| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package k8s_apiserver |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | "math" |
| 8 | "sort" |
| 9 | "strings" |
| 10 | |
| 11 | "github.com/prometheus/common/model" |
| 12 | |
| 13 | "github.com/netdata/netdata/go/plugins/pkg/prometheus" |
| 14 | "github.com/netdata/netdata/go/plugins/pkg/stm" |
| 15 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 16 | mtx "github.com/netdata/netdata/go/plugins/plugin/go.d/pkg/oldmetrix" |
| 17 | ) |
| 18 | |
| 19 | const ( |
| 20 | precision = 1000 // for counters (e.g., CPU seconds) |
| 21 | |
| 22 | // For latency: multiply seconds by 1e6 to get microseconds, then chart Div: 1000 gives milliseconds |
| 23 | latencyPrecision = 1000000 |
| 24 | |
| 25 | // Default cardinality limits to prevent unbounded memory growth |
| 26 | // Once these limits are reached, new dimensions are silently ignored |
| 27 | defaultMaxResources = 500 |
| 28 | defaultMaxWorkqueues = 100 |
| 29 | defaultMaxAdmCtrl = 100 |
| 30 | defaultMaxAdmWebhooks = 50 |
| 31 | |
| 32 | // Cleanup: dimensions not seen for this many cycles are removed |
| 33 | staleThresholdCycles = 300 // ~5 minutes at 1s update interval |
| 34 | ) |
| 35 | |
| 36 | // K8s admission histogram bucket bounds (seconds) |
| 37 | // Ref: https://github.com/kubernetes/kubernetes/blob/master/staging/src/k8s.io/apiserver/pkg/endpoints/metrics/metrics.go |
| 38 | var admissionBucketBounds = []float64{0.005, 0.025, 0.1, 0.5, 1.0, 2.5} |
| 39 | |
| 40 | // idReplacer sanitizes label values for use in chart/dimension IDs |
| 41 | // Replaces characters that could cause issues in Netdata with underscores |
| 42 | var idReplacer = strings.NewReplacer( |
| 43 | ".", "_", |
| 44 | " ", "_", |
| 45 | "/", "_", |
| 46 | ":", "_", |
| 47 | ) |
| 48 | |
| 49 | // cleanID sanitizes a string for use in chart or dimension IDs |
| 50 | func cleanID(s string) string { |
| 51 | return strings.ToLower(idReplacer.Replace(s)) |
| 52 | } |
| 53 | |
| 54 | func (c *Collector) collect() (map[string]int64, error) { |
| 55 | mfs, err := c.prom.Scrape() |
| 56 | if err != nil { |
| 57 | return nil, err |
| 58 | } |
| 59 | |
| 60 | c.collectCycle++ |
| 61 | |
| 62 | mx := newMetrics() |
| 63 | |
| 64 | c.collectRequests(mfs, mx) |
| 65 | c.collectInflight(mfs, mx) |
| 66 | c.collectRESTClient(mfs, mx) |
| 67 | c.collectAdmission(mfs, mx) |
| 68 | c.collectEtcd(mfs, mx) |
| 69 | c.collectWorkqueues(mfs, mx) |
| 70 | c.collectProcess(mfs, mx) |
| 71 | c.collectAudit(mfs, mx) |
| 72 | c.collectAuth(mfs, mx) |
| 73 | |
| 74 | // Periodically cleanup stale dimensions (every 100 cycles to avoid overhead) |
| 75 | if c.collectCycle%100 == 0 { |
| 76 | c.cleanupStaleDimensions() |
| 77 | } |
| 78 | |
| 79 | return stm.ToMap(mx), nil |
| 80 | } |
| 81 | |
| 82 | // collectRequests collects apiserver_request_total, apiserver_request_duration_seconds, etc. |
| 83 | func (c *Collector) collectRequests(mfs prometheus.MetricFamilies, mx *metrics) { |
| 84 | // Total requests and by verb/code/resource |
| 85 | // Prefer apiserver_request_total (newer), fall back to apiserver_request_count (legacy) |
| 86 | mf := mfs.Get("apiserver_request_total") |
| 87 | if mf == nil { |
| 88 | mf = mfs.Get("apiserver_request_count") |
| 89 | } |
| 90 | |
| 91 | if mf != nil { |
| 92 | for _, m := range mf.Metrics() { |
| 93 | value := metricValue(mf, m) |
| 94 | if math.IsNaN(value) { |
| 95 | continue |
| 96 | } |
| 97 | |
| 98 | verb := m.Labels().Get("verb") |
| 99 | code := m.Labels().Get("code") |
| 100 | resource := m.Labels().Get("resource") |
| 101 | |
| 102 | mx.Request.Total.Add(value) |
| 103 | |
| 104 | // By verb |
| 105 | if verb != "" { |
| 106 | c.addVerbDimension(verb) |
| 107 | verbID := cleanID(verb) |
| 108 | mx.Request.ByVerb[verbID] = mtx.Gauge(mx.Request.ByVerb[verbID].Value() + value) |
| 109 | } |
| 110 | |
| 111 | // By code |
| 112 | if code != "" { |
| 113 | c.addCodeDimension(code) |
| 114 | codeID := cleanID(code) |
| 115 | mx.Request.ByCode[codeID] = mtx.Gauge(mx.Request.ByCode[codeID].Value() + value) |
| 116 | } |
| 117 | |
| 118 | // By resource (with cardinality limit) |
| 119 | if resource != "" { |
| 120 | resourceID := cleanID(resource) |
| 121 | _, seen := c.collectedResources[resourceID] |
| 122 | if seen || len(c.collectedResources) < defaultMaxResources { |
| 123 | c.addResourceDimension(resource) |
| 124 | c.collectedResources[resourceID] = c.collectCycle |
| 125 | mx.Request.ByResource[resourceID] = mtx.Gauge(mx.Request.ByResource[resourceID].Value() + value) |
| 126 | } |
| 127 | } |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | // Dropped/rejected requests - sum both legacy and APF (flow control) metrics |
| 132 | // Both can be present and meaningful on different K8s versions |
| 133 | var dropped float64 |
| 134 | if mf := mfs.Get("apiserver_dropped_requests_total"); mf != nil { |
| 135 | for _, m := range mf.Metrics() { |
| 136 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 137 | dropped += v |
| 138 | } |
| 139 | } |
| 140 | } |
| 141 | if mf := mfs.Get("apiserver_flowcontrol_rejected_requests_total"); mf != nil { |
| 142 | for _, m := range mf.Metrics() { |
| 143 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 144 | dropped += v |
| 145 | } |
| 146 | } |
| 147 | } |
