Line data Source code
1 : //!
2 : //! This provides an abstraction to store PostgreSQL relations and other files
3 : //! in the key-value store that implements the Repository interface.
4 : //!
5 : //! (TODO: The line between PUT-functions here and walingest.rs is a bit blurry, as
6 : //! walingest.rs handles a few things like implicit relation creation and extension.
7 : //! Clarify that)
8 : //!
9 : use std::collections::{HashMap, HashSet, hash_map};
10 : use std::ops::{ControlFlow, Range};
11 :
12 : use crate::walingest::{WalIngestError, WalIngestErrorKind};
13 : use crate::{PERF_TRACE_TARGET, ensure_walingest};
14 : use anyhow::Context;
15 : use bytes::{Buf, Bytes, BytesMut};
16 : use enum_map::Enum;
17 : use pageserver_api::key::{
18 : AUX_FILES_KEY, CHECKPOINT_KEY, CONTROLFILE_KEY, CompactKey, DBDIR_KEY, Key, RelDirExists,
19 : TWOPHASEDIR_KEY, dbdir_key_range, rel_block_to_key, rel_dir_to_key, rel_key_range,
20 : rel_size_to_key, rel_tag_sparse_key, rel_tag_sparse_key_range, relmap_file_key,
21 : repl_origin_key, repl_origin_key_range, slru_block_to_key, slru_dir_to_key,
22 : slru_segment_key_range, slru_segment_size_to_key, twophase_file_key, twophase_key_range,
23 : };
24 : use pageserver_api::keyspace::{KeySpaceRandomAccum, SparseKeySpace};
25 : use pageserver_api::models::RelSizeMigration;
26 : use pageserver_api::record::NeonWalRecord;
27 : use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
28 : use pageserver_api::shard::ShardIdentity;
29 : use pageserver_api::value::Value;
30 : use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
31 : use postgres_ffi::{BLCKSZ, Oid, RepOriginId, TimestampTz, TransactionId};
32 : use serde::{Deserialize, Serialize};
33 : use strum::IntoEnumIterator;
34 : use tokio_util::sync::CancellationToken;
35 : use tracing::{debug, info, info_span, trace, warn};
36 : use utils::bin_ser::{BeSer, DeserializeError};
37 : use utils::lsn::Lsn;
38 : use utils::pausable_failpoint;
39 : use wal_decoder::serialized_batch::{SerializedValueBatch, ValueMeta};
40 :
41 : use super::tenant::{PageReconstructError, Timeline};
42 : use crate::aux_file;
43 : use crate::context::{PerfInstrumentFutureExt, RequestContext, RequestContextBuilder};
44 : use crate::keyspace::{KeySpace, KeySpaceAccum};
45 : use crate::metrics::{
46 : RELSIZE_CACHE_MISSES_OLD, RELSIZE_LATEST_CACHE_ENTRIES, RELSIZE_LATEST_CACHE_HITS,
47 : RELSIZE_LATEST_CACHE_MISSES, RELSIZE_SNAPSHOT_CACHE_ENTRIES, RELSIZE_SNAPSHOT_CACHE_HITS,
48 : RELSIZE_SNAPSHOT_CACHE_MISSES,
49 : };
50 : use crate::span::{
51 : debug_assert_current_span_has_tenant_and_timeline_id,
52 : debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id,
53 : };
54 : use crate::tenant::storage_layer::IoConcurrency;
55 : use crate::tenant::timeline::{GetVectoredError, VersionedKeySpaceQuery};
56 :
57 : /// Max delta records appended to the AUX_FILES_KEY (for aux v1). The write path will write a full image once this threshold is reached.
58 : pub const MAX_AUX_FILE_DELTAS: usize = 1024;
59 :
60 : /// Max number of aux-file-related delta layers. The compaction will create a new image layer once this threshold is reached.
