Line data Source code
1 : use enum_map::EnumMap;
2 : use metrics::{
3 : register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
4 : register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
5 : register_int_gauge, register_int_gauge_vec, register_uint_gauge, register_uint_gauge_vec,
6 : Counter, CounterVec, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
7 : IntCounterPairVec, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
8 : };
9 : use once_cell::sync::Lazy;
10 : use pageserver_api::shard::TenantShardId;
11 : use strum::{EnumCount, VariantNames};
12 : use strum_macros::{EnumVariantNames, IntoStaticStr};
13 : use tracing::warn;
14 : use utils::id::TimelineId;
15 :
16 : /// Prometheus histogram buckets (in seconds) for operations in the critical
17 : /// path. In other words, operations that directly affect that latency of user
18 : /// queries.
19 : ///
20 : /// The buckets capture the majority of latencies in the microsecond and
21 : /// millisecond range but also extend far enough up to distinguish "bad" from
22 : /// "really bad".
23 : const CRITICAL_OP_BUCKETS: &[f64] = &[
24 : 0.000_001, 0.000_010, 0.000_100, // 1 us, 10 us, 100 us
25 : 0.001_000, 0.010_000, 0.100_000, // 1 ms, 10 ms, 100 ms
26 : 1.0, 10.0, 100.0, // 1 s, 10 s, 100 s
27 : ];
28 :
29 : // Metrics collected on operations on the storage repository.
30 3248 : #[derive(Debug, EnumVariantNames, IntoStaticStr)]
31 : #[strum(serialize_all = "kebab_case")]
32 : pub(crate) enum StorageTimeOperation {
33 : #[strum(serialize = "layer flush")]
34 : LayerFlush,
35 :
36 : #[strum(serialize = "compact")]
37 : Compact,
38 :
39 : #[strum(serialize = "create images")]
40 : CreateImages,
41 :
42 : #[strum(serialize = "logical size")]
43 : LogicalSize,
44 :
45 : #[strum(serialize = "imitate logical size")]
46 : ImitateLogicalSize,
47 :
48 : #[strum(serialize = "load layer map")]
49 : LoadLayerMap,
50 :
51 : #[strum(serialize = "gc")]
52 : Gc,
53 :
54 : #[strum(serialize = "find gc cutoffs")]
55 : FindGcCutoffs,
56 : }
57 :
58 162 : pub(crate) static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
59 : register_counter_vec!(
60 : "pageserver_storage_operations_seconds_sum",
61 : "Total time spent on storage operations with operation, tenant and timeline dimensions",
62 : &["operation", "tenant_id", "shard_id", "timeline_id"],
63 : )
64 162 : .expect("failed to define a metric")
65 162 : });
66 :
67 162 : pub(crate) static STORAGE_TIME_COUNT_PER_TIMELINE: Lazy<IntCounterVec> = Lazy::new(|| {
68 : register_int_counter_vec!(
69 : "pageserver_storage_operations_seconds_count",
70 : "Count of storage operations with operation, tenant and timeline dimensions",
71 : &["operation", "tenant_id", "shard_id", "timeline_id"],
72 : )
73 162 : .expect("failed to define a metric")
74 162 : });
75 :
76 : // Buckets for background operations like compaction, GC, size calculation
77 : const STORAGE_OP_BUCKETS: &[f64] = &[0.010, 0.100, 1.0, 10.0, 100.0, 1000.0];
78 :
79 162 : pub(crate) static STORAGE_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
80 : register_histogram_vec!(
81 : "pageserver_storage_operations_seconds_global",
82 : "Time spent on storage operations",
83 : &["operation"],
84 : STORAGE_OP_BUCKETS.into(),
85 : )
86 162 : .expect("failed to define a metric")
87 162 : });
88 :
89 0 : pub(crate) static READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
90 : register_histogram!(
91 : "pageserver_layers_visited_per_read_global",
92 : "Number of layers visited to reconstruct one key",
93 : vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0],
94 : )
95 0 : .expect("failed to define a metric")
96 0 : });
97 :
98 158 : pub(crate) static VEC_READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
99 : register_histogram!(
100 : "pageserver_layers_visited_per_vectored_read_global",
101 : "Average number of layers visited to reconstruct one key",
102 : vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0],
103 : )
104 158 : .expect("failed to define a metric")
105 158 : });
106 :
107 : // Metrics collected on operations on the storage repository.
108 : #[derive(
109 640 : Clone, Copy, enum_map::Enum, strum_macros::EnumString, strum_macros::Display, IntoStaticStr,
110 : )]
111 : pub(crate) enum GetKind {
112 : Singular,
113 : Vectored,
114 : }
115 :
116 : pub(crate) struct ReconstructTimeMetrics {
117 : singular: Histogram,
118 : vectored: Histogram,
119 : }
120 :
121 160 : pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
122 160 : let inner = register_histogram_vec!(
123 : "pageserver_getpage_reconstruct_seconds",
124 : "Time spent in reconstruct_value (reconstruct a page from deltas)",
125 : &["get_kind"],
126 : CRITICAL_OP_BUCKETS.into(),
127 : )
128 160 : .expect("failed to define a metric");
129 160 :
130 160 : ReconstructTimeMetrics {
131 160 : singular: inner.with_label_values(&[GetKind::Singular.into()]),
132 160 : vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
133 160 : }
134 160 : });
135 :
136 : impl ReconstructTimeMetrics {
137 626167 : pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
138 626167 : match get_kind {
139 625919 : GetKind::Singular => &self.singular,
140 248 : GetKind::Vectored => &self.vectored,
141 : }
142 626167 : }
143 : }
144 :
145 : pub(crate) struct ReconstructDataTimeMetrics {
146 : singular: Histogram,
147 : vectored: Histogram,
148 : }
149 :
150 : impl ReconstructDataTimeMetrics {
151 626183 : pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
152 626183 : match get_kind {
153 625935 : GetKind::Singular => &self.singular,
154 248 : GetKind::Vectored => &self.vectored,
155 : }
156 626183 : }
157 : }
158 :
159 160 : pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<ReconstructDataTimeMetrics> = Lazy::new(|| {
160 160 : let inner = register_histogram_vec!(
161 : "pageserver_getpage_get_reconstruct_data_seconds",
162 : "Time spent in get_reconstruct_value_data",
163 : &["get_kind"],
164 : CRITICAL_OP_BUCKETS.into(),
165 : )
166 160 : .expect("failed to define a metric");
167 160 :
168 160 : ReconstructDataTimeMetrics {
169 160 : singular: inner.with_label_values(&[GetKind::Singular.into()]),
170 160 : vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
171 160 : }
172 160 : });
173 :
174 : pub(crate) struct GetVectoredLatency {
175 : map: EnumMap<TaskKind, Option<Histogram>>,
176 : }
177 :
178 : #[allow(dead_code)]
179 : pub(crate) struct ScanLatency {
180 : map: EnumMap<TaskKind, Option<Histogram>>,
181 : }
182 :
183 : impl GetVectoredLatency {
184 : // Only these task types perform vectored gets. Filter all other tasks out to reduce total
185 : // cardinality of the metric.
186 : const TRACKED_TASK_KINDS: [TaskKind; 2] = [TaskKind::Compaction, TaskKind::PageRequestHandler];
187 :
188 1084 : pub(crate) fn for_task_kind(&self, task_kind: TaskKind) -> Option<&Histogram> {
189 1084 : self.map[task_kind].as_ref()
190 1084 : }
191 : }
192 :
193 : impl ScanLatency {
194 : // Only these task types perform vectored gets. Filter all other tasks out to reduce total
195 : // cardinality of the metric.
196 : const TRACKED_TASK_KINDS: [TaskKind; 1] = [TaskKind::PageRequestHandler];
197 :
198 12 : pub(crate) fn for_task_kind(&self, task_kind: TaskKind) -> Option<&Histogram> {
199 12 : self.map[task_kind].as_ref()
200 12 : }
201 : }
202 :
203 : pub(crate) struct ScanLatencyOngoingRecording<'a> {
204 : parent: &'a Histogram,
205 : start: std::time::Instant,
206 : }
207 :
208 : impl<'a> ScanLatencyOngoingRecording<'a> {
209 0 : pub(crate) fn start_recording(parent: &'a Histogram) -> ScanLatencyOngoingRecording<'a> {
210 0 : let start = Instant::now();
211 0 : ScanLatencyOngoingRecording { parent, start }
212 0 : }
213 :
214 0 : pub(crate) fn observe(self, throttled: Option<Duration>) {
215 0 : let elapsed = self.start.elapsed();
216 0 : let ex_throttled = if let Some(throttled) = throttled {
217 0 : elapsed.checked_sub(throttled)
218 : } else {
219 0 : Some(elapsed)
220 : };
221 0 : if let Some(ex_throttled) = ex_throttled {
222 0 : self.parent.observe(ex_throttled.as_secs_f64());
223 0 : } else {
224 0 : use utils::rate_limit::RateLimit;
225 0 : static LOGGED: Lazy<Mutex<RateLimit>> =
226 0 : Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
227 0 : let mut rate_limit = LOGGED.lock().unwrap();
228 0 : rate_limit.call(|| {
229 0 : warn!("error deducting time spent throttled; this message is logged at a global rate limit");
230 0 : });
231 0 : }
232 0 : }
233 : }
234 :
235 154 : pub(crate) static GET_VECTORED_LATENCY: Lazy<GetVectoredLatency> = Lazy::new(|| {
236 154 : let inner = register_histogram_vec!(
237 : "pageserver_get_vectored_seconds",
238 : "Time spent in get_vectored, excluding time spent in timeline_get_throttle.",
239 : &["task_kind"],
240 : CRITICAL_OP_BUCKETS.into(),
241 : )
242 154 : .expect("failed to define a metric");
243 154 :
244 154 : GetVectoredLatency {
245 4620 : map: EnumMap::from_array(std::array::from_fn(|task_kind_idx| {
246 4620 : let task_kind = <TaskKind as enum_map::Enum>::from_usize(task_kind_idx);
247 4620 :
248 4620 : if GetVectoredLatency::TRACKED_TASK_KINDS.contains(&task_kind) {
249 308 : let task_kind = task_kind.into();
250 308 : Some(inner.with_label_values(&[task_kind]))
251 : } else {
252 4312 : None
253 : }
254 4620 : })),
255 154 : }
256 154 : });
257 :
258 4 : pub(crate) static SCAN_LATENCY: Lazy<ScanLatency> = Lazy::new(|| {
259 4 : let inner = register_histogram_vec!(
260 : "pageserver_scan_seconds",
261 : "Time spent in scan, excluding time spent in timeline_get_throttle.",
262 : &["task_kind"],
263 : CRITICAL_OP_BUCKETS.into(),
264 : )
265 4 : .expect("failed to define a metric");
266 4 :
267 4 : ScanLatency {
268 120 : map: EnumMap::from_array(std::array::from_fn(|task_kind_idx| {
269 120 : let task_kind = <TaskKind as enum_map::Enum>::from_usize(task_kind_idx);
270 120 :
271 120 : if ScanLatency::TRACKED_TASK_KINDS.contains(&task_kind) {
272 4 : let task_kind = task_kind.into();
273 4 : Some(inner.with_label_values(&[task_kind]))
274 : } else {
275 116 : None
276 : }
277 120 : })),
278 4 : }
279 4 : });
280 :
281 : pub(crate) struct PageCacheMetricsForTaskKind {
282 : pub read_accesses_immutable: IntCounter,
283 : pub read_hits_immutable: IntCounter,
284 : }
285 :
286 : pub(crate) struct PageCacheMetrics {
287 : map: EnumMap<TaskKind, EnumMap<PageContentKind, PageCacheMetricsForTaskKind>>,
288 : }
289 :
290 86 : static PAGE_CACHE_READ_HITS: Lazy<IntCounterVec> = Lazy::new(|| {
291 : register_int_counter_vec!(
292 : "pageserver_page_cache_read_hits_total",
293 : "Number of read accesses to the page cache that hit",
294 : &["task_kind", "key_kind", "content_kind", "hit_kind"]
295 : )
296 86 : .expect("failed to define a metric")
297 86 : });
298 :
299 86 : static PAGE_CACHE_READ_ACCESSES: Lazy<IntCounterVec> = Lazy::new(|| {
300 : register_int_counter_vec!(
301 : "pageserver_page_cache_read_accesses_total",
302 : "Number of read accesses to the page cache",
303 : &["task_kind", "key_kind", "content_kind"]
304 : )
305 86 : .expect("failed to define a metric")
306 86 : });
307 :
308 86 : pub(crate) static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMetrics {
309 2580 : map: EnumMap::from_array(std::array::from_fn(|task_kind| {
310 2580 : let task_kind = <TaskKind as enum_map::Enum>::from_usize(task_kind);
311 2580 : let task_kind: &'static str = task_kind.into();
312 15480 : EnumMap::from_array(std::array::from_fn(|content_kind| {
313 15480 : let content_kind = <PageContentKind as enum_map::Enum>::from_usize(content_kind);
314 15480 : let content_kind: &'static str = content_kind.into();
315 15480 : PageCacheMetricsForTaskKind {
316 15480 : read_accesses_immutable: {
317 15480 : PAGE_CACHE_READ_ACCESSES
318 15480 : .get_metric_with_label_values(&[task_kind, "immutable", content_kind])
319 15480 : .unwrap()
320 15480 : },
321 15480 :
322 15480 : read_hits_immutable: {
323 15480 : PAGE_CACHE_READ_HITS
324 15480 : .get_metric_with_label_values(&[task_kind, "immutable", content_kind, "-"])
325 15480 : .unwrap()
326 15480 : },
327 15480 : }
328 15480 : }))
329 2580 : })),
330 86 : });
331 :
332 : impl PageCacheMetrics {
333 5672778 : pub(crate) fn for_ctx(&self, ctx: &RequestContext) -> &PageCacheMetricsForTaskKind {
334 5672778 : &self.map[ctx.task_kind()][ctx.page_content_kind()]
335 5672778 : }
336 : }
337 :
338 : pub(crate) struct PageCacheSizeMetrics {
339 : pub max_bytes: UIntGauge,
340 :
341 : pub current_bytes_immutable: UIntGauge,
342 : }
343 :
344 86 : static PAGE_CACHE_SIZE_CURRENT_BYTES: Lazy<UIntGaugeVec> = Lazy::new(|| {
345 : register_uint_gauge_vec!(
346 : "pageserver_page_cache_size_current_bytes",
347 : "Current size of the page cache in bytes, by key kind",
348 : &["key_kind"]
349 : )
350 86 : .expect("failed to define a metric")
351 86 : });
352 :
353 : pub(crate) static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> =
354 86 : Lazy::new(|| PageCacheSizeMetrics {
355 : max_bytes: {
356 : register_uint_gauge!(
357 : "pageserver_page_cache_size_max_bytes",
358 : "Maximum size of the page cache in bytes"
359 : )
360 86 : .expect("failed to define a metric")
361 86 : },
362 86 : current_bytes_immutable: {
363 86 : PAGE_CACHE_SIZE_CURRENT_BYTES
364 86 : .get_metric_with_label_values(&["immutable"])
365 86 : .unwrap()
366 86 : },
367 86 : });
368 :
369 : pub(crate) mod page_cache_eviction_metrics {
370 : use std::num::NonZeroUsize;
371 :
372 : use metrics::{register_int_counter_vec, IntCounter, IntCounterVec};
373 : use once_cell::sync::Lazy;
374 :
375 : #[derive(Clone, Copy)]
376 : pub(crate) enum Outcome {
377 : FoundSlotUnused { iters: NonZeroUsize },
378 : FoundSlotEvicted { iters: NonZeroUsize },
379 : ItersExceeded { iters: NonZeroUsize },
380 : }
381 :
382 86 : static ITERS_TOTAL_VEC: Lazy<IntCounterVec> = Lazy::new(|| {
383 : register_int_counter_vec!(
384 : "pageserver_page_cache_find_victim_iters_total",
385 : "Counter for the number of iterations in the find_victim loop",
386 : &["outcome"],
387 : )
388 86 : .expect("failed to define a metric")
389 86 : });
390 :
391 86 : static CALLS_VEC: Lazy<IntCounterVec> = Lazy::new(|| {
392 : register_int_counter_vec!(
393 : "pageserver_page_cache_find_victim_calls",
394 : "Incremented at the end of each find_victim() call.\
395 : Filter by outcome to get e.g., eviction rate.",
396 : &["outcome"]
397 : )
398 86 : .unwrap()
399 86 : });
400 :
401 61860 : pub(crate) fn observe(outcome: Outcome) {
402 61860 : macro_rules! dry {
403 61860 : ($label:literal, $iters:expr) => {{
404 61860 : static LABEL: &'static str = $label;
405 61860 : static ITERS_TOTAL: Lazy<IntCounter> =
406 61860 : Lazy::new(|| ITERS_TOTAL_VEC.with_label_values(&[LABEL]));
407 61860 : static CALLS: Lazy<IntCounter> =
408 61860 : Lazy::new(|| CALLS_VEC.with_label_values(&[LABEL]));
409 61860 : ITERS_TOTAL.inc_by(($iters.get()) as u64);
410 61860 : CALLS.inc();
411 61860 : }};
412 61860 : }
413 61860 : match outcome {
414 1902 : Outcome::FoundSlotUnused { iters } => dry!("found_empty", iters),
415 59958 : Outcome::FoundSlotEvicted { iters } => {
416 59958 : dry!("found_evicted", iters)
417 : }
418 0 : Outcome::ItersExceeded { iters } => {
419 0 : dry!("err_iters_exceeded", iters);
420 0 : super::page_cache_errors_inc(super::PageCacheErrorKind::EvictIterLimit);
421 0 : }
422 : }
423 61860 : }
424 : }
425 :
426 0 : static PAGE_CACHE_ERRORS: Lazy<IntCounterVec> = Lazy::new(|| {
427 : register_int_counter_vec!(
428 : "page_cache_errors_total",
429 : "Number of timeouts while acquiring a pinned slot in the page cache",
430 : &["error_kind"]
431 : )
432 0 : .expect("failed to define a metric")
433 0 : });
434 :
435 0 : #[derive(IntoStaticStr)]
436 : #[strum(serialize_all = "kebab_case")]
437 : pub(crate) enum PageCacheErrorKind {
438 : AcquirePinnedSlotTimeout,
439 : EvictIterLimit,
440 : }
441 :
442 0 : pub(crate) fn page_cache_errors_inc(error_kind: PageCacheErrorKind) {
443 0 : PAGE_CACHE_ERRORS
444 0 : .get_metric_with_label_values(&[error_kind.into()])
445 0 : .unwrap()
446 0 : .inc();
447 0 : }
448 :
449 18 : pub(crate) static WAIT_LSN_TIME: Lazy<Histogram> = Lazy::new(|| {
450 : register_histogram!(
451 : "pageserver_wait_lsn_seconds",
452 : "Time spent waiting for WAL to arrive",
453 : CRITICAL_OP_BUCKETS.into(),
454 : )
455 18 : .expect("failed to define a metric")
456 18 : });
457 :
458 162 : static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
459 : register_int_gauge_vec!(
460 : "pageserver_last_record_lsn",
461 : "Last record LSN grouped by timeline",
462 : &["tenant_id", "shard_id", "timeline_id"]
463 : )
464 162 : .expect("failed to define a metric")
465 162 : });
466 :
467 162 : static PITR_HISTORY_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
468 : register_uint_gauge_vec!(
469 : "pageserver_pitr_history_size",
470 : "Data written since PITR cutoff on this timeline",
471 : &["tenant_id", "shard_id", "timeline_id"]
472 : )
473 162 : .expect("failed to define a metric")
474 162 : });
475 :
476 1656 : #[derive(strum_macros::EnumString, strum_macros::Display, strum_macros::IntoStaticStr)]
477 : #[strum(serialize_all = "kebab_case")]
478 : pub(crate) enum MetricLayerKind {
479 : Delta,
480 : Image,
481 : }
482 :
483 162 : static TIMELINE_LAYER_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
484 : register_uint_gauge_vec!(
485 : "pageserver_layer_bytes",
486 : "Sum of layer physical sizes in bytes",
487 : &["tenant_id", "shard_id", "timeline_id", "kind"]
488 : )
489 162 : .expect("failed to define a metric")
490 162 : });
491 :
492 162 : static TIMELINE_LAYER_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
493 : register_uint_gauge_vec!(
494 : "pageserver_layer_count",
495 : "Number of layers that exist",
496 : &["tenant_id", "shard_id", "timeline_id", "kind"]
497 : )
498 162 : .expect("failed to define a metric")
499 162 : });
500 :
501 162 : static TIMELINE_ARCHIVE_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
502 : register_uint_gauge_vec!(
503 : "pageserver_archive_size",
504 : "Timeline's logical size if it is considered eligible for archival (outside PITR window), else zero",
505 : &["tenant_id", "shard_id", "timeline_id"]
506 : )
507 162 : .expect("failed to define a metric")
508 162 : });
509 :
510 162 : static STANDBY_HORIZON: Lazy<IntGaugeVec> = Lazy::new(|| {
511 : register_int_gauge_vec!(
512 : "pageserver_standby_horizon",
513 : "Standby apply LSN for which GC is hold off, by timeline.",
514 : &["tenant_id", "shard_id", "timeline_id"]
515 : )
516 162 : .expect("failed to define a metric")
517 162 : });
518 :
519 162 : static RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
520 : register_uint_gauge_vec!(
521 : "pageserver_resident_physical_size",
522 : "The size of the layer files present in the pageserver's filesystem, for attached locations.",
523 : &["tenant_id", "shard_id", "timeline_id"]
524 : )
525 162 : .expect("failed to define a metric")
526 162 : });
527 :
528 162 : static VISIBLE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
529 : register_uint_gauge_vec!(
530 : "pageserver_visible_physical_size",
531 : "The size of the layer files present in the pageserver's filesystem.",
532 : &["tenant_id", "shard_id", "timeline_id"]
533 : )
534 162 : .expect("failed to define a metric")
535 162 : });
536 :
537 158 : pub(crate) static RESIDENT_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(|| {
538 : register_uint_gauge!(
539 : "pageserver_resident_physical_size_global",
540 : "Like `pageserver_resident_physical_size`, but without tenant/timeline dimensions."