| 148 | mx.Request.Dropped.Set(dropped) |
| 149 | |
| 150 | // Request latency and response size (histograms) |
| 151 | c.collectRequestLatency(mfs, mx) |
| 152 | c.collectResponseSize(mfs, mx) |
| 153 | } |
| 154 | |
| 155 | func (c *Collector) collectRequestLatency(mfs prometheus.MetricFamilies, mx *metrics) { |
| 156 | mf := mfs.Get("apiserver_request_duration_seconds") |
| 157 | if mf == nil || mf.Type() != model.MetricTypeHistogram { |
| 158 | return |
| 159 | } |
| 160 | |
| 161 | hd := collectHistogramBucketsFromMF(mf, nil) |
| 162 | if len(hd.buckets) > 0 { |
| 163 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 164 | mx.Request.Latency.P50.Set(p50 * latencyPrecision) |
| 165 | } |
| 166 | if p90 := histogramPercentile(hd, 0.9); !math.IsNaN(p90) { |
| 167 | mx.Request.Latency.P90.Set(p90 * latencyPrecision) |
| 168 | } |
| 169 | if p99 := histogramPercentile(hd, 0.99); !math.IsNaN(p99) { |
| 170 | mx.Request.Latency.P99.Set(p99 * latencyPrecision) |
| 171 | } |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | func (c *Collector) collectResponseSize(mfs prometheus.MetricFamilies, mx *metrics) { |
| 176 | mf := mfs.Get("apiserver_response_sizes") |
| 177 | if mf == nil || mf.Type() != model.MetricTypeHistogram { |
| 178 | return |
| 179 | } |
| 180 | |
| 181 | hd := collectHistogramBucketsFromMF(mf, nil) |
| 182 | if len(hd.buckets) > 0 { |
| 183 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 184 | mx.Request.ResponseSize.P50.Set(p50) |
| 185 | } |
| 186 | if p90 := histogramPercentile(hd, 0.9); !math.IsNaN(p90) { |
| 187 | mx.Request.ResponseSize.P90.Set(p90) |
| 188 | } |
| 189 | if p99 := histogramPercentile(hd, 0.99); !math.IsNaN(p99) { |
| 190 | mx.Request.ResponseSize.P99.Set(p99) |
| 191 | } |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | // collectInflight collects apiserver_current_inflight_requests and apiserver_longrunning_requests |
| 196 | func (c *Collector) collectInflight(mfs prometheus.MetricFamilies, mx *metrics) { |
| 197 | if mf := mfs.Get("apiserver_current_inflight_requests"); mf != nil { |
| 198 | for _, m := range mf.Metrics() { |
| 199 | value := metricValue(mf, m) |
| 200 | if math.IsNaN(value) { |
| 201 | continue |
| 202 | } |
| 203 | kind := m.Labels().Get("request_kind") |
| 204 | switch kind { |
| 205 | case "mutating": |
| 206 | mx.Inflight.Mutating.Set(value) |
| 207 | case "readOnly": |
| 208 | mx.Inflight.ReadOnly.Set(value) |
| 209 | } |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | // Sum all long-running requests across all label combinations |
| 214 | if mf := mfs.Get("apiserver_longrunning_requests"); mf != nil { |
| 215 | var total float64 |
| 216 | for _, m := range mf.Metrics() { |
| 217 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 218 | total += v |
| 219 | } |
| 220 | } |
| 221 | mx.Inflight.Longrunning.Set(total) |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | // collectRESTClient collects rest_client_requests_total and rest_client_request_duration_seconds |
| 226 | func (c *Collector) collectRESTClient(mfs prometheus.MetricFamilies, mx *metrics) { |
| 227 | codeChart := c.charts.Get("rest_client_requests_by_code") |
| 228 | methodChart := c.charts.Get("rest_client_requests_by_method") |
| 229 | |
| 230 | if mf := mfs.Get("rest_client_requests_total"); mf != nil { |
| 231 | for _, m := range mf.Metrics() { |
| 232 | value := metricValue(mf, m) |
| 233 | if math.IsNaN(value) { |
| 234 | continue |
| 235 | } |
| 236 | |
| 237 | code := m.Labels().Get("code") |
| 238 | method := m.Labels().Get("method") |
| 239 | |
| 240 | // By code (track for cleanup) |
| 241 | if code != "" { |
| 242 | codeID := cleanID(code) |
| 243 | _, seen := c.collectedRESTCodes[codeID] |
| 244 | c.collectedRESTCodes[codeID] = c.collectCycle |
| 245 | if !seen { |
| 246 | dimID := "rest_client_by_code_" + codeID |
| 247 | if codeChart != nil && !codeChart.HasDim(dimID) { |
| 248 | if err := codeChart.AddDim(&Dim{ID: dimID, Name: code, Algo: collectorapi.Incremental}); err != nil { |
| 249 | c.Warningf("failed to add REST client code dimension %s: %v", code, err) |
| 250 | } else { |
| 251 | codeChart.MarkNotCreated() |
| 252 | } |
| 253 | } |
| 254 | } |
| 255 | mx.RESTClient.ByCode[codeID] = mtx.Gauge(mx.RESTClient.ByCode[codeID].Value() + value) |
| 256 | } |
| 257 | |
| 258 | // By method (track for cleanup) |
| 259 | if method != "" { |
| 260 | methodID := cleanID(method) |
| 261 | _, seen := c.collectedRESTMethods[methodID] |
| 262 | c.collectedRESTMethods[methodID] = c.collectCycle |
| 263 | if !seen { |
| 264 | dimID := "rest_client_by_method_" + methodID |
| 265 | if methodChart != nil && !methodChart.HasDim(dimID) { |
| 266 | if err := methodChart.AddDim(&Dim{ID: dimID, Name: method, Algo: collectorapi.Incremental}); err != nil { |
| 267 | c.Warningf("failed to add REST client method dimension %s: %v", method, err) |
| 268 | } else { |
| 269 | methodChart.MarkNotCreated() |
| 270 | } |
| 271 | } |
| 272 | } |
| 273 | mx.RESTClient.ByMethod[methodID] = mtx.Gauge(mx.RESTClient.ByMethod[methodID].Value() + value) |
| 274 | } |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | // REST client latency |
| 279 | if mf := mfs.Get("rest_client_request_duration_seconds"); mf != nil && mf.Type() == model.MetricTypeHistogram { |
| 280 | hd := collectHistogramBucketsFromMF(mf, nil) |