61 : pub const MAX_AUX_FILE_V2_DELTAS: usize = 16;
62 :
63 : #[derive(Debug)]
64 : pub enum LsnForTimestamp {
65 : /// Found commits both before and after the given timestamp
66 : Present(Lsn),
67 :
68 : /// Found no commits after the given timestamp, this means
69 : /// that the newest data in the branch is older than the given
70 : /// timestamp.
71 : ///
72 : /// All commits <= LSN happened before the given timestamp
73 : Future(Lsn),
74 :
75 : /// The queried timestamp is past our horizon we look back at (PITR)
76 : ///
77 : /// All commits > LSN happened after the given timestamp,
78 : /// but any commits < LSN might have happened before or after
79 : /// the given timestamp. We don't know because no data before
80 : /// the given lsn is available.
81 : Past(Lsn),
82 :
83 : /// We have found no commit with a timestamp,
84 : /// so we can't return anything meaningful.
85 : ///
86 : /// The associated LSN is the lower bound value we can safely
87 : /// create branches on, but no statement is made if it is
88 : /// older or newer than the timestamp.
89 : ///
90 : /// This variant can e.g. be returned right after a
91 : /// cluster import.
92 : NoData(Lsn),
93 : }
94 :
95 : /// Each request to page server contains LSN range: `not_modified_since..request_lsn`.
96 : /// See comments libs/pageserver_api/src/models.rs.
97 : /// Based on this range and `last_record_lsn` PS calculates `effective_lsn`.
98 : /// But to distinguish requests from primary and replicas we need also to pass `request_lsn`.
99 : #[derive(Debug, Clone, Copy, Default)]
100 : pub struct LsnRange {
101 : pub effective_lsn: Lsn,
102 : pub request_lsn: Lsn,
103 : }
104 :
105 : impl LsnRange {
106 0 : pub fn at(lsn: Lsn) -> LsnRange {
107 0 : LsnRange {
108 0 : effective_lsn: lsn,
109 0 : request_lsn: lsn,
110 0 : }
111 0 : }
112 16 : pub fn is_latest(&self) -> bool {
113 16 : self.request_lsn == Lsn::MAX
114 16 : }
115 : }
116 :
117 : #[derive(Debug, thiserror::Error)]
118 : pub(crate) enum CalculateLogicalSizeError {
119 : #[error("cancelled")]
120 : Cancelled,
121 :
122 : /// Something went wrong while reading the metadata we use to calculate logical size
123 : /// Note that cancellation variants of `PageReconstructError` are transformed to [`Self::Cancelled`]
124 : /// in the `From` implementation for this variant.
125 : #[error(transparent)]
126 : PageRead(PageReconstructError),
127 :
128 : /// Something went wrong deserializing metadata that we read to calculate logical size
129 : #[error("decode error: {0}")]
130 : Decode(#[from] DeserializeError),
131 : }
132 :
133 : #[derive(Debug, thiserror::Error)]
134 : pub(crate) enum CollectKeySpaceError {
135 : #[error(transparent)]
136 : Decode(#[from] DeserializeError),
137 : #[error(transparent)]
138 : PageRead(PageReconstructError),
139 : #[error("cancelled")]
140 : Cancelled,
141 : }
142 :
143 : impl From<PageReconstructError> for CollectKeySpaceError {
144 0 : fn from(err: PageReconstructError) -> Self {
145 0 : match err {
146 0 : PageReconstructError::Cancelled => Self::Cancelled,
147 0 : err => Self::PageRead(err),
148 : }
149 0 : }
150 : }
151 :
152 : impl From<PageReconstructError> for CalculateLogicalSizeError {
153 0 : fn from(pre: PageReconstructError) -> Self {
154 0 : match pre {
155 0 : PageReconstructError::Cancelled => Self::Cancelled,
156 0 : _ => Self::PageRead(pre),
157 : }
158 0 : }
159 : }
160 :
161 : #[derive(Debug, thiserror::Error)]
162 : pub enum RelationError {
163 : #[error("invalid relnode")]
164 : InvalidRelnode,
165 : }
166 :
167 : ///
168 : /// This impl provides all the functionality to store PostgreSQL relations, SLRUs,
169 : /// and other special kinds of files, in a versioned key-value store. The
170 : /// Timeline struct provides the key-value store.