541 : )
542 158 : .expect("failed to define a metric")
543 158 : });
544 :
545 162 : static REMOTE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
546 : register_uint_gauge_vec!(
547 : "pageserver_remote_physical_size",
548 : "The size of the layer files present in the remote storage that are listed in the remote index_part.json.",
549 : // Corollary: If any files are missing from the index part, they won't be included here.
550 : &["tenant_id", "shard_id", "timeline_id"]
551 : )
552 162 : .expect("failed to define a metric")
553 162 : });
554 :
555 162 : static REMOTE_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(|| {
556 : register_uint_gauge!(
557 : "pageserver_remote_physical_size_global",
558 : "Like `pageserver_remote_physical_size`, but without tenant/timeline dimensions."
559 : )
560 162 : .expect("failed to define a metric")
561 162 : });
562 :
563 4 : pub(crate) static REMOTE_ONDEMAND_DOWNLOADED_LAYERS: Lazy<IntCounter> = Lazy::new(|| {
564 : register_int_counter!(
565 : "pageserver_remote_ondemand_downloaded_layers_total",
566 : "Total on-demand downloaded layers"
567 : )
568 4 : .unwrap()
569 4 : });
570 :
571 4 : pub(crate) static REMOTE_ONDEMAND_DOWNLOADED_BYTES: Lazy<IntCounter> = Lazy::new(|| {
572 : register_int_counter!(
573 : "pageserver_remote_ondemand_downloaded_bytes_total",
574 : "Total bytes of layers on-demand downloaded",
575 : )
576 4 : .unwrap()
577 4 : });
578 :
579 162 : static CURRENT_LOGICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
580 : register_uint_gauge_vec!(
581 : "pageserver_current_logical_size",
582 : "Current logical size grouped by timeline",
583 : &["tenant_id", "shard_id", "timeline_id"]
584 : )
585 162 : .expect("failed to define current logical size metric")
586 162 : });
587 :
588 162 : static AUX_FILE_SIZE: Lazy<IntGaugeVec> = Lazy::new(|| {
589 : register_int_gauge_vec!(
590 : "pageserver_aux_file_estimated_size",
591 : "The size of all aux files for a timeline in aux file v2 store.",
592 : &["tenant_id", "shard_id", "timeline_id"]
593 : )
594 162 : .expect("failed to define a metric")
595 162 : });
596 :
597 162 : static VALID_LSN_LEASE_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
598 : register_uint_gauge_vec!(
599 : "pageserver_valid_lsn_lease_count",
600 : "The number of valid leases after refreshing gc info.",
601 : &["tenant_id", "shard_id", "timeline_id"],
602 : )
603 162 : .expect("failed to define a metric")
604 162 : });
605 :
606 0 : pub(crate) static CIRCUIT_BREAKERS_BROKEN: Lazy<IntCounter> = Lazy::new(|| {
607 : register_int_counter!(
608 : "pageserver_circuit_breaker_broken",
609 : "How many times a circuit breaker has broken"
610 : )
611 0 : .expect("failed to define a metric")
612 0 : });
613 :
614 0 : pub(crate) static CIRCUIT_BREAKERS_UNBROKEN: Lazy<IntCounter> = Lazy::new(|| {
615 : register_int_counter!(
616 : "pageserver_circuit_breaker_unbroken",
617 : "How many times a circuit breaker has been un-broken (recovered)"
618 : )
619 0 : .expect("failed to define a metric")
620 0 : });
621 :
622 154 : pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES: Lazy<IntCounter> = Lazy::new(|| {
623 : register_int_counter!(
624 : "pageserver_compression_image_in_bytes_total",
625 : "Size of data written into image layers before compression"
626 : )
627 154 : .expect("failed to define a metric")
628 154 : });
629 :
630 154 : pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES_CONSIDERED: Lazy<IntCounter> = Lazy::new(|| {
631 : register_int_counter!(
632 : "pageserver_compression_image_in_bytes_considered",
633 : "Size of potentially compressible data written into image layers before compression"
634 : )
635 154 : .expect("failed to define a metric")
636 154 : });
637 :
638 154 : pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES_CHOSEN: Lazy<IntCounter> = Lazy::new(|| {
639 : register_int_counter!(
640 : "pageserver_compression_image_in_bytes_chosen",
641 : "Size of data whose compressed form was written into image layers"
642 : )
643 154 : .expect("failed to define a metric")
644 154 : });
645 :
646 154 : pub(crate) static COMPRESSION_IMAGE_OUTPUT_BYTES: Lazy<IntCounter> = Lazy::new(|| {
647 : register_int_counter!(
648 : "pageserver_compression_image_out_bytes_total",
649 : "Size of compressed image layer written"
650 : )
651 154 : .expect("failed to define a metric")
652 154 : });
653 :
654 : pub(crate) mod initial_logical_size {
655 : use metrics::{register_int_counter, register_int_counter_vec, IntCounter, IntCounterVec};
656 : use once_cell::sync::Lazy;
657 :
658 : pub(crate) struct StartCalculation(IntCounterVec);
659 162 : pub(crate) static START_CALCULATION: Lazy<StartCalculation> = Lazy::new(|| {
660 162 : StartCalculation(
661 162 : register_int_counter_vec!(
662 : "pageserver_initial_logical_size_start_calculation",
663 : "Incremented each time we start an initial logical size calculation attempt. \
664 : The `circumstances` label provides some additional details.",
665 : &["attempt", "circumstances"]
666 162 : )
667 162 : .unwrap(),
668 162 : )
669 162 : });
670 :
671 : struct DropCalculation {
672 : first: IntCounter,
673 : retry: IntCounter,
674 : }
675 :
676 162 : static DROP_CALCULATION: Lazy<DropCalculation> = Lazy::new(|| {
677 162 : let vec = register_int_counter_vec!(
678 : "pageserver_initial_logical_size_drop_calculation",
679 : "Incremented each time we abort a started size calculation attmpt.",
680 : &["attempt"]
681 : )
682 162 : .unwrap();
683 162 : DropCalculation {
684 162 : first: vec.with_label_values(&["first"]),
685 162 : retry: vec.with_label_values(&["retry"]),
686 162 : }
687 162 : });
688 :
689 : pub(crate) struct Calculated {
690 : pub(crate) births: IntCounter,
691 : pub(crate) deaths: IntCounter,
692 : }
693 :
694 162 : pub(crate) static CALCULATED: Lazy<Calculated> = Lazy::new(|| Calculated {
695 : births: register_int_counter!(
696 : "pageserver_initial_logical_size_finish_calculation",
697 : "Incremented every time we finish calculation of initial logical size.\
698 : If everything is working well, this should happen at most once per Timeline object."
699 : )
700 162 : .unwrap(),
701 : deaths: register_int_counter!(
702 : "pageserver_initial_logical_size_drop_finished_calculation",
703 : "Incremented when we drop a finished initial logical size calculation result.\
704 : Mainly useful to turn pageserver_initial_logical_size_finish_calculation into a gauge."
705 : )
706 162 : .unwrap(),
707 162 : });
708 :
709 : pub(crate) struct OngoingCalculationGuard {
710 : inc_drop_calculation: Option<IntCounter>,
711 : }
712 :
713 174 : #[derive(strum_macros::IntoStaticStr)]
714 : pub(crate) enum StartCircumstances {
715 : EmptyInitial,
716 : SkippedConcurrencyLimiter,
717 : AfterBackgroundTasksRateLimit,
718 : }
719 :
720 : impl StartCalculation {
721 174 : pub(crate) fn first(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
722 174 : let circumstances_label: &'static str = circumstances.into();
723 174 : self.0
724 174 : .with_label_values(&["first", circumstances_label])
725 174 : .inc();
726 174 : OngoingCalculationGuard {
727 174 : inc_drop_calculation: Some(DROP_CALCULATION.first.clone()),
728 174 : }
729 174 : }
730 0 : pub(crate) fn retry(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
731 0 : let circumstances_label: &'static str = circumstances.into();
732 0 : self.0
733 0 : .with_label_values(&["retry", circumstances_label])
734 0 : .inc();
735 0 : OngoingCalculationGuard {
736 0 : inc_drop_calculation: Some(DROP_CALCULATION.retry.clone()),
737 0 : }
738 0 : }
739 : }
740 :
741 : impl Drop for OngoingCalculationGuard {
742 174 : fn drop(&mut self) {
743 174 : if let Some(counter) = self.inc_drop_calculation.take() {
744 0 : counter.inc();
745 174 : }
746 174 : }
747 : }
748 :
749 : impl OngoingCalculationGuard {
750 174 : pub(crate) fn calculation_result_saved(mut self) -> FinishedCalculationGuard {
751 174 : drop(self.inc_drop_calculation.take());
752 174 : CALCULATED.births.inc();
753 174 : FinishedCalculationGuard {
754 174 : inc_on_drop: CALCULATED.deaths.clone(),
755 174 : }
756 174 : }
757 : }
758 :
759 : pub(crate) struct FinishedCalculationGuard {
760 : inc_on_drop: IntCounter,
761 : }
762 :
763 : impl Drop for FinishedCalculationGuard {
764 6 : fn drop(&mut self) {
765 6 : self.inc_on_drop.inc();
766 6 : }
767 : }
768 :
769 : // context: https://github.com/neondatabase/neon/issues/5963
770 : pub(crate) static TIMELINES_WHERE_WALRECEIVER_GOT_APPROXIMATE_SIZE: Lazy<IntCounter> =
771 0 : Lazy::new(|| {
772 : register_int_counter!(
773 : "pageserver_initial_logical_size_timelines_where_walreceiver_got_approximate_size",
774 : "Counter for the following event: walreceiver calls\
775 : Timeline::get_current_logical_size() and it returns `Approximate` for the first time."
776 : )
777 0 : .unwrap()
778 0 : });
779 : }
780 :
781 0 : static DIRECTORY_ENTRIES_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
782 : register_uint_gauge_vec!(
783 : "pageserver_directory_entries_count",
784 : "Sum of the entries in pageserver-stored directory listings",
785 : &["tenant_id", "shard_id", "timeline_id"]
786 : )
787 0 : .expect("failed to define a metric")
788 0 : });
789 :
790 164 : pub(crate) static TENANT_STATE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
791 : register_uint_gauge_vec!(
792 : "pageserver_tenant_states_count",
793 : "Count of tenants per state",
794 : &["state"]
795 : )
796 164 : .expect("Failed to register pageserver_tenant_states_count metric")
797 164 : });
798 :
799 : /// A set of broken tenants.
800 : ///
801 : /// These are expected to be so rare that a set is fine. Set as in a new timeseries per each broken
802 : /// tenant.
803 12 : pub(crate) static BROKEN_TENANTS_SET: Lazy<UIntGaugeVec> = Lazy::new(|| {
804 : register_uint_gauge_vec!(
805 : "pageserver_broken_tenants_count",
806 : "Set of broken tenants",
807 : &["tenant_id", "shard_id"]
808 : )
809 12 : .expect("Failed to register pageserver_tenant_states_count metric")
810 12 : });
811 :
812 6 : pub(crate) static TENANT_SYNTHETIC_SIZE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
813 : register_uint_gauge_vec!(
814 : "pageserver_tenant_synthetic_cached_size_bytes",
815 : "Synthetic size of each tenant in bytes",
816 : &["tenant_id"]
817 : )
818 6 : .expect("Failed to register pageserver_tenant_synthetic_cached_size_bytes metric")
819 6 : });
820 :
821 0 : pub(crate) static EVICTION_ITERATION_DURATION: Lazy<HistogramVec> = Lazy::new(|| {
822 : register_histogram_vec!(
823 : "pageserver_eviction_iteration_duration_seconds_global",
824 : "Time spent on a single eviction iteration",
825 : &["period_secs", "threshold_secs"],
826 : STORAGE_OP_BUCKETS.into(),
827 : )
828 0 : .expect("failed to define a metric")
829 0 : });
830 :
831 162 : static EVICTIONS: Lazy<IntCounterVec> = Lazy::new(|| {
832 : register_int_counter_vec!(
833 : "pageserver_evictions",
834 : "Number of layers evicted from the pageserver",
835 : &["tenant_id", "shard_id", "timeline_id"]
836 : )
837 162 : .expect("failed to define a metric")
838 162 : });
839 :
840 162 : static EVICTIONS_WITH_LOW_RESIDENCE_DURATION: Lazy<IntCounterVec> = Lazy::new(|| {
841 : register_int_counter_vec!(
842 : "pageserver_evictions_with_low_residence_duration",
843 : "If a layer is evicted that was resident for less than `low_threshold`, it is counted to this counter. \
844 : Residence duration is determined using the `residence_duration_data_source`.",
845 : &["tenant_id", "shard_id", "timeline_id", "residence_duration_data_source", "low_threshold_secs"]
846 : )
847 162 : .expect("failed to define a metric")
848 162 : });
849 :
850 0 : pub(crate) static UNEXPECTED_ONDEMAND_DOWNLOADS: Lazy<IntCounter> = Lazy::new(|| {
851 : register_int_counter!(
852 : "pageserver_unexpected_ondemand_downloads_count",
853 : "Number of unexpected on-demand downloads. \
854 : We log more context for each increment, so, forgo any labels in this metric.",
855 : )
856 0 : .expect("failed to define a metric")
857 0 : });
858 :
859 : /// How long did we take to start up? Broken down by labels to describe
860 : /// different phases of startup.