| 281 | if len(hd.buckets) > 0 { |
| 282 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 283 | mx.RESTClient.Latency.P50.Set(p50 * latencyPrecision) |
| 284 | } |
| 285 | if p90 := histogramPercentile(hd, 0.9); !math.IsNaN(p90) { |
| 286 | mx.RESTClient.Latency.P90.Set(p90 * latencyPrecision) |
| 287 | } |
| 288 | if p99 := histogramPercentile(hd, 0.99); !math.IsNaN(p99) { |
| 289 | mx.RESTClient.Latency.P99.Set(p99 * latencyPrecision) |
| 290 | } |
| 291 | } |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | // collectAdmission collects admission controller and webhook metrics |
| 296 | func (c *Collector) collectAdmission(mfs prometheus.MetricFamilies, mx *metrics) { |
| 297 | // Admission step latency - aggregate buckets across all operation/rejected label combinations |
| 298 | if mf := mfs.Get("apiserver_admission_step_admission_duration_seconds"); mf != nil && mf.Type() == model.MetricTypeHistogram { |
| 299 | // Use map[type][le] -> count for proper aggregation |
| 300 | stepBuckets := make(map[string]map[float64]float64) |
| 301 | stepTotals := make(map[string]float64) |
| 302 | |
| 303 | for _, m := range mf.Metrics() { |
| 304 | if m.Histogram() == nil { |
| 305 | continue |
| 306 | } |
| 307 | stepType := m.Labels().Get("type") |
| 308 | if stepType == "" { |
| 309 | continue |
| 310 | } |
| 311 | |
| 312 | if stepBuckets[stepType] == nil { |
| 313 | stepBuckets[stepType] = make(map[float64]float64) |
| 314 | } |
| 315 | |
| 316 | for _, b := range m.Histogram().Buckets() { |
| 317 | if math.IsInf(b.UpperBound(), 0) { |
| 318 | stepTotals[stepType] += b.CumulativeCount() |
| 319 | continue |
| 320 | } |
| 321 | // Aggregate by summing counts for same le across all label combinations |
| 322 | stepBuckets[stepType][b.UpperBound()] += b.CumulativeCount() |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | // Convert aggregated buckets to histogramData for percentile calculation |
| 327 | for stepType, bucketMap := range stepBuckets { |
| 328 | hd := histogramData{total: stepTotals[stepType]} |
| 329 | for le, count := range bucketMap { |
| 330 | hd.buckets = append(hd.buckets, histogramBucket{le: le, count: count}) |
| 331 | } |
| 332 | sortBuckets(hd.buckets) |
| 333 | |
| 334 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 335 | switch stepType { |
| 336 | case "validate": |
| 337 | mx.Admission.StepLatency.Validate.Set(p50 * latencyPrecision) |
| 338 | case "admit": |
| 339 | mx.Admission.StepLatency.Admit.Set(p50 * latencyPrecision) |
| 340 | } |
| 341 | } |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | // Admission controller latency (dynamic charts) |
| 346 | c.collectAdmissionControllerLatency(mfs, mx) |
| 347 | c.collectAdmissionWebhookLatency(mfs, mx) |
| 348 | } |
| 349 | |
| 350 | func (c *Collector) collectAdmissionControllerLatency(mfs prometheus.MetricFamilies, mx *metrics) { |
| 351 | mf := mfs.Get("apiserver_admission_controller_admission_duration_seconds") |
| 352 | if mf == nil || mf.Type() != model.MetricTypeHistogram { |
| 353 | return |
| 354 | } |
| 355 | |
| 356 | // Collect as heatmap with non-cumulative bucket counts |
| 357 | // Must aggregate by controller name, then by le bucket |
| 358 | controllerBucketMaps := make(map[string]map[float64]float64) // name -> le -> cumulative_count |
| 359 | |
| 360 | for _, m := range mf.Metrics() { |
| 361 | if m.Histogram() == nil { |
| 362 | continue |
| 363 | } |
| 364 | name := m.Labels().Get("name") |
| 365 | if name == "" { |
| 366 | continue |
| 367 | } |
| 368 | |
| 369 | if controllerBucketMaps[name] == nil { |
| 370 | controllerBucketMaps[name] = make(map[float64]float64) |
| 371 | } |
| 372 | |
| 373 | for _, b := range m.Histogram().Buckets() { |
| 374 | controllerBucketMaps[name][b.UpperBound()] += float64(b.CumulativeCount()) |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | // Collect names into slice first to avoid modifying maps during iteration |
| 379 | controllerNames := make([]string, 0, len(controllerBucketMaps)) |
| 380 | for name := range controllerBucketMaps { |
| 381 | controllerNames = append(controllerNames, name) |
| 382 | } |
| 383 | |
| 384 | for _, name := range controllerNames { |
| 385 | bucketMap := controllerBucketMaps[name] |
| 386 | |
| 387 | // Sanitize name for use in chart/dimension IDs |
| 388 | cleanName := cleanID(name) |
| 389 | |
| 390 | // Cardinality limit: only accept new items if under limit |
| 391 | _, seen := c.collectedAdmissionCtrl[cleanName] |
| 392 | if !seen && len(c.collectedAdmissionCtrl) >= defaultMaxAdmCtrl { |
| 393 | continue |
| 394 | } |
| 395 | |
| 396 | // Track last-seen cycle |
| 397 | c.collectedAdmissionCtrl[cleanName] = c.collectCycle |
| 398 | |
| 399 | if !seen { |
| 400 | if err := c.charts.Add(newAdmissionControllerLatencyChart(cleanName)); err != nil { |
| 401 | c.Warningf("failed to add admission controller chart %s: %v", name, err) |
| 402 | } |
| 403 | } |
| 404 | |
| 405 | if mx.Admission.Controllers[cleanName] == nil { |
| 406 | mx.Admission.Controllers[cleanName] = &admissionControllerMetrics{} |
| 407 | } |
| 408 | |
| 409 | // Use cumulative bucket values directly - chart uses Incremental algorithm |
| 410 | // This avoids negative values when Prometheus counters reset |
