Line data Source code
1 : //! Common traits and structs for layers
2 :
3 : pub mod batch_split_writer;
4 : pub mod delta_layer;
5 : pub mod filter_iterator;
6 : pub mod image_layer;
7 : pub mod inmemory_layer;
8 : pub(crate) mod layer;
9 : mod layer_desc;
10 : mod layer_name;
11 : pub mod merge_iterator;
12 :
13 : use std::cmp::Ordering;
14 : use std::collections::hash_map::Entry;
15 : use std::collections::{BinaryHeap, HashMap};
16 : use std::future::Future;
17 : use std::ops::Range;
18 : use std::pin::Pin;
19 : use std::sync::Arc;
20 : use std::sync::atomic::AtomicUsize;
21 : use std::time::{Duration, SystemTime, UNIX_EPOCH};
22 :
23 : pub use batch_split_writer::{BatchLayerWriter, SplitDeltaLayerWriter, SplitImageLayerWriter};
24 : use bytes::Bytes;
25 : pub use delta_layer::{DeltaLayer, DeltaLayerWriter, ValueRef};
26 : use futures::StreamExt;
27 : use futures::stream::FuturesUnordered;
28 : pub use image_layer::{ImageLayer, ImageLayerWriter};
29 : pub use inmemory_layer::InMemoryLayer;
30 : pub(crate) use layer::{EvictionError, Layer, ResidentLayer};
31 : pub use layer_desc::{PersistentLayerDesc, PersistentLayerKey};
32 : pub use layer_name::{DeltaLayerName, ImageLayerName, LayerName};
33 : use pageserver_api::key::Key;
34 : use pageserver_api::keyspace::{KeySpace, KeySpaceRandomAccum};
35 : use pageserver_api::record::NeonWalRecord;
36 : use pageserver_api::value::Value;
37 : use tracing::{Instrument, trace};
38 : use utils::lsn::Lsn;
39 : use utils::sync::gate::GateGuard;
40 :
41 : use self::inmemory_layer::InMemoryLayerFileId;
42 : use super::PageReconstructError;
43 : use super::layer_map::InMemoryLayerDesc;
44 : use super::timeline::{GetVectoredError, ReadPath};
45 : use crate::config::PageServerConf;
46 : use crate::context::{AccessStatsBehavior, RequestContext};
47 :
48 0 : pub fn range_overlaps<T>(a: &Range<T>, b: &Range<T>) -> bool
49 0 : where
50 0 : T: PartialOrd<T>,
51 0 : {
52 0 : if a.start < b.start {
53 0 : a.end > b.start
54 : } else {
55 0 : b.end > a.start
56 : }
57 0 : }
58 :
59 : /// Struct used to communicate across calls to 'get_value_reconstruct_data'.
60 : ///
61 : /// Before first call, you can fill in 'page_img' if you have an older cached
62 : /// version of the page available. That can save work in
63 : /// 'get_value_reconstruct_data', as it can stop searching for page versions
64 : /// when all the WAL records going back to the cached image have been collected.
65 : ///
66 : /// When get_value_reconstruct_data returns Complete, 'img' is set to an image
67 : /// of the page, or the oldest WAL record in 'records' is a will_init-type
68 : /// record that initializes the page without requiring a previous image.
69 : ///
70 : /// If 'get_page_reconstruct_data' returns Continue, some 'records' may have
71 : /// been collected, but there are more records outside the current layer. Pass
72 : /// the same ValueReconstructState struct in the next 'get_value_reconstruct_data'
73 : /// call, to collect more records.
74 : ///
75 : #[derive(Debug, Default)]
76 : pub(crate) struct ValueReconstructState {
77 : pub(crate) records: Vec<(Lsn, NeonWalRecord)>,
78 : pub(crate) img: Option<(Lsn, Bytes)>,
79 : }
80 :
81 : impl ValueReconstructState {
82 : /// Returns the number of page deltas applied to the page image.
83 1335978 : pub fn num_deltas(&self) -> usize {
84 1335978 : match self.img {
85 1335882 : Some(_) => self.records.len(),
86 96 : None => self.records.len() - 1, // omit will_init record
87 : }
88 1335978 : }
89 : }
90 :
91 : #[derive(Clone, Copy, Debug, Default, Eq, PartialEq)]
92 : pub(crate) enum ValueReconstructSituation {
93 : Complete,
94 : #[default]
95 : Continue,
96 : }
97 :
98 : /// On disk representation of a value loaded in a buffer
99 : #[derive(Debug)]
100 : pub(crate) enum OnDiskValue {
101 : /// Unencoded [`Value::Image`]
102 : RawImage(Bytes),
103 : /// Encoded [`Value`]. Can deserialize into an image or a WAL record
104 : WalRecordOrImage(Bytes),
105 : }
106 :
107 : /// Reconstruct data accumulated for a single key during a vectored get
108 : #[derive(Debug, Default)]
109 : pub(crate) struct VectoredValueReconstructState {
110 : pub(crate) on_disk_values: Vec<(Lsn, OnDiskValueIoWaiter)>,
111 :
112 : pub(crate) situation: ValueReconstructSituation,
113 : }
114 :
115 : #[derive(Debug)]
116 : pub(crate) struct OnDiskValueIoWaiter {
117 : rx: tokio::sync::oneshot::Receiver<OnDiskValueIoResult>,
118 : }
119 :
120 : #[derive(Debug)]
121 : #[must_use]
122 : pub(crate) enum OnDiskValueIo {
123 : /// Traversal identified this IO as required to complete the vectored get.
124 : Required {
125 : num_active_ios: Arc<AtomicUsize>,
126 : tx: tokio::sync::oneshot::Sender<OnDiskValueIoResult>,
127 : },
128 : /// Sparse keyspace reads always read all the values for a given key,
129 : /// even though only the first value is needed.
130 : ///
131 : /// This variant represents the unnecessary IOs for those values at lower LSNs
132 : /// that aren't needed, but are currently still being done.
133 : ///
134 : /// The execution of unnecessary IOs was a pre-existing behavior before concurrent IO.
135 : /// We added this explicit representation here so that we can drop
136 : /// unnecessary IO results immediately, instead of buffering them in
137 : /// `oneshot` channels inside [`VectoredValueReconstructState`] until
138 : /// [`VectoredValueReconstructState::collect_pending_ios`] gets called.
139 : Unnecessary,
140 : }
141 :
142 : type OnDiskValueIoResult = Result<OnDiskValue, std::io::Error>;
143 :
144 : impl OnDiskValueIo {
145 1482487 : pub(crate) fn complete(self, res: OnDiskValueIoResult) {
146 1482487 : match self {
147 1337504 : OnDiskValueIo::Required { num_active_ios, tx } => {
148 1337504 : num_active_ios.fetch_sub(1, std::sync::atomic::Ordering::Release);
149 1337504 : let _ = tx.send(res);
150 1337504 : }
151 144983 : OnDiskValueIo::Unnecessary => {
152 144983 : // Nobody cared, see variant doc comment.