861 0 : pub static STARTUP_DURATION: Lazy<GaugeVec> = Lazy::new(|| {
862 : register_gauge_vec!(
863 : "pageserver_startup_duration_seconds",
864 : "Time taken by phases of pageserver startup, in seconds",
865 : &["phase"]
866 : )
867 0 : .expect("Failed to register pageserver_startup_duration_seconds metric")
868 0 : });
869 :
870 0 : pub static STARTUP_IS_LOADING: Lazy<UIntGauge> = Lazy::new(|| {
871 : register_uint_gauge!(
872 : "pageserver_startup_is_loading",
873 : "1 while in initial startup load of tenants, 0 at other times"
874 : )
875 0 : .expect("Failed to register pageserver_startup_is_loading")
876 0 : });
877 :
878 158 : pub(crate) static TIMELINE_EPHEMERAL_BYTES: Lazy<UIntGauge> = Lazy::new(|| {
879 : register_uint_gauge!(
880 : "pageserver_timeline_ephemeral_bytes",
881 : "Total number of bytes in ephemeral layers, summed for all timelines. Approximate, lazily updated."
882 : )
883 158 : .expect("Failed to register metric")
884 158 : });
885 :
886 : /// Metrics related to the lifecycle of a [`crate::tenant::Tenant`] object: things
887 : /// like how long it took to load.
888 : ///
889 : /// Note that these are process-global metrics, _not_ per-tenant metrics. Per-tenant
890 : /// metrics are rather expensive, and usually fine grained stuff makes more sense
891 : /// at a timeline level than tenant level.
892 : pub(crate) struct TenantMetrics {
893 : /// How long did tenants take to go from construction to active state?
894 : pub(crate) activation: Histogram,
895 : pub(crate) preload: Histogram,
896 : pub(crate) attach: Histogram,
897 :
898 : /// How many tenants are included in the initial startup of the pagesrever?
899 : pub(crate) startup_scheduled: IntCounter,
900 : pub(crate) startup_complete: IntCounter,
901 : }
902 :
903 0 : pub(crate) static TENANT: Lazy<TenantMetrics> = Lazy::new(|| {
904 0 : TenantMetrics {
905 0 : activation: register_histogram!(
906 : "pageserver_tenant_activation_seconds",
907 : "Time taken by tenants to activate, in seconds",
908 : CRITICAL_OP_BUCKETS.into()
909 0 : )
910 0 : .expect("Failed to register metric"),
911 0 : preload: register_histogram!(
912 : "pageserver_tenant_preload_seconds",
913 : "Time taken by tenants to load remote metadata on startup/attach, in seconds",
914 : CRITICAL_OP_BUCKETS.into()
915 0 : )
916 0 : .expect("Failed to register metric"),
917 0 : attach: register_histogram!(
918 : "pageserver_tenant_attach_seconds",
919 : "Time taken by tenants to intialize, after remote metadata is already loaded",
920 : CRITICAL_OP_BUCKETS.into()
921 0 : )
922 0 : .expect("Failed to register metric"),
923 0 : startup_scheduled: register_int_counter!(
924 : "pageserver_tenant_startup_scheduled",
925 : "Number of tenants included in pageserver startup (doesn't count tenants attached later)"
926 0 : ).expect("Failed to register metric"),
927 0 : startup_complete: register_int_counter!(
928 : "pageserver_tenant_startup_complete",
929 : "Number of tenants that have completed warm-up, or activated on-demand during initial startup: \
930 : should eventually reach `pageserver_tenant_startup_scheduled_total`. Does not include broken \
931 : tenants: such cases will lead to this metric never reaching the scheduled count."
932 0 : ).expect("Failed to register metric"),
933 0 : }
934 0 : });
935 :
936 : /// Each `Timeline`'s [`EVICTIONS_WITH_LOW_RESIDENCE_DURATION`] metric.
937 : #[derive(Debug)]
938 : pub(crate) struct EvictionsWithLowResidenceDuration {
939 : data_source: &'static str,
940 : threshold: Duration,
941 : counter: Option<IntCounter>,
942 : }
943 :
944 : pub(crate) struct EvictionsWithLowResidenceDurationBuilder {
945 : data_source: &'static str,
946 : threshold: Duration,
947 : }
948 :
949 : impl EvictionsWithLowResidenceDurationBuilder {
950 406 : pub fn new(data_source: &'static str, threshold: Duration) -> Self {
951 406 : Self {
952 406 : data_source,
953 406 : threshold,
954 406 : }
955 406 : }
956 :
957 406 : fn build(
958 406 : &self,
959 406 : tenant_id: &str,
960 406 : shard_id: &str,
961 406 : timeline_id: &str,
962 406 : ) -> EvictionsWithLowResidenceDuration {
963 406 : let counter = EVICTIONS_WITH_LOW_RESIDENCE_DURATION
964 406 : .get_metric_with_label_values(&[
965 406 : tenant_id,
966 406 : shard_id,
967 406 : timeline_id,
968 406 : self.data_source,
969 406 : &EvictionsWithLowResidenceDuration::threshold_label_value(self.threshold),
970 406 : ])
971 406 : .unwrap();
972 406 : EvictionsWithLowResidenceDuration {
973 406 : data_source: self.data_source,
974 406 : threshold: self.threshold,
975 406 : counter: Some(counter),
976 406 : }
977 406 : }
978 : }
979 :
980 : impl EvictionsWithLowResidenceDuration {
981 414 : fn threshold_label_value(threshold: Duration) -> String {
982 414 : format!("{}", threshold.as_secs())
983 414 : }
984 :
985 4 : pub fn observe(&self, observed_value: Duration) {
986 4 : if observed_value < self.threshold {
987 4 : self.counter
988 4 : .as_ref()
989 4 : .expect("nobody calls this function after `remove_from_vec`")
990 4 : .inc();
991 4 : }
992 4 : }
993 :
994 8 : pub fn change_threshold(
995 8 : &mut self,
996 8 : tenant_id: &str,
997 8 : shard_id: &str,
998 8 : timeline_id: &str,
999 8 : new_threshold: Duration,
1000 8 : ) {
1001 8 : if new_threshold == self.threshold {
1002 8 : return;
1003 0 : }
1004 0 : let mut with_new = EvictionsWithLowResidenceDurationBuilder::new(
1005 0 : self.data_source,
1006 0 : new_threshold,
1007 0 : )
1008 0 : .build(tenant_id, shard_id, timeline_id);
1009 0 : std::mem::swap(self, &mut with_new);
1010 0 : with_new.remove(tenant_id, shard_id, timeline_id);
1011 8 : }
1012 :
1013 : // This could be a `Drop` impl, but, we need the `tenant_id` and `timeline_id`.
1014 8 : fn remove(&mut self, tenant_id: &str, shard_id: &str, timeline_id: &str) {
1015 8 : let Some(_counter) = self.counter.take() else {
1016 0 : return;
1017 : };
1018 :
1019 8 : let threshold = Self::threshold_label_value(self.threshold);
1020 8 :
1021 8 : let removed = EVICTIONS_WITH_LOW_RESIDENCE_DURATION.remove_label_values(&[
1022 8 : tenant_id,
1023 8 : shard_id,
1024 8 : timeline_id,
1025 8 : self.data_source,
1026 8 : &threshold,
1027 8 : ]);
1028 8 :
1029 8 : match removed {
1030 0 : Err(e) => {
1031 0 : // this has been hit in staging as
1032 0 : // <https://neondatabase.sentry.io/issues/4142396994/>, but we don't know how.
1033 0 : // because we can be in the drop path already, don't risk:
1034 0 : // - "double-panic => illegal instruction" or
1035 0 : // - future "drop panick => abort"
1036 0 : //
1037 0 : // so just nag: (the error has the labels)
1038 0 : tracing::warn!("failed to remove EvictionsWithLowResidenceDuration, it was already removed? {e:#?}");
1039 : }
1040 : Ok(()) => {
1041 : // to help identify cases where we double-remove the same values, let's log all
1042 : // deletions?
1043 8 : tracing::info!("removed EvictionsWithLowResidenceDuration with {tenant_id}, {timeline_id}, {}, {threshold}", self.data_source);
1044 : }
1045 : }
1046 8 : }
1047 : }
1048 :
1049 : // Metrics collected on disk IO operations
1050 : //
1051 : // Roughly logarithmic scale.
1052 : const STORAGE_IO_TIME_BUCKETS: &[f64] = &[
1053 : 0.000030, // 30 usec
1054 : 0.001000, // 1000 usec
1055 : 0.030, // 30 ms
1056 : 1.000, // 1000 ms
1057 : 30.000, // 30000 ms
1058 : ];
1059 :
1060 : /// VirtualFile fs operation variants.
1061 : ///
1062 : /// Operations:
1063 : /// - open ([`std::fs::OpenOptions::open`])
1064 : /// - close (dropping [`crate::virtual_file::VirtualFile`])
1065 : /// - close-by-replace (close by replacement algorithm)
1066 : /// - read (`read_at`)
1067 : /// - write (`write_at`)
1068 : /// - seek (modify internal position or file length query)
1069 : /// - fsync ([`std::fs::File::sync_all`])
1070 : /// - metadata ([`std::fs::File::metadata`])
1071 : #[derive(
1072 1692 : Debug, Clone, Copy, strum_macros::EnumCount, strum_macros::EnumIter, strum_macros::FromRepr,
1073 : )]
1074 : pub(crate) enum StorageIoOperation {
1075 : Open,
1076 : OpenAfterReplace,
1077 : Close,
1078 : CloseByReplace,
1079 : Read,
1080 : Write,
1081 : Seek,
1082 : Fsync,
1083 : Metadata,
1084 : }
1085 :
1086 : impl StorageIoOperation {
1087 1692 : pub fn as_str(&self) -> &'static str {
1088 1692 : match self {
1089 188 : StorageIoOperation::Open => "open",
1090 188 : StorageIoOperation::OpenAfterReplace => "open-after-replace",
1091 188 : StorageIoOperation::Close => "close",
1092 188 : StorageIoOperation::CloseByReplace => "close-by-replace",
1093 188 : StorageIoOperation::Read => "read",
1094 188 : StorageIoOperation::Write => "write",
1095 188 : StorageIoOperation::Seek => "seek",
1096 188 : StorageIoOperation::Fsync => "fsync",
1097 188 : StorageIoOperation::Metadata => "metadata",
1098 : }
1099 1692 : }
1100 : }
1101 :
1102 : /// Tracks time taken by fs operations near VirtualFile.
1103 : #[derive(Debug)]
1104 : pub(crate) struct StorageIoTime {
1105 : metrics: [Histogram; StorageIoOperation::COUNT],
1106 : }
1107 :
1108 : impl StorageIoTime {
1109 188 : fn new() -> Self {
1110 188 : let storage_io_histogram_vec = register_histogram_vec!(
1111 : "pageserver_io_operations_seconds",
1112 : "Time spent in IO operations",
1113 : &["operation"],
1114 : STORAGE_IO_TIME_BUCKETS.into()
1115 : )
1116 188 : .expect("failed to define a metric");
1117 1692 : let metrics = std::array::from_fn(|i| {
1118 1692 : let op = StorageIoOperation::from_repr(i).unwrap();
1119 1692 : storage_io_histogram_vec
1120 1692 : .get_metric_with_label_values(&[op.as_str()])
1121 1692 : .unwrap()
1122 1692 : });
1123 188 : Self { metrics }
1124 188 : }
1125 :
1126 2009425 : pub(crate) fn get(&self, op: StorageIoOperation) -> &Histogram {
1127 2009425 : &self.metrics[op as usize]
1128 2009425 : }
1129 : }
1130 :
1131 : pub(crate) static STORAGE_IO_TIME_METRIC: Lazy<StorageIoTime> = Lazy::new(StorageIoTime::new);
1132 :
1133 : const STORAGE_IO_SIZE_OPERATIONS: &[&str] = &["read", "write"];
1134 :
1135 : // Needed for the https://neonprod.grafana.net/d/5uK9tHL4k/picking-tenant-for-relocation?orgId=1
1136 184 : pub(crate) static STORAGE_IO_SIZE: Lazy<IntGaugeVec> = Lazy::new(|| {
1137 : register_int_gauge_vec!(
1138 : "pageserver_io_operations_bytes_total",
1139 : "Total amount of bytes read/written in IO operations",
1140 : &["operation", "tenant_id", "shard_id", "timeline_id"]
1141 : )
1142 184 : .expect("failed to define a metric")
1143 184 : });
1144 :
1145 : #[cfg(not(test))]
1146 : pub(crate) mod virtual_file_descriptor_cache {
1147 : use super::*;
1148 :
1149 0 : pub(crate) static SIZE_MAX: Lazy<UIntGauge> = Lazy::new(|| {
1150 : register_uint_gauge!(
1151 : "pageserver_virtual_file_descriptor_cache_size_max",
1152 : "Maximum number of open file descriptors in the cache."
1153 : )
1154 0 : .unwrap()
1155 0 : });
1156 :
1157 : // SIZE_CURRENT: derive it like so:
1158 : // ```
1159 : // sum (pageserver_io_operations_seconds_count{operation=~"^(open|open-after-replace)$")
1160 : // -ignoring(operation)
1161 : // sum(pageserver_io_operations_seconds_count{operation=~"^(close|close-by-replace)$"}
1162 : // ```
1163 : }
1164 :
1165 : #[cfg(not(test))]
1166 : pub(crate) mod virtual_file_io_engine {
1167 : use super::*;
1168 :
1169 0 : pub(crate) static KIND: Lazy<UIntGaugeVec> = Lazy::new(|| {
1170 : register_uint_gauge_vec!(
1171 : "pageserver_virtual_file_io_engine_kind",
1172 : "The configured io engine for VirtualFile",
1173 : &["kind"],
1174 : )
1175 0 : .unwrap()
1176 0 : });
1177 : }
1178 :
1179 : struct GlobalAndPerTimelineHistogramTimer<'a, 'c> {
1180 : global_metric: &'a Histogram,
1181 :
1182 : // Optional because not all op types are tracked per-timeline
1183 : timeline_metric: Option<&'a Histogram>,
1184 :
1185 : ctx: &'c RequestContext,
1186 : start: std::time::Instant,
1187 : op: SmgrQueryType,
1188 : }
1189 :
1190 : impl<'a, 'c> Drop for GlobalAndPerTimelineHistogramTimer<'a, 'c> {
1191 10 : fn drop(&mut self) {
1192 10 : let elapsed = self.start.elapsed();
1193 10 : let ex_throttled = self
1194 10 : .ctx
1195 10 : .micros_spent_throttled
1196 10 : .close_and_checked_sub_from(elapsed);
1197 10 : let ex_throttled = match ex_throttled {
1198 10 : Ok(res) => res,
1199 0 : Err(error) => {
1200 0 : use utils::rate_limit::RateLimit;
1201 0 : static LOGGED: Lazy<Mutex<enum_map::EnumMap<SmgrQueryType, RateLimit>>> =
1202 0 : Lazy::new(|| {
1203 0 : Mutex::new(enum_map::EnumMap::from_array(std::array::from_fn(|_| {
1204 0 : RateLimit::new(Duration::from_secs(10))
1205 0 : })))
1206 0 : });
1207 0 : let mut guard = LOGGED.lock().unwrap();
1208 0 : let rate_limit = &mut guard[self.op];
1209 0 : rate_limit.call(|| {
1210 0 : warn!(op=?self.op, error, "error deducting time spent throttled; this message is logged at a global rate limit");
1211 0 : });
1212 0 : elapsed
1213 : }
1214 : };
1215 10 : self.global_metric.observe(ex_throttled.as_secs_f64());
1216 10 : if let Some(timeline_metric) = self.timeline_metric {
1217 2 : timeline_metric.observe(ex_throttled.as_secs_f64());
1218 8 : }
1219 10 : }
1220 : }
1221 :
1222 : #[derive(
1223 : Debug,
1224 : Clone,
1225 : Copy,
1226 2514 : IntoStaticStr,
1227 : strum_macros::EnumCount,
1228 24 : strum_macros::EnumIter,
1229 2080 : strum_macros::FromRepr,
1230 : enum_map::Enum,
1231 : )]
1232 : #[strum(serialize_all = "snake_case")]
1233 : pub enum SmgrQueryType {
1234 : GetRelExists,
1235 : GetRelSize,
1236 : GetPageAtLsn,
1237 : GetDbSize,
1238 : GetSlruSegment,
1239 : }
1240 :
1241 : #[derive(Debug)]
1242 : pub(crate) struct SmgrQueryTimePerTimeline {
1243 : global_metrics: [Histogram; SmgrQueryType::COUNT],
1244 : per_timeline_getpage: Histogram,
1245 : }
1246 :
1247 164 : static SMGR_QUERY_TIME_PER_TENANT_TIMELINE: Lazy<HistogramVec> = Lazy::new(|| {
1248 : register_histogram_vec!(
1249 : "pageserver_smgr_query_seconds",
1250 : "Time spent on smgr query handling, aggegated by query type and tenant/timeline.",
1251 : &["smgr_query_type", "tenant_id", "shard_id", "timeline_id"],
1252 : CRITICAL_OP_BUCKETS.into(),
1253 : )
1254 164 : .expect("failed to define a metric")
1255 164 : });
1256 :
1257 164 : static SMGR_QUERY_TIME_GLOBAL_BUCKETS: Lazy<Vec<f64>> = Lazy::new(|| {
1258 164 : [
1259 164 : 1,
1260 164 : 10,
1261 164 : 20,
1262 164 : 40,
1263 164 : 60,
1264 164 : 80,
1265 164 : 100,
1266 164 : 200,
1267 164 : 300,
1268 164 : 400,
1269 164 : 500,
1270 164 : 600,
1271 164 : 700,
1272 164 : 800,
1273 164 : 900,
1274 164 : 1_000, // 1ms
1275 164 : 2_000,
1276 164 : 4_000,
1277 164 : 6_000,
1278 164 : 8_000,
1279 164 : 10_000, // 10ms
1280 164 : 20_000,
1281 164 : 40_000,
1282 164 : 60_000,
1283 164 : 80_000,
1284 164 : 100_000,
1285 164 : 200_000,
1286 164 : 400_000,
1287 164 : 600_000,
1288 164 : 800_000,
1289 164 : 1_000_000, // 1s
1290 164 : 2_000_000,
1291 164 : 4_000_000,
1292 164 : 6_000_000,
1293 164 : 8_000_000,
1294 164 : 10_000_000, // 10s
1295 164 : 20_000_000,
1296 164 : 50_000_000,
1297 164 : 100_000_000,
1298 164 : 200_000_000,
1299 164 : 1_000_000_000, // 1000s
1300 164 : ]
1301 164 : .into_iter()
1302 164 : .map(Duration::from_micros)
1303 6724 : .map(|d| d.as_secs_f64())
1304 164 : .collect()
1305 164 : });
1306 :
1307 164 : static SMGR_QUERY_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
1308 : register_histogram_vec!(
1309 : "pageserver_smgr_query_seconds_global",
1310 : "Time spent on smgr query handling, aggregated by query type.",
1311 : &["smgr_query_type"],
1312 : SMGR_QUERY_TIME_GLOBAL_BUCKETS.clone(),
1313 : )
1314 164 : .expect("failed to define a metric")
1315 164 : });
1316 :
1317 : impl SmgrQueryTimePerTimeline {
1318 416 : pub(crate) fn new(tenant_shard_id: &TenantShardId, timeline_id: &TimelineId) -> Self {
1319 416 : let tenant_id = tenant_shard_id.tenant_id.to_string();
1320 416 : let shard_slug = format!("{}", tenant_shard_id.shard_slug());
1321 416 : let timeline_id = timeline_id.to_string();
1322 2080 : let global_metrics = std::array::from_fn(|i| {
1323 2080 : let op = SmgrQueryType::from_repr(i).unwrap();
1324 2080 : SMGR_QUERY_TIME_GLOBAL
1325 2080 : .get_metric_with_label_values(&[op.into()])
1326 2080 : .unwrap()
1327 2080 : });
1328 416 :
1329 416 : let per_timeline_getpage = SMGR_QUERY_TIME_PER_TENANT_TIMELINE
1330 416 : .get_metric_with_label_values(&[
1331 416 : SmgrQueryType::GetPageAtLsn.into(),
1332 416 : &tenant_id,
1333 416 : &shard_slug,
1334 416 : &timeline_id,
1335 416 : ])
1336 416 : .unwrap();
1337 416 : Self {
1338 416 : global_metrics,
1339 416 : per_timeline_getpage,
1340 416 : }
1341 416 : }
1342 10 : pub(crate) fn start_timer<'c: 'a, 'a>(
1343 10 : &'a self,
1344 10 : op: SmgrQueryType,
1345 10 : ctx: &'c RequestContext,
1346 10 : ) -> Option<impl Drop + '_> {
1347 10 : let global_metric = &self.global_metrics[op as usize];
1348 10 : let start = Instant::now();
1349 10 : match ctx.micros_spent_throttled.open() {
1350 10 : Ok(()) => (),
1351 0 : Err(error) => {
1352 0 : use utils::rate_limit::RateLimit;
1353 0 : static LOGGED: Lazy<Mutex<enum_map::EnumMap<SmgrQueryType, RateLimit>>> =
1354 0 : Lazy::new(|| {
1355 0 : Mutex::new(enum_map::EnumMap::from_array(std::array::from_fn(|_| {
1356 0 : RateLimit::new(Duration::from_secs(10))
1357 0 : })))
1358 0 : });
1359 0 : let mut guard = LOGGED.lock().unwrap();
1360 0 : let rate_limit = &mut guard[op];
1361 0 : rate_limit.call(|| {
1362 0 : warn!(?op, error, "error opening micros_spent_throttled; this message is logged at a global rate limit");
1363 0 : });
1364 0 : }
1365 : }
1366 :
1367 10 : let timeline_metric = if matches!(op, SmgrQueryType::GetPageAtLsn) {
1368 2 : Some(&self.per_timeline_getpage)
1369 : } else {
1370 8 : None
1371 : };
1372 :
1373 10 : Some(GlobalAndPerTimelineHistogramTimer {
1374 10 : global_metric,
1375 10 : timeline_metric,
1376 10 : ctx,
1377 10 : start,
1378 10 : op,
1379 10 : })
1380 10 : }
1381 : }
1382 :
1383 : #[cfg(test)]
1384 : mod smgr_query_time_tests {
1385 : use pageserver_api::shard::TenantShardId;
1386 : use strum::IntoEnumIterator;
1387 : use utils::id::{TenantId, TimelineId};
1388 :
1389 : use crate::{
1390 : context::{DownloadBehavior, RequestContext},
1391 : task_mgr::TaskKind,
1392 : };
1393 :
1394 : // Regression test, we used hard-coded string constants before using an enum.