| 411 | c.setAdmissionBucketMetrics(bucketMap, mx.Admission.Controllers[cleanName]) |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | func (c *Collector) collectAdmissionWebhookLatency(mfs prometheus.MetricFamilies, mx *metrics) { |
| 416 | mf := mfs.Get("apiserver_admission_webhook_admission_duration_seconds") |
| 417 | if mf == nil || mf.Type() != model.MetricTypeHistogram { |
| 418 | return |
| 419 | } |
| 420 | |
| 421 | // Collect as heatmap with non-cumulative bucket counts |
| 422 | webhookBucketMaps := make(map[string]map[float64]float64) // name -> le -> cumulative_count |
| 423 | |
| 424 | for _, m := range mf.Metrics() { |
| 425 | if m.Histogram() == nil { |
| 426 | continue |
| 427 | } |
| 428 | name := m.Labels().Get("name") |
| 429 | if name == "" { |
| 430 | continue |
| 431 | } |
| 432 | |
| 433 | if webhookBucketMaps[name] == nil { |
| 434 | webhookBucketMaps[name] = make(map[float64]float64) |
| 435 | } |
| 436 | |
| 437 | for _, b := range m.Histogram().Buckets() { |
| 438 | webhookBucketMaps[name][b.UpperBound()] += float64(b.CumulativeCount()) |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | // Collect names into slice first to avoid modifying maps during iteration |
| 443 | webhookNames := make([]string, 0, len(webhookBucketMaps)) |
| 444 | for name := range webhookBucketMaps { |
| 445 | webhookNames = append(webhookNames, name) |
| 446 | } |
| 447 | |
| 448 | for _, name := range webhookNames { |
| 449 | bucketMap := webhookBucketMaps[name] |
| 450 | |
| 451 | // Sanitize name for use in chart/dimension IDs |
| 452 | cleanName := cleanID(name) |
| 453 | |
| 454 | // Cardinality limit: only accept new items if under limit |
| 455 | _, seen := c.collectedAdmissionWH[cleanName] |
| 456 | if !seen && len(c.collectedAdmissionWH) >= defaultMaxAdmWebhooks { |
| 457 | continue |
| 458 | } |
| 459 | |
| 460 | // Track last-seen cycle |
| 461 | c.collectedAdmissionWH[cleanName] = c.collectCycle |
| 462 | |
| 463 | if !seen { |
| 464 | if err := c.charts.Add(newAdmissionWebhookLatencyChart(cleanName)); err != nil { |
| 465 | c.Warningf("failed to add admission webhook chart %s: %v", name, err) |
| 466 | } |
| 467 | } |
| 468 | |
| 469 | if mx.Admission.Webhooks[cleanName] == nil { |
| 470 | mx.Admission.Webhooks[cleanName] = &admissionWebhookMetrics{} |
| 471 | } |
| 472 | |
| 473 | // Use cumulative bucket values directly - chart uses Incremental algorithm |
| 474 | // This avoids negative values when Prometheus counters reset |
| 475 | c.setAdmissionBucketMetrics(bucketMap, mx.Admission.Webhooks[cleanName]) |
| 476 | } |
| 477 | } |
| 478 | |
| 479 | // collectEtcd collects etcd/storage object count metrics |
| 480 | func (c *Collector) collectEtcd(mfs prometheus.MetricFamilies, mx *metrics) { |
| 481 | chart := c.charts.Get("etcd_object_counts") |
| 482 | if chart == nil { |
| 483 | chart = newEtcdObjectCountsChart() |
| 484 | if err := c.charts.Add(chart); err != nil { |
| 485 | c.Warningf("failed to add etcd object counts chart: %v", err) |
| 486 | } |
| 487 | } |
| 488 | |
| 489 | // Try apiserver_storage_objects first (newer k8s), then etcd_object_counts (older) |
| 490 | mf := mfs.Get("apiserver_storage_objects") |
| 491 | if mf == nil { |
| 492 | mf = mfs.Get("etcd_object_counts") |
| 493 | } |
| 494 | if mf == nil { |
| 495 | return |
| 496 | } |
| 497 | |
| 498 | for _, m := range mf.Metrics() { |
| 499 | value := metricValue(mf, m) |
| 500 | if math.IsNaN(value) { |
| 501 | continue |
| 502 | } |
| 503 | |
| 504 | resource := m.Labels().Get("resource") |
| 505 | if resource == "" { |
| 506 | continue |
| 507 | } |
| 508 | |
| 509 | resourceName := simplifyResourceName(resource) |
| 510 | resourceID := cleanID(resourceName) |
| 511 | dimID := "etcd_objects_" + resourceID |
| 512 | |
| 513 | if chart != nil && !chart.HasDim(dimID) { |
| 514 | if err := chart.AddDim(&Dim{ID: dimID, Name: resourceName}); err != nil { |
| 515 | c.Debugf("failed to add etcd object dimension %s: %v", resourceName, err) |
| 516 | } else { |
| 517 | chart.MarkNotCreated() |
| 518 | } |
| 519 | } |
| 520 | mx.Etcd.ObjectCounts[resourceID] = mtx.Gauge(value) |
| 521 | } |
| 522 | } |
| 523 | |
| 524 | // collectWorkqueues collects controller work queue metrics |
| 525 | func (c *Collector) collectWorkqueues(mfs prometheus.MetricFamilies, mx *metrics) { |
| 526 | // Pre-index metrics by queue name |
| 527 | depthByName := make(map[string]float64) |
| 528 | addsByName := make(map[string]float64) |
| 529 | retriesByName := make(map[string]float64) |
| 530 | |
| 531 | if mf := mfs.Get("workqueue_depth"); mf != nil { |
| 532 | for _, m := range mf.Metrics() { |
| 533 | if name := m.Labels().Get("name"); name != "" { |
| 534 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 535 | depthByName[name] = v |
| 536 | } |
| 537 | } |
| 538 | } |
| 539 | } |
| 540 | if mf := mfs.Get("workqueue_adds_total"); mf != nil { |
| 541 | for _, m := range mf.Metrics() { |
| 542 | if name := m.Labels().Get("name"); name != "" { |
| 543 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 544 | addsByName[name] = v |
| 545 | } |
| 546 | } |
| 547 | } |
| 548 | } |
| 549 | if mf := mfs.Get("workqueue_retries_total"); mf != nil { |
| 550 | for _, m := range mf.Metrics() { |
| 551 | if name := m.Labels().Get("name"); name != "" { |
| 552 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 553 | retriesByName[name] = v |