153 144983 : }
154 : }
155 1482487 : }
156 : }
157 :
158 : #[derive(Debug, thiserror::Error)]
159 : pub(crate) enum WaitCompletionError {
160 : #[error("OnDiskValueIo was dropped without completing, likely the sidecar task panicked")]
161 : IoDropped,
162 : }
163 :
164 : impl OnDiskValueIoWaiter {
165 1337502 : pub(crate) async fn wait_completion(self) -> Result<OnDiskValueIoResult, WaitCompletionError> {
166 1337502 : // NB: for Unnecessary IOs, this method never gets called because we don't add them to `on_disk_values`.
167 1337502 : self.rx.await.map_err(|_| WaitCompletionError::IoDropped)
168 1337502 : }
169 : }
170 :
171 : impl VectoredValueReconstructState {
172 : /// # Cancel-Safety
173 : ///
174 : /// Technically fine to stop polling this future, but, the IOs will still
175 : /// be executed to completion by the sidecar task and hold on to / consume resources.
176 : /// Better not do it to make reasonsing about the system easier.
177 1336118 : pub(crate) async fn collect_pending_ios(
178 1336118 : self,
179 1336118 : ) -> Result<ValueReconstructState, PageReconstructError> {
180 : use utils::bin_ser::BeSer;
181 :
182 1336118 : let mut res = Ok(ValueReconstructState::default());
183 :
184 : // We should try hard not to bail early, so that by the time we return from this
185 : // function, all IO for this value is done. It's not required -- we could totally
186 : // stop polling the IO futures in the sidecar task, they need to support that,
187 : // but just stopping to poll doesn't reduce the IO load on the disk. It's easier
188 : // to reason about the system if we just wait for all IO to complete, even if
189 : // we're no longer interested in the result.
190 : //
191 : // Revisit this when IO futures are replaced with a more sophisticated IO system
192 : // and an IO scheduler, where we know which IOs were submitted and which ones
193 : // just queued. Cf the comment on IoConcurrency::spawn_io.
194 2673620 : for (lsn, waiter) in self.on_disk_values {
195 1337502 : let value_recv_res = waiter
196 1337502 : .wait_completion()
197 1337502 : // we rely on the caller to poll us to completion, so this is not a bail point
198 1337502 : .await;
199 : // Force not bailing early by wrapping the code into a closure.
200 : #[allow(clippy::redundant_closure_call)]
201 1337502 : let _: () = (|| {
202 1337502 : match (&mut res, value_recv_res) {
203 0 : (Err(_), _) => {
204 0 : // We've already failed, no need to process more.
205 0 : }
206 0 : (Ok(_), Err(wait_err)) => {
207 0 : // This shouldn't happen - likely the sidecar task panicked.
208 0 : res = Err(PageReconstructError::Other(wait_err.into()));
209 0 : }
210 0 : (Ok(_), Ok(Err(err))) => {
211 0 : let err: std::io::Error = err;
212 0 : // TODO: returning IO error here will fail a compute query.
213 0 : // Probably not what we want, we're not doing `maybe_fatal_err`
214 0 : // in the IO futures.
215 0 : // But it's been like that for a long time, not changing it
216 0 : // as part of concurrent IO.
217 0 : // => https://github.com/neondatabase/neon/issues/10454
218 0 : res = Err(PageReconstructError::Other(err.into()));
219 0 : }
220 38217 : (Ok(ok), Ok(Ok(OnDiskValue::RawImage(img)))) => {
221 38217 : assert!(ok.img.is_none());
222 38217 : ok.img = Some((lsn, img));
223 : }
224 1299285 : (Ok(ok), Ok(Ok(OnDiskValue::WalRecordOrImage(buf)))) => {
225 1299285 : match Value::des(&buf) {
226 1480 : Ok(Value::WalRecord(rec)) => {
227 1480 : ok.records.push((lsn, rec));
228 1480 : }
229 1297805 : Ok(Value::Image(img)) => {
230 1297805 : assert!(ok.img.is_none());
231 1297805 : ok.img = Some((lsn, img));
232 : }
233 0 : Err(err) => {
234 0 : res = Err(PageReconstructError::Other(err.into()));
235 0 : }
236 : }
237 : }
238 : }
239 1337502 : })();
240 1337502 : }
241 :
242 1336118 : res
243 1336118 : }
244 : }
245 :
246 : /// Bag of data accumulated during a vectored get..
247 : pub(crate) struct ValuesReconstructState {
248 : /// The keys will be removed after `get_vectored` completes. The caller outside `Timeline`
249 : /// should not expect to get anything from this hashmap.
250 : pub(crate) keys: HashMap<Key, VectoredValueReconstructState>,
251 : /// The keys which are already retrieved
252 : keys_done: KeySpaceRandomAccum,
253 :
254 : /// The keys covered by the image layers
255 : keys_with_image_coverage: Option<Range<Key>>,
256 :
257 : // Statistics that are still accessible as a caller of `get_vectored_impl`.
258 : layers_visited: u32,
259 : delta_layers_visited: u32,
260 :
261 : pub(crate) io_concurrency: IoConcurrency,
262 : num_active_ios: Arc<AtomicUsize>,
263 :
264 : pub(crate) read_path: Option<ReadPath>,
265 : }
266 :
267 : /// The level of IO concurrency to be used on the read path
268 : ///
269 : /// The desired end state is that we always do parallel IO.
270 : /// This struct and the dispatching in the impl will be removed once
271 : /// we've built enough confidence.
272 : pub(crate) enum IoConcurrency {
273 : Sequential,
274 : SidecarTask {
275 : task_id: usize,
276 : ios_tx: tokio::sync::mpsc::UnboundedSender<IoFuture>,
277 : },
278 : }
279 :
280 : type IoFuture = Pin<Box<dyn Send + Future<Output = ()>>>;
281 :
282 : pub(crate) enum SelectedIoConcurrency {
283 : Sequential,
284 : SidecarTask(GateGuard),
285 : }
286 :
287 : impl std::fmt::Debug for IoConcurrency {
288 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
289 0 : match self {
290 0 : IoConcurrency::Sequential => write!(f, "Sequential"),
291 0 : IoConcurrency::SidecarTask { .. } => write!(f, "SidecarTask"),
292 : }
293 0 : }
294 : }
295 :
296 : impl std::fmt::Debug for SelectedIoConcurrency {
297 64 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
298 64 : match self {
299 32 : SelectedIoConcurrency::Sequential => write!(f, "Sequential"),
300 32 : SelectedIoConcurrency::SidecarTask(_) => write!(f, "SidecarTask"),
301 : }
302 64 : }
303 : }
304 :
305 : impl IoConcurrency {
306 : /// Force sequential IO. This is a temporary workaround until we have
307 : /// moved plumbing-through-the-call-stack
308 : /// of IoConcurrency into `RequestContextq.
309 : ///
310 : /// DO NOT USE for new code.
311 : ///
312 : /// Tracking issue: <https://github.com/neondatabase/neon/issues/10460>.