1395 : #[test]
1396 2 : fn op_label_name() {
1397 2 : use super::SmgrQueryType::*;
1398 2 : let expect: [(super::SmgrQueryType, &'static str); 5] = [
1399 2 : (GetRelExists, "get_rel_exists"),
1400 2 : (GetRelSize, "get_rel_size"),
1401 2 : (GetPageAtLsn, "get_page_at_lsn"),
1402 2 : (GetDbSize, "get_db_size"),
1403 2 : (GetSlruSegment, "get_slru_segment"),
1404 2 : ];
1405 12 : for (op, expect) in expect {
1406 10 : let actual: &'static str = op.into();
1407 10 : assert_eq!(actual, expect);
1408 : }
1409 2 : }
1410 :
1411 : #[test]
1412 2 : fn basic() {
1413 2 : let ops: Vec<_> = super::SmgrQueryType::iter().collect();
1414 :
1415 12 : for op in &ops {
1416 10 : let tenant_id = TenantId::generate();
1417 10 : let timeline_id = TimelineId::generate();
1418 10 : let metrics = super::SmgrQueryTimePerTimeline::new(
1419 10 : &TenantShardId::unsharded(tenant_id),
1420 10 : &timeline_id,
1421 10 : );
1422 10 :
1423 20 : let get_counts = || {
1424 20 : let global: u64 = ops
1425 20 : .iter()
1426 100 : .map(|op| metrics.global_metrics[*op as usize].get_sample_count())
1427 20 : .sum();
1428 20 : (global, metrics.per_timeline_getpage.get_sample_count())
1429 20 : };
1430 :
1431 10 : let (pre_global, pre_per_tenant_timeline) = get_counts();
1432 10 : assert_eq!(pre_per_tenant_timeline, 0);
1433 :
1434 10 : let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Download);
1435 10 : let timer = metrics.start_timer(*op, &ctx);
1436 10 : drop(timer);
1437 10 :
1438 10 : let (post_global, post_per_tenant_timeline) = get_counts();
1439 10 : if matches!(op, super::SmgrQueryType::GetPageAtLsn) {
1440 : // getpage ops are tracked per-timeline, others aren't
1441 2 : assert_eq!(post_per_tenant_timeline, 1);
1442 : } else {
1443 8 : assert_eq!(post_per_tenant_timeline, 0);
1444 : }
1445 10 : assert!(post_global > pre_global);
1446 : }
1447 2 : }
1448 : }
1449 :
1450 : // keep in sync with control plane Go code so that we can validate
1451 : // compute's basebackup_ms metric with our perspective in the context of SLI/SLO.
1452 0 : static COMPUTE_STARTUP_BUCKETS: Lazy<[f64; 28]> = Lazy::new(|| {
1453 0 : // Go code uses milliseconds. Variable is called `computeStartupBuckets`
1454 0 : [
1455 0 : 5, 10, 20, 30, 50, 70, 100, 120, 150, 200, 250, 300, 350, 400, 450, 500, 600, 800, 1000,
1456 0 : 1500, 2000, 2500, 3000, 5000, 10000, 20000, 40000, 60000,
1457 0 : ]
1458 0 : .map(|ms| (ms as f64) / 1000.0)
1459 0 : });
1460 :
1461 : pub(crate) struct BasebackupQueryTime {
1462 : ok: Histogram,
1463 : error: Histogram,
1464 : }
1465 :
1466 0 : pub(crate) static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|| {
1467 0 : let vec = register_histogram_vec!(
1468 : "pageserver_basebackup_query_seconds",
1469 : "Histogram of basebackup queries durations, by result type",
1470 : &["result"],
1471 : COMPUTE_STARTUP_BUCKETS.to_vec(),
1472 : )
1473 0 : .expect("failed to define a metric");
1474 0 : BasebackupQueryTime {
1475 0 : ok: vec.get_metric_with_label_values(&["ok"]).unwrap(),
1476 0 : error: vec.get_metric_with_label_values(&["error"]).unwrap(),
1477 0 : }
1478 0 : });
1479 :
1480 : pub(crate) struct BasebackupQueryTimeOngoingRecording<'a, 'c> {
1481 : parent: &'a BasebackupQueryTime,
1482 : ctx: &'c RequestContext,
1483 : start: std::time::Instant,
1484 : }
1485 :
1486 : impl BasebackupQueryTime {
1487 0 : pub(crate) fn start_recording<'c: 'a, 'a>(
1488 0 : &'a self,
1489 0 : ctx: &'c RequestContext,
1490 0 : ) -> BasebackupQueryTimeOngoingRecording<'_, '_> {
1491 0 : let start = Instant::now();
1492 0 : match ctx.micros_spent_throttled.open() {
1493 0 : Ok(()) => (),
1494 0 : Err(error) => {
1495 0 : use utils::rate_limit::RateLimit;
1496 0 : static LOGGED: Lazy<Mutex<RateLimit>> =
1497 0 : Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
1498 0 : let mut rate_limit = LOGGED.lock().unwrap();
1499 0 : rate_limit.call(|| {
1500 0 : warn!(error, "error opening micros_spent_throttled; this message is logged at a global rate limit");
1501 0 : });
1502 0 : }
1503 : }
1504 0 : BasebackupQueryTimeOngoingRecording {
1505 0 : parent: self,
1506 0 : ctx,
1507 0 : start,
1508 0 : }
1509 0 : }
1510 : }
1511 :
1512 : impl<'a, 'c> BasebackupQueryTimeOngoingRecording<'a, 'c> {
1513 0 : pub(crate) fn observe<T, E>(self, res: &Result<T, E>) {
1514 0 : let elapsed = self.start.elapsed();
1515 0 : let ex_throttled = self
1516 0 : .ctx
1517 0 : .micros_spent_throttled
1518 0 : .close_and_checked_sub_from(elapsed);
1519 0 : let ex_throttled = match ex_throttled {
1520 0 : Ok(ex_throttled) => ex_throttled,
1521 0 : Err(error) => {
1522 0 : use utils::rate_limit::RateLimit;
1523 0 : static LOGGED: Lazy<Mutex<RateLimit>> =
1524 0 : Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
1525 0 : let mut rate_limit = LOGGED.lock().unwrap();
1526 0 : rate_limit.call(|| {
1527 0 : warn!(error, "error deducting time spent throttled; this message is logged at a global rate limit");
1528 0 : });
1529 0 : elapsed
1530 : }
1531 : };
1532 0 : let metric = if res.is_ok() {
1533 0 : &self.parent.ok
1534 : } else {
1535 0 : &self.parent.error
1536 : };
1537 0 : metric.observe(ex_throttled.as_secs_f64());
1538 0 : }
1539 : }
1540 :
1541 0 : pub(crate) static LIVE_CONNECTIONS: Lazy<IntCounterPairVec> = Lazy::new(|| {
1542 : register_int_counter_pair_vec!(
1543 : "pageserver_live_connections_started",
1544 : "Number of network connections that we started handling",
1545 : "pageserver_live_connections_finished",
1546 : "Number of network connections that we finished handling",
1547 : &["pageserver_connection_kind"]
1548 : )
1549 0 : .expect("failed to define a metric")
1550 0 : });
1551 :
1552 0 : #[derive(Clone, Copy, enum_map::Enum, IntoStaticStr)]
1553 : pub(crate) enum ComputeCommandKind {
1554 : PageStreamV2,
1555 : PageStream,
1556 : Basebackup,
1557 : Fullbackup,
1558 : LeaseLsn,
1559 : }
1560 :
1561 : pub(crate) struct ComputeCommandCounters {
1562 : map: EnumMap<ComputeCommandKind, IntCounter>,
1563 : }
1564 :
1565 0 : pub(crate) static COMPUTE_COMMANDS_COUNTERS: Lazy<ComputeCommandCounters> = Lazy::new(|| {
1566 0 : let inner = register_int_counter_vec!(
1567 : "pageserver_compute_commands",
1568 : "Number of compute -> pageserver commands processed",
1569 : &["command"]
1570 : )
1571 0 : .expect("failed to define a metric");
1572 0 :
1573 0 : ComputeCommandCounters {
1574 0 : map: EnumMap::from_array(std::array::from_fn(|i| {
1575 0 : let command = <ComputeCommandKind as enum_map::Enum>::from_usize(i);
1576 0 : let command_str: &'static str = command.into();
1577 0 : inner.with_label_values(&[command_str])
1578 0 : })),
1579 0 : }
1580 0 : });
1581 :
1582 : impl ComputeCommandCounters {
1583 0 : pub(crate) fn for_command(&self, command: ComputeCommandKind) -> &IntCounter {
1584 0 : &self.map[command]
1585 0 : }
1586 : }
1587 :
1588 : // remote storage metrics
1589 :
1590 160 : static REMOTE_TIMELINE_CLIENT_CALLS: Lazy<IntCounterPairVec> = Lazy::new(|| {
1591 : register_int_counter_pair_vec!(
1592 : "pageserver_remote_timeline_client_calls_started",
1593 : "Number of started calls to remote timeline client.",
1594 : "pageserver_remote_timeline_client_calls_finished",
1595 : "Number of finshed calls to remote timeline client.",
1596 : &[
1597 : "tenant_id",
1598 : "shard_id",
1599 : "timeline_id",
1600 : "file_kind",
1601 : "op_kind"
1602 : ],
1603 : )
1604 160 : .unwrap()
1605 160 : });
1606 :
1607 : static REMOTE_TIMELINE_CLIENT_BYTES_STARTED_COUNTER: Lazy<IntCounterVec> =
1608 156 : Lazy::new(|| {
1609 : register_int_counter_vec!(
1610 : "pageserver_remote_timeline_client_bytes_started",
1611 : "Incremented by the number of bytes associated with a remote timeline client operation. \
1612 : The increment happens when the operation is scheduled.",
1613 : &["tenant_id", "shard_id", "timeline_id", "file_kind", "op_kind"],
1614 : )
1615 156 : .expect("failed to define a metric")
1616 156 : });
1617 :
1618 156 : static REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER: Lazy<IntCounterVec> = Lazy::new(|| {
1619 : register_int_counter_vec!(
1620 : "pageserver_remote_timeline_client_bytes_finished",
1621 : "Incremented by the number of bytes associated with a remote timeline client operation. \
1622 : The increment happens when the operation finishes (regardless of success/failure/shutdown).",
1623 : &["tenant_id", "shard_id", "timeline_id", "file_kind", "op_kind"],
1624 : )
1625 156 : .expect("failed to define a metric")
1626 156 : });
1627 :
1628 : pub(crate) struct TenantManagerMetrics {
1629 : tenant_slots_attached: UIntGauge,
1630 : tenant_slots_secondary: UIntGauge,
1631 : tenant_slots_inprogress: UIntGauge,
1632 : pub(crate) tenant_slot_writes: IntCounter,
1633 : pub(crate) unexpected_errors: IntCounter,
1634 : }
1635 :
1636 : impl TenantManagerMetrics {
1637 : /// Helpers for tracking slots. Note that these do not track the lifetime of TenantSlot objects
1638 : /// exactly: they track the lifetime of the slots _in the tenant map_.
1639 2 : pub(crate) fn slot_inserted(&self, slot: &TenantSlot) {
1640 2 : match slot {
1641 0 : TenantSlot::Attached(_) => {
1642 0 : self.tenant_slots_attached.inc();
1643 0 : }
1644 0 : TenantSlot::Secondary(_) => {
1645 0 : self.tenant_slots_secondary.inc();
1646 0 : }
1647 2 : TenantSlot::InProgress(_) => {
1648 2 : self.tenant_slots_inprogress.inc();
1649 2 : }
1650 : }
1651 2 : }
1652 :
1653 2 : pub(crate) fn slot_removed(&self, slot: &TenantSlot) {
1654 2 : match slot {
1655 2 : TenantSlot::Attached(_) => {
1656 2 : self.tenant_slots_attached.dec();
1657 2 : }
1658 0 : TenantSlot::Secondary(_) => {
1659 0 : self.tenant_slots_secondary.dec();
1660 0 : }
1661 0 : TenantSlot::InProgress(_) => {
1662 0 : self.tenant_slots_inprogress.dec();
1663 0 : }
1664 : }
1665 2 : }
1666 :
1667 : #[cfg(all(debug_assertions, not(test)))]
1668 0 : pub(crate) fn slots_total(&self) -> u64 {
1669 0 : self.tenant_slots_attached.get()
1670 0 : + self.tenant_slots_secondary.get()
1671 0 : + self.tenant_slots_inprogress.get()
1672 0 : }
1673 : }
1674 :
1675 2 : pub(crate) static TENANT_MANAGER: Lazy<TenantManagerMetrics> = Lazy::new(|| {
1676 2 : let tenant_slots = register_uint_gauge_vec!(
1677 : "pageserver_tenant_manager_slots",
1678 : "How many slots currently exist, including all attached, secondary and in-progress operations",
1679 : &["mode"]
1680 : )
1681 2 : .expect("failed to define a metric");
1682 2 : TenantManagerMetrics {
1683 2 : tenant_slots_attached: tenant_slots
1684 2 : .get_metric_with_label_values(&["attached"])
1685 2 : .unwrap(),
1686 2 : tenant_slots_secondary: tenant_slots
1687 2 : .get_metric_with_label_values(&["secondary"])
1688 2 : .unwrap(),
1689 2 : tenant_slots_inprogress: tenant_slots
1690 2 : .get_metric_with_label_values(&["inprogress"])
1691 2 : .unwrap(),
1692 2 : tenant_slot_writes: register_int_counter!(
1693 : "pageserver_tenant_manager_slot_writes",
1694 : "Writes to a tenant slot, including all of create/attach/detach/delete"
1695 2 : )
1696 2 : .expect("failed to define a metric"),
1697 2 : unexpected_errors: register_int_counter!(
1698 : "pageserver_tenant_manager_unexpected_errors_total",
1699 : "Number of unexpected conditions encountered: nonzero value indicates a non-fatal bug."