| 554 | } |
| 555 | } |
| 556 | } |
| 557 | } |
| 558 | |
| 559 | queueLatencyMF := mfs.Get("workqueue_queue_duration_seconds") |
| 560 | workDurationMF := mfs.Get("workqueue_work_duration_seconds") |
| 561 | |
| 562 | // Merge all queue names from depth, adds, and retries to avoid missing workqueues |
| 563 | // that may not have depth metric but have other metrics |
| 564 | allQueueNames := make(map[string]struct{}) |
| 565 | for name := range depthByName { |
| 566 | allQueueNames[name] = struct{}{} |
| 567 | } |
| 568 | for name := range addsByName { |
| 569 | allQueueNames[name] = struct{}{} |
| 570 | } |
| 571 | for name := range retriesByName { |
| 572 | allQueueNames[name] = struct{}{} |
| 573 | } |
| 574 | |
| 575 | // Collect names into slice to avoid modifying maps during iteration |
| 576 | queueNames := make([]string, 0, len(allQueueNames)) |
| 577 | for name := range allQueueNames { |
| 578 | queueNames = append(queueNames, name) |
| 579 | } |
| 580 | |
| 581 | for _, queueName := range queueNames { |
| 582 | // Sanitize name for use in chart/dimension IDs |
| 583 | cleanName := cleanID(queueName) |
| 584 | |
| 585 | // Cardinality limit: only accept new items if under limit |
| 586 | _, seen := c.collectedWorkqueues[cleanName] |
| 587 | if !seen && len(c.collectedWorkqueues) >= defaultMaxWorkqueues { |
| 588 | continue |
| 589 | } |
| 590 | |
| 591 | // Track last-seen cycle |
| 592 | c.collectedWorkqueues[cleanName] = c.collectCycle |
| 593 | |
| 594 | if !seen { |
| 595 | if err := c.charts.Add(newWorkqueueDepthChart(cleanName)); err != nil { |
| 596 | c.Warningf("failed to add workqueue depth chart %s: %v", queueName, err) |
| 597 | } |
| 598 | if err := c.charts.Add(newWorkqueueLatencyChart(cleanName)); err != nil { |
| 599 | c.Warningf("failed to add workqueue latency chart %s: %v", queueName, err) |
| 600 | } |
| 601 | if err := c.charts.Add(newWorkqueueAddsChart(cleanName)); err != nil { |
| 602 | c.Warningf("failed to add workqueue adds chart %s: %v", queueName, err) |
| 603 | } |
| 604 | if err := c.charts.Add(newWorkqueueDurationChart(cleanName)); err != nil { |
| 605 | c.Warningf("failed to add workqueue duration chart %s: %v", queueName, err) |
| 606 | } |
| 607 | } |
| 608 | |
| 609 | if mx.Workqueue.Controllers[cleanName] == nil { |
| 610 | mx.Workqueue.Controllers[cleanName] = &workqueueMetrics{} |
| 611 | } |
| 612 | |
| 613 | wq := mx.Workqueue.Controllers[cleanName] |
| 614 | wq.Depth.Set(depthByName[queueName]) |
| 615 | wq.Adds.Set(addsByName[queueName]) |
| 616 | wq.Retries.Set(retriesByName[queueName]) |
| 617 | |
| 618 | // Queue latency (histogram) |
| 619 | if queueLatencyMF != nil && queueLatencyMF.Type() == model.MetricTypeHistogram { |
| 620 | hd := collectHistogramBucketsFromMF(queueLatencyMF, func(m prometheus.Metric) bool { |
| 621 | return m.Labels().Get("name") == queueName |
| 622 | }) |
| 623 | if len(hd.buckets) > 0 { |
| 624 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 625 | wq.LatencyP50.Set(p50 * latencyPrecision) |
| 626 | } |
| 627 | if p90 := histogramPercentile(hd, 0.9); !math.IsNaN(p90) { |
| 628 | wq.LatencyP90.Set(p90 * latencyPrecision) |
| 629 | } |
| 630 | if p99 := histogramPercentile(hd, 0.99); !math.IsNaN(p99) { |
| 631 | wq.LatencyP99.Set(p99 * latencyPrecision) |
| 632 | } |
| 633 | } |
| 634 | } |
| 635 | |
| 636 | // Work duration (histogram) |
| 637 | if workDurationMF != nil && workDurationMF.Type() == model.MetricTypeHistogram { |
| 638 | hd := collectHistogramBucketsFromMF(workDurationMF, func(m prometheus.Metric) bool { |
| 639 | return m.Labels().Get("name") == queueName |
| 640 | }) |
| 641 | if len(hd.buckets) > 0 { |
| 642 | if p50 := histogramPercentile(hd, 0.5); !math.IsNaN(p50) { |
| 643 | wq.DurationP50.Set(p50 * latencyPrecision) |
| 644 | } |
| 645 | if p90 := histogramPercentile(hd, 0.9); !math.IsNaN(p90) { |
| 646 | wq.DurationP90.Set(p90 * latencyPrecision) |
| 647 | } |
| 648 | if p99 := histogramPercentile(hd, 0.99); !math.IsNaN(p99) { |
| 649 | wq.DurationP99.Set(p99 * latencyPrecision) |
| 650 | } |
| 651 | } |
| 652 | } |
| 653 | } |
| 654 | } |
| 655 | |
| 656 | // collectProcess collects Go runtime and process metrics |
| 657 | func (c *Collector) collectProcess(mfs prometheus.MetricFamilies, mx *metrics) { |
| 658 | mx.Process.Goroutines.Set(getMaxValue(mfs, "go_goroutines")) |
| 659 | mx.Process.Threads.Set(getMaxValue(mfs, "go_threads")) |
| 660 | mx.Process.CPUSeconds.Set(getMaxValue(mfs, "process_cpu_seconds_total") * precision) |
| 661 | mx.Process.ResidentMemory.Set(getMaxValue(mfs, "process_resident_memory_bytes")) |
| 662 | mx.Process.VirtualMemory.Set(getMaxValue(mfs, "process_virtual_memory_bytes")) |
| 663 | mx.Process.OpenFDs.Set(getMaxValue(mfs, "process_open_fds")) |
| 664 | mx.Process.MaxFDs.Set(getMaxValue(mfs, "process_max_fds")) |
| 665 | mx.Process.HeapAlloc.Set(getMaxValue(mfs, "go_memstats_heap_alloc_bytes")) |
| 666 | mx.Process.HeapInuse.Set(getMaxValue(mfs, "go_memstats_heap_inuse_bytes")) |
| 667 | mx.Process.StackInuse.Set(getMaxValue(mfs, "go_memstats_stack_inuse_bytes")) |
| 668 | |
| 669 | // GC duration (summary with quantile labels) |
| 670 | if mf := mfs.Get("go_gc_duration_seconds"); mf != nil && mf.Type() == model.MetricTypeSummary { |