313 1215138 : pub(crate) fn sequential() -> Self {
314 1215138 : Self::spawn(SelectedIoConcurrency::Sequential)
315 1215138 : }
316 :
317 1036 : pub(crate) fn spawn_from_conf(
318 1036 : conf: &'static PageServerConf,
319 1036 : gate_guard: GateGuard,
320 1036 : ) -> IoConcurrency {
321 : use pageserver_api::config::GetVectoredConcurrentIo;
322 1036 : let selected = match conf.get_vectored_concurrent_io {
323 1036 : GetVectoredConcurrentIo::Sequential => SelectedIoConcurrency::Sequential,
324 0 : GetVectoredConcurrentIo::SidecarTask => SelectedIoConcurrency::SidecarTask(gate_guard),
325 : };
326 1036 : Self::spawn(selected)
327 1036 : }
328 :
329 1216238 : pub(crate) fn spawn(io_concurrency: SelectedIoConcurrency) -> Self {
330 1216238 : match io_concurrency {
331 1216206 : SelectedIoConcurrency::Sequential => IoConcurrency::Sequential,
332 32 : SelectedIoConcurrency::SidecarTask(gate_guard) => {
333 32 : let (ios_tx, ios_rx) = tokio::sync::mpsc::unbounded_channel();
334 : static TASK_ID: AtomicUsize = AtomicUsize::new(0);
335 32 : let task_id = TASK_ID.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
336 : // TODO: enrich the span with more context (tenant,shard,timeline) + (basebackup|pagestream|...)
337 32 : let span =
338 32 : tracing::info_span!(parent: None, "IoConcurrency_sidecar", task_id = task_id);
339 32 : trace!(task_id, "spawning sidecar task");
340 32 : tokio::spawn(async move {
341 32 : trace!("start");
342 32 : scopeguard::defer!{ trace!("end") };
343 : type IosRx = tokio::sync::mpsc::UnboundedReceiver<IoFuture>;
344 : enum State {
345 : Waiting {
346 : // invariant: is_empty(), but we recycle the allocation
347 : empty_futures: FuturesUnordered<IoFuture>,
348 : ios_rx: IosRx,
349 : },
350 : Executing {
351 : futures: FuturesUnordered<IoFuture>,
352 : ios_rx: IosRx,
353 : },
354 : ShuttingDown {
355 : futures: FuturesUnordered<IoFuture>,
356 : },
357 : }
358 32 : let mut state = State::Waiting {
359 32 : empty_futures: FuturesUnordered::new(),
360 32 : ios_rx,
361 32 : };
362 : loop {
363 39110 : match state {
364 : State::Waiting {
365 18563 : empty_futures,
366 18563 : mut ios_rx,
367 18563 : } => {
368 18563 : assert!(empty_futures.is_empty());
369 18563 : tokio::select! {
370 18563 : fut = ios_rx.recv() => {
371 18531 : if let Some(fut) = fut {
372 18531 : trace!("received new io future");
373 18531 : empty_futures.push(fut);
374 18531 : state = State::Executing { futures: empty_futures, ios_rx };
375 : } else {
376 0 : state = State::ShuttingDown { futures: empty_futures }
377 : }
378 : }
379 : }
380 : }
381 : State::Executing {
382 20547 : mut futures,
383 20547 : mut ios_rx,
384 20547 : } => {
385 20547 : tokio::select! {
386 20547 : res = futures.next() => {
387 19539 : trace!("io future completed");
388 19539 : assert!(res.is_some());
389 19539 : if futures.is_empty() {
390 18531 : state = State::Waiting { empty_futures: futures, ios_rx};
391 18531 : } else {
392 1008 : state = State::Executing { futures, ios_rx };
393 1008 : }
394 : }
395 20547 : fut = ios_rx.recv() => {
396 1008 : if let Some(fut) = fut {
397 1008 : trace!("received new io future");
398 1008 : futures.push(fut);
399 1008 : state = State::Executing { futures, ios_rx};
400 0 : } else {
401 0 : state = State::ShuttingDown { futures };
402 0 : }
403 : }
404 : }
405 : }
406 : State::ShuttingDown {
407 0 : mut futures,
408 0 : } => {
409 0 : trace!("shutting down");
410 0 : while let Some(()) = futures.next().await {
411 0 : trace!("io future completed (shutdown)");
412 : // drain
413 : }
414 0 : trace!("shutdown complete");
415 0 : break;
416 0 : }
417 0 : }
418 0 : }
419 0 : drop(gate_guard); // drop it right before we exit
420 32 : }.instrument(span));
421 32 : IoConcurrency::SidecarTask { task_id, ios_tx }
422 : }
423 : }
424 1216238 : }
425 :
426 76410 : pub(crate) fn clone(&self) -> Self {
427 76410 : match self {
428 39532 : IoConcurrency::Sequential => IoConcurrency::Sequential,
429 36878 : IoConcurrency::SidecarTask { task_id, ios_tx } => IoConcurrency::SidecarTask {
430 36878 : task_id: *task_id,
431 36878 : ios_tx: ios_tx.clone(),
432 36878 : },
433 : }
434 76410 : }
435 :
436 : /// Submit an IO to be executed in the background. DEADLOCK RISK, read the full doc string.
437 : ///
438 : /// The IO is represented as an opaque future.
439 : /// IO completion must be handled inside the future, e.g., through a oneshot channel.
440 : ///
441 : /// The API seems simple but there are multiple **pitfalls** involving
442 : /// DEADLOCK RISK.
443 : ///
444 : /// First, there are no guarantees about the exexecution of the IO.
445 : /// It may be `await`ed in-place before this function returns.
446 : /// It may be polled partially by this task and handed off to another task to be finished.
447 : /// It may be polled and then dropped before returning ready.
448 : ///
449 : /// This means that submitted IOs must not be interedependent.
450 : /// Interdependence may be through shared limited resources, e.g.,
451 : /// - VirtualFile file descriptor cache slot acquisition
452 : /// - tokio-epoll-uring slot
453 : ///
454 : /// # Why current usage is safe from deadlocks
455 : ///
456 : /// Textbook condition for a deadlock is that _all_ of the following be given
457 : /// - Mutual exclusion
458 : /// - Hold and wait
459 : /// - No preemption
460 : /// - Circular wait
461 : ///
462 : /// The current usage is safe because:
463 : /// - Mutual exclusion: IO futures definitely use mutexes, no way around that for now
464 : /// - Hold and wait: IO futures currently hold two kinds of locks/resources while waiting
465 : /// for acquisition of other resources:
466 : /// - VirtualFile file descriptor cache slot tokio mutex
467 : /// - tokio-epoll-uring slot (uses tokio notify => wait queue, much like mutex)
468 : /// - No preemption: there's no taking-away of acquired locks/resources => given
469 : /// - Circular wait: this is the part of the condition that isn't met: all IO futures
470 : /// first acquire VirtualFile mutex, then tokio-epoll-uring slot.
471 : /// There is no IO future that acquires slot before VirtualFile.
472 : /// Hence there can be no circular waiting.
473 : /// Hence there cannot be a deadlock.
474 : ///
475 : /// This is a very fragile situation and must be revisited whenver any code called from
476 : /// inside the IO futures is changed.
