Line data Source code
1 : //! This module contains functions to serve per-tenant background processes,
2 : //! such as compaction and GC
3 :
4 : use std::ops::ControlFlow;
5 : use std::sync::Arc;
6 : use std::time::{Duration, Instant};
7 :
8 : use crate::context::{DownloadBehavior, RequestContext};
9 : use crate::metrics::TENANT_TASK_EVENTS;
10 : use crate::task_mgr;
11 : use crate::task_mgr::{TaskKind, BACKGROUND_RUNTIME};
12 : use crate::tenant::throttle::Stats;
13 : use crate::tenant::timeline::CompactionError;
14 : use crate::tenant::{Tenant, TenantState};
15 : use tokio_util::sync::CancellationToken;
16 : use tracing::*;
17 : use utils::{backoff, completion};
18 :
19 : static CONCURRENT_BACKGROUND_TASKS: once_cell::sync::Lazy<tokio::sync::Semaphore> =
20 10 : once_cell::sync::Lazy::new(|| {
21 10 : let total_threads = *task_mgr::BACKGROUND_RUNTIME_WORKER_THREADS;
22 10 : let permits = usize::max(
23 10 : 1,
24 10 : // while a lot of the work is done on spawn_blocking, we still do
25 10 : // repartitioning in the async context. this should give leave us some workers
26 10 : // unblocked to be blocked on other work, hopefully easing any outside visible
27 10 : // effects of restarts.
28 10 : //
29 10 : // 6/8 is a guess; previously we ran with unlimited 8 and more from
30 10 : // spawn_blocking.
31 10 : (total_threads * 3).checked_div(4).unwrap_or(0),
32 10 : );
33 10 : assert_ne!(permits, 0, "we will not be adding in permits later");
34 10 : assert!(
35 10 : permits < total_threads,
36 0 : "need threads avail for shorter work"
37 : );
38 10 : tokio::sync::Semaphore::new(permits)
39 10 : });
40 :
41 408 : #[derive(Debug, PartialEq, Eq, Clone, Copy, strum_macros::IntoStaticStr)]
42 : #[strum(serialize_all = "snake_case")]
43 : pub(crate) enum BackgroundLoopKind {
44 : Compaction,
45 : Gc,
46 : Eviction,
47 : ConsumptionMetricsCollectMetrics,
48 : ConsumptionMetricsSyntheticSizeWorker,
49 : InitialLogicalSizeCalculation,
50 : HeatmapUpload,
51 : SecondaryDownload,
52 : }
53 :
54 : impl BackgroundLoopKind {
55 408 : fn as_static_str(&self) -> &'static str {
56 408 : let s: &'static str = self.into();
57 408 : s
58 408 : }
59 : }
60 :
61 : /// Cancellation safe.
62 408 : pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
63 408 : loop_kind: BackgroundLoopKind,
64 408 : _ctx: &RequestContext,
65 408 : ) -> impl Drop {
66 408 : let _guard = crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_GAUGE
67 408 : .with_label_values(&[loop_kind.as_static_str()])
68 408 : .guard();
69 408 :
70 408 : pausable_failpoint!(
71 0 : "initial-size-calculation-permit-pause",
72 0 : loop_kind == BackgroundLoopKind::InitialLogicalSizeCalculation
73 0 : );
74 :
75 408 : match CONCURRENT_BACKGROUND_TASKS.acquire().await {
76 408 : Ok(permit) => permit,
77 0 : Err(_closed) => unreachable!("we never close the semaphore"),
78 : }
79 408 : }
80 :
81 : /// Start per tenant background loops: compaction and gc.