1700 2 : )
1701 2 : .expect("failed to define a metric"),
1702 2 : }
1703 2 : });
1704 :
1705 : pub(crate) struct DeletionQueueMetrics {
1706 : pub(crate) keys_submitted: IntCounter,
1707 : pub(crate) keys_dropped: IntCounter,
1708 : pub(crate) keys_executed: IntCounter,
1709 : pub(crate) keys_validated: IntCounter,
1710 : pub(crate) dropped_lsn_updates: IntCounter,
1711 : pub(crate) unexpected_errors: IntCounter,
1712 : pub(crate) remote_errors: IntCounterVec,
1713 : }
1714 28 : pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
1715 28 : DeletionQueueMetrics{
1716 28 :
1717 28 : keys_submitted: register_int_counter!(
1718 : "pageserver_deletion_queue_submitted_total",
1719 : "Number of objects submitted for deletion"
1720 28 : )
1721 28 : .expect("failed to define a metric"),
1722 28 :
1723 28 : keys_dropped: register_int_counter!(
1724 : "pageserver_deletion_queue_dropped_total",
1725 : "Number of object deletions dropped due to stale generation."
1726 28 : )
1727 28 : .expect("failed to define a metric"),
1728 28 :
1729 28 : keys_executed: register_int_counter!(
1730 : "pageserver_deletion_queue_executed_total",
1731 : "Number of objects deleted. Only includes objects that we actually deleted, sum with pageserver_deletion_queue_dropped_total for the total number of keys processed to completion"
1732 28 : )
1733 28 : .expect("failed to define a metric"),
1734 28 :
1735 28 : keys_validated: register_int_counter!(
1736 : "pageserver_deletion_queue_validated_total",
1737 : "Number of keys validated for deletion. Sum with pageserver_deletion_queue_dropped_total for the total number of keys that have passed through the validation stage."
1738 28 : )
1739 28 : .expect("failed to define a metric"),
1740 28 :
1741 28 : dropped_lsn_updates: register_int_counter!(
1742 : "pageserver_deletion_queue_dropped_lsn_updates_total",
1743 : "Updates to remote_consistent_lsn dropped due to stale generation number."
1744 28 : )
1745 28 : .expect("failed to define a metric"),
1746 28 : unexpected_errors: register_int_counter!(
1747 : "pageserver_deletion_queue_unexpected_errors_total",
1748 : "Number of unexpected condiions that may stall the queue: any value above zero is unexpected."
1749 28 : )
1750 28 : .expect("failed to define a metric"),
1751 28 : remote_errors: register_int_counter_vec!(
1752 : "pageserver_deletion_queue_remote_errors_total",
1753 : "Retryable remote I/O errors while executing deletions, for example 503 responses to DeleteObjects",
1754 : &["op_kind"],
1755 28 : )
1756 28 : .expect("failed to define a metric")
1757 28 : }
1758 28 : });
1759 :
1760 : pub(crate) struct SecondaryModeMetrics {
1761 : pub(crate) upload_heatmap: IntCounter,
1762 : pub(crate) upload_heatmap_errors: IntCounter,
1763 : pub(crate) upload_heatmap_duration: Histogram,
1764 : pub(crate) download_heatmap: IntCounter,
1765 : pub(crate) download_layer: IntCounter,
1766 : }
1767 0 : pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| {
1768 0 : SecondaryModeMetrics {
1769 0 : upload_heatmap: register_int_counter!(
1770 : "pageserver_secondary_upload_heatmap",
1771 : "Number of heatmaps written to remote storage by attached tenants"
1772 0 : )
1773 0 : .expect("failed to define a metric"),
1774 0 : upload_heatmap_errors: register_int_counter!(
1775 : "pageserver_secondary_upload_heatmap_errors",
1776 : "Failures writing heatmap to remote storage"
1777 0 : )
1778 0 : .expect("failed to define a metric"),
1779 0 : upload_heatmap_duration: register_histogram!(
1780 : "pageserver_secondary_upload_heatmap_duration",
1781 : "Time to build and upload a heatmap, including any waiting inside the S3 client"
1782 0 : )
1783 0 : .expect("failed to define a metric"),
1784 0 : download_heatmap: register_int_counter!(
1785 : "pageserver_secondary_download_heatmap",
1786 : "Number of downloads of heatmaps by secondary mode locations, including when it hasn't changed"
1787 0 : )
1788 0 : .expect("failed to define a metric"),
1789 0 : download_layer: register_int_counter!(
1790 : "pageserver_secondary_download_layer",
1791 : "Number of downloads of layers by secondary mode locations"
1792 0 : )
1793 0 : .expect("failed to define a metric"),
1794 0 : }
1795 0 : });
1796 :
1797 0 : pub(crate) static SECONDARY_RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
1798 : register_uint_gauge_vec!(
1799 : "pageserver_secondary_resident_physical_size",
1800 : "The size of the layer files present in the pageserver's filesystem, for secondary locations.",
1801 : &["tenant_id", "shard_id"]
1802 : )
1803 0 : .expect("failed to define a metric")
1804 0 : });
1805 :
1806 0 : pub(crate) static SECONDARY_HEATMAP_TOTAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
1807 : register_uint_gauge_vec!(
1808 : "pageserver_secondary_heatmap_total_size",
1809 : "The total size in bytes of all layers in the most recently downloaded heatmap.",
1810 : &["tenant_id", "shard_id"]
1811 : )
1812 0 : .expect("failed to define a metric")
1813 0 : });
1814 :
1815 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
1816 : pub enum RemoteOpKind {
1817 : Upload,
1818 : Download,
1819 : Delete,
1820 : }
1821 : impl RemoteOpKind {
1822 13314 : pub fn as_str(&self) -> &'static str {
1823 13314 : match self {
1824 12547 : Self::Upload => "upload",
1825 52 : Self::Download => "download",
1826 715 : Self::Delete => "delete",
1827 : }
1828 13314 : }
1829 : }
1830 :
1831 : #[derive(Debug, Clone, Copy, Hash, PartialEq, Eq)]
1832 : pub enum RemoteOpFileKind {
1833 : Layer,
1834 : Index,
1835 : }
1836 : impl RemoteOpFileKind {
1837 13314 : pub fn as_str(&self) -> &'static str {
1838 13314 : match self {
1839 9022 : Self::Layer => "layer",
1840 4292 : Self::Index => "index",
1841 : }
1842 13314 : }
1843 : }
1844 :
1845 157 : pub(crate) static REMOTE_OPERATION_TIME: Lazy<HistogramVec> = Lazy::new(|| {
1846 : register_histogram_vec!(
1847 : "pageserver_remote_operation_seconds",
1848 : "Time spent on remote storage operations. \
1849 : Grouped by tenant, timeline, operation_kind and status. \
1850 : Does not account for time spent waiting in remote timeline client's queues.",
1851 : &["file_kind", "op_kind", "status"]
1852 : )
1853 157 : .expect("failed to define a metric")
1854 157 : });
1855 :
1856 0 : pub(crate) static TENANT_TASK_EVENTS: Lazy<IntCounterVec> = Lazy::new(|| {
1857 : register_int_counter_vec!(
1858 : "pageserver_tenant_task_events",
1859 : "Number of task start/stop/fail events.",
1860 : &["event"],
1861 : )
1862 0 : .expect("Failed to register tenant_task_events metric")
1863 0 : });
1864 :
1865 : pub struct BackgroundLoopSemaphoreMetrics {
1866 : counters: EnumMap<BackgroundLoopKind, IntCounterPair>,
1867 : durations: EnumMap<BackgroundLoopKind, Counter>,
1868 : }
1869 :
1870 : pub(crate) static BACKGROUND_LOOP_SEMAPHORE: Lazy<BackgroundLoopSemaphoreMetrics> = Lazy::new(
1871 20 : || {
1872 20 : let counters = register_int_counter_pair_vec!(
1873 : "pageserver_background_loop_semaphore_wait_start_count",
1874 : "Counter for background loop concurrency-limiting semaphore acquire calls started",
1875 : "pageserver_background_loop_semaphore_wait_finish_count",
1876 : "Counter for background loop concurrency-limiting semaphore acquire calls finished",
1877 : &["task"],
1878 : )
1879 20 : .unwrap();
1880 20 :
1881 20 : let durations = register_counter_vec!(
1882 : "pageserver_background_loop_semaphore_wait_duration_seconds",
1883 : "Sum of wall clock time spent waiting on the background loop concurrency-limiting semaphore acquire calls",
1884 : &["task"],
1885 : )
1886 20 : .unwrap();
1887 20 :
1888 20 : BackgroundLoopSemaphoreMetrics {
1889 180 : counters: enum_map::EnumMap::from_array(std::array::from_fn(|i| {
1890 180 : let kind = <BackgroundLoopKind as enum_map::Enum>::from_usize(i);
1891 180 : counters.with_label_values(&[kind.into()])
1892 180 : })),
1893 180 : durations: enum_map::EnumMap::from_array(std::array::from_fn(|i| {
1894 180 : let kind = <BackgroundLoopKind as enum_map::Enum>::from_usize(i);
1895 180 : durations.with_label_values(&[kind.into()])
1896 180 : })),
1897 20 : }
1898 20 : },
1899 : );
1900 :
1901 : impl BackgroundLoopSemaphoreMetrics {
1902 364 : pub(crate) fn measure_acquisition(&self, task: BackgroundLoopKind) -> impl Drop + '_ {
1903 364 : struct Record<'a> {
1904 364 : metrics: &'a BackgroundLoopSemaphoreMetrics,
1905 364 : task: BackgroundLoopKind,
1906 364 : _counter_guard: metrics::IntCounterPairGuard,
1907 364 : start: Instant,
1908 364 : }
1909 364 : impl Drop for Record<'_> {
1910 364 : fn drop(&mut self) {
1911 364 : let elapsed = self.start.elapsed().as_secs_f64();
1912 364 : self.metrics.durations[self.task].inc_by(elapsed);
1913 364 : }
1914 364 : }
1915 364 : Record {
1916 364 : metrics: self,
1917 364 : task,
1918 364 : _counter_guard: self.counters[task].guard(),
1919 364 : start: Instant::now(),
1920 364 : }
1921 364 : }
1922 : }
1923 :
1924 0 : pub(crate) static BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT: Lazy<IntCounterVec> = Lazy::new(|| {
1925 : register_int_counter_vec!(
1926 : "pageserver_background_loop_period_overrun_count",
1927 : "Incremented whenever warn_when_period_overrun() logs a warning.",
1928 : &["task", "period"],
1929 : )
1930 0 : .expect("failed to define a metric")
1931 0 : });
1932 :
1933 : // walreceiver metrics
1934 :
1935 0 : pub(crate) static WALRECEIVER_STARTED_CONNECTIONS: Lazy<IntCounter> = Lazy::new(|| {
1936 : register_int_counter!(
1937 : "pageserver_walreceiver_started_connections_total",
1938 : "Number of started walreceiver connections"
1939 : )
1940 0 : .expect("failed to define a metric")
1941 0 : });
1942 :
1943 0 : pub(crate) static WALRECEIVER_ACTIVE_MANAGERS: Lazy<IntGauge> = Lazy::new(|| {
1944 : register_int_gauge!(
1945 : "pageserver_walreceiver_active_managers",
1946 : "Number of active walreceiver managers"
1947 : )
1948 0 : .expect("failed to define a metric")
1949 0 : });
1950 :
1951 0 : pub(crate) static WALRECEIVER_SWITCHES: Lazy<IntCounterVec> = Lazy::new(|| {
1952 : register_int_counter_vec!(
1953 : "pageserver_walreceiver_switches_total",
1954 : "Number of walreceiver manager change_connection calls",
1955 : &["reason"]
1956 : )
1957 0 : .expect("failed to define a metric")
1958 0 : });
1959 :
1960 0 : pub(crate) static WALRECEIVER_BROKER_UPDATES: Lazy<IntCounter> = Lazy::new(|| {
1961 : register_int_counter!(
1962 : "pageserver_walreceiver_broker_updates_total",
1963 : "Number of received broker updates in walreceiver"
1964 : )
1965 0 : .expect("failed to define a metric")
1966 0 : });
1967 :
1968 2 : pub(crate) static WALRECEIVER_CANDIDATES_EVENTS: Lazy<IntCounterVec> = Lazy::new(|| {
1969 : register_int_counter_vec!(
1970 : "pageserver_walreceiver_candidates_events_total",
1971 : "Number of walreceiver candidate events",
1972 : &["event"]
1973 : )
1974 2 : .expect("failed to define a metric")
1975 2 : });
1976 :
1977 : pub(crate) static WALRECEIVER_CANDIDATES_ADDED: Lazy<IntCounter> =
1978 0 : Lazy::new(|| WALRECEIVER_CANDIDATES_EVENTS.with_label_values(&["add"]));
1979 :
1980 : pub(crate) static WALRECEIVER_CANDIDATES_REMOVED: Lazy<IntCounter> =
1981 2 : Lazy::new(|| WALRECEIVER_CANDIDATES_EVENTS.with_label_values(&["remove"]));
1982 :
1983 : // Metrics collected on WAL redo operations
1984 : //
1985 : // We collect the time spent in actual WAL redo ('redo'), and time waiting
1986 : // for access to the postgres process ('wait') since there is only one for
1987 : // each tenant.
1988 :
1989 : /// Time buckets are small because we want to be able to measure the
1990 : /// smallest redo processing times. These buckets allow us to measure down
1991 : /// to 5us, which equates to 200'000 pages/sec, which equates to 1.6GB/sec.
1992 : /// This is much better than the previous 5ms aka 200 pages/sec aka 1.6MB/sec.
1993 : ///
1994 : /// Values up to 1s are recorded because metrics show that we have redo
1995 : /// durations and lock times larger than 0.250s.
1996 : macro_rules! redo_histogram_time_buckets {
1997 : () => {
1998 : vec![
1999 : 0.000_005, 0.000_010, 0.000_025, 0.000_050, 0.000_100, 0.000_250, 0.000_500, 0.001_000,
2000 : 0.002_500, 0.005_000, 0.010_000, 0.025_000, 0.050_000, 0.100_000, 0.250_000, 0.500_000,
2001 : 1.000_000,
2002 : ]
2003 : };
2004 : }
2005 :
2006 : /// While we're at it, also measure the amount of records replayed in each
2007 : /// operation. We have a global 'total replayed' counter, but that's not
2008 : /// as useful as 'what is the skew for how many records we replay in one
2009 : /// operation'.
2010 : macro_rules! redo_histogram_count_buckets {
2011 : () => {
2012 : vec![0.0, 1.0, 2.0, 5.0, 10.0, 25.0, 50.0, 100.0, 250.0, 500.0]
2013 : };
2014 : }
2015 :
2016 : macro_rules! redo_bytes_histogram_count_buckets {
2017 : () => {
2018 : // powers of (2^.5), from 2^4.5 to 2^15 (22 buckets)
2019 : // rounded up to the next multiple of 8 to capture any MAXALIGNed record of that size, too.
2020 : vec![
2021 : 24.0, 32.0, 48.0, 64.0, 96.0, 128.0, 184.0, 256.0, 368.0, 512.0, 728.0, 1024.0, 1456.0,
2022 : 2048.0, 2904.0, 4096.0, 5800.0, 8192.0, 11592.0, 16384.0, 23176.0, 32768.0,
2023 : ]
2024 : };
2025 : }
2026 :
2027 : pub(crate) struct WalIngestMetrics {
2028 : pub(crate) bytes_received: IntCounter,
2029 : pub(crate) records_received: IntCounter,
2030 : pub(crate) records_committed: IntCounter,
2031 : pub(crate) records_filtered: IntCounter,
2032 : }
2033 :
2034 2 : pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
2035 : bytes_received: register_int_counter!(
2036 : "pageserver_wal_ingest_bytes_received",
2037 : "Bytes of WAL ingested from safekeepers",
2038 : )
2039 2 : .unwrap(),
2040 : records_received: register_int_counter!(
2041 : "pageserver_wal_ingest_records_received",
2042 : "Number of WAL records received from safekeepers"
2043 : )
2044 2 : .expect("failed to define a metric"),
2045 : records_committed: register_int_counter!(
2046 : "pageserver_wal_ingest_records_committed",
2047 : "Number of WAL records which resulted in writes to pageserver storage"
2048 : )
2049 2 : .expect("failed to define a metric"),
2050 : records_filtered: register_int_counter!(
2051 : "pageserver_wal_ingest_records_filtered",
2052 : "Number of WAL records filtered out due to sharding"
2053 : )
2054 2 : .expect("failed to define a metric"),
2055 2 : });
2056 :
2057 6 : pub(crate) static WAL_REDO_TIME: Lazy<Histogram> = Lazy::new(|| {
2058 : register_histogram!(
2059 : "pageserver_wal_redo_seconds",
2060 : "Time spent on WAL redo",
2061 : redo_histogram_time_buckets!()
2062 : )
2063 6 : .expect("failed to define a metric")
2064 6 : });
2065 :
2066 6 : pub(crate) static WAL_REDO_RECORDS_HISTOGRAM: Lazy<Histogram> = Lazy::new(|| {
2067 : register_histogram!(
2068 : "pageserver_wal_redo_records_histogram",
2069 : "Histogram of number of records replayed per redo in the Postgres WAL redo process",
2070 : redo_histogram_count_buckets!(),
2071 : )
2072 6 : .expect("failed to define a metric")
2073 6 : });
2074 :
2075 6 : pub(crate) static WAL_REDO_BYTES_HISTOGRAM: Lazy<Histogram> = Lazy::new(|| {
2076 : register_histogram!(
2077 : "pageserver_wal_redo_bytes_histogram",
2078 : "Histogram of number of records replayed per redo sent to Postgres",
2079 : redo_bytes_histogram_count_buckets!(),
2080 : )
2081 6 : .expect("failed to define a metric")
2082 6 : });
2083 :
2084 : // FIXME: isn't this already included by WAL_REDO_RECORDS_HISTOGRAM which has _count?