477 : ///
478 : /// We will move away from opaque IO futures towards well-defined IOs at some point in
479 : /// the future when we have shipped this first version of concurrent IO to production
480 : /// and are ready to retire the Sequential mode which runs the futures in place.
481 : /// Right now, while brittle, the opaque IO approach allows us to ship the feature
482 : /// with minimal changes to the code and minimal changes to existing behavior in Sequential mode.
483 : ///
484 : /// Also read the comment in `collect_pending_ios`.
485 1525669 : pub(crate) async fn spawn_io<F>(&mut self, fut: F)
486 1525669 : where
487 1525669 : F: std::future::Future<Output = ()> + Send + 'static,
488 1525669 : {
489 1525669 : match self {
490 1506130 : IoConcurrency::Sequential => fut.await,
491 19539 : IoConcurrency::SidecarTask { ios_tx, .. } => {
492 19539 : let fut = Box::pin(fut);
493 19539 : // NB: experiments showed that doing an opportunistic poll of `fut` here was bad for throughput
494 19539 : // while insignificant for latency.
495 19539 : // It would make sense to revisit the tokio-epoll-uring API in the future such that we can try
496 19539 : // a submission here, but never poll the future. That way, io_uring can make proccess while
497 19539 : // the future sits in the ios_tx queue.
498 19539 : match ios_tx.send(fut) {
499 19539 : Ok(()) => {}
500 : Err(_) => {
501 0 : unreachable!("the io task must have exited, likely it panicked")
502 : }
503 : }
504 : }
505 : }
506 1525669 : }
507 :
508 : #[cfg(test)]
509 64 : pub(crate) fn spawn_for_test() -> impl std::ops::DerefMut<Target = Self> {
510 : use std::ops::{Deref, DerefMut};
511 :
512 : use tracing::info;
513 : use utils::sync::gate::Gate;
514 :
515 : // Spawn needs a Gate, give it one.
516 : struct Wrapper {
517 : inner: IoConcurrency,
518 : #[allow(dead_code)]
519 : gate: Box<Gate>,
520 : }
521 : impl Deref for Wrapper {
522 : type Target = IoConcurrency;
523 :
524 36974 : fn deref(&self) -> &Self::Target {
525 36974 : &self.inner
526 36974 : }
527 : }
528 : impl DerefMut for Wrapper {
529 0 : fn deref_mut(&mut self) -> &mut Self::Target {
530 0 : &mut self.inner
531 0 : }
532 : }
533 64 : let gate = Box::new(Gate::default());
534 :
535 : // The default behavior when running Rust unit tests without any further
536 : // flags is to use the new behavior.
537 : // The CI uses the following environment variable to unit test both old
538 : // and new behavior.
539 : // NB: the Python regression & perf tests take the `else` branch
540 : // below and have their own defaults management.
541 64 : let selected = {
542 : // The pageserver_api::config type is unsuitable because it's internally tagged.
543 64 : #[derive(serde::Deserialize)]
544 : #[serde(rename_all = "kebab-case")]
545 : enum TestOverride {
546 : Sequential,
547 : SidecarTask,
548 : }
549 : use once_cell::sync::Lazy;
550 64 : static TEST_OVERRIDE: Lazy<TestOverride> = Lazy::new(|| {
551 64 : utils::env::var_serde_json_string(
552 64 : "NEON_PAGESERVER_UNIT_TEST_GET_VECTORED_CONCURRENT_IO",
553 64 : )
554 64 : .unwrap_or(TestOverride::SidecarTask)
555 64 : });
556 :
557 64 : match *TEST_OVERRIDE {
558 32 : TestOverride::Sequential => SelectedIoConcurrency::Sequential,
559 : TestOverride::SidecarTask => {
560 32 : SelectedIoConcurrency::SidecarTask(gate.enter().expect("just created it"))
561 : }
562 : }
563 : };
564 :
565 64 : info!(?selected, "get_vectored_concurrent_io test");
566 :
567 64 : Wrapper {
568 64 : inner: Self::spawn(selected),
569 64 : gate,
570 64 : }
571 64 : }
572 : }
573 :
574 : /// Make noise in case the [`ValuesReconstructState`] gets dropped while
575 : /// there are still IOs in flight.
576 : /// Refer to `collect_pending_ios` for why we prefer not to do that.
577 : //
578 : /// We log from here instead of from the sidecar task because the [`ValuesReconstructState`]
579 : /// gets dropped in a tracing span with more context.
580 : /// We repeat the sidecar tasks's `task_id` so we can correlate what we emit here with
581 : /// the logs / panic handler logs from the sidecar task, which also logs the `task_id`.
582 : impl Drop for ValuesReconstructState {
583 1255332 : fn drop(&mut self) {
584 1255332 : let num_active_ios = self
585 1255332 : .num_active_ios
586 1255332 : .load(std::sync::atomic::Ordering::Acquire);
587 1255332 : if num_active_ios == 0 {
588 1255330 : return;
589 2 : }
590 2 : let sidecar_task_id = match &self.io_concurrency {
591 0 : IoConcurrency::Sequential => None,
592 2 : IoConcurrency::SidecarTask { task_id, .. } => Some(*task_id),
593 : };
594 2 : tracing::warn!(
595 : num_active_ios,
596 : ?sidecar_task_id,
597 0 : backtrace=%std::backtrace::Backtrace::force_capture(),
598 0 : "dropping ValuesReconstructState while some IOs have not been completed",
599 : );
600 1255332 : }
601 : }
602 :
603 : impl ValuesReconstructState {
604 1255332 : pub(crate) fn new(io_concurrency: IoConcurrency) -> Self {
605 1255332 : Self {
606 1255332 : keys: HashMap::new(),
607 1255332 : keys_done: KeySpaceRandomAccum::new(),
608 1255332 : keys_with_image_coverage: None,
609 1255332 : layers_visited: 0,
610 1255332 : delta_layers_visited: 0,
611 1255332 : io_concurrency,
612 1255332 : num_active_ios: Arc::new(AtomicUsize::new(0)),
613 1255332 : read_path: None,
614 1255332 : }
615 1255332 : }
616 :
617 : /// Absolutely read [`IoConcurrency::spawn_io`] to learn about assumptions & pitfalls.
618 1525669 : pub(crate) async fn spawn_io<F>(&mut self, fut: F)
619 1525669 : where
620 1525669 : F: std::future::Future<Output = ()> + Send + 'static,
621 1525669 : {
622 1525669 : self.io_concurrency.spawn_io(fut).await;
623 1525669 : }
624 :
625 1693718 : pub(crate) fn on_layer_visited(&mut self, layer: &ReadableLayer) {
626 1693718 : self.layers_visited += 1;
627 1693718 : if let ReadableLayer::PersistentLayer(layer) = layer {
628 480346 : if layer.layer_desc().is_delta() {
629 435246 : self.delta_layers_visited += 1;
630 435246 : }
631 1213372 : }
632 1693718 : }
633 :
634 468 : pub(crate) fn get_delta_layers_visited(&self) -> u32 {
635 468 : self.delta_layers_visited
636 468 : }
637 :
638 1255274 : pub(crate) fn get_layers_visited(&self) -> u32 {
639 1255274 : self.layers_visited
640 1255274 : }
641 :
642 : /// On hitting image layer, we can mark all keys in this range as done, because
643 : /// if the image layer does not contain a key, it is deleted/never added.
