| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package cassandra |
| 4 | |
| 5 | import ( |
| 6 | "errors" |
| 7 | "strings" |
| 8 | |
| 9 | "github.com/netdata/netdata/go/plugins/pkg/prometheus" |
| 10 | ) |
| 11 | |
| 12 | const ( |
| 13 | suffixCount = "_count" |
| 14 | suffixValue = "_value" |
| 15 | ) |
| 16 | |
| 17 | func (c *Collector) collect() (map[string]int64, error) { |
| 18 | pms, err := c.prom.ScrapeSeries() |
| 19 | if err != nil { |
| 20 | return nil, err |
| 21 | } |
| 22 | |
| 23 | if c.validateMetrics { |
| 24 | if !isCassandraMetrics(pms) { |
| 25 | return nil, errors.New("collected metrics aren't Collector metrics") |
| 26 | } |
| 27 | c.validateMetrics = false |
| 28 | } |
| 29 | |
| 30 | mx := make(map[string]int64) |
| 31 | |
| 32 | c.resetMetrics() |
| 33 | c.collectMetrics(pms) |
| 34 | c.processMetric(mx) |
| 35 | |
| 36 | return mx, nil |
| 37 | } |
| 38 | |
| 39 | func (c *Collector) resetMetrics() { |
| 40 | cm := newCassandraMetrics() |
| 41 | for key, p := range c.mx.threadPools { |
| 42 | cm.threadPools[key] = &threadPoolMetrics{ |
| 43 | name: p.name, |
| 44 | hasCharts: p.hasCharts, |
| 45 | } |
| 46 | } |
| 47 | c.mx = cm |
| 48 | } |
| 49 | |
| 50 | func (c *Collector) processMetric(mx map[string]int64) { |
| 51 | c.mx.clientReqTotalLatencyReads.write(mx, "client_request_total_latency_reads") |
| 52 | c.mx.clientReqTotalLatencyWrites.write(mx, "client_request_total_latency_writes") |
| 53 | c.mx.clientReqLatencyReads.write(mx, "client_request_latency_reads") |
| 54 | c.mx.clientReqLatencyWrites.write(mx, "client_request_latency_writes") |
| 55 | c.mx.clientReqTimeoutsReads.write(mx, "client_request_timeouts_reads") |
| 56 | c.mx.clientReqTimeoutsWrites.write(mx, "client_request_timeouts_writes") |
| 57 | c.mx.clientReqUnavailablesReads.write(mx, "client_request_unavailables_reads") |
| 58 | c.mx.clientReqUnavailablesWrites.write(mx, "client_request_unavailables_writes") |
| 59 | c.mx.clientReqFailuresReads.write(mx, "client_request_failures_reads") |
| 60 | c.mx.clientReqFailuresWrites.write(mx, "client_request_failures_writes") |
| 61 | |
| 62 | c.mx.clientReqReadLatencyP50.write(mx, "client_request_read_latency_p50") |
| 63 | c.mx.clientReqReadLatencyP75.write(mx, "client_request_read_latency_p75") |
| 64 | c.mx.clientReqReadLatencyP95.write(mx, "client_request_read_latency_p95") |
| 65 | c.mx.clientReqReadLatencyP98.write(mx, "client_request_read_latency_p98") |
| 66 | c.mx.clientReqReadLatencyP99.write(mx, "client_request_read_latency_p99") |
| 67 | c.mx.clientReqReadLatencyP999.write(mx, "client_request_read_latency_p999") |
| 68 | c.mx.clientReqWriteLatencyP50.write(mx, "client_request_write_latency_p50") |
| 69 | c.mx.clientReqWriteLatencyP75.write(mx, "client_request_write_latency_p75") |
| 70 | c.mx.clientReqWriteLatencyP95.write(mx, "client_request_write_latency_p95") |
| 71 | c.mx.clientReqWriteLatencyP98.write(mx, "client_request_write_latency_p98") |
| 72 | c.mx.clientReqWriteLatencyP99.write(mx, "client_request_write_latency_p99") |
