rate", |
| 434 | Title: "Storage exceptions rate", |
| 435 | Units: "exceptions/s", |
| 436 | Fam: "errors", |
| 437 | Ctx: "cassandra.storage_exceptions_rate", |
| 438 | Priority: prioStorageExceptionsRate, |
| 439 | Dims: collectorapi.Dims{ |
| 440 | {ID: "storage_exceptions", Name: "storage", Algo: collectorapi.Incremental}, |
| 441 | }, |
| 442 | } |
| 443 | ) |
| 444 | |
| 445 | func (c *Collector) addThreadPoolCharts(pool *threadPoolMetrics) { |
| 446 | charts := chartsTmplThreadPool.Copy() |
| 447 | |
| 448 | for _, chart := range *charts { |
| 449 | chart.ID = fmt.Sprintf(chart.ID, pool.name) |
| 450 | chart.Labels = []collectorapi.Label{ |
| 451 | {Key: "thread_pool", Value: pool.name}, |
| 452 | } |
| 453 | for _, dim := range chart.Dims { |
| 454 | dim.ID = fmt.Sprintf(dim.ID, pool.name) |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | if err := c.Charts().Add(*charts...); err != nil { |
| 459 | c.Warning(err) |
| 460 | } |
| 461 | } |