644 45124 : pub(crate) fn on_image_layer_visited(&mut self, key_range: &Range<Key>) {
645 45124 : let prev_val = self.keys_with_image_coverage.replace(key_range.clone());
646 45124 : assert_eq!(
647 : prev_val, None,
648 0 : "should consume the keyspace before the next iteration"
649 : );
650 45124 : }
651 :
652 : /// Update the state collected for a given key.
653 : /// Returns true if this was the last value needed for the key and false otherwise.
654 : ///
655 : /// If the key is done after the update, mark it as such.
656 : ///
657 : /// If the key is in the sparse keyspace (i.e., aux files), we do not track them in
658 : /// `key_done`.
659 : // TODO: rename this method & update description.
660 1482487 : pub(crate) fn update_key(&mut self, key: &Key, lsn: Lsn, completes: bool) -> OnDiskValueIo {
661 1482487 : let state = self.keys.entry(*key).or_default();
662 1482487 :
663 1482487 : let is_sparse_key = key.is_sparse();
664 :
665 1482487 : let required_io = match state.situation {
666 : ValueReconstructSituation::Complete => {
667 144983 : if is_sparse_key {
668 : // Sparse keyspace might be visited multiple times because
669 : // we don't track unmapped keyspaces.
670 144983 : return OnDiskValueIo::Unnecessary;
671 : } else {
672 0 : unreachable!()
673 : }
674 : }
675 : ValueReconstructSituation::Continue => {
676 1337504 : self.num_active_ios
677 1337504 : .fetch_add(1, std::sync::atomic::Ordering::Release);
678 1337504 : let (tx, rx) = tokio::sync::oneshot::channel();
679 1337504 : state.on_disk_values.push((lsn, OnDiskValueIoWaiter { rx }));
680 1337504 : OnDiskValueIo::Required {
681 1337504 : tx,
682 1337504 : num_active_ios: Arc::clone(&self.num_active_ios),
683 1337504 : }
684 1337504 : }
685 1337504 : };
686 1337504 :
687 1337504 : if completes && state.situation == ValueReconstructSituation::Continue {
688 1336120 : state.situation = ValueReconstructSituation::Complete;
689 1336120 : if !is_sparse_key {
690 1208604 : self.keys_done.add_key(*key);
691 1208604 : }
692 1384 : }
693 :
694 1337504 : required_io
695 1482487 : }
696 :
697 : /// Returns the key space describing the keys that have
698 : /// been marked as completed since the last call to this function.
699 : /// Returns individual keys done, and the image layer coverage.
700 3397353 : pub(crate) fn consume_done_keys(&mut self) -> (KeySpace, Option<Range<Key>>) {
701 3397353 : (
702 3397353 : self.keys_done.consume_keyspace(),
703 3397353 : self.keys_with_image_coverage.take(),
704 3397353 : )
705 3397353 : }
706 : }
707 :
708 : /// A key that uniquely identifies a layer in a timeline
709 : #[derive(Debug, PartialEq, Eq, Clone, Hash)]
710 : pub(crate) enum LayerId {
711 : PersitentLayerId(PersistentLayerKey),
712 : InMemoryLayerId(InMemoryLayerFileId),
713 : }
714 :
715 : /// Uniquely identify a layer visit by the layer
716 : /// and LSN floor (or start LSN) of the reads.
717 : /// The layer itself is not enough since we may
718 : /// have different LSN lower bounds for delta layer reads.
719 : #[derive(Debug, PartialEq, Eq, Clone, Hash)]
720 : struct LayerToVisitId {
721 : layer_id: LayerId,
722 : lsn_floor: Lsn,
723 : }
724 :
725 : #[derive(Debug, PartialEq, Eq, Hash)]
726 : pub enum ReadableLayerWeak {
727 : PersistentLayer(Arc<PersistentLayerDesc>),
728 : InMemoryLayer(InMemoryLayerDesc),
729 : }
730 :
731 : /// Layer wrapper for the read path. Note that it is valid
732 : /// to use these layers even after external operations have
733 : /// been performed on them (compaction, freeze, etc.).
734 : #[derive(Debug)]
735 : pub(crate) enum ReadableLayer {
736 : PersistentLayer(Layer),
737 : InMemoryLayer(Arc<InMemoryLayer>),
738 : }
739 :
740 : /// A partial description of a read to be done.
741 : #[derive(Debug, Clone)]
742 : struct LayerVisit {
743 : /// An id used to resolve the readable layer within the fringe
744 : layer_to_visit_id: LayerToVisitId,
745 : /// Lsn range for the read, used for selecting the next read
746 : lsn_range: Range<Lsn>,
747 : }
748 :
749 : /// Data structure which maintains a fringe of layers for the
750 : /// read path. The fringe is the set of layers which intersects
751 : /// the current keyspace that the search is descending on.
752 : /// Each layer tracks the keyspace that intersects it.
753 : ///
754 : /// The fringe must appear sorted by Lsn. Hence, it uses
755 : /// a two layer indexing scheme.