171 : ///
172 : /// This is a separate impl, so that we can easily include all these functions in a Timeline
173 : /// implementation, and might be moved into a separate struct later.
174 : impl Timeline {
175 : /// Start ingesting a WAL record, or other atomic modification of
176 : /// the timeline.
177 : ///
178 : /// This provides a transaction-like interface to perform a bunch
179 : /// of modifications atomically.
180 : ///
181 : /// To ingest a WAL record, call begin_modification(lsn) to get a
182 : /// DatadirModification object. Use the functions in the object to
183 : /// modify the repository state, updating all the pages and metadata
184 : /// that the WAL record affects. When you're done, call commit() to
185 : /// commit the changes.
186 : ///
187 : /// Lsn stored in modification is advanced by `ingest_record` and
188 : /// is used by `commit()` to update `last_record_lsn`.
189 : ///
190 : /// Calling commit() will flush all the changes and reset the state,
191 : /// so the `DatadirModification` struct can be reused to perform the next modification.
192 : ///
193 : /// Note that any pending modifications you make through the
194 : /// modification object won't be visible to calls to the 'get' and list
195 : /// functions of the timeline until you finish! And if you update the
196 : /// same page twice, the last update wins.
197 : ///
198 134214 : pub fn begin_modification(&self, lsn: Lsn) -> DatadirModification
199 134214 : where
200 134214 : Self: Sized,
201 134214 : {
202 134214 : DatadirModification {
203 134214 : tline: self,
204 134214 : pending_lsns: Vec::new(),
205 134214 : pending_metadata_pages: HashMap::new(),
206 134214 : pending_data_batch: None,
207 134214 : pending_deletions: Vec::new(),
208 134214 : pending_nblocks: 0,
209 134214 : pending_directory_entries: Vec::new(),
210 134214 : pending_metadata_bytes: 0,
211 134214 : lsn,
212 134214 : }
213 134214 : }
214 :
215 : //------------------------------------------------------------------------------
216 : // Public GET functions
217 : //------------------------------------------------------------------------------
218 :
219 : /// Look up given page version.
220 9192 : pub(crate) async fn get_rel_page_at_lsn(
221 9192 : &self,
222 9192 : tag: RelTag,
223 9192 : blknum: BlockNumber,
224 9192 : version: Version<'_>,
225 9192 : ctx: &RequestContext,
226 9192 : io_concurrency: IoConcurrency,
227 9192 : ) -> Result<Bytes, PageReconstructError> {
228 9192 : match version {
229 9192 : Version::LsnRange(lsns) => {
230 9192 : let pages: smallvec::SmallVec<[_; 1]> = smallvec::smallvec![(tag, blknum)];
231 9192 : let res = self
232 9192 : .get_rel_page_at_lsn_batched(
233 9192 : pages
234 9192 : .iter()
235 9192 : .map(|(tag, blknum)| (tag, blknum, lsns, ctx.attached_child())),
236 9192 : io_concurrency.clone(),
237 9192 : ctx,
238 9192 : )
239 9192 : .await;
240 9192 : assert_eq!(res.len(), 1);
241 9192 : res.into_iter().next().unwrap()
242 : }
243 0 : Version::Modified(modification) => {
244 0 : if tag.relnode == 0 {
245 0 : return Err(PageReconstructError::Other(
246 0 : RelationError::InvalidRelnode.into(),
247 0 : ));
248 0 : }
249 :
250 0 : let nblocks = self.get_rel_size(tag, version, ctx).await?;
251 0 : if blknum >= nblocks {
252 0 : debug!(
253 0 : "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
254 0 : tag,
255 0 : blknum,
256 0 : version.get_lsn(),
257 : nblocks
258 : );
259 0 : return Ok(ZERO_PAGE.clone());
260 0 : }
261 0 :
262 0 : let key = rel_block_to_key(tag, blknum);
263 0 : modification.get(key, ctx).await
264 : }
265 : }
266 9192 : }
267 :
268 : /// Like [`Self::get_rel_page_at_lsn`], but returns a batch of pages.