| 671 | for _, m := range mf.Metrics() { |
| 672 | if m.Summary() == nil { |
| 673 | continue |
| 674 | } |
| 675 | for _, q := range m.Summary().Quantiles() { |
| 676 | if math.IsNaN(q.Value()) { |
| 677 | continue |
| 678 | } |
| 679 | switch q.Quantile() { |
| 680 | case 0: |
| 681 | mx.Process.GCDurationMin.Set(q.Value() * latencyPrecision) |
| 682 | case 0.25: |
| 683 | mx.Process.GCDurationP25.Set(q.Value() * latencyPrecision) |
| 684 | case 0.5: |
| 685 | mx.Process.GCDurationP50.Set(q.Value() * latencyPrecision) |
| 686 | case 0.75: |
| 687 | mx.Process.GCDurationP75.Set(q.Value() * latencyPrecision) |
| 688 | case 1: |
| 689 | mx.Process.GCDurationMax.Set(q.Value() * latencyPrecision) |
| 690 | } |
| 691 | } |
| 692 | } |
| 693 | } |
| 694 | } |
| 695 | |
| 696 | // collectAudit collects audit event metrics |
| 697 | func (c *Collector) collectAudit(mfs prometheus.MetricFamilies, mx *metrics) { |
| 698 | mx.Audit.EventsTotal.Set(getMaxValue(mfs, "apiserver_audit_event_total")) |
| 699 | mx.Audit.RejectedTotal.Set(getMaxValue(mfs, "apiserver_audit_requests_rejected_total")) |
| 700 | } |
| 701 | |
| 702 | // collectAuth collects authentication metrics |
| 703 | func (c *Collector) collectAuth(mfs prometheus.MetricFamilies, mx *metrics) { |
| 704 | // Sum across all usernames |
| 705 | if mf := mfs.Get("authenticated_user_requests"); mf != nil { |
| 706 | for _, m := range mf.Metrics() { |
| 707 | if v := metricValue(mf, m); !math.IsNaN(v) { |
| 708 | mx.Auth.AuthenticatedRequests.Add(v) |
| 709 | } |
| 710 | } |
| 711 | } |
| 712 | |
| 713 | // Client certificate expiration - histogram, track count of certs expiring within 24h |
| 714 | if mf := mfs.Get("apiserver_client_certificate_expiration_seconds"); mf != nil && mf.Type() == model.MetricTypeHistogram { |
| 715 | for _, m := range mf.Metrics() { |
| 716 | if m.Histogram() == nil { |
| 717 | continue |
| 718 | } |
| 719 | for _, b := range m.Histogram().Buckets() { |
| 720 | if b.UpperBound() == 86400 { // 1 day bucket |
| 721 | mx.Auth.CertExpirationSeconds.Set(float64(b.CumulativeCount())) |
| 722 | break |
| 723 | } |
| 724 | } |
| 725 | } |
| 726 | } |
| 727 | } |
| 728 | |
| 729 | // Helper functions for dynamic dimension creation |
| 730 | |
| 731 | func (c *Collector) addVerbDimension(verb string) { |
| 732 | _, seen := c.collectedVerbs[verb] |
| 733 | c.collectedVerbs[verb] = c.collectCycle |
| 734 | |
| 735 | if seen { |
| 736 | return |
| 737 | } |
| 738 | |
| 739 | chart := c.charts.Get("requests_by_verb") |
| 740 | if chart == nil { |
| 741 | c.Warningf("chart 'requests_by_verb' not found, cannot add dimension for verb: %s", verb) |
| 742 | return |
| 743 | } |
| 744 | dimID := "request_by_verb_" + cleanID(verb) |
| 745 | if !chart.HasDim(dimID) { |
| 746 | if err := chart.AddDim(&Dim{ID: dimID, Name: verb, Algo: collectorapi.Incremental}); err != nil { |
| 747 | c.Warningf("failed to add verb dimension %s: %v", verb, err) |
| 748 | } else { |
| 749 | chart.MarkNotCreated() |
| 750 | } |
| 751 | } |
| 752 | } |
| 753 | |
| 754 | func (c *Collector) addCodeDimension(code string) { |
| 755 | _, seen := c.collectedCodes[code] |
| 756 | c.collectedCodes[code] = c.collectCycle |
| 757 | |
| 758 | if seen { |
| 759 | return |
| 760 | } |
| 761 | |
| 762 | chart := c.charts.Get("requests_by_code") |
| 763 | if chart == nil { |
| 764 | c.Warningf("chart 'requests_by_code' not found, cannot add dimension for code: %s", code) |
| 765 | return |
| 766 | } |
| 767 | dimID := "request_by_code_" + cleanID(code) |
| 768 | if !chart.HasDim(dimID) { |
| 769 | if err := chart.AddDim(&Dim{ID: dimID, Name: code, Algo: collectorapi.Incremental}); err != nil { |
| 770 | c.Warningf("failed to add code dimension %s: %v", code, err) |
| 771 | } else { |
| 772 | chart.MarkNotCreated() |
| 773 | } |
| 774 | } |
| 775 | } |
| 776 | |
| 777 | func (c *Collector) addResourceDimension(resource string) { |
| 778 | _, seen := c.collectedResources[resource] |
| 779 | // Note: cycle tracking for resources is done in collectRequests due to cardinality limit check |
| 780 | if seen { |
| 781 | return |
| 782 | } |
| 783 | |
| 784 | chart := c.charts.Get("requests_by_resource") |
| 785 | if chart == nil { |
| 786 | c.Warningf("chart 'requests_by_resource' not found, cannot add dimension for resource: %s", resource) |
| 787 | return |
| 788 | } |
| 789 | dimID := "request_by_resource_" + cleanID(resource) |
| 790 | if !chart.HasDim(dimID) { |
| 791 | if err := chart.AddDim(&Dim{ID: dimID, Name: resource, Algo: collectorapi.Incremental}); err != nil { |
| 792 | c.Warningf("failed to add resource dimension %s: %v", resource, err) |
| 793 | } else { |
| 794 | chart.MarkNotCreated() |
| 795 | } |
| 796 | } |
| 797 | } |
| 798 | |
| 799 | // simplifyResourceName removes API group suffix from resource names |
| 800 | func simplifyResourceName(resource string) string { |
| 801 | parts := strings.SplitN(resource, ".", 2) |
| 802 | return parts[0] |
| 803 | } |
| 804 | |
| 805 | // metricValue extracts the value from a metric based on its type |
| 806 | func metricValue(mf *prometheus.MetricFamily, m prometheus.Metric) float64 { |
| 807 | switch mf.Type() { |
| 808 | case model.MetricTypeGauge: |
| 809 | if m.Gauge() != nil { |
| 810 | return m.Gauge().Value() |
| 811 | } |
| 812 | case model.MetricTypeCounter: |