756 : #[derive(Debug)]
757 : pub(crate) struct LayerFringe {
758 : planned_visits_by_lsn: BinaryHeap<LayerVisit>,
759 : visit_reads: HashMap<LayerToVisitId, LayerVisitReads>,
760 : }
761 :
762 : #[derive(Debug)]
763 : struct LayerVisitReads {
764 : layer: ReadableLayer,
765 : target_keyspace: KeySpaceRandomAccum,
766 : }
767 :
768 : impl LayerFringe {
769 1703635 : pub(crate) fn new() -> Self {
770 1703635 : LayerFringe {
771 1703635 : planned_visits_by_lsn: BinaryHeap::new(),
772 1703635 : visit_reads: HashMap::new(),
773 1703635 : }
774 1703635 : }
775 :
776 3397353 : pub(crate) fn next_layer(&mut self) -> Option<(ReadableLayer, KeySpace, Range<Lsn>)> {
777 3397353 : let read_desc = self.planned_visits_by_lsn.pop()?;
778 :
779 1693718 : let removed = self.visit_reads.remove_entry(&read_desc.layer_to_visit_id);
780 1693718 :
781 1693718 : match removed {
782 : Some((
783 : _,
784 : LayerVisitReads {
785 1693718 : layer,
786 1693718 : mut target_keyspace,
787 1693718 : },
788 1693718 : )) => Some((
789 1693718 : layer,
790 1693718 : target_keyspace.consume_keyspace(),
791 1693718 : read_desc.lsn_range,
792 1693718 : )),
793 0 : None => unreachable!("fringe internals are always consistent"),
794 : }
795 3397353 : }
796 :
797 1694654 : pub(crate) fn update(
798 1694654 : &mut self,
799 1694654 : layer: ReadableLayer,
800 1694654 : keyspace: KeySpace,
801 1694654 : lsn_range: Range<Lsn>,
802 1694654 : ) {
803 1694654 : let layer_to_visit_id = LayerToVisitId {
804 1694654 : layer_id: layer.id(),
805 1694654 : lsn_floor: lsn_range.start,
806 1694654 : };
807 1694654 :
808 1694654 : let entry = self.visit_reads.entry(layer_to_visit_id.clone());
809 1694654 : match entry {
810 936 : Entry::Occupied(mut entry) => {
811 936 : entry.get_mut().target_keyspace.add_keyspace(keyspace);
812 936 : }
813 1693718 : Entry::Vacant(entry) => {
814 1693718 : self.planned_visits_by_lsn.push(LayerVisit {
815 1693718 : lsn_range,
816 1693718 : layer_to_visit_id: layer_to_visit_id.clone(),
817 1693718 : });
818 1693718 : let mut accum = KeySpaceRandomAccum::new();
819 1693718 : accum.add_keyspace(keyspace);
820 1693718 : entry.insert(LayerVisitReads {
821 1693718 : layer,
822 1693718 : target_keyspace: accum,
823 1693718 : });
824 1693718 : }
825 : }
826 1694654 : }
827 : }
828 :
829 : impl Default for LayerFringe {
830 0 : fn default() -> Self {
831 0 : Self::new()
832 0 : }
833 : }
834 :
835 : impl Ord for LayerVisit {
836 72 : fn cmp(&self, other: &Self) -> Ordering {
837 72 : let ord = self.lsn_range.end.cmp(&other.lsn_range.end);
838 72 : if ord == std::cmp::Ordering::Equal {
839 48 : self.lsn_range.start.cmp(&other.lsn_range.start).reverse()
840 : } else {
841 24 : ord
842 : }
843 72 : }
844 : }
845 :
846 : impl PartialOrd for LayerVisit {
847 72 : fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
848 72 : Some(self.cmp(other))
849 72 : }
850 : }
851 :
852 : impl PartialEq for LayerVisit {
853 0 : fn eq(&self, other: &Self) -> bool {
854 0 : self.lsn_range == other.lsn_range
855 0 : }
856 : }
857 :
858 : impl Eq for LayerVisit {}
859 :
860 : impl ReadableLayer {
861 1694654 : pub(crate) fn id(&self) -> LayerId {
862 1694654 : match self {
863 480442 : Self::PersistentLayer(layer) => LayerId::PersitentLayerId(layer.layer_desc().key()),
864 1214212 : Self::InMemoryLayer(layer) => LayerId::InMemoryLayerId(layer.file_id()),
865 : }
866 1694654 : }
867 :
868 1693718 : pub(crate) async fn get_values_reconstruct_data(
869 1693718 : &self,
870 1693718 : keyspace: KeySpace,
871 1693718 : lsn_range: Range<Lsn>,
872 1693718 : reconstruct_state: &mut ValuesReconstructState,
873 1693718 : ctx: &RequestContext,
874 1693718 : ) -> Result<(), GetVectoredError> {
875 1693718 : match self {
876 480346 : ReadableLayer::PersistentLayer(layer) => {
877 480346 : layer
878 480346 : .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_state, ctx)
879 480346 : .await
880 : }
881 1213372 : ReadableLayer::InMemoryLayer(layer) => {
882 1213372 : layer
883 1213372 : .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_state, ctx)
884 1213372 : .await
885 : }
886 : }
887 1693718 : }
888 : }
889 :
890 : /// Layers contain a hint indicating whether they are likely to be used for reads.
891 : ///
892 : /// This is a hint rather than an authoritative value, so that we do not have to update it synchronously
893 : /// when changing the visibility of layers (for example when creating a branch that makes some previously
894 : /// covered layers visible). It should be used for cache management but not for correctness-critical checks.
895 : #[derive(Debug, Clone, PartialEq, Eq)]
896 : pub enum LayerVisibilityHint {
897 : /// A Visible layer might be read while serving a read, because there is not an image layer between it
898 : /// and a readable LSN (the tip of the branch or a child's branch point)
899 : Visible,
900 : /// A Covered layer probably won't be read right now, but _can_ be read in future if someone creates
901 : /// a branch or ephemeral endpoint at an LSN below the layer that covers this.
902 : Covered,
903 : }
904 :
905 : pub(crate) struct LayerAccessStats(std::sync::atomic::AtomicU64);
906 :
907 0 : #[derive(Clone, Copy, strum_macros::EnumString)]
908 : pub(crate) enum LayerAccessStatsReset {
909 : NoReset,
910 : AllStats,
911 : }
912 :
913 : impl Default for LayerAccessStats {
914 3864 : fn default() -> Self {
915 3864 : // Default value is to assume resident since creation time, and visible.
916 3864 : let (_mask, mut value) = Self::to_low_res_timestamp(Self::RTIME_SHIFT, SystemTime::now());
917 3864 : value |= 0x1 << Self::VISIBILITY_SHIFT;
918 3864 :
919 3864 : Self(std::sync::atomic::AtomicU64::new(value))
920 3864 : }
921 : }
922 :
923 : // Efficient store of two very-low-resolution timestamps and some bits. Used for storing last access time and
924 : // last residence change time.
925 : impl LayerAccessStats {
926 : // How many high bits to drop from a u32 timestamp?
927 : // - Only storing up to a u32 timestamp will work fine until 2038 (if this code is still in use
928 : // after that, this software has been very successful!)
929 : // - Dropping the top bit is implicitly safe because unix timestamps are meant to be
930 : // stored in an i32, so they never used it.
931 : // - Dropping the next two bits is safe because this code is only running on systems in
932 : // years >= 2024, and these bits have been 1 since 2021
933 : //
934 : // Therefore we may store only 28 bits for a timestamp with one second resolution. We do
935 : // this truncation to make space for some flags in the high bits of our u64.
936 : const TS_DROP_HIGH_BITS: u32 = u32::count_ones(Self::TS_ONES) + 1;
937 : const TS_MASK: u32 = 0x1f_ff_ff_ff;
938 : const TS_ONES: u32 = 0x60_00_00_00;
939 :
940 : const ATIME_SHIFT: u32 = 0;
941 : const RTIME_SHIFT: u32 = 32 - Self::TS_DROP_HIGH_BITS;
942 : const VISIBILITY_SHIFT: u32 = 64 - 2 * Self::TS_DROP_HIGH_BITS;
943 :
944 480654 : fn write_bits(&self, mask: u64, value: u64) -> u64 {
945 480654 : self.0
946 480654 : .fetch_update(
947 480654 : // TODO: decide what orderings are correct
948 480654 : std::sync::atomic::Ordering::Relaxed,
949 480654 : std::sync::atomic::Ordering::Relaxed,
950 480654 : |v| Some((v & !mask) | (value & mask)),
951 480654 : )
952 480654 : .expect("Inner function is infallible")
953 480654 : }
954 :
955 483774 : fn to_low_res_timestamp(shift: u32, time: SystemTime) -> (u64, u64) {
956 483774 : // Drop the low three bits of the timestamp, for an ~8s accuracy
957 483774 : let timestamp = time.duration_since(UNIX_EPOCH).unwrap().as_secs() & (Self::TS_MASK as u64);
958 483774 :
959 483774 : ((Self::TS_MASK as u64) << shift, timestamp << shift)
960 483774 : }
961 :
962 292 : fn read_low_res_timestamp(&self, shift: u32) -> Option<SystemTime> {
963 292 : let read = self.0.load(std::sync::atomic::Ordering::Relaxed);
964 292 :
965 292 : let ts_bits = (read & ((Self::TS_MASK as u64) << shift)) >> shift;
966 292 : if ts_bits == 0 {
967 132 : None
968 : } else {
969 160 : Some(UNIX_EPOCH + Duration::from_secs(ts_bits | (Self::TS_ONES as u64)))
970 : }
971 292 : }
972 :
973 : /// Record a change in layer residency.