269 : ///
270 : /// The ordering of the returned vec corresponds to the ordering of `pages`.
271 9192 : pub(crate) async fn get_rel_page_at_lsn_batched(
272 9192 : &self,
273 9192 : pages: impl ExactSizeIterator<Item = (&RelTag, &BlockNumber, LsnRange, RequestContext)>,
274 9192 : io_concurrency: IoConcurrency,
275 9192 : ctx: &RequestContext,
276 9192 : ) -> Vec<Result<Bytes, PageReconstructError>> {
277 9192 : debug_assert_current_span_has_tenant_and_timeline_id();
278 9192 :
279 9192 : let mut slots_filled = 0;
280 9192 : let page_count = pages.len();
281 9192 :
282 9192 : // Would be nice to use smallvec here but it doesn't provide the spare_capacity_mut() API.
283 9192 : let mut result = Vec::with_capacity(pages.len());
284 9192 : let result_slots = result.spare_capacity_mut();
285 9192 :
286 9192 : let mut keys_slots: HashMap<Key, smallvec::SmallVec<[(usize, RequestContext); 1]>> =
287 9192 : HashMap::with_capacity(pages.len());
288 9192 :
289 9192 : let mut req_keyspaces: HashMap<Lsn, KeySpaceRandomAccum> =
290 9192 : HashMap::with_capacity(pages.len());
291 :
292 9192 : for (response_slot_idx, (tag, blknum, lsns, ctx)) in pages.enumerate() {
293 9192 : if tag.relnode == 0 {
294 0 : result_slots[response_slot_idx].write(Err(PageReconstructError::Other(
295 0 : RelationError::InvalidRelnode.into(),
296 0 : )));
297 0 :
298 0 : slots_filled += 1;
299 0 : continue;
300 9192 : }
301 9192 : let lsn = lsns.effective_lsn;
302 9192 : let nblocks = {
303 9192 : let ctx = RequestContextBuilder::from(&ctx)
304 9192 : .perf_span(|crnt_perf_span| {
305 0 : info_span!(
306 : target: PERF_TRACE_TARGET,
307 0 : parent: crnt_perf_span,
308 : "GET_REL_SIZE",
309 : reltag=%tag,
310 : lsn=%lsn,
311 : )
312 9192 : })
313 9192 : .attached_child();
314 9192 :
315 9192 : match self
316 9192 : .get_rel_size(*tag, Version::LsnRange(lsns), &ctx)
317 9192 : .maybe_perf_instrument(&ctx, |crnt_perf_span| crnt_perf_span.clone())
318 9192 : .await
319 : {
320 9192 : Ok(nblocks) => nblocks,
321 0 : Err(err) => {
322 0 : result_slots[response_slot_idx].write(Err(err));
323 0 : slots_filled += 1;
324 0 : continue;
325 : }
326 : }
327 : };
328 :
329 9192 : if *blknum >= nblocks {
330 0 : debug!(
331 0 : "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
332 : tag, blknum, lsn, nblocks
333 : );
334 0 : result_slots[response_slot_idx].write(Ok(ZERO_PAGE.clone()));
335 0 : slots_filled += 1;
336 0 : continue;
337 9192 : }
338 9192 :
339 9192 : let key = rel_block_to_key(*tag, *blknum);
340 9192 :
341 9192 : let ctx = RequestContextBuilder::from(&ctx)
342 9192 : .perf_span(|crnt_perf_span| {