| 73 | c.mx.clientReqWriteLatencyP999.write(mx, "client_request_write_latency_p999") |
| 74 | |
| 75 | c.mx.rowCacheHits.write(mx, "row_cache_hits") |
| 76 | c.mx.rowCacheMisses.write(mx, "row_cache_misses") |
| 77 | c.mx.rowCacheSize.write(mx, "row_cache_size") |
| 78 | if c.mx.rowCacheHits.isSet && c.mx.rowCacheMisses.isSet { |
| 79 | if s := c.mx.rowCacheHits.value + c.mx.rowCacheMisses.value; s > 0 { |
| 80 | mx["row_cache_hit_ratio"] = int64((c.mx.rowCacheHits.value * 100 / s) * 1000) |
| 81 | } else { |
| 82 | mx["row_cache_hit_ratio"] = 0 |
| 83 | } |
| 84 | } |
| 85 | if c.mx.rowCacheCapacity.isSet && c.mx.rowCacheSize.isSet { |
| 86 | if s := c.mx.rowCacheCapacity.value; s > 0 { |
| 87 | mx["row_cache_utilization"] = int64((c.mx.rowCacheSize.value * 100 / s) * 1000) |
| 88 | } else { |
| 89 | mx["row_cache_utilization"] = 0 |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | c.mx.keyCacheHits.write(mx, "key_cache_hits") |
| 94 | c.mx.keyCacheMisses.write(mx, "key_cache_misses") |
| 95 | c.mx.keyCacheSize.write(mx, "key_cache_size") |
| 96 | if c.mx.keyCacheHits.isSet && c.mx.keyCacheMisses.isSet { |
| 97 | if s := c.mx.keyCacheHits.value + c.mx.keyCacheMisses.value; s > 0 { |
| 98 | mx["key_cache_hit_ratio"] = int64((c.mx.keyCacheHits.value * 100 / s) * 1000) |
| 99 | } else { |
| 100 | mx["key_cache_hit_ratio"] = 0 |
| 101 | } |
| 102 | } |
| 103 | if c.mx.keyCacheCapacity.isSet && c.mx.keyCacheSize.isSet { |
| 104 | if s := c.mx.keyCacheCapacity.value; s > 0 { |
| 105 | mx["key_cache_utilization"] = int64((c.mx.keyCacheSize.value * 100 / s) * 1000) |
| 106 | } else { |
| 107 | mx["key_cache_utilization"] = 0 |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | c.mx.droppedMessages.write1k(mx, "dropped_messages") |
| 112 | |
| 113 | c.mx.storageLoad.write(mx, "storage_load") |
| 114 | c.mx.storageExceptions.write(mx, "storage_exceptions") |
| 115 | |
| 116 | c.mx.compactionBytesCompacted.write(mx, "compaction_bytes_compacted") |
| 117 | c.mx.compactionPendingTasks.write(mx, "compaction_pending_tasks") |
| 118 | c.mx.compactionCompletedTasks.write(mx, "compaction_completed_tasks") |
| 119 | |
| 120 | c.mx.jvmMemoryHeapUsed.write(mx, "jvm_memory_heap_used") |
| 121 | c.mx.jvmMemoryNonHeapUsed.write(mx, "jvm_memory_nonheap_used") |
| 122 | c.mx.jvmGCParNewCount.write(mx, "jvm_gc_parnew_count") |
| 123 | c.mx.jvmGCParNewTime.write1k(mx, "jvm_gc_parnew_time") |
| 124 | c.mx.jvmGCCMSCount.write(mx, "jvm_gc_cms_count") |
| 125 | c.mx.jvmGCCMSTime.write1k(mx, "jvm_gc_cms_time") |
| 126 | |
| 127 | for _, p := range c.mx.threadPools { |
| 128 | if !p.hasCharts { |
| 129 | p.hasCharts = true |
| 130 | c.addThreadPoolCharts(p) |
| 131 | } |
| 132 | |
| 133 | px := "thread_pool_" + p.name + "_" |
| 134 | p.activeTasks.write(mx, px+"active_tasks") |
| 135 | p.pendingTasks.write(mx, px+"pending_tasks") |
| 136 | p.blockedTasks.write(mx, px+"blocked_tasks") |
| 137 | p.totalBlockedTasks.write(mx, px+"total_blocked_tasks") |
| 138 | } |
| 139 | } |
| 140 | |
| 141 | func (c *Collector) collectMetrics(pms prometheus.Series) { |