974 : ///
975 : /// Recording the event must happen while holding the layer map lock to
976 : /// ensure that latest-activity-threshold-based layer eviction (eviction_task.rs)
977 : /// can do an "imitate access" to this layer, before it observes `now-latest_activity() > threshold`.
978 : ///
979 : /// If we instead recorded the residence event with a timestamp from before grabbing the layer map lock,
980 : /// the following race could happen:
981 : ///
982 : /// - Compact: Write out an L1 layer from several L0 layers. This records residence event LayerCreate with the current timestamp.
983 : /// - Eviction: imitate access logical size calculation. This accesses the L0 layers because the L1 layer is not yet in the layer map.
984 : /// - Compact: Grab layer map lock, add the new L1 to layer map and remove the L0s, release layer map lock.
985 : /// - Eviction: observes the new L1 layer whose only activity timestamp is the LayerCreate event.
986 100 : pub(crate) fn record_residence_event_at(&self, now: SystemTime) {
987 100 : let (mask, value) = Self::to_low_res_timestamp(Self::RTIME_SHIFT, now);
988 100 : self.write_bits(mask, value);
989 100 : }
990 :
991 96 : pub(crate) fn record_residence_event(&self) {
992 96 : self.record_residence_event_at(SystemTime::now())
993 96 : }
994 :
995 479810 : fn record_access_at(&self, now: SystemTime) -> bool {
996 479810 : let (mut mask, mut value) = Self::to_low_res_timestamp(Self::ATIME_SHIFT, now);
997 479810 :
998 479810 : // A layer which is accessed must be visible.
999 479810 : mask |= 0x1 << Self::VISIBILITY_SHIFT;
1000 479810 : value |= 0x1 << Self::VISIBILITY_SHIFT;
1001 479810 :
1002 479810 : let old_bits = self.write_bits(mask, value);
1003 4 : !matches!(
1004 479810 : self.decode_visibility(old_bits),
1005 : LayerVisibilityHint::Visible
1006 : )
1007 479810 : }
1008 :
1009 : /// Returns true if we modified the layer's visibility to set it to Visible implicitly
1010 : /// as a result of this access
1011 480370 : pub(crate) fn record_access(&self, ctx: &RequestContext) -> bool {
1012 480370 : if ctx.access_stats_behavior() == AccessStatsBehavior::Skip {
1013 572 : return false;
1014 479798 : }
1015 479798 :
1016 479798 : self.record_access_at(SystemTime::now())
1017 480370 : }
1018 :
1019 0 : fn as_api_model(
1020 0 : &self,
1021 0 : reset: LayerAccessStatsReset,
1022 0 : ) -> pageserver_api::models::LayerAccessStats {
1023 0 : let ret = pageserver_api::models::LayerAccessStats {
1024 0 : access_time: self
1025 0 : .read_low_res_timestamp(Self::ATIME_SHIFT)
1026 0 : .unwrap_or(UNIX_EPOCH),
1027 0 : residence_time: self
1028 0 : .read_low_res_timestamp(Self::RTIME_SHIFT)
1029 0 : .unwrap_or(UNIX_EPOCH),
1030 0 : visible: matches!(self.visibility(), LayerVisibilityHint::Visible),
1031 : };
1032 0 : match reset {
1033 0 : LayerAccessStatsReset::NoReset => {}
1034 0 : LayerAccessStatsReset::AllStats => {
1035 0 : self.write_bits((Self::TS_MASK as u64) << Self::ATIME_SHIFT, 0x0);
1036 0 : self.write_bits((Self::TS_MASK as u64) << Self::RTIME_SHIFT, 0x0);
1037 0 : }
1038 : }
1039 0 : ret
1040 0 : }
1041 :
1042 : /// Get the latest access timestamp, falling back to latest residence event. The latest residence event
1043 : /// will be this Layer's construction time, if its residence hasn't changed since then.
1044 84 : pub(crate) fn latest_activity(&self) -> SystemTime {
1045 84 : if let Some(t) = self.read_low_res_timestamp(Self::ATIME_SHIFT) {
1046 12 : t
1047 : } else {
1048 72 : self.read_low_res_timestamp(Self::RTIME_SHIFT)
1049 72 : .expect("Residence time is set on construction")
1050 : }
1051 84 : }
1052 :
1053 : /// Whether this layer has been accessed (excluding in [`AccessStatsBehavior::Skip`]).
1054 : ///
1055 : /// This indicates whether the layer has been used for some purpose that would motivate
1056 : /// us to keep it on disk, such as for serving a getpage request.
1057 68 : fn accessed(&self) -> bool {
1058 68 : // Consider it accessed if the most recent access is more recent than
1059 68 : // the most recent change in residence status.
1060 68 : match (
1061 68 : self.read_low_res_timestamp(Self::ATIME_SHIFT),
1062 68 : self.read_low_res_timestamp(Self::RTIME_SHIFT),
1063 : ) {
1064 60 : (None, _) => false,
1065 0 : (Some(_), None) => true,
1066 8 : (Some(a), Some(r)) => a >= r,
1067 : }
1068 68 : }
1069 :
1070 : /// Helper for extracting the visibility hint from the literal value of our inner u64
1071 482148 : fn decode_visibility(&self, bits: u64) -> LayerVisibilityHint {
1072 482148 : match (bits >> Self::VISIBILITY_SHIFT) & 0x1 {
1073 482100 : 1 => LayerVisibilityHint::Visible,
1074 48 : 0 => LayerVisibilityHint::Covered,
1075 0 : _ => unreachable!(),
1076 : }
1077 482148 : }
1078 :
1079 : /// Returns the old value which has been replaced
1080 744 : pub(crate) fn set_visibility(&self, visibility: LayerVisibilityHint) -> LayerVisibilityHint {
1081 744 : let value = match visibility {
1082 640 : LayerVisibilityHint::Visible => 0x1 << Self::VISIBILITY_SHIFT,
1083 104 : LayerVisibilityHint::Covered => 0x0,
1084 : };
1085 :
1086 744 : let old_bits = self.write_bits(0x1 << Self::VISIBILITY_SHIFT, value);
1087 744 : self.decode_visibility(old_bits)
1088 744 : }
1089 :
1090 1594 : pub(crate) fn visibility(&self) -> LayerVisibilityHint {
1091 1594 : let read = self.0.load(std::sync::atomic::Ordering::Relaxed);
1092 1594 : self.decode_visibility(read)
1093 1594 : }
1094 : }
1095 :
1096 : /// Get a layer descriptor from a layer.