343 0 : info_span!(
344 : target: PERF_TRACE_TARGET,
345 0 : parent: crnt_perf_span,
346 : "GET_BATCH",
347 : batch_size = %page_count,
348 : )
349 9192 : })
350 9192 : .attached_child();
351 9192 :
352 9192 : let key_slots = keys_slots.entry(key).or_default();
353 9192 : key_slots.push((response_slot_idx, ctx));
354 9192 :
355 9192 : let acc = req_keyspaces.entry(lsn).or_default();
356 9192 : acc.add_key(key);
357 9192 : }
358 :
359 9192 : let query: Vec<(Lsn, KeySpace)> = req_keyspaces
360 9192 : .into_iter()
361 9192 : .map(|(lsn, acc)| (lsn, acc.to_keyspace()))
362 9192 : .collect();
363 9192 :
364 9192 : let query = VersionedKeySpaceQuery::scattered(query);
365 9192 : let res = self
366 9192 : .get_vectored(query, io_concurrency, ctx)
367 9192 : .maybe_perf_instrument(ctx, |current_perf_span| current_perf_span.clone())
368 9192 : .await;
369 :
370 9192 : match res {
371 9192 : Ok(results) => {
372 18384 : for (key, res) in results {
373 9192 : let mut key_slots = keys_slots.remove(&key).unwrap().into_iter();
374 9192 : let (first_slot, first_req_ctx) = key_slots.next().unwrap();
375 :
376 9192 : for (slot, req_ctx) in key_slots {
377 0 : let clone = match &res {
378 0 : Ok(buf) => Ok(buf.clone()),
379 0 : Err(err) => Err(match err {
380 0 : PageReconstructError::Cancelled => PageReconstructError::Cancelled,
381 :
382 0 : x @ PageReconstructError::Other(_)
383 0 : | x @ PageReconstructError::AncestorLsnTimeout(_)
384 0 : | x @ PageReconstructError::WalRedo(_)
385 0 : | x @ PageReconstructError::MissingKey(_) => {
386 0 : PageReconstructError::Other(anyhow::anyhow!(
387 0 : "there was more than one request for this key in the batch, error logged once: {x:?}"
388 0 : ))
389 : }
390 : }),
391 : };
392 :
393 0 : result_slots[slot].write(clone);
394 0 : // There is no standardized way to express that the batched span followed from N request spans.
395 0 : // So, abuse the system and mark the request contexts as follows_from the batch span, so we get
396 0 : // some linkage in our trace viewer. It allows us to answer: which GET_VECTORED did this GET_PAGE wait for.
397 0 : req_ctx.perf_follows_from(ctx);
398 0 : slots_filled += 1;
399 : }
400 :
401 9192 : result_slots[first_slot].write(res);
402 9192 : first_req_ctx.perf_follows_from(ctx);
403 9192 : slots_filled += 1;
404 : }
405 : }
406 0 : Err(err) => {
407 : // this cannot really happen because get_vectored only errors globally on invalid LSN or too large batch size
408 : // (We enforce the max batch size outside of this function, in the code that constructs the batch request.)