2085 6 : pub(crate) static WAL_REDO_RECORD_COUNTER: Lazy<IntCounter> = Lazy::new(|| {
2086 : register_int_counter!(
2087 : "pageserver_replayed_wal_records_total",
2088 : "Number of WAL records replayed in WAL redo process"
2089 : )
2090 6 : .unwrap()
2091 6 : });
2092 :
2093 : #[rustfmt::skip]
2094 6 : pub(crate) static WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM: Lazy<Histogram> = Lazy::new(|| {
2095 : register_histogram!(
2096 : "pageserver_wal_redo_process_launch_duration",
2097 : "Histogram of the duration of successful WalRedoProcess::launch calls",
2098 : vec![
2099 : 0.0002, 0.0004, 0.0006, 0.0008, 0.0010,
2100 : 0.0020, 0.0040, 0.0060, 0.0080, 0.0100,
2101 : 0.0200, 0.0400, 0.0600, 0.0800, 0.1000,
2102 : 0.2000, 0.4000, 0.6000, 0.8000, 1.0000,
2103 : 1.5000, 2.0000, 2.5000, 3.0000, 4.0000, 10.0000
2104 : ],
2105 : )
2106 6 : .expect("failed to define a metric")
2107 6 : });
2108 :
2109 : pub(crate) struct WalRedoProcessCounters {
2110 : pub(crate) started: IntCounter,
2111 : pub(crate) killed_by_cause: enum_map::EnumMap<WalRedoKillCause, IntCounter>,
2112 : pub(crate) active_stderr_logger_tasks_started: IntCounter,
2113 : pub(crate) active_stderr_logger_tasks_finished: IntCounter,
2114 : }
2115 :
2116 18 : #[derive(Debug, enum_map::Enum, strum_macros::IntoStaticStr)]
2117 : pub(crate) enum WalRedoKillCause {
2118 : WalRedoProcessDrop,
2119 : NoLeakChildDrop,
2120 : Startup,
2121 : }
2122 :
2123 : impl Default for WalRedoProcessCounters {
2124 6 : fn default() -> Self {
2125 6 : let started = register_int_counter!(
2126 : "pageserver_wal_redo_process_started_total",
2127 : "Number of WAL redo processes started",
2128 : )
2129 6 : .unwrap();
2130 6 :
2131 6 : let killed = register_int_counter_vec!(
2132 : "pageserver_wal_redo_process_stopped_total",
2133 : "Number of WAL redo processes stopped",
2134 : &["cause"],
2135 : )
2136 6 : .unwrap();
2137 6 :
2138 6 : let active_stderr_logger_tasks_started = register_int_counter!(
2139 : "pageserver_walredo_stderr_logger_tasks_started_total",
2140 : "Number of active walredo stderr logger tasks that have started",
2141 : )
2142 6 : .unwrap();
2143 6 :
2144 6 : let active_stderr_logger_tasks_finished = register_int_counter!(
2145 : "pageserver_walredo_stderr_logger_tasks_finished_total",
2146 : "Number of active walredo stderr logger tasks that have finished",
2147 : )
2148 6 : .unwrap();
2149 6 :
2150 6 : Self {
2151 6 : started,
2152 18 : killed_by_cause: EnumMap::from_array(std::array::from_fn(|i| {
2153 18 : let cause = <WalRedoKillCause as enum_map::Enum>::from_usize(i);
2154 18 : let cause_str: &'static str = cause.into();
2155 18 : killed.with_label_values(&[cause_str])
2156 18 : })),
2157 6 : active_stderr_logger_tasks_started,
2158 6 : active_stderr_logger_tasks_finished,
2159 6 : }
2160 6 : }
2161 : }
2162 :
2163 : pub(crate) static WAL_REDO_PROCESS_COUNTERS: Lazy<WalRedoProcessCounters> =
2164 : Lazy::new(WalRedoProcessCounters::default);
2165 :
2166 : /// Similar to `prometheus::HistogramTimer` but does not record on drop.
2167 : pub(crate) struct StorageTimeMetricsTimer {
2168 : metrics: StorageTimeMetrics,
2169 : start: Instant,
2170 : }
2171 :
2172 : impl StorageTimeMetricsTimer {
2173 4664 : fn new(metrics: StorageTimeMetrics) -> Self {
2174 4664 : Self {
2175 4664 : metrics,
2176 4664 : start: Instant::now(),
2177 4664 : }
2178 4664 : }
2179 :
2180 : /// Record the time from creation to now.
2181 3542 : pub fn stop_and_record(self) {
2182 3542 : let duration = self.start.elapsed().as_secs_f64();
2183 3542 : self.metrics.timeline_sum.inc_by(duration);
2184 3542 : self.metrics.timeline_count.inc();
2185 3542 : self.metrics.global_histogram.observe(duration);
2186 3542 : }
2187 :
2188 : /// Turns this timer into a timer, which will always record -- usually this means recording
2189 : /// regardless an early `?` path was taken in a function.
2190 754 : pub(crate) fn record_on_drop(self) -> AlwaysRecordingStorageTimeMetricsTimer {
2191 754 : AlwaysRecordingStorageTimeMetricsTimer(Some(self))
2192 754 : }
2193 : }
2194 :
2195 : pub(crate) struct AlwaysRecordingStorageTimeMetricsTimer(Option<StorageTimeMetricsTimer>);
2196 :
2197 : impl Drop for AlwaysRecordingStorageTimeMetricsTimer {
2198 754 : fn drop(&mut self) {
2199 754 : if let Some(inner) = self.0.take() {
2200 754 : inner.stop_and_record();
2201 754 : }
2202 754 : }
2203 : }
2204 :
2205 : /// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
2206 : /// timeline total sum and count.
2207 : #[derive(Clone, Debug)]
2208 : pub(crate) struct StorageTimeMetrics {
2209 : /// Sum of f64 seconds, per operation, tenant_id and timeline_id
2210 : timeline_sum: Counter,
2211 : /// Number of oeprations, per operation, tenant_id and timeline_id
2212 : timeline_count: IntCounter,
2213 : /// Global histogram having only the "operation" label.
2214 : global_histogram: Histogram,
2215 : }
2216 :
2217 : impl StorageTimeMetrics {
2218 3248 : pub fn new(
2219 3248 : operation: StorageTimeOperation,
2220 3248 : tenant_id: &str,
2221 3248 : shard_id: &str,
2222 3248 : timeline_id: &str,
2223 3248 : ) -> Self {
2224 3248 : let operation: &'static str = operation.into();
2225 3248 :
2226 3248 : let timeline_sum = STORAGE_TIME_SUM_PER_TIMELINE
2227 3248 : .get_metric_with_label_values(&[operation, tenant_id, shard_id, timeline_id])
2228 3248 : .unwrap();
2229 3248 : let timeline_count = STORAGE_TIME_COUNT_PER_TIMELINE
2230 3248 : .get_metric_with_label_values(&[operation, tenant_id, shard_id, timeline_id])
2231 3248 : .unwrap();
2232 3248 : let global_histogram = STORAGE_TIME_GLOBAL
2233 3248 : .get_metric_with_label_values(&[operation])
2234 3248 : .unwrap();
2235 3248 :
2236 3248 : StorageTimeMetrics {
2237 3248 : timeline_sum,
2238 3248 : timeline_count,
2239 3248 : global_histogram,
2240 3248 : }
2241 3248 : }
2242 :
2243 : /// Starts timing a new operation.
2244 : ///
2245 : /// Note: unlike `prometheus::HistogramTimer` the returned timer does not record on drop.
2246 4664 : pub fn start_timer(&self) -> StorageTimeMetricsTimer {
2247 4664 : StorageTimeMetricsTimer::new(self.clone())
2248 4664 : }
2249 : }
2250 :
2251 : #[derive(Debug)]
2252 : pub(crate) struct TimelineMetrics {
2253 : tenant_id: String,
2254 : shard_id: String,
2255 : timeline_id: String,
2256 : pub flush_time_histo: StorageTimeMetrics,
2257 : pub compact_time_histo: StorageTimeMetrics,
2258 : pub create_images_time_histo: StorageTimeMetrics,
2259 : pub logical_size_histo: StorageTimeMetrics,
2260 : pub imitate_logical_size_histo: StorageTimeMetrics,
2261 : pub load_layer_map_histo: StorageTimeMetrics,
2262 : pub garbage_collect_histo: StorageTimeMetrics,
2263 : pub find_gc_cutoffs_histo: StorageTimeMetrics,
2264 : pub last_record_gauge: IntGauge,
2265 : pub pitr_history_size: UIntGauge,
2266 : pub archival_size: UIntGauge,
2267 : pub(crate) layer_size_image: UIntGauge,
2268 : pub(crate) layer_count_image: UIntGauge,
2269 : pub(crate) layer_size_delta: UIntGauge,
2270 : pub(crate) layer_count_delta: UIntGauge,
2271 : pub standby_horizon_gauge: IntGauge,
2272 : pub resident_physical_size_gauge: UIntGauge,
2273 : pub visible_physical_size_gauge: UIntGauge,
2274 : /// copy of LayeredTimeline.current_logical_size
2275 : pub current_logical_size_gauge: UIntGauge,
2276 : pub aux_file_size_gauge: IntGauge,
2277 : pub directory_entries_count_gauge: Lazy<UIntGauge, Box<dyn Send + Fn() -> UIntGauge>>,
2278 : pub evictions: IntCounter,
2279 : pub evictions_with_low_residence_duration: std::sync::RwLock<EvictionsWithLowResidenceDuration>,
2280 : /// Number of valid LSN leases.
2281 : pub valid_lsn_lease_count_gauge: UIntGauge,
2282 : shutdown: std::sync::atomic::AtomicBool,
2283 : }
2284 :
2285 : impl TimelineMetrics {
2286 406 : pub fn new(
2287 406 : tenant_shard_id: &TenantShardId,
2288 406 : timeline_id_raw: &TimelineId,
2289 406 : evictions_with_low_residence_duration_builder: EvictionsWithLowResidenceDurationBuilder,
2290 406 : ) -> Self {
2291 406 : let tenant_id = tenant_shard_id.tenant_id.to_string();
2292 406 : let shard_id = format!("{}", tenant_shard_id.shard_slug());
2293 406 : let timeline_id = timeline_id_raw.to_string();
2294 406 : let flush_time_histo = StorageTimeMetrics::new(
2295 406 : StorageTimeOperation::LayerFlush,
2296 406 : &tenant_id,
2297 406 : &shard_id,
2298 406 : &timeline_id,
2299 406 : );
2300 406 : let compact_time_histo = StorageTimeMetrics::new(
2301 406 : StorageTimeOperation::Compact,
2302 406 : &tenant_id,
2303 406 : &shard_id,
2304 406 : &timeline_id,
2305 406 : );
2306 406 : let create_images_time_histo = StorageTimeMetrics::new(
2307 406 : StorageTimeOperation::CreateImages,
2308 406 : &tenant_id,
2309 406 : &shard_id,
2310 406 : &timeline_id,
2311 406 : );
2312 406 : let logical_size_histo = StorageTimeMetrics::new(
2313 406 : StorageTimeOperation::LogicalSize,
2314 406 : &tenant_id,
2315 406 : &shard_id,
2316 406 : &timeline_id,
2317 406 : );
2318 406 : let imitate_logical_size_histo = StorageTimeMetrics::new(
2319 406 : StorageTimeOperation::ImitateLogicalSize,
2320 406 : &tenant_id,
2321 406 : &shard_id,
2322 406 : &timeline_id,
2323 406 : );
2324 406 : let load_layer_map_histo = StorageTimeMetrics::new(
2325 406 : StorageTimeOperation::LoadLayerMap,
2326 406 : &tenant_id,
2327 406 : &shard_id,
2328 406 : &timeline_id,
2329 406 : );
2330 406 : let garbage_collect_histo = StorageTimeMetrics::new(
2331 406 : StorageTimeOperation::Gc,
2332 406 : &tenant_id,
2333 406 : &shard_id,
2334 406 : &timeline_id,
2335 406 : );
2336 406 : let find_gc_cutoffs_histo = StorageTimeMetrics::new(
2337 406 : StorageTimeOperation::FindGcCutoffs,
2338 406 : &tenant_id,
2339 406 : &shard_id,
2340 406 : &timeline_id,
2341 406 : );
2342 406 : let last_record_gauge = LAST_RECORD_LSN
2343 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2344 406 : .unwrap();
2345 406 :
2346 406 : let pitr_history_size = PITR_HISTORY_SIZE
2347 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2348 406 : .unwrap();
2349 406 :
2350 406 : let archival_size = TIMELINE_ARCHIVE_SIZE
2351 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2352 406 : .unwrap();
2353 406 :
2354 406 : let layer_size_image = TIMELINE_LAYER_SIZE
2355 406 : .get_metric_with_label_values(&[
2356 406 : &tenant_id,
2357 406 : &shard_id,
2358 406 : &timeline_id,
2359 406 : MetricLayerKind::Image.into(),
2360 406 : ])
2361 406 : .unwrap();
2362 406 :
2363 406 : let layer_count_image = TIMELINE_LAYER_COUNT
2364 406 : .get_metric_with_label_values(&[
2365 406 : &tenant_id,
2366 406 : &shard_id,
2367 406 : &timeline_id,
2368 406 : MetricLayerKind::Image.into(),
2369 406 : ])
2370 406 : .unwrap();
2371 406 :
2372 406 : let layer_size_delta = TIMELINE_LAYER_SIZE
2373 406 : .get_metric_with_label_values(&[
2374 406 : &tenant_id,
2375 406 : &shard_id,
2376 406 : &timeline_id,
2377 406 : MetricLayerKind::Delta.into(),
2378 406 : ])
2379 406 : .unwrap();
2380 406 :
2381 406 : let layer_count_delta = TIMELINE_LAYER_COUNT
2382 406 : .get_metric_with_label_values(&[
2383 406 : &tenant_id,
2384 406 : &shard_id,
2385 406 : &timeline_id,
2386 406 : MetricLayerKind::Delta.into(),
2387 406 : ])
2388 406 : .unwrap();
2389 406 :
2390 406 : let standby_horizon_gauge = STANDBY_HORIZON
2391 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2392 406 : .unwrap();
2393 406 : let resident_physical_size_gauge = RESIDENT_PHYSICAL_SIZE
2394 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2395 406 : .unwrap();
2396 406 : let visible_physical_size_gauge = VISIBLE_PHYSICAL_SIZE
2397 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2398 406 : .unwrap();
2399 406 : // TODO: we shouldn't expose this metric
2400 406 : let current_logical_size_gauge = CURRENT_LOGICAL_SIZE
2401 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2402 406 : .unwrap();
2403 406 : let aux_file_size_gauge = AUX_FILE_SIZE
2404 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2405 406 : .unwrap();
2406 406 : // TODO use impl Trait syntax here once we have ability to use it: https://github.com/rust-lang/rust/issues/63065
2407 406 : let directory_entries_count_gauge_closure = {
2408 406 : let tenant_shard_id = *tenant_shard_id;
2409 406 : let timeline_id_raw = *timeline_id_raw;
2410 0 : move || {
2411 0 : let tenant_id = tenant_shard_id.tenant_id.to_string();
2412 0 : let shard_id = format!("{}", tenant_shard_id.shard_slug());
2413 0 : let timeline_id = timeline_id_raw.to_string();
2414 0 : let gauge: UIntGauge = DIRECTORY_ENTRIES_COUNT
2415 0 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2416 0 : .unwrap();
2417 0 : gauge
2418 0 : }
2419 : };
2420 406 : let directory_entries_count_gauge: Lazy<UIntGauge, Box<dyn Send + Fn() -> UIntGauge>> =
2421 406 : Lazy::new(Box::new(directory_entries_count_gauge_closure));
2422 406 : let evictions = EVICTIONS
2423 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2424 406 : .unwrap();
2425 406 : let evictions_with_low_residence_duration = evictions_with_low_residence_duration_builder
2426 406 : .build(&tenant_id, &shard_id, &timeline_id);
2427 406 :
2428 406 : let valid_lsn_lease_count_gauge = VALID_LSN_LEASE_COUNT
2429 406 : .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
2430 406 : .unwrap();
2431 406 :
2432 406 : TimelineMetrics {
2433 406 : tenant_id,
2434 406 : shard_id,
2435 406 : timeline_id,
2436 406 : flush_time_histo,
2437 406 : compact_time_histo,
2438 406 : create_images_time_histo,
2439 406 : logical_size_histo,
2440 406 : imitate_logical_size_histo,
2441 406 : garbage_collect_histo,
2442 406 : find_gc_cutoffs_histo,
2443 406 : load_layer_map_histo,
2444 406 : last_record_gauge,
2445 406 : pitr_history_size,
2446 406 : archival_size,
2447 406 : layer_size_image,
2448 406 : layer_count_image,
2449 406 : layer_size_delta,
2450 406 : layer_count_delta,
2451 406 : standby_horizon_gauge,
2452 406 : resident_physical_size_gauge,
2453 406 : visible_physical_size_gauge,
2454 406 : current_logical_size_gauge,
2455 406 : aux_file_size_gauge,
2456 406 : directory_entries_count_gauge,
2457 406 : evictions,
2458 406 : evictions_with_low_residence_duration: std::sync::RwLock::new(
2459 406 : evictions_with_low_residence_duration,
2460 406 : ),
2461 406 : valid_lsn_lease_count_gauge,
2462 406 : shutdown: std::sync::atomic::AtomicBool::default(),
2463 406 : }
2464 406 : }
2465 :
2466 1516 : pub(crate) fn record_new_file_metrics(&self, sz: u64) {
2467 1516 : self.resident_physical_size_add(sz);
2468 1516 : }
2469 :
2470 472 : pub(crate) fn resident_physical_size_sub(&self, sz: u64) {
2471 472 : self.resident_physical_size_gauge.sub(sz);
2472 472 : crate::metrics::RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(sz);
2473 472 : }
2474 :
2475 1546 : pub(crate) fn resident_physical_size_add(&self, sz: u64) {
2476 1546 : self.resident_physical_size_gauge.add(sz);
2477 1546 : crate::metrics::RESIDENT_PHYSICAL_SIZE_GLOBAL.add(sz);
2478 1546 : }
2479 :
2480 8 : pub(crate) fn resident_physical_size_get(&self) -> u64 {
2481 8 : self.resident_physical_size_gauge.get()
2482 8 : }
2483 :
2484 8 : pub(crate) fn shutdown(&self) {
2485 8 : let was_shutdown = self
2486 8 : .shutdown
2487 8 : .swap(true, std::sync::atomic::Ordering::Relaxed);
2488 8 :
2489 8 : if was_shutdown {
2490 : // this happens on tenant deletion because tenant first shuts down timelines, then
2491 : // invokes timeline deletion which first shuts down the timeline again.