82 0 : pub fn start_background_loops(
83 0 : tenant: &Arc<Tenant>,
84 0 : background_jobs_can_start: Option<&completion::Barrier>,
85 0 : ) {
86 0 : let tenant_shard_id = tenant.tenant_shard_id;
87 0 : task_mgr::spawn(
88 0 : BACKGROUND_RUNTIME.handle(),
89 0 : TaskKind::Compaction,
90 0 : Some(tenant_shard_id),
91 0 : None,
92 0 : &format!("compactor for tenant {tenant_shard_id}"),
93 0 : false,
94 0 : {
95 0 : let tenant = Arc::clone(tenant);
96 0 : let background_jobs_can_start = background_jobs_can_start.cloned();
97 0 : async move {
98 0 : let cancel = task_mgr::shutdown_token();
99 0 : tokio::select! {
100 0 : _ = cancel.cancelled() => { return Ok(()) },
101 0 : _ = completion::Barrier::maybe_wait(background_jobs_can_start) => {}
102 0 : };
103 0 : compaction_loop(tenant, cancel)
104 0 : .instrument(info_span!("compaction_loop", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
105 0 : .await;
106 0 : Ok(())
107 0 : }
108 0 : },
109 0 : );
110 0 : task_mgr::spawn(
111 0 : BACKGROUND_RUNTIME.handle(),
112 0 : TaskKind::GarbageCollector,
113 0 : Some(tenant_shard_id),
114 0 : None,
115 0 : &format!("garbage collector for tenant {tenant_shard_id}"),
116 0 : false,
117 0 : {
118 0 : let tenant = Arc::clone(tenant);
119 0 : let background_jobs_can_start = background_jobs_can_start.cloned();
120 0 : async move {
121 0 : let cancel = task_mgr::shutdown_token();
122 0 : tokio::select! {
123 0 : _ = cancel.cancelled() => { return Ok(()) },
124 0 : _ = completion::Barrier::maybe_wait(background_jobs_can_start) => {}
125 0 : };
126 0 : gc_loop(tenant, cancel)
127 0 : .instrument(info_span!("gc_loop", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
128 0 : .await;
129 0 : Ok(())
130 0 : }
131 0 : },
132 0 : );
133 0 : }
134 :
135 : ///
136 : /// Compaction task's main loop
137 : ///
138 0 : async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
139 0 : const MAX_BACKOFF_SECS: f64 = 300.0;
140 0 : // How many errors we have seen consequtively
141 0 : let mut error_run_count = 0;
142 0 :
143 0 : let mut last_throttle_flag_reset_at = Instant::now();
144 0 :
145 0 : TENANT_TASK_EVENTS.with_label_values(&["start"]).inc();
146 0 : async {
147 0 : let ctx = RequestContext::todo_child(TaskKind::Compaction, DownloadBehavior::Download);
148 0 : let mut first = true;
149 0 : loop {
150 0 : tokio::select! {
151 : _ = cancel.cancelled() => {
152 : return;
153 : },
154 0 : tenant_wait_result = wait_for_active_tenant(&tenant) => match tenant_wait_result {
155 : ControlFlow::Break(()) => return,
156 : ControlFlow::Continue(()) => (),
157 : },
158 0 : }
159 :
160 0 : let period = tenant.get_compaction_period();
161 0 :
162 0 : // TODO: we shouldn't need to await to find tenant and this could be moved outside of
163 0 : // loop, #3501. There are also additional "allowed_errors" in tests.
164 0 : if first {
165 0 : first = false;
166 0 : if random_init_delay(period, &cancel).await.is_err() {
167 0 : break;
168 0 : }
169 0 : }
170 :
171 0 : let started_at = Instant::now();
172 :
173 0 : let sleep_duration = if period == Duration::ZERO {
174 : #[cfg(not(feature = "testing"))]
175 : info!("automatic compaction is disabled");
176 : // check again in 10 seconds, in case it's been enabled again.
177 0 : Duration::from_secs(10)
178 : } else {
179 : // Run compaction
180 0 : if let Err(e) = tenant.compaction_iteration(&cancel, &ctx).await {
181 0 : let wait_duration = backoff::exponential_backoff_duration_seconds(
182 0 : error_run_count + 1,
183 0 : 1.0,
184 0 : MAX_BACKOFF_SECS,
185 0 : );
186 0 : error_run_count += 1;
187 0 : let wait_duration = Duration::from_secs_f64(wait_duration);
188 0 : log_compaction_error(
189 0 : &e,
190 0 : error_run_count,
191 0 : &wait_duration,
192 0 : cancel.is_cancelled(),
193 0 : );
194 0 : wait_duration
195 : } else {
196 0 : error_run_count = 0;
197 0 : period
198 : }
199 : };
200 :
201 0 : warn_when_period_overrun(started_at.elapsed(), period, BackgroundLoopKind::Compaction);
202 :
203 : // Perhaps we did no work and the walredo process has been idle for some time:
204 : // give it a chance to shut down to avoid leaving walredo process running indefinitely.
205 0 : if let Some(walredo_mgr) = &tenant.walredo_mgr {
206 0 : walredo_mgr.maybe_quiesce(period * 10);
207 0 : }
208 :
209 : // TODO: move this (and walredo quiesce) to a separate task that isn't affected by the back-off,
210 : // so we get some upper bound guarantee on when walredo quiesce / this throttling reporting here happens.