| 813 | if m.Counter() != nil { |
| 814 | return m.Counter().Value() |
| 815 | } |
| 816 | case model.MetricTypeUnknown: |
| 817 | if m.Gauge() != nil { |
| 818 | return m.Gauge().Value() |
| 819 | } |
| 820 | if m.Counter() != nil { |
| 821 | return m.Counter().Value() |
| 822 | } |
| 823 | } |
| 824 | return math.NaN() |
| 825 | } |
| 826 | |
| 827 | // getMaxValue gets the maximum value across all metrics in a family |
| 828 | func getMaxValue(mfs prometheus.MetricFamilies, name string) float64 { |
| 829 | mf := mfs.Get(name) |
| 830 | if mf == nil { |
| 831 | return 0 |
| 832 | } |
| 833 | |
| 834 | var maxVal float64 |
| 835 | for _, m := range mf.Metrics() { |
| 836 | if v := metricValue(mf, m); !math.IsNaN(v) && v > maxVal { |
| 837 | maxVal = v |
| 838 | } |
| 839 | } |
| 840 | return maxVal |
| 841 | } |
| 842 | |
| 843 | // Histogram percentile calculation utilities |
| 844 | |
| 845 | type histogramBucket struct { |
| 846 | le float64 |
| 847 | count float64 |
| 848 | } |
| 849 | |
| 850 | type histogramData struct { |
| 851 | buckets []histogramBucket |
| 852 | total float64 // from +Inf bucket |
| 853 | } |
| 854 | |
| 855 | // collectHistogramBucketsFromMF collects histogram buckets from a metric family |
| 856 | func collectHistogramBucketsFromMF(mf *prometheus.MetricFamily, filter func(prometheus.Metric) bool) histogramData { |
| 857 | bucketMap := make(map[float64]float64) |
| 858 | var total float64 |
| 859 | |
| 860 | for _, m := range mf.Metrics() { |
| 861 | if filter != nil && !filter(m) { |
| 862 | continue |
| 863 | } |
| 864 | if m.Histogram() == nil { |
| 865 | continue |
| 866 | } |
| 867 | |
| 868 | for _, b := range m.Histogram().Buckets() { |
| 869 | if math.IsInf(b.UpperBound(), 0) { |
| 870 | total += b.CumulativeCount() |
| 871 | continue |
| 872 | } |
| 873 | bucketMap[b.UpperBound()] += b.CumulativeCount() |
| 874 | } |
| 875 | } |
| 876 | |
| 877 | buckets := make([]histogramBucket, 0, len(bucketMap)) |
| 878 | for le, count := range bucketMap { |
| 879 | buckets = append(buckets, histogramBucket{le: le, count: count}) |
| 880 | } |
| 881 | |
| 882 | sortBuckets(buckets) |
| 883 | return histogramData{buckets: buckets, total: total} |
| 884 | } |
| 885 | |
| 886 | func sortBuckets(buckets []histogramBucket) { |
| 887 | sort.Slice(buckets, func(i, j int) bool { |
| 888 | return buckets[i].le < buckets[j].le |
| 889 | }) |
| 890 | } |
| 891 | |
| 892 | // histogramPercentile estimates the percentile value from histogram buckets |
| 893 | // total should be from the +Inf bucket which contains the true total count |
| 894 | func histogramPercentile(hd histogramData, percentile float64) float64 { |
| 895 | if len(hd.buckets) == 0 || hd.total == 0 { |
| 896 | return math.NaN() |
| 897 | } |
| 898 | |
| 899 | target := percentile * hd.total |
| 900 | |
| 901 | var prevBound, prevCount float64 |
| 902 | for _, b := range hd.buckets { |
| 903 | if b.count >= target { |
| 904 | bucketWidth := b.le - prevBound |
| 905 | bucketCount := b.count - prevCount |
| 906 | if bucketCount == 0 { |
| 907 | return b.le |
| 908 | } |
| 909 | fraction := (target - prevCount) / bucketCount |
| 910 | return prevBound + fraction*bucketWidth |
| 911 | } |
| 912 | prevBound = b.le |
| 913 | prevCount = b.count |
| 914 | } |
| 915 | |
| 916 | return hd.buckets[len(hd.buckets)-1].le |
| 917 | } |
| 918 | |
| 919 | // bucketSetter is implemented by admission controller and webhook metrics |
| 920 | type bucketSetter interface { |
| 921 | setBuckets(b5ms, b25ms, b100ms, b500ms, b1s, b2500ms, bInf float64) |
| 922 | } |
| 923 | |
| 924 | func (m *admissionControllerMetrics) setBuckets(b5ms, b25ms, b100ms, b500ms, b1s, b2500ms, bInf float64) { |
| 925 | m.Bucket5ms.Set(b5ms) |
| 926 | m.Bucket25ms.Set(b25ms) |
| 927 | m.Bucket100ms.Set(b100ms) |
| 928 | m.Bucket500ms.Set(b500ms) |
| 929 | m.Bucket1s.Set(b1s) |
| 930 | m.Bucket2500ms.Set(b2500ms) |
| 931 | m.BucketInf.Set(bInf) |
| 932 | } |
| 933 | |
| 934 | func (m *admissionWebhookMetrics) setBuckets(b5ms, b25ms, b100ms, b500ms, b1s, b2500ms, bInf float64) { |
| 935 | m.Bucket5ms.Set(b5ms) |
| 936 | m.Bucket25ms.Set(b25ms) |
| 937 | m.Bucket100ms.Set(b100ms) |
| 938 | m.Bucket500ms.Set(b500ms) |
| 939 | m.Bucket1s.Set(b1s) |
| 940 | m.Bucket2500ms.Set(b2500ms) |
| 941 | m.BucketInf.Set(bInf) |
| 942 | } |
| 943 | |
| 944 | // setAdmissionBucketMetrics extracts bucket values and sets them on the metrics |
| 945 | // Converts Prometheus cumulative buckets to non-cumulative for heatmap display |
| 946 | // Uses max(0, diff) to protect against negative values from counter resets |
| 947 | func (c *Collector) setAdmissionBucketMetrics(bucketMap map[float64]float64, m bucketSetter) { |
| 948 | // Validate bucket presence and log warnings for missing buckets |
| 949 | var missingBuckets []string |
| 950 | for _, bound := range admissionBucketBounds { |
| 951 | if _, ok := bucketMap[bound]; !ok { |
| 952 | missingBuckets = append(missingBuckets, formatBucketBound(bound)) |
| 953 | } |
| 954 | } |
| 955 | if _, ok := bucketMap[math.Inf(1)]; !ok { |
| 956 | missingBuckets = append(missingBuckets, "+Inf") |
| 957 | } |
| 958 | if len(missingBuckets) > 0 { |
| 959 | c.Debugf("missing histogram buckets: %v", missingBuckets) |
| 960 | } |
| 961 | |
| 962 | // Extract cumulative bucket values (0 if missing) |