1097 : pub(crate) trait AsLayerDesc {
1098 : /// Get the layer descriptor.
1099 : fn layer_desc(&self) -> &PersistentLayerDesc;
1100 : }
1101 :
1102 : pub mod tests {
1103 : use pageserver_api::shard::TenantShardId;
1104 : use utils::id::TimelineId;
1105 :
1106 : use super::*;
1107 :
1108 : impl From<DeltaLayerName> for PersistentLayerDesc {
1109 44 : fn from(value: DeltaLayerName) -> Self {
1110 44 : PersistentLayerDesc::new_delta(
1111 44 : TenantShardId::from([0; 18]),
1112 44 : TimelineId::from_array([0; 16]),
1113 44 : value.key_range,
1114 44 : value.lsn_range,
1115 44 : 233,
1116 44 : )
1117 44 : }
1118 : }
1119 :
1120 : impl From<ImageLayerName> for PersistentLayerDesc {
1121 48 : fn from(value: ImageLayerName) -> Self {
1122 48 : PersistentLayerDesc::new_img(
1123 48 : TenantShardId::from([0; 18]),
1124 48 : TimelineId::from_array([0; 16]),
1125 48 : value.key_range,
1126 48 : value.lsn,
1127 48 : 233,
1128 48 : )
1129 48 : }
1130 : }
1131 :
1132 : impl From<LayerName> for PersistentLayerDesc {
1133 92 : fn from(value: LayerName) -> Self {
1134 92 : match value {
1135 44 : LayerName::Delta(d) => Self::from(d),
1136 48 : LayerName::Image(i) => Self::from(i),
1137 : }
1138 92 : }
1139 : }
1140 : }
1141 :
1142 : /// Range wrapping newtype, which uses display to render Debug.
1143 : ///
1144 : /// Useful with `Key`, which has too verbose `{:?}` for printing multiple layers.
1145 : struct RangeDisplayDebug<'a, T: std::fmt::Display>(&'a Range<T>);
1146 :
1147 : impl<T: std::fmt::Display> std::fmt::Debug for RangeDisplayDebug<'_, T> {
1148 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
1149 0 : write!(f, "{}..{}", self.0.start, self.0.end)
1150 0 : }
1151 : }
1152 :
1153 : #[cfg(test)]
1154 : mod tests2 {
1155 : use pageserver_api::key::DBDIR_KEY;
1156 : use tracing::info;
1157 :
1158 : use super::*;
1159 : use crate::tenant::storage_layer::IoConcurrency;
1160 :
1161 : /// TODO: currently this test relies on manual visual inspection of the --no-capture output.
1162 : /// Should look like so:
1163 : /// ```text
1164 : /// RUST_LOG=trace cargo nextest run --features testing --no-capture test_io_concurrency_noise
1165 : /// running 1 test
1166 : /// 2025-01-21T17:42:01.335679Z INFO get_vectored_concurrent_io test selected=SidecarTask
1167 : /// 2025-01-21T17:42:01.335680Z TRACE spawning sidecar task task_id=0
1168 : /// 2025-01-21T17:42:01.335937Z TRACE IoConcurrency_sidecar{task_id=0}: start
1169 : /// 2025-01-21T17:42:01.335972Z TRACE IoConcurrency_sidecar{task_id=0}: received new io future
1170 : /// 2025-01-21T17:42:01.335999Z INFO IoConcurrency_sidecar{task_id=0}: waiting for signal to complete IO
1171 : /// 2025-01-21T17:42:01.336229Z WARN dropping ValuesReconstructState while some IOs have not been completed num_active_ios=1 sidecar_task_id=Some(0) backtrace= 0: <pageserver::tenant::storage_layer::ValuesReconstructState as core::ops::drop::Drop>::drop
1172 : /// at ./src/tenant/storage_layer.rs:553:24
1173 : /// 1: core::ptr::drop_in_place<pageserver::tenant::storage_layer::ValuesReconstructState>
1174 : /// at /home/christian/.rustup/toolchains/1.84.0-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:521:1
1175 : /// 2: core::mem::drop
1176 : /// at /home/christian/.rustup/toolchains/1.84.0-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/mem/mod.rs:942:24
1177 : /// 3: pageserver::tenant::storage_layer::tests2::test_io_concurrency_noise::{{closure}}
1178 : /// at ./src/tenant/storage_layer.rs:1159:9
1179 : /// ...
1180 : /// 49: <unknown>
1181 : /// 2025-01-21T17:42:01.452293Z INFO IoConcurrency_sidecar{task_id=0}: completing IO
1182 : /// 2025-01-21T17:42:01.452357Z TRACE IoConcurrency_sidecar{task_id=0}: io future completed
1183 : /// 2025-01-21T17:42:01.452473Z TRACE IoConcurrency_sidecar{task_id=0}: end
1184 : /// test tenant::storage_layer::tests2::test_io_concurrency_noise ... ok
1185 : ///
1186 : /// ```
1187 : #[tokio::test]
1188 4 : async fn test_io_concurrency_noise() {
1189 4 : crate::tenant::harness::setup_logging();
1190 4 :
1191 4 : let io_concurrency = IoConcurrency::spawn_for_test();
1192 4 : match *io_concurrency {
1193 4 : IoConcurrency::Sequential => {
1194 4 : // This test asserts behavior in sidecar mode, doesn't make sense in sequential mode.
1195 4 : return;
1196 4 : }
1197 4 : IoConcurrency::SidecarTask { .. } => {}
1198 2 : }
1199 2 : let mut reconstruct_state = ValuesReconstructState::new(io_concurrency.clone());
1200 2 :
1201 2 : let (io_fut_is_waiting_tx, io_fut_is_waiting) = tokio::sync::oneshot::channel();
1202 2 : let (do_complete_io, should_complete_io) = tokio::sync::oneshot::channel();
1203 2 : let (io_fut_exiting_tx, io_fut_exiting) = tokio::sync::oneshot::channel();
1204 2 :
1205 2 : let io = reconstruct_state.update_key(&DBDIR_KEY, Lsn(8), true);
1206 2 : reconstruct_state
1207 2 : .spawn_io(async move {
1208 2 : info!("waiting for signal to complete IO");
1209 4 : io_fut_is_waiting_tx.send(()).unwrap();
1210 2 : should_complete_io.await.unwrap();
1211 2 : info!("completing IO");
1212 4 : io.complete(Ok(OnDiskValue::RawImage(Bytes::new())));
1213 2 : io_fut_exiting_tx.send(()).unwrap();
1214 2 : })
1215 2 : .await;
1216 4 :
1217 4 : io_fut_is_waiting.await.unwrap();
1218 2 :
1219 2 : // this is what makes the noise
1220 2 : drop(reconstruct_state);
1221 2 :
1222 2 : do_complete_io.send(()).unwrap();
1223 2 :
1224 2 : io_fut_exiting.await.unwrap();
1225 4 : }
1226 : }
|