211 0 : info_span!(parent: None, "timeline_get_throttle", tenant_id=%tenant.tenant_shard_id, shard_id=%tenant.tenant_shard_id.shard_slug()).in_scope(|| {
212 0 : let now = Instant::now();
213 0 : let prev = std::mem::replace(&mut last_throttle_flag_reset_at, now);
214 0 : let Stats { count_accounted, count_throttled, sum_throttled_usecs } = tenant.timeline_get_throttle.reset_stats();
215 0 : if count_throttled == 0 {
216 0 : return;
217 0 : }
218 0 : let allowed_rps = tenant.timeline_get_throttle.steady_rps();
219 0 : let delta = now - prev;
220 0 : warn!(
221 0 : n_seconds=%format_args!("{:.3}",
222 0 : delta.as_secs_f64()),
223 0 : count_accounted,
224 0 : count_throttled,
225 0 : sum_throttled_usecs,
226 0 : allowed_rps=%format_args!("{allowed_rps:.0}"),
227 0 : "shard was throttled in the last n_seconds")
228 0 : });
229 0 :
230 0 : // Sleep
231 0 : if tokio::time::timeout(sleep_duration, cancel.cancelled())
232 0 : .await
233 0 : .is_ok()
234 : {
235 0 : break;
236 0 : }
237 : }
238 0 : }
239 0 : .await;
240 0 : TENANT_TASK_EVENTS.with_label_values(&["stop"]).inc();
241 0 : }
242 :
243 0 : fn log_compaction_error(
244 0 : e: &CompactionError,
245 0 : error_run_count: u32,
246 0 : sleep_duration: &std::time::Duration,
247 0 : task_cancelled: bool,
248 0 : ) {
249 : use crate::tenant::upload_queue::NotInitialized;
250 : use crate::tenant::PageReconstructError;
251 : use CompactionError::*;
252 :
253 : enum LooksLike {
254 : Info,
255 : Error,
256 : }
257 :
258 0 : let decision = match e {
259 0 : ShuttingDown => None,
260 0 : _ if task_cancelled => Some(LooksLike::Info),
261 0 : Other(e) => {
262 0 : let root_cause = e.root_cause();
263 :
264 0 : let is_stopping = {
265 0 : let upload_queue = root_cause
266 0 : .downcast_ref::<NotInitialized>()
267 0 : .is_some_and(|e| e.is_stopping());
268 0 :
269 0 : let timeline = root_cause
270 0 : .downcast_ref::<PageReconstructError>()
271 0 : .is_some_and(|e| e.is_stopping());
272 0 :
273 0 : upload_queue || timeline
274 : };
275 :
276 0 : if is_stopping {
277 0 : Some(LooksLike::Info)
278 : } else {
279 0 : Some(LooksLike::Error)
280 : }
281 : }
282 : };
283 :
284 0 : match decision {
285 0 : Some(LooksLike::Info) => info!(
286 0 : "Compaction failed {error_run_count} times, retrying in {sleep_duration:?}: {e:#}",
287 0 : ),
288 0 : Some(LooksLike::Error) => error!(
289 0 : "Compaction failed {error_run_count} times, retrying in {sleep_duration:?}: {e:?}",
290 0 : ),
291 0 : None => {}
292 : }
293 0 : }
294 :
295 : ///
296 : /// GC task's main loop
297 : ///
298 0 : async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
299 0 : const MAX_BACKOFF_SECS: f64 = 300.0;
300 0 : // How many errors we have seen consequtively
301 0 : let mut error_run_count = 0;
302 0 :
303 0 : TENANT_TASK_EVENTS.with_label_values(&["start"]).inc();
304 0 : async {
305 0 : // GC might require downloading, to find the cutoff LSN that corresponds to the
306 0 : // cutoff specified as time.
307 0 : let ctx =
308 0 : RequestContext::todo_child(TaskKind::GarbageCollector, DownloadBehavior::Download);
309 0 : let mut first = true;
310 0 : loop {
311 0 : tokio::select! {
312 : _ = cancel.cancelled() => {
313 : return;
314 : },
315 0 : tenant_wait_result = wait_for_active_tenant(&tenant) => match tenant_wait_result {
316 : ControlFlow::Break(()) => return,
317 : ControlFlow::Continue(()) => (),
318 : },
319 0 : }
320 :
321 0 : let period = tenant.get_gc_period();
322 0 :
323 0 : if first {
324 0 : first = false;
325 0 : if random_init_delay(period, &cancel).await.is_err() {
326 0 : break;
327 0 : }
328 0 : }
329 :
330 0 : let started_at = Instant::now();
331 0 :
332 0 : let gc_horizon = tenant.get_gc_horizon();
333 0 : let sleep_duration = if period == Duration::ZERO || gc_horizon == 0 {
334 : #[cfg(not(feature = "testing"))]
335 : info!("automatic GC is disabled");
336 : // check again in 10 seconds, in case it's been enabled again.