| 142 | c.collectClientRequestMetrics(pms) |
| 143 | c.collectDroppedMessagesMetrics(pms) |
| 144 | c.collectThreadPoolsMetrics(pms) |
| 145 | c.collectStorageMetrics(pms) |
| 146 | c.collectCacheMetrics(pms) |
| 147 | c.collectJVMMetrics(pms) |
| 148 | c.collectCompactionMetrics(pms) |
| 149 | } |
| 150 | |
| 151 | func (c *Collector) collectClientRequestMetrics(pms prometheus.Series) { |
| 152 | const metric = "org_apache_cassandra_metrics_clientrequest" |
| 153 | |
| 154 | var rw struct{ read, write *metricValue } |
| 155 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 156 | name := pm.Labels.Get("name") |
| 157 | scope := pm.Labels.Get("scope") |
| 158 | |
| 159 | switch name { |
| 160 | case "TotalLatency": |
| 161 | rw.read, rw.write = &c.mx.clientReqTotalLatencyReads, &c.mx.clientReqTotalLatencyWrites |
| 162 | case "Latency": |
| 163 | rw.read, rw.write = &c.mx.clientReqLatencyReads, &c.mx.clientReqLatencyWrites |
| 164 | case "Timeouts": |
| 165 | rw.read, rw.write = &c.mx.clientReqTimeoutsReads, &c.mx.clientReqTimeoutsWrites |
| 166 | case "Unavailables": |
| 167 | rw.read, rw.write = &c.mx.clientReqUnavailablesReads, &c.mx.clientReqUnavailablesWrites |
| 168 | case "Failures": |
| 169 | rw.read, rw.write = &c.mx.clientReqFailuresReads, &c.mx.clientReqFailuresWrites |
| 170 | default: |
| 171 | continue |
| 172 | } |
| 173 | |
| 174 | switch scope { |
| 175 | case "Read": |
| 176 | rw.read.add(pm.Value) |
| 177 | case "Write": |
| 178 | rw.write.add(pm.Value) |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | rw = struct{ read, write *metricValue }{} |
| 183 | |
| 184 | for _, pm := range pms.FindByNames( |
| 185 | metric+"_50thpercentile", |
| 186 | metric+"_75thpercentile", |
| 187 | metric+"_95thpercentile", |
| 188 | metric+"_98thpercentile", |
| 189 | metric+"_99thpercentile", |
| 190 | metric+"_999thpercentile", |
| 191 | ) { |
| 192 | name := pm.Labels.Get("name") |
| 193 | scope := pm.Labels.Get("scope") |
| 194 | |
| 195 | if name != "Latency" { |
| 196 | continue |
| 197 | } |
| 198 | |
| 199 | switch { |
| 200 | case strings.HasSuffix(pm.Name(), "_50thpercentile"): |
| 201 | rw.read, rw.write = &c.mx.clientReqReadLatencyP50, &c.mx.clientReqWriteLatencyP50 |
| 202 | case strings.HasSuffix(pm.Name(), "_75thpercentile"): |
| 203 | rw.read, rw.write = &c.mx.clientReqReadLatencyP75, &c.mx.clientReqWriteLatencyP75 |
| 204 | case strings.HasSuffix(pm.Name(), "_95thpercentile"): |
| 205 | rw.read, rw.write = &c.mx.clientReqReadLatencyP95, &c.mx.clientReqWriteLatencyP95 |
| 206 | case strings.HasSuffix(pm.Name(), "_98thpercentile"): |
| 207 | rw.read, rw.write = &c.mx.clientReqReadLatencyP98, &c.mx.clientReqWriteLatencyP98 |
| 208 | case strings.HasSuffix(pm.Name(), "_99thpercentile"): |
| 209 | rw.read, rw.write = &c.mx.clientReqReadLatencyP99, &c.mx.clientReqWriteLatencyP99 |
| 210 | case strings.HasSuffix(pm.Name(), "_999thpercentile"): |
| 211 | rw.read, rw.write = &c.mx.clientReqReadLatencyP999, &c.mx.clientReqWriteLatencyP999 |
| 212 | default: |