409 0 : for (slot, req_ctx) in keys_slots.values().flatten() {
410 : // this whole `match` is a lot like `From<GetVectoredError> for PageReconstructError`
411 : // but without taking ownership of the GetVectoredError
412 0 : let err = match &err {
413 0 : GetVectoredError::Cancelled => Err(PageReconstructError::Cancelled),
414 : // TODO: restructure get_vectored API to make this error per-key
415 0 : GetVectoredError::MissingKey(err) => {
416 0 : Err(PageReconstructError::Other(anyhow::anyhow!(
417 0 : "whole vectored get request failed because one or more of the requested keys were missing: {err:?}"
418 0 : )))
419 : }
420 : // TODO: restructure get_vectored API to make this error per-key
421 0 : GetVectoredError::GetReadyAncestorError(err) => {
422 0 : Err(PageReconstructError::Other(anyhow::anyhow!(
423 0 : "whole vectored get request failed because one or more key required ancestor that wasn't ready: {err:?}"
424 0 : )))
425 : }
426 : // TODO: restructure get_vectored API to make this error per-key
427 0 : GetVectoredError::Other(err) => Err(PageReconstructError::Other(
428 0 : anyhow::anyhow!("whole vectored get request failed: {err:?}"),
429 0 : )),
430 : // TODO: we can prevent this error class by moving this check into the type system
431 0 : GetVectoredError::InvalidLsn(e) => {
432 0 : Err(anyhow::anyhow!("invalid LSN: {e:?}").into())
433 : }
434 : // NB: this should never happen in practice because we limit batch size to be smaller than max_get_vectored_keys
435 : // TODO: we can prevent this error class by moving this check into the type system
436 0 : GetVectoredError::Oversized(err, max) => {
437 0 : Err(anyhow::anyhow!("batching oversized: {err} > {max}").into())
438 : }
439 : };
440 :
441 0 : req_ctx.perf_follows_from(ctx);
442 0 : result_slots[*slot].write(err);
443 : }
444 :
445 0 : slots_filled += keys_slots.values().map(|slots| slots.len()).sum::<usize>();
446 0 : }
447 : };
448 :
449 9192 : assert_eq!(slots_filled, page_count);
450 : // SAFETY:
451 : // 1. `result` and any of its uninint members are not read from until this point
452 : // 2. The length below is tracked at run-time and matches the number of requested pages.
453 9192 : unsafe {
454 9192 : result.set_len(page_count);
455 9192 : }
456 9192 :
457 9192 : result
458 9192 : }
459 :
460 : /// Get size of a database in blocks. This is only accurate on shard 0. It will undercount on
461 : /// other shards, by only accounting for relations the shard has pages for, and only accounting
462 : /// for pages up to the highest page number it has stored.
463 0 : pub(crate) async fn get_db_size(
464 0 : &self,
465 0 : spcnode: Oid,
466 0 : dbnode: Oid,
467 0 : version: Version<'_>,
468 0 : ctx: &RequestContext,
469 0 : ) -> Result<usize, PageReconstructError> {
470 0 : let mut total_blocks = 0;
471 :
472 0 : let rels = self.list_rels(spcnode, dbnode, version, ctx).await?;
473 :
474 0 : if rels.is_empty() {
475 0 : return Ok(0);
476 0 : }
477 0 :
478 0 : // Pre-deserialize the rel directory to avoid duplicated work in `get_relsize_cached`.
479 0 : let reldir_key = rel_dir_to_key(spcnode, dbnode);
480 0 : let buf = version.get(self, reldir_key, ctx).await?;
481 0 : let reldir = RelDirectory::des(&buf)?;
482 :
483 0 : for rel in rels {
484 0 : let n_blocks = self
485 0 : .get_rel_size_in_reldir(rel, version, Some((reldir_key, &reldir)), ctx)
486 0 : .await?;
487 0 : total_blocks += n_blocks as usize;
488 : }
489 0 : Ok(total_blocks)
490 0 : }
491 :
492 : /// Get size of a relation file. The relation must exist, otherwise an error is returned.
493 : ///
494 : /// This is only accurate on shard 0. On other shards, it will return the size up to the highest
495 : /// page number stored in the shard.
496 12217 : pub(crate) async fn get_rel_size(
497 12217 : &self,
498 12217 : tag: RelTag,
499 12217 : version: Version<'_>,
500 12217 : ctx: &RequestContext,
501 12217 : ) -> Result<BlockNumber, PageReconstructError> {
502 12217 : self.get_rel_size_in_reldir(tag, version, None, ctx).await
503 12217 : }
504 :
505 : /// Get size of a relation file. The relation must exist, otherwise an error is returned.
506 : ///
507 : /// See [`Self::get_rel_exists_in_reldir`] on why we need `deserialized_reldir_v1`.