2492 : // TODO: this can be removed once https://github.com/neondatabase/neon/issues/5080
2493 0 : return;
2494 8 : }
2495 8 :
2496 8 : let tenant_id = &self.tenant_id;
2497 8 : let timeline_id = &self.timeline_id;
2498 8 : let shard_id = &self.shard_id;
2499 8 : let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2500 8 : let _ = STANDBY_HORIZON.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2501 8 : {
2502 8 : RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
2503 8 : let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2504 8 : }
2505 8 : let _ = VISIBLE_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2506 8 : let _ = CURRENT_LOGICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2507 8 : if let Some(metric) = Lazy::get(&DIRECTORY_ENTRIES_COUNT) {
2508 0 : let _ = metric.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2509 8 : }
2510 :
2511 8 : let _ = TIMELINE_ARCHIVE_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2512 8 : let _ = PITR_HISTORY_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2513 8 :
2514 8 : let _ = TIMELINE_LAYER_SIZE.remove_label_values(&[
2515 8 : tenant_id,
2516 8 : shard_id,
2517 8 : timeline_id,
2518 8 : MetricLayerKind::Image.into(),
2519 8 : ]);
2520 8 : let _ = TIMELINE_LAYER_COUNT.remove_label_values(&[
2521 8 : tenant_id,
2522 8 : shard_id,
2523 8 : timeline_id,
2524 8 : MetricLayerKind::Image.into(),
2525 8 : ]);
2526 8 : let _ = TIMELINE_LAYER_SIZE.remove_label_values(&[
2527 8 : tenant_id,
2528 8 : shard_id,
2529 8 : timeline_id,
2530 8 : MetricLayerKind::Delta.into(),
2531 8 : ]);
2532 8 : let _ = TIMELINE_LAYER_COUNT.remove_label_values(&[
2533 8 : tenant_id,
2534 8 : shard_id,
2535 8 : timeline_id,
2536 8 : MetricLayerKind::Delta.into(),
2537 8 : ]);
2538 8 :
2539 8 : let _ = EVICTIONS.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2540 8 : let _ = AUX_FILE_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2541 8 : let _ = VALID_LSN_LEASE_COUNT.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2542 8 :
2543 8 : self.evictions_with_low_residence_duration
2544 8 : .write()
2545 8 : .unwrap()
2546 8 : .remove(tenant_id, shard_id, timeline_id);
2547 :
2548 : // The following metrics are born outside of the TimelineMetrics lifecycle but still
2549 : // removed at the end of it. The idea is to have the metrics outlive the
2550 : // entity during which they're observed, e.g., the smgr metrics shall
2551 : // outlive an individual smgr connection, but not the timeline.
2552 :
2553 72 : for op in StorageTimeOperation::VARIANTS {
2554 64 : let _ = STORAGE_TIME_SUM_PER_TIMELINE.remove_label_values(&[
2555 64 : op,
2556 64 : tenant_id,
2557 64 : shard_id,
2558 64 : timeline_id,
2559 64 : ]);
2560 64 : let _ = STORAGE_TIME_COUNT_PER_TIMELINE.remove_label_values(&[
2561 64 : op,
2562 64 : tenant_id,
2563 64 : shard_id,
2564 64 : timeline_id,
2565 64 : ]);
2566 64 : }
2567 :
2568 24 : for op in STORAGE_IO_SIZE_OPERATIONS {
2569 16 : let _ = STORAGE_IO_SIZE.remove_label_values(&[op, tenant_id, shard_id, timeline_id]);
2570 16 : }
2571 :
2572 8 : let _ = SMGR_QUERY_TIME_PER_TENANT_TIMELINE.remove_label_values(&[
2573 8 : SmgrQueryType::GetPageAtLsn.into(),
2574 8 : tenant_id,
2575 8 : shard_id,
2576 8 : timeline_id,
2577 8 : ]);
2578 8 : }
2579 : }
2580 :
2581 6 : pub(crate) fn remove_tenant_metrics(tenant_shard_id: &TenantShardId) {
2582 6 : // Only shard zero deals in synthetic sizes
2583 6 : if tenant_shard_id.is_shard_zero() {
2584 6 : let tid = tenant_shard_id.tenant_id.to_string();
2585 6 : let _ = TENANT_SYNTHETIC_SIZE_METRIC.remove_label_values(&[&tid]);
2586 6 : }
2587 :
2588 : // we leave the BROKEN_TENANTS_SET entry if any
2589 6 : }
2590 :
2591 : use futures::Future;
2592 : use pin_project_lite::pin_project;
2593 : use std::collections::HashMap;
2594 : use std::num::NonZeroUsize;
2595 : use std::pin::Pin;
2596 : use std::sync::atomic::AtomicU64;
2597 : use std::sync::{Arc, Mutex};
2598 : use std::task::{Context, Poll};
2599 : use std::time::{Duration, Instant};
2600 :
2601 : use crate::context::{PageContentKind, RequestContext};
2602 : use crate::task_mgr::TaskKind;
2603 : use crate::tenant::mgr::TenantSlot;
2604 : use crate::tenant::tasks::BackgroundLoopKind;
2605 :
2606 : /// Maintain a per timeline gauge in addition to the global gauge.
2607 : pub(crate) struct PerTimelineRemotePhysicalSizeGauge {
2608 : last_set: AtomicU64,
2609 : gauge: UIntGauge,
2610 : }
2611 :
2612 : impl PerTimelineRemotePhysicalSizeGauge {
2613 416 : fn new(per_timeline_gauge: UIntGauge) -> Self {
2614 416 : Self {
2615 416 : last_set: AtomicU64::new(0),
2616 416 : gauge: per_timeline_gauge,
2617 416 : }
2618 416 : }
2619 1808 : pub(crate) fn set(&self, sz: u64) {
2620 1808 : self.gauge.set(sz);
2621 1808 : let prev = self.last_set.swap(sz, std::sync::atomic::Ordering::Relaxed);
2622 1808 : if sz < prev {
2623 27 : REMOTE_PHYSICAL_SIZE_GLOBAL.sub(prev - sz);
2624 1781 : } else {
2625 1781 : REMOTE_PHYSICAL_SIZE_GLOBAL.add(sz - prev);
2626 1781 : };
2627 1808 : }
2628 2 : pub(crate) fn get(&self) -> u64 {
2629 2 : self.gauge.get()
2630 2 : }
2631 : }
2632 :
2633 : impl Drop for PerTimelineRemotePhysicalSizeGauge {
2634 18 : fn drop(&mut self) {
2635 18 : REMOTE_PHYSICAL_SIZE_GLOBAL.sub(self.last_set.load(std::sync::atomic::Ordering::Relaxed));
2636 18 : }
2637 : }
2638 :
2639 : pub(crate) struct RemoteTimelineClientMetrics {
2640 : tenant_id: String,
2641 : shard_id: String,
2642 : timeline_id: String,
2643 : pub(crate) remote_physical_size_gauge: PerTimelineRemotePhysicalSizeGauge,
2644 : calls: Mutex<HashMap<(&'static str, &'static str), IntCounterPair>>,
2645 : bytes_started_counter: Mutex<HashMap<(&'static str, &'static str), IntCounter>>,
2646 : bytes_finished_counter: Mutex<HashMap<(&'static str, &'static str), IntCounter>>,
2647 : }
2648 :
2649 : impl RemoteTimelineClientMetrics {
2650 416 : pub fn new(tenant_shard_id: &TenantShardId, timeline_id: &TimelineId) -> Self {
2651 416 : let tenant_id_str = tenant_shard_id.tenant_id.to_string();
2652 416 : let shard_id_str = format!("{}", tenant_shard_id.shard_slug());
2653 416 : let timeline_id_str = timeline_id.to_string();
2654 416 :
2655 416 : let remote_physical_size_gauge = PerTimelineRemotePhysicalSizeGauge::new(
2656 416 : REMOTE_PHYSICAL_SIZE
2657 416 : .get_metric_with_label_values(&[&tenant_id_str, &shard_id_str, &timeline_id_str])
2658 416 : .unwrap(),
2659 416 : );
2660 416 :
2661 416 : RemoteTimelineClientMetrics {
2662 416 : tenant_id: tenant_id_str,
2663 416 : shard_id: shard_id_str,
2664 416 : timeline_id: timeline_id_str,
2665 416 : calls: Mutex::new(HashMap::default()),
2666 416 : bytes_started_counter: Mutex::new(HashMap::default()),
2667 416 : bytes_finished_counter: Mutex::new(HashMap::default()),
2668 416 : remote_physical_size_gauge,
2669 416 : }
2670 416 : }
2671 :
2672 2665 : pub fn remote_operation_time(
2673 2665 : &self,
2674 2665 : file_kind: &RemoteOpFileKind,
2675 2665 : op_kind: &RemoteOpKind,
2676 2665 : status: &'static str,
2677 2665 : ) -> Histogram {
2678 2665 : let key = (file_kind.as_str(), op_kind.as_str(), status);
2679 2665 : REMOTE_OPERATION_TIME
2680 2665 : .get_metric_with_label_values(&[key.0, key.1, key.2])
2681 2665 : .unwrap()
2682 2665 : }
2683 :
2684 6352 : fn calls_counter_pair(
2685 6352 : &self,
2686 6352 : file_kind: &RemoteOpFileKind,
2687 6352 : op_kind: &RemoteOpKind,
2688 6352 : ) -> IntCounterPair {
2689 6352 : let mut guard = self.calls.lock().unwrap();
2690 6352 : let key = (file_kind.as_str(), op_kind.as_str());
2691 6352 : let metric = guard.entry(key).or_insert_with(move || {
2692 730 : REMOTE_TIMELINE_CLIENT_CALLS
2693 730 : .get_metric_with_label_values(&[
2694 730 : &self.tenant_id,
2695 730 : &self.shard_id,
2696 730 : &self.timeline_id,
2697 730 : key.0,
2698 730 : key.1,
2699 730 : ])
2700 730 : .unwrap()
2701 6352 : });
2702 6352 : metric.clone()
2703 6352 : }
2704 :
2705 1524 : fn bytes_started_counter(
2706 1524 : &self,
2707 1524 : file_kind: &RemoteOpFileKind,
2708 1524 : op_kind: &RemoteOpKind,
2709 1524 : ) -> IntCounter {
2710 1524 : let mut guard = self.bytes_started_counter.lock().unwrap();
2711 1524 : let key = (file_kind.as_str(), op_kind.as_str());
2712 1524 : let metric = guard.entry(key).or_insert_with(move || {
2713 280 : REMOTE_TIMELINE_CLIENT_BYTES_STARTED_COUNTER
2714 280 : .get_metric_with_label_values(&[
2715 280 : &self.tenant_id,
2716 280 : &self.shard_id,
2717 280 : &self.timeline_id,
2718 280 : key.0,
2719 280 : key.1,
2720 280 : ])
2721 280 : .unwrap()
2722 1524 : });
2723 1524 : metric.clone()
2724 1524 : }
2725 :
2726 2761 : fn bytes_finished_counter(
2727 2761 : &self,
2728 2761 : file_kind: &RemoteOpFileKind,
2729 2761 : op_kind: &RemoteOpKind,
2730 2761 : ) -> IntCounter {
2731 2761 : let mut guard = self.bytes_finished_counter.lock().unwrap();
2732 2761 : let key = (file_kind.as_str(), op_kind.as_str());
2733 2761 : let metric = guard.entry(key).or_insert_with(move || {
2734 280 : REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER
2735 280 : .get_metric_with_label_values(&[
2736 280 : &self.tenant_id,
2737 280 : &self.shard_id,
2738 280 : &self.timeline_id,
2739 280 : key.0,
2740 280 : key.1,
2741 280 : ])
2742 280 : .unwrap()
2743 2761 : });
2744 2761 : metric.clone()
2745 2761 : }
2746 : }
2747 :
2748 : #[cfg(test)]
2749 : impl RemoteTimelineClientMetrics {
2750 6 : pub fn get_bytes_started_counter_value(
2751 6 : &self,
2752 6 : file_kind: &RemoteOpFileKind,
2753 6 : op_kind: &RemoteOpKind,
2754 6 : ) -> Option<u64> {
2755 6 : let guard = self.bytes_started_counter.lock().unwrap();
2756 6 : let key = (file_kind.as_str(), op_kind.as_str());
2757 6 : guard.get(&key).map(|counter| counter.get())
2758 6 : }
2759 :
2760 6 : pub fn get_bytes_finished_counter_value(
2761 6 : &self,
2762 6 : file_kind: &RemoteOpFileKind,
2763 6 : op_kind: &RemoteOpKind,
2764 6 : ) -> Option<u64> {
2765 6 : let guard = self.bytes_finished_counter.lock().unwrap();
2766 6 : let key = (file_kind.as_str(), op_kind.as_str());
2767 6 : guard.get(&key).map(|counter| counter.get())
2768 6 : }
2769 : }
2770 :
2771 : /// See [`RemoteTimelineClientMetrics::call_begin`].
2772 : #[must_use]
2773 : pub(crate) struct RemoteTimelineClientCallMetricGuard {
2774 : /// Decremented on drop.
2775 : calls_counter_pair: Option<IntCounterPair>,
2776 : /// If Some(), this references the bytes_finished metric, and we increment it by the given `u64` on drop.
2777 : bytes_finished: Option<(IntCounter, u64)>,
2778 : }
2779 :
2780 : impl RemoteTimelineClientCallMetricGuard {
2781 : /// Consume this guard object without performing the metric updates it would do on `drop()`.
2782 : /// The caller vouches to do the metric updates manually.
2783 3430 : pub fn will_decrement_manually(mut self) {
2784 3430 : let RemoteTimelineClientCallMetricGuard {
2785 3430 : calls_counter_pair,
2786 3430 : bytes_finished,
2787 3430 : } = &mut self;
2788 3430 : calls_counter_pair.take();
2789 3430 : bytes_finished.take();
2790 3430 : }
2791 : }
2792 :
2793 : impl Drop for RemoteTimelineClientCallMetricGuard {
2794 3456 : fn drop(&mut self) {
2795 3456 : let RemoteTimelineClientCallMetricGuard {
2796 3456 : calls_counter_pair,
2797 3456 : bytes_finished,
2798 3456 : } = self;
2799 3456 : if let Some(guard) = calls_counter_pair.take() {
2800 26 : guard.dec();
2801 3430 : }
2802 3456 : if let Some((bytes_finished_metric, value)) = bytes_finished {
2803 0 : bytes_finished_metric.inc_by(*value);
2804 3456 : }
2805 3456 : }
2806 : }
2807 :
2808 : /// The enum variants communicate to the [`RemoteTimelineClientMetrics`] whether to
2809 : /// track the byte size of this call in applicable metric(s).
2810 : pub(crate) enum RemoteTimelineClientMetricsCallTrackSize {
2811 : /// Do not account for this call's byte size in any metrics.
2812 : /// The `reason` field is there to make the call sites self-documenting
2813 : /// about why they don't need the metric.
2814 : DontTrackSize { reason: &'static str },
2815 : /// Track the byte size of the call in applicable metric(s).
2816 : Bytes(u64),
2817 : }
2818 :
2819 : impl RemoteTimelineClientMetrics {
2820 : /// Update the metrics that change when a call to the remote timeline client instance starts.
2821 : ///
2822 : /// Drop the returned guard object once the operation is finished to updates corresponding metrics that track completions.
2823 : /// Or, use [`RemoteTimelineClientCallMetricGuard::will_decrement_manually`] and [`call_end`](Self::call_end) if that
2824 : /// is more suitable.
2825 : /// Never do both.
2826 3456 : pub(crate) fn call_begin(
2827 3456 : &self,
2828 3456 : file_kind: &RemoteOpFileKind,
2829 3456 : op_kind: &RemoteOpKind,
2830 3456 : size: RemoteTimelineClientMetricsCallTrackSize,
2831 3456 : ) -> RemoteTimelineClientCallMetricGuard {
2832 3456 : let calls_counter_pair = self.calls_counter_pair(file_kind, op_kind);
2833 3456 : calls_counter_pair.inc();
2834 :
2835 3456 : let bytes_finished = match size {
2836 1932 : RemoteTimelineClientMetricsCallTrackSize::DontTrackSize { reason: _reason } => {
2837 1932 : // nothing to do
2838 1932 : None
2839 : }
2840 1524 : RemoteTimelineClientMetricsCallTrackSize::Bytes(size) => {
2841 1524 : self.bytes_started_counter(file_kind, op_kind).inc_by(size);
2842 1524 : let finished_counter = self.bytes_finished_counter(file_kind, op_kind);
2843 1524 : Some((finished_counter, size))
2844 : }
2845 : };
2846 3456 : RemoteTimelineClientCallMetricGuard {
2847 3456 : calls_counter_pair: Some(calls_counter_pair),
2848 3456 : bytes_finished,
2849 3456 : }
2850 3456 : }
2851 :
2852 : /// Manually udpate the metrics that track completions, instead of using the guard object.