| 963 | b5ms := bucketMap[admissionBucketBounds[0]] |
| 964 | b25ms := bucketMap[admissionBucketBounds[1]] |
| 965 | b100ms := bucketMap[admissionBucketBounds[2]] |
| 966 | b500ms := bucketMap[admissionBucketBounds[3]] |
| 967 | b1s := bucketMap[admissionBucketBounds[4]] |
| 968 | b2500ms := bucketMap[admissionBucketBounds[5]] |
| 969 | bInf := bucketMap[math.Inf(1)] |
| 970 | |
| 971 | // Convert cumulative to non-cumulative (differential) bucket counts |
| 972 | // Use max(0, diff) to handle Prometheus counter resets gracefully |
| 973 | // When a counter resets, the cumulative value decreases, causing negative diffs |
| 974 | m.setBuckets( |
| 975 | b5ms, |
| 976 | math.Max(0, b25ms-b5ms), |
| 977 | math.Max(0, b100ms-b25ms), |
| 978 | math.Max(0, b500ms-b100ms), |
| 979 | math.Max(0, b1s-b500ms), |
| 980 | math.Max(0, b2500ms-b1s), |
| 981 | math.Max(0, bInf-b2500ms), |
| 982 | ) |
| 983 | } |
| 984 | |
| 985 | // formatBucketBound formats a bucket bound for logging |
| 986 | func formatBucketBound(bound float64) string { |
| 987 | if bound < 1 { |
| 988 | return fmt.Sprintf("%.0fms", bound*1000) |
| 989 | } |
| 990 | return fmt.Sprintf("%.1fs", bound) |
| 991 | } |
| 992 | |
| 993 | // cleanupStaleDimensions removes dimensions that haven't been seen for staleThresholdCycles |
| 994 | func (c *Collector) cleanupStaleDimensions() { |
| 995 | threshold := c.collectCycle - staleThresholdCycles |
| 996 | |
| 997 | // Cleanup resources |
| 998 | for name, lastSeen := range c.collectedResources { |
| 999 | if lastSeen < threshold { |
| 1000 | delete(c.collectedResources, name) |
| 1001 | if chart := c.charts.Get("requests_by_resource"); chart != nil { |
| 1002 | dimID := "request_by_resource_" + cleanID(name) |
| 1003 | _ = chart.RemoveDim(dimID) |
| 1004 | chart.MarkNotCreated() |
| 1005 | } |
| 1006 | c.Debugf("removed stale resource dimension: %s", name) |
| 1007 | } |
| 1008 | } |
| 1009 | |
| 1010 | // Cleanup verbs |
| 1011 | for name, lastSeen := range c.collectedVerbs { |
| 1012 | if lastSeen < threshold { |
| 1013 | delete(c.collectedVerbs, name) |
| 1014 | if chart := c.charts.Get("requests_by_verb"); chart != nil { |
| 1015 | dimID := "request_by_verb_" + cleanID(name) |
| 1016 | _ = chart.RemoveDim(dimID) |
| 1017 | chart.MarkNotCreated() |
| 1018 | } |
| 1019 | c.Debugf("removed stale verb dimension: %s", name) |
| 1020 | } |
| 1021 | } |
| 1022 | |
| 1023 | // Cleanup codes |
| 1024 | for name, lastSeen := range c.collectedCodes { |
| 1025 | if lastSeen < threshold { |
| 1026 | delete(c.collectedCodes, name) |
| 1027 | if chart := c.charts.Get("requests_by_code"); chart != nil { |
| 1028 | dimID := "request_by_code_" + cleanID(name) |
| 1029 | _ = chart.RemoveDim(dimID) |
| 1030 | chart.MarkNotCreated() |
| 1031 | } |
| 1032 | c.Debugf("removed stale code dimension: %s", name) |
| 1033 | } |
| 1034 | } |
| 1035 | |
| 1036 | // Cleanup REST client codes |
| 1037 | for name, lastSeen := range c.collectedRESTCodes { |
| 1038 | if lastSeen < threshold { |
| 1039 | delete(c.collectedRESTCodes, name) |
| 1040 | if chart := c.charts.Get("rest_client_requests_by_code"); chart != nil { |
| 1041 | dimID := "rest_client_by_code_" + cleanID(name) |
| 1042 | _ = chart.RemoveDim(dimID) |
| 1043 | chart.MarkNotCreated() |
| 1044 | } |
| 1045 | c.Debugf("removed stale REST client code dimension: %s", name) |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | // Cleanup REST client methods |
| 1050 | for name, lastSeen := range c.collectedRESTMethods { |
| 1051 | if lastSeen < threshold { |
| 1052 | delete(c.collectedRESTMethods, name) |
| 1053 | if chart := c.charts.Get("rest_client_requests_by_method"); chart != nil { |
| 1054 | dimID := "rest_client_by_method_" + cleanID(name) |
| 1055 | _ = chart.RemoveDim(dimID) |
| 1056 | chart.MarkNotCreated() |
| 1057 | } |
| 1058 | c.Debugf("removed stale REST client method dimension: %s", name) |
| 1059 | } |
| 1060 | } |
| 1061 | |
| 1062 | // Cleanup workqueues (remove charts) |
| 1063 | for name, lastSeen := range c.collectedWorkqueues { |
| 1064 | if lastSeen < threshold { |
| 1065 | delete(c.collectedWorkqueues, name) |
| 1066 | _ = c.charts.Remove("workqueue_depth_" + cleanID(name)) |
| 1067 | _ = c.charts.Remove("workqueue_latency_" + cleanID(name)) |
| 1068 | _ = c.charts.Remove("workqueue_adds_" + cleanID(name)) |
| 1069 | _ = c.charts.Remove("workqueue_duration_" + cleanID(name)) |
| 1070 | c.Debugf("removed stale workqueue charts: %s", name) |
| 1071 | } |
| 1072 | } |
| 1073 | |
| 1074 | // Cleanup admission controllers (remove charts) |
| 1075 | for name, lastSeen := range c.collectedAdmissionCtrl { |
| 1076 | if lastSeen < threshold { |
| 1077 | delete(c.collectedAdmissionCtrl, name) |
| 1078 | _ = c.charts.Remove("admission_controller_latency_" + cleanID(name)) |
| 1079 | c.Debugf("removed stale admission controller chart: %s", name) |
| 1080 | } |
| 1081 | } |
| 1082 | |
| 1083 | // Cleanup admission webhooks (remove charts) |
| 1084 | for name, lastSeen := range c.collectedAdmissionWH { |
| 1085 | if lastSeen < threshold { |
| 1086 | delete(c.collectedAdmissionWH, name) |
| 1087 | _ = c.charts.Remove("admission_webhook_latency_" + cleanID(name)) |
| 1088 | c.Debugf("removed stale admission webhook chart: %s", name) |
| 1089 | } |
| 1090 | } |
| 1091 | } |