508 12217 : pub(crate) async fn get_rel_size_in_reldir(
509 12217 : &self,
510 12217 : tag: RelTag,
511 12217 : version: Version<'_>,
512 12217 : deserialized_reldir_v1: Option<(Key, &RelDirectory)>,
513 12217 : ctx: &RequestContext,
514 12217 : ) -> Result<BlockNumber, PageReconstructError> {
515 12217 : if tag.relnode == 0 {
516 0 : return Err(PageReconstructError::Other(
517 0 : RelationError::InvalidRelnode.into(),
518 0 : ));
519 12217 : }
520 :
521 12217 : if let Some(nblocks) = self.get_cached_rel_size(&tag, version) {
522 12210 : return Ok(nblocks);
523 7 : }
524 7 :
525 7 : if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
526 0 : && !self
527 0 : .get_rel_exists_in_reldir(tag, version, deserialized_reldir_v1, ctx)
528 0 : .await?
529 : {
530 : // FIXME: Postgres sometimes calls smgrcreate() to create
531 : // FSM, and smgrnblocks() on it immediately afterwards,
532 : // without extending it. Tolerate that by claiming that
533 : // any non-existent FSM fork has size 0.
534 0 : return Ok(0);
535 7 : }
536 7 :
537 7 : let key = rel_size_to_key(tag);
538 7 : let mut buf = version.get(self, key, ctx).await?;
539 5 : let nblocks = buf.get_u32_le();
540 5 :
541 5 : self.update_cached_rel_size(tag, version, nblocks);
542 5 :
543 5 : Ok(nblocks)
544 12217 : }
545 :
546 : /// Does the relation exist?
547 : ///
548 : /// Only shard 0 has a full view of the relations. Other shards only know about relations that
549 : /// the shard stores pages for.
550 : ///
551 3025 : pub(crate) async fn get_rel_exists(
552 3025 : &self,
553 3025 : tag: RelTag,
554 3025 : version: Version<'_>,
555 3025 : ctx: &RequestContext,
556 3025 : ) -> Result<bool, PageReconstructError> {
557 3025 : self.get_rel_exists_in_reldir(tag, version, None, ctx).await
558 3025 : }
559 :
560 : /// Does the relation exist? With a cached deserialized `RelDirectory`.
561 : ///
562 : /// There are some cases where the caller loops across all relations. In that specific case,
563 : /// the caller should obtain the deserialized `RelDirectory` first and then call this function
564 : /// to avoid duplicated work of deserliazation. This is a hack and should be removed by introducing
565 : /// a new API (e.g., `get_rel_exists_batched`).
566 3025 : pub(crate) async fn get_rel_exists_in_reldir(
567 3025 : &self,
568 3025 : tag: RelTag,
569 3025 : version: Version<'_>,
570 3025 : deserialized_reldir_v1: Option<(Key, &RelDirectory)>,
571 3025 : ctx: &RequestContext,
572 3025 : ) -> Result<bool, PageReconstructError> {
573 3025 : if tag.relnode == 0 {
574 0 : return Err(PageReconstructError::Other(
575 0 : RelationError::InvalidRelnode.into(),
576 0 : ));
577 3025 : }
578 :
579 : // first try to lookup relation in cache
580 3025 : if let Some(_nblocks) = self.get_cached_rel_size(&tag, version) {
581 3016 : return Ok(true);
582 9 : }
583 : // then check if the database was already initialized.
584 : // get_rel_exists can be called before dbdir is created.
585 9 : let buf = version.get(self, DBDIR_KEY, ctx).await?;
586 9 : let dbdirs = DbDirectory::des(&buf)?.dbdirs;
587 9 : if !dbdirs.contains_key(&(tag.spcnode, tag.dbnode)) {
588 0 : return Ok(false);
589 9 : }
590 9 :
591 9 : // Read path: first read the new reldir keyspace. Early return if the relation exists.
592 9 : // Otherwise, read the old reldir keyspace.