2853 : /// Using the guard object is generally preferable.
2854 : /// See [`call_begin`](Self::call_begin) for more context.
2855 2896 : pub(crate) fn call_end(
2856 2896 : &self,
2857 2896 : file_kind: &RemoteOpFileKind,
2858 2896 : op_kind: &RemoteOpKind,
2859 2896 : size: RemoteTimelineClientMetricsCallTrackSize,
2860 2896 : ) {
2861 2896 : let calls_counter_pair = self.calls_counter_pair(file_kind, op_kind);
2862 2896 : calls_counter_pair.dec();
2863 2896 : match size {
2864 1659 : RemoteTimelineClientMetricsCallTrackSize::DontTrackSize { reason: _reason } => {}
2865 1237 : RemoteTimelineClientMetricsCallTrackSize::Bytes(size) => {
2866 1237 : self.bytes_finished_counter(file_kind, op_kind).inc_by(size);
2867 1237 : }
2868 : }
2869 2896 : }
2870 : }
2871 :
2872 : impl Drop for RemoteTimelineClientMetrics {
2873 18 : fn drop(&mut self) {
2874 18 : let RemoteTimelineClientMetrics {
2875 18 : tenant_id,
2876 18 : shard_id,
2877 18 : timeline_id,
2878 18 : remote_physical_size_gauge,
2879 18 : calls,
2880 18 : bytes_started_counter,
2881 18 : bytes_finished_counter,
2882 18 : } = self;
2883 24 : for ((a, b), _) in calls.get_mut().unwrap().drain() {
2884 24 : let mut res = [Ok(()), Ok(())];
2885 24 : REMOTE_TIMELINE_CLIENT_CALLS
2886 24 : .remove_label_values(&mut res, &[tenant_id, shard_id, timeline_id, a, b]);
2887 24 : // don't care about results
2888 24 : }
2889 18 : for ((a, b), _) in bytes_started_counter.get_mut().unwrap().drain() {
2890 6 : let _ = REMOTE_TIMELINE_CLIENT_BYTES_STARTED_COUNTER.remove_label_values(&[
2891 6 : tenant_id,
2892 6 : shard_id,
2893 6 : timeline_id,
2894 6 : a,
2895 6 : b,
2896 6 : ]);
2897 6 : }
2898 18 : for ((a, b), _) in bytes_finished_counter.get_mut().unwrap().drain() {
2899 6 : let _ = REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER.remove_label_values(&[
2900 6 : tenant_id,
2901 6 : shard_id,
2902 6 : timeline_id,
2903 6 : a,
2904 6 : b,
2905 6 : ]);
2906 6 : }
2907 18 : {
2908 18 : let _ = remote_physical_size_gauge; // use to avoid 'unused' warning in desctructuring above
2909 18 : let _ = REMOTE_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
2910 18 : }
2911 18 : }
2912 : }
2913 :
2914 : /// Wrapper future that measures the time spent by a remote storage operation,
2915 : /// and records the time and success/failure as a prometheus metric.
2916 : pub(crate) trait MeasureRemoteOp: Sized {
2917 2807 : fn measure_remote_op(
2918 2807 : self,
2919 2807 : file_kind: RemoteOpFileKind,
2920 2807 : op: RemoteOpKind,
2921 2807 : metrics: Arc<RemoteTimelineClientMetrics>,
2922 2807 : ) -> MeasuredRemoteOp<Self> {
2923 2807 : let start = Instant::now();
2924 2807 : MeasuredRemoteOp {
2925 2807 : inner: self,
2926 2807 : file_kind,
2927 2807 : op,
2928 2807 : start,
2929 2807 : metrics,
2930 2807 : }
2931 2807 : }
2932 : }
2933 :
2934 : impl<T: Sized> MeasureRemoteOp for T {}
2935 :
2936 : pin_project! {
2937 : pub(crate) struct MeasuredRemoteOp<F>
2938 : {
2939 : #[pin]
2940 : inner: F,
2941 : file_kind: RemoteOpFileKind,
2942 : op: RemoteOpKind,
2943 : start: Instant,
2944 : metrics: Arc<RemoteTimelineClientMetrics>,
2945 : }
2946 : }
2947 :
2948 : impl<F: Future<Output = Result<O, E>>, O, E> Future for MeasuredRemoteOp<F> {
2949 : type Output = Result<O, E>;
2950 :
2951 40832 : fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
2952 40832 : let this = self.project();
2953 40832 : let poll_result = this.inner.poll(cx);
2954 40832 : if let Poll::Ready(ref res) = poll_result {
2955 2665 : let duration = this.start.elapsed();
2956 2665 : let status = if res.is_ok() { &"success" } else { &"failure" };
2957 2665 : this.metrics
2958 2665 : .remote_operation_time(this.file_kind, this.op, status)
2959 2665 : .observe(duration.as_secs_f64());
2960 38167 : }
2961 40832 : poll_result
2962 40832 : }
2963 : }
2964 :
2965 : pub mod tokio_epoll_uring {
2966 : use metrics::{register_int_counter, UIntGauge};
2967 : use once_cell::sync::Lazy;
2968 :
2969 : pub struct Collector {
2970 : descs: Vec<metrics::core::Desc>,
2971 : systems_created: UIntGauge,
2972 : systems_destroyed: UIntGauge,
2973 : }
2974 :
2975 : impl metrics::core::Collector for Collector {
2976 0 : fn desc(&self) -> Vec<&metrics::core::Desc> {
2977 0 : self.descs.iter().collect()
2978 0 : }
2979 :
2980 0 : fn collect(&self) -> Vec<metrics::proto::MetricFamily> {
2981 0 : let mut mfs = Vec::with_capacity(Self::NMETRICS);
2982 0 : let tokio_epoll_uring::metrics::Metrics {
2983 0 : systems_created,
2984 0 : systems_destroyed,
2985 0 : } = tokio_epoll_uring::metrics::global();
2986 0 : self.systems_created.set(systems_created);
2987 0 : mfs.extend(self.systems_created.collect());
2988 0 : self.systems_destroyed.set(systems_destroyed);
2989 0 : mfs.extend(self.systems_destroyed.collect());
2990 0 : mfs
2991 0 : }
2992 : }
2993 :
2994 : impl Collector {
2995 : const NMETRICS: usize = 2;
2996 :
2997 : #[allow(clippy::new_without_default)]
2998 0 : pub fn new() -> Self {
2999 0 : let mut descs = Vec::new();
3000 0 :
3001 0 : let systems_created = UIntGauge::new(
3002 0 : "pageserver_tokio_epoll_uring_systems_created",
3003 0 : "counter of tokio-epoll-uring systems that were created",
3004 0 : )
3005 0 : .unwrap();
3006 0 : descs.extend(
3007 0 : metrics::core::Collector::desc(&systems_created)
3008 0 : .into_iter()
3009 0 : .cloned(),
3010 0 : );
3011 0 :
3012 0 : let systems_destroyed = UIntGauge::new(
3013 0 : "pageserver_tokio_epoll_uring_systems_destroyed",
3014 0 : "counter of tokio-epoll-uring systems that were destroyed",
3015 0 : )
3016 0 : .unwrap();
3017 0 : descs.extend(
3018 0 : metrics::core::Collector::desc(&systems_destroyed)
3019 0 : .into_iter()
3020 0 : .cloned(),
3021 0 : );
3022 0 :
3023 0 : Self {
3024 0 : descs,
3025 0 : systems_created,
3026 0 : systems_destroyed,
3027 0 : }
3028 0 : }
3029 : }
3030 :
3031 95 : pub(crate) static THREAD_LOCAL_LAUNCH_SUCCESSES: Lazy<metrics::IntCounter> = Lazy::new(|| {
3032 : register_int_counter!(
3033 : "pageserver_tokio_epoll_uring_pageserver_thread_local_launch_success_count",
3034 : "Number of times where thread_local_system creation spanned multiple executor threads",
3035 : )
3036 95 : .unwrap()
3037 95 : });
3038 :
3039 0 : pub(crate) static THREAD_LOCAL_LAUNCH_FAILURES: Lazy<metrics::IntCounter> = Lazy::new(|| {
3040 : register_int_counter!(
3041 : "pageserver_tokio_epoll_uring_pageserver_thread_local_launch_failures_count",
3042 : "Number of times thread_local_system creation failed and was retried after back-off.",
3043 : )
3044 0 : .unwrap()
3045 0 : });
3046 : }
3047 :
3048 : pub(crate) mod tenant_throttling {
3049 : use metrics::{register_int_counter_vec, IntCounter};
3050 : use once_cell::sync::Lazy;
3051 :
3052 : use crate::tenant::{self, throttle::Metric};
3053 :
3054 : pub(crate) struct TimelineGet {
3055 : wait_time: IntCounter,
3056 : count: IntCounter,
3057 : }
3058 :
3059 164 : pub(crate) static TIMELINE_GET: Lazy<TimelineGet> = Lazy::new(|| {
3060 164 : static WAIT_USECS: Lazy<metrics::IntCounterVec> = Lazy::new(|| {
3061 164 : register_int_counter_vec!(
3062 164 : "pageserver_tenant_throttling_wait_usecs_sum_global",
3063 164 : "Sum of microseconds that tenants spent waiting for a tenant throttle of a given kind.",
3064 164 : &["kind"]
3065 164 : )
3066 164 : .unwrap()
3067 164 : });
3068 164 :
3069 164 : static WAIT_COUNT: Lazy<metrics::IntCounterVec> = Lazy::new(|| {
3070 164 : register_int_counter_vec!(
3071 164 : "pageserver_tenant_throttling_count_global",
3072 164 : "Count of tenant throttlings, by kind of throttle.",
3073 164 : &["kind"]
3074 164 : )
3075 164 : .unwrap()
3076 164 : });
3077 164 :
3078 164 : let kind = "timeline_get";
3079 164 : TimelineGet {
3080 164 : wait_time: WAIT_USECS.with_label_values(&[kind]),
3081 164 : count: WAIT_COUNT.with_label_values(&[kind]),
3082 164 : }
3083 164 : });
3084 :
3085 : impl Metric for &'static TimelineGet {
3086 : #[inline(always)]
3087 0 : fn observe_throttling(
3088 0 : &self,
3089 0 : tenant::throttle::Observation { wait_time }: &tenant::throttle::Observation,
3090 0 : ) {
3091 0 : let val = u64::try_from(wait_time.as_micros()).unwrap();
3092 0 : self.wait_time.inc_by(val);
3093 0 : self.count.inc();
3094 0 : }
3095 : }
3096 : }
3097 :
3098 : pub(crate) mod disk_usage_based_eviction {
3099 : use super::*;
3100 :
3101 : pub(crate) struct Metrics {
3102 : pub(crate) tenant_collection_time: Histogram,
3103 : pub(crate) tenant_layer_count: Histogram,
3104 : pub(crate) layers_collected: IntCounter,
3105 : pub(crate) layers_selected: IntCounter,
3106 : pub(crate) layers_evicted: IntCounter,
3107 : }
3108 :
3109 : impl Default for Metrics {
3110 0 : fn default() -> Self {
3111 0 : let tenant_collection_time = register_histogram!(
3112 : "pageserver_disk_usage_based_eviction_tenant_collection_seconds",
3113 : "Time spent collecting layers from a tenant -- not normalized by collected layer amount",
3114 : vec![0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1.0, 5.0, 10.0]
3115 : )
3116 0 : .unwrap();
3117 0 :
3118 0 : let tenant_layer_count = register_histogram!(
3119 : "pageserver_disk_usage_based_eviction_tenant_collected_layers",
3120 : "Amount of layers gathered from a tenant",
3121 : vec![5.0, 50.0, 500.0, 5000.0, 50000.0]
3122 : )
3123 0 : .unwrap();
3124 0 :
3125 0 : let layers_collected = register_int_counter!(
3126 : "pageserver_disk_usage_based_eviction_collected_layers_total",
3127 : "Amount of layers collected"
3128 : )
3129 0 : .unwrap();
3130 0 :
3131 0 : let layers_selected = register_int_counter!(
3132 : "pageserver_disk_usage_based_eviction_select_layers_total",
3133 : "Amount of layers selected"
3134 : )
3135 0 : .unwrap();
3136 0 :
3137 0 : let layers_evicted = register_int_counter!(
3138 : "pageserver_disk_usage_based_eviction_evicted_layers_total",
3139 : "Amount of layers successfully evicted"
3140 : )
3141 0 : .unwrap();
3142 0 :
3143 0 : Self {
3144 0 : tenant_collection_time,
3145 0 : tenant_layer_count,
3146 0 : layers_collected,
3147 0 : layers_selected,
3148 0 : layers_evicted,
3149 0 : }
3150 0 : }
3151 : }
3152 :
3153 : pub(crate) static METRICS: Lazy<Metrics> = Lazy::new(Metrics::default);
3154 : }
3155 :
3156 158 : static TOKIO_EXECUTOR_THREAD_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
3157 : register_uint_gauge_vec!(
3158 : "pageserver_tokio_executor_thread_configured_count",
3159 : "Total number of configued tokio executor threads in the process.
3160 : The `setup` label denotes whether we're running with multiple runtimes or a single runtime.",
3161 : &["setup"],
3162 : )
3163 158 : .unwrap()
3164 158 : });
3165 :
3166 158 : pub(crate) fn set_tokio_runtime_setup(setup: &str, num_threads: NonZeroUsize) {
3167 158 : static SERIALIZE: std::sync::Mutex<()> = std::sync::Mutex::new(());
3168 158 : let _guard = SERIALIZE.lock().unwrap();
3169 158 : TOKIO_EXECUTOR_THREAD_COUNT.reset();
3170 158 : TOKIO_EXECUTOR_THREAD_COUNT
3171 158 : .get_metric_with_label_values(&[setup])
3172 158 : .unwrap()
3173 158 : .set(u64::try_from(num_threads.get()).unwrap());
3174 158 : }
3175 :
3176 0 : pub fn preinitialize_metrics() {
3177 0 : // Python tests need these and on some we do alerting.
3178 0 : //
3179 0 : // FIXME(4813): make it so that we have no top level metrics as this fn will easily fall out of
3180 0 : // order:
3181 0 : // - global metrics reside in a Lazy<PageserverMetrics>
3182 0 : // - access via crate::metrics::PS_METRICS.some_metric.inc()
3183 0 : // - could move the statics into TimelineMetrics::new()?
3184 0 :
3185 0 : // counters
3186 0 : [
3187 0 : &UNEXPECTED_ONDEMAND_DOWNLOADS,
3188 0 : &WALRECEIVER_STARTED_CONNECTIONS,
3189 0 : &WALRECEIVER_BROKER_UPDATES,
3190 0 : &WALRECEIVER_CANDIDATES_ADDED,
3191 0 : &WALRECEIVER_CANDIDATES_REMOVED,
3192 0 : &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_FAILURES,
3193 0 : &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_SUCCESSES,
3194 0 : &REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
3195 0 : &REMOTE_ONDEMAND_DOWNLOADED_BYTES,
3196 0 : &CIRCUIT_BREAKERS_BROKEN,
3197 0 : &CIRCUIT_BREAKERS_UNBROKEN,
3198 0 : ]
3199 0 : .into_iter()
3200 0 : .for_each(|c| {
3201 0 : Lazy::force(c);
3202 0 : });
3203 0 :
3204 0 : // Deletion queue stats
3205 0 : Lazy::force(&DELETION_QUEUE);
3206 0 :
3207 0 : // Tenant stats
3208 0 : Lazy::force(&TENANT);
3209 0 :
3210 0 : // Tenant manager stats
3211 0 : Lazy::force(&TENANT_MANAGER);
3212 0 :
3213 0 : Lazy::force(&crate::tenant::storage_layer::layer::LAYER_IMPL_METRICS);
3214 0 : Lazy::force(&disk_usage_based_eviction::METRICS);
3215 :
3216 0 : for state_name in pageserver_api::models::TenantState::VARIANTS {
3217 0 : // initialize the metric for all gauges, otherwise the time series might seemingly show
3218 0 : // values from last restart.
3219 0 : TENANT_STATE_METRIC.with_label_values(&[state_name]).set(0);
3220 0 : }
3221 :
3222 : // countervecs
3223 0 : [&BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT]
3224 0 : .into_iter()
3225 0 : .for_each(|c| {
3226 0 : Lazy::force(c);
3227 0 : });
3228 0 :
3229 0 : // gauges
3230 0 : WALRECEIVER_ACTIVE_MANAGERS.get();
3231 0 :
3232 0 : // histograms
3233 0 : [
3234 0 : &READ_NUM_LAYERS_VISITED,
3235 0 : &VEC_READ_NUM_LAYERS_VISITED,
3236 0 : &WAIT_LSN_TIME,
3237 0 : &WAL_REDO_TIME,
3238 0 : &WAL_REDO_RECORDS_HISTOGRAM,
3239 0 : &WAL_REDO_BYTES_HISTOGRAM,
3240 0 : &WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM,
3241 0 : ]
3242 0 : .into_iter()
3243 0 : .for_each(|h| {
3244 0 : Lazy::force(h);
3245 0 : });
3246 0 :
3247 0 : // Custom
3248 0 : Lazy::force(&RECONSTRUCT_TIME);
3249 0 : Lazy::force(&tenant_throttling::TIMELINE_GET);
3250 0 : Lazy::force(&BASEBACKUP_QUERY_TIME);
3251 0 : Lazy::force(&COMPUTE_COMMANDS_COUNTERS);
3252 0 : }
|