337 0 : Duration::from_secs(10)
338 : } else {
339 : // Run gc
340 0 : let res = tenant
341 0 : .gc_iteration(None, gc_horizon, tenant.get_pitr_interval(), &cancel, &ctx)
342 0 : .await;
343 0 : if let Err(e) = res {
344 0 : let wait_duration = backoff::exponential_backoff_duration_seconds(
345 0 : error_run_count + 1,
346 0 : 1.0,
347 0 : MAX_BACKOFF_SECS,
348 0 : );
349 0 : error_run_count += 1;
350 0 : let wait_duration = Duration::from_secs_f64(wait_duration);
351 0 : error!(
352 0 : "Gc failed {error_run_count} times, retrying in {wait_duration:?}: {e:?}",
353 0 : );
354 0 : wait_duration
355 : } else {
356 0 : error_run_count = 0;
357 0 : period
358 : }
359 : };
360 :
361 0 : warn_when_period_overrun(started_at.elapsed(), period, BackgroundLoopKind::Gc);
362 0 :
363 0 : // Sleep
364 0 : if tokio::time::timeout(sleep_duration, cancel.cancelled())
365 0 : .await
366 0 : .is_ok()
367 : {
368 0 : break;
369 0 : }
370 : }
371 0 : }
372 0 : .await;
373 0 : TENANT_TASK_EVENTS.with_label_values(&["stop"]).inc();
374 0 : }
375 :
376 0 : async fn wait_for_active_tenant(tenant: &Arc<Tenant>) -> ControlFlow<()> {
377 0 : // if the tenant has a proper status already, no need to wait for anything
378 0 : if tenant.current_state() == TenantState::Active {
379 0 : ControlFlow::Continue(())
380 : } else {
381 0 : let mut tenant_state_updates = tenant.subscribe_for_state_updates();
382 : loop {
383 0 : match tenant_state_updates.changed().await {
384 : Ok(()) => {
385 0 : let new_state = &*tenant_state_updates.borrow();
386 0 : match new_state {
387 : TenantState::Active => {
388 0 : debug!("Tenant state changed to active, continuing the task loop");
389 0 : return ControlFlow::Continue(());
390 : }
391 0 : state => {
392 0 : debug!("Not running the task loop, tenant is not active: {state:?}");
393 0 : continue;
394 : }
395 : }
396 : }
397 0 : Err(_sender_dropped_error) => {
398 0 : return ControlFlow::Break(());
399 : }
400 : }
401 : }
402 : }
403 0 : }
404 :
405 0 : #[derive(thiserror::Error, Debug)]
406 : #[error("cancelled")]
407 : pub(crate) struct Cancelled;
408 :
409 : /// Provide a random delay for background task initialization.
410 : ///
411 : /// This delay prevents a thundering herd of background tasks and will likely keep them running on
412 : /// different periods for more stable load.
413 0 : pub(crate) async fn random_init_delay(
414 0 : period: Duration,
415 0 : cancel: &CancellationToken,
416 0 : ) -> Result<(), Cancelled> {
417 0 : use rand::Rng;
418 0 :
419 0 : if period == Duration::ZERO {
420 0 : return Ok(());
421 0 : }
422 0 :
423 0 : let d = {
424 0 : let mut rng = rand::thread_rng();
425 0 : rng.gen_range(Duration::ZERO..=period)
426 0 : };
427 0 :
428 0 : match tokio::time::timeout(d, cancel.cancelled()).await {
429 0 : Ok(_) => Err(Cancelled),
430 0 : Err(_) => Ok(()),
431 : }
432 0 : }
433 :
434 : /// Attention: the `task` and `period` beocme labels of a pageserver-wide prometheus metric.
435 0 : pub(crate) fn warn_when_period_overrun(
436 0 : elapsed: Duration,
437 0 : period: Duration,
438 0 : task: BackgroundLoopKind,
439 0 : ) {
440 0 : // Duration::ZERO will happen because it's the "disable [bgtask]" value.
441 0 : if elapsed >= period && period != Duration::ZERO {
442 : // humantime does no significant digits clamping whereas Duration's debug is a bit more
443 : // intelligent. however it makes sense to keep the "configuration format" for period, even
444 : // though there's no way to output the actual config value.
445 0 : info!(
446 0 : ?elapsed,
447 0 : period = %humantime::format_duration(period),
448 0 : ?task,
449 0 : "task iteration took longer than the configured period"
450 0 : );
451 0 : crate::metrics::BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT
452 0 : .with_label_values(&[task.as_static_str(), &format!("{}", period.as_secs())])
453 0 : .inc();
454 0 : }
455 0 : }
|