| 213 | continue |
| 214 | } |
| 215 | |
| 216 | switch scope { |
| 217 | case "Read": |
| 218 | rw.read.add(pm.Value) |
| 219 | case "Write": |
| 220 | rw.write.add(pm.Value) |
| 221 | } |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | func (c *Collector) collectCacheMetrics(pms prometheus.Series) { |
| 226 | const metric = "org_apache_cassandra_metrics_cache" |
| 227 | |
| 228 | var hm struct{ hits, misses *metricValue } |
| 229 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 230 | name := pm.Labels.Get("name") |
| 231 | scope := pm.Labels.Get("scope") |
| 232 | |
| 233 | switch scope { |
| 234 | case "KeyCache": |
| 235 | hm.hits, hm.misses = &c.mx.keyCacheHits, &c.mx.keyCacheMisses |
| 236 | case "RowCache": |
| 237 | hm.hits, hm.misses = &c.mx.rowCacheHits, &c.mx.rowCacheMisses |
| 238 | default: |
| 239 | continue |
| 240 | } |
| 241 | |
| 242 | switch name { |
| 243 | case "Hits": |
| 244 | hm.hits.add(pm.Value) |
| 245 | case "Misses": |
| 246 | hm.misses.add(pm.Value) |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | var cs struct{ cap, size *metricValue } |
| 251 | for _, pm := range pms.FindByName(metric + suffixValue) { |
| 252 | name := pm.Labels.Get("name") |
| 253 | scope := pm.Labels.Get("scope") |
| 254 | |
| 255 | switch scope { |
| 256 | case "KeyCache": |
| 257 | cs.cap, cs.size = &c.mx.keyCacheCapacity, &c.mx.keyCacheSize |
| 258 | case "RowCache": |
| 259 | cs.cap, cs.size = &c.mx.rowCacheCapacity, &c.mx.rowCacheSize |
| 260 | default: |
| 261 | continue |
| 262 | } |
| 263 | |
| 264 | switch name { |
| 265 | case "Capacity": |
| 266 | cs.cap.add(pm.Value) |
| 267 | case "Size": |
| 268 | cs.size.add(pm.Value) |
| 269 | } |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | func (c *Collector) collectThreadPoolsMetrics(pms prometheus.Series) { |
| 274 | const metric = "org_apache_cassandra_metrics_threadpools" |
| 275 | |
| 276 | for _, pm := range pms.FindByName(metric + suffixValue) { |
| 277 | name := pm.Labels.Get("name") |
| 278 | scope := pm.Labels.Get("scope") |
| 279 | pool := c.getThreadPoolMetrics(scope) |
| 280 | |
| 281 | switch name { |
| 282 | case "ActiveTasks": |
| 283 | pool.activeTasks.add(pm.Value) |
| 284 | case "PendingTasks": |
| 285 | pool.pendingTasks.add(pm.Value) |
| 286 | } |
| 287 | } |
| 288 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 289 | name := pm.Labels.Get("name") |
| 290 | scope := pm.Labels.Get("scope") |
| 291 | pool := c.getThreadPoolMetrics(scope) |
| 292 | |
| 293 | switch name { |
| 294 | case "CompletedTasks": |
| 295 | pool.totalBlockedTasks.add(pm.Value) |
| 296 | case "TotalBlockedTasks": |
| 297 | pool.totalBlockedTasks.add(pm.Value) |
| 298 | case "CurrentlyBlockedTasks": |
| 299 | pool.blockedTasks.add(pm.Value) |
| 300 | } |
| 301 | } |
| 302 | } |
| 303 | |
| 304 | func (c *Collector) collectStorageMetrics(pms prometheus.Series) { |
| 305 | const metric = "org_apache_cassandra_metrics_storage" |
| 306 | |
| 307 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 308 | name := pm.Labels.Get("name") |