593 9 : // TODO: if IndexPart::rel_size_migration is `Migrated`, we only need to read from v2.
594 9 :
595 9 : if let RelSizeMigration::Migrated | RelSizeMigration::Migrating =
596 9 : self.get_rel_size_v2_status()
597 : {
598 : // fetch directory listing (new)
599 0 : let key = rel_tag_sparse_key(tag.spcnode, tag.dbnode, tag.relnode, tag.forknum);
600 0 : let buf = RelDirExists::decode_option(version.sparse_get(self, key, ctx).await?)
601 0 : .map_err(|_| PageReconstructError::Other(anyhow::anyhow!("invalid reldir key")))?;
602 0 : let exists_v2 = buf == RelDirExists::Exists;
603 0 : // Fast path: if the relation exists in the new format, return true.
604 0 : // TODO: we should have a verification mode that checks both keyspaces
605 0 : // to ensure the relation only exists in one of them.
606 0 : if exists_v2 {
607 0 : return Ok(true);
608 0 : }
609 9 : }
610 :
611 : // fetch directory listing (old)
612 :
613 9 : let key = rel_dir_to_key(tag.spcnode, tag.dbnode);
614 :
615 9 : if let Some((cached_key, dir)) = deserialized_reldir_v1 {
616 0 : if cached_key == key {
617 0 : return Ok(dir.rels.contains(&(tag.relnode, tag.forknum)));
618 0 : } else if cfg!(test) || cfg!(feature = "testing") {
619 0 : panic!("cached reldir key mismatch: {cached_key} != {key}");
620 : } else {
621 0 : warn!("cached reldir key mismatch: {cached_key} != {key}");
622 : }
623 : // Fallback to reading the directory from the datadir.
624 9 : }
625 9 : let buf = version.get(self, key, ctx).await?;
626 :
627 9 : let dir = RelDirectory::des(&buf)?;
628 9 : let exists_v1 = dir.rels.contains(&(tag.relnode, tag.forknum));
629 9 : Ok(exists_v1)
630 3025 : }
631 :
632 : /// Get a list of all existing relations in given tablespace and database.
633 : ///
634 : /// Only shard 0 has a full view of the relations. Other shards only know about relations that
635 : /// the shard stores pages for.
636 : ///
637 : /// # Cancel-Safety
638 : ///
639 : /// This method is cancellation-safe.
640 0 : pub(crate) async fn list_rels(
641 0 : &self,
642 0 : spcnode: Oid,
643 0 : dbnode: Oid,
644 0 : version: Version<'_>,
645 0 : ctx: &RequestContext,
646 0 : ) -> Result<HashSet<RelTag>, PageReconstructError> {
647 0 : // fetch directory listing (old)
648 0 : let key = rel_dir_to_key(spcnode, dbnode);
649 0 : let buf = version.get(self, key, ctx).await?;
650 :
651 0 : let dir = RelDirectory::des(&buf)?;
652 0 : let rels_v1: HashSet<RelTag> =
653 0 : HashSet::from_iter(dir.rels.iter().map(|(relnode, forknum)| RelTag {
654 0 : spcnode,
655 0 : dbnode,
656 0 : relnode: *relnode,
657 0 : forknum: *forknum,
658 0 : }));
659 0 :
660 0 : if let RelSizeMigration::Legacy = self.get_rel_size_v2_status() {
661 0 : return Ok(rels_v1);
662 0 : }
663 0 :
664 0 : // scan directory listing (new), merge with the old results
665 0 : let key_range = rel_tag_sparse_key_range(spcnode, dbnode);
666 0 : let io_concurrency = IoConcurrency::spawn_from_conf(
667 0 : self.conf.get_vectored_concurrent_io,
668 0 : self.gate
669 0 : .enter()
670 0 : .map_err(|_| PageReconstructError::Cancelled)?,
671 : );
672 0 : let results = self
673 0 : .scan(
674 0 : KeySpace::single(key_range),
675 0 : version.get