| 309 | |
| 310 | switch name { |
| 311 | case "Load": |
| 312 | c.mx.storageLoad.add(pm.Value) |
| 313 | case "Exceptions": |
| 314 | c.mx.storageExceptions.add(pm.Value) |
| 315 | } |
| 316 | } |
| 317 | } |
| 318 | |
| 319 | func (c *Collector) collectDroppedMessagesMetrics(pms prometheus.Series) { |
| 320 | const metric = "org_apache_cassandra_metrics_droppedmessage" |
| 321 | |
| 322 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 323 | c.mx.droppedMessages.add(pm.Value) |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | func (c *Collector) collectJVMMetrics(pms prometheus.Series) { |
| 328 | const metricMemUsed = "jvm_memory_bytes_used" |
| 329 | const metricGC = "jvm_gc_collection_seconds" |
| 330 | |
| 331 | for _, pm := range pms.FindByName(metricMemUsed) { |
| 332 | area := pm.Labels.Get("area") |
| 333 | |
| 334 | switch area { |
| 335 | case "heap": |
| 336 | c.mx.jvmMemoryHeapUsed.add(pm.Value) |
| 337 | case "nonheap": |
| 338 | c.mx.jvmMemoryNonHeapUsed.add(pm.Value) |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | for _, pm := range pms.FindByName(metricGC + suffixCount) { |
| 343 | gc := pm.Labels.Get("gc") |
| 344 | |
| 345 | switch gc { |
| 346 | case "ParNew": |
| 347 | c.mx.jvmGCParNewCount.add(pm.Value) |
| 348 | case "ConcurrentMarkSweep": |
| 349 | c.mx.jvmGCCMSCount.add(pm.Value) |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | for _, pm := range pms.FindByName(metricGC + "_sum") { |
| 354 | gc := pm.Labels.Get("gc") |
| 355 | |
| 356 | switch gc { |
| 357 | case "ParNew": |
| 358 | c.mx.jvmGCParNewTime.add(pm.Value) |
| 359 | case "ConcurrentMarkSweep": |
| 360 | c.mx.jvmGCCMSTime.add(pm.Value) |
| 361 | } |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | func (c *Collector) collectCompactionMetrics(pms prometheus.Series) { |
| 366 | const metric = "org_apache_cassandra_metrics_compaction" |
| 367 | |
| 368 | for _, pm := range pms.FindByName(metric + suffixValue) { |
| 369 | name := pm.Labels.Get("name") |
| 370 | |
| 371 | switch name { |
| 372 | case "CompletedTasks": |
| 373 | c.mx.compactionCompletedTasks.add(pm.Value) |
| 374 | case "PendingTasks": |
| 375 | c.mx.compactionPendingTasks.add(pm.Value) |
| 376 | } |
| 377 | } |
| 378 | for _, pm := range pms.FindByName(metric + suffixCount) { |
| 379 | name := pm.Labels.Get("name") |
| 380 | |
| 381 | switch name { |
| 382 | case "BytesCompacted": |
| 383 | c.mx.compactionBytesCompacted.add(pm.Value) |
| 384 | } |
| 385 | } |
| 386 | } |
| 387 | |
| 388 | func (c *Collector) getThreadPoolMetrics(name string) *threadPoolMetrics { |
| 389 | pool, ok := c.mx.threadPools[name] |
| 390 | if !ok { |
| 391 | pool = &threadPoolMetrics{name: name} |
| 392 | c.mx.threadPools[name] = pool |
| 393 | } |
| 394 | return pool |
| 395 | } |
| 396 | |
| 397 | func isCassandraMetrics(pms prometheus.Series) bool { |
| 398 | for _, pm := range pms { |
| 399 | if strings.HasPrefix(pm.Name(), "org_apache_cassandra_metrics") { |
| 400 | return true |
| 401 | } |
| 402 | } |
| 403 | return false |
| 404 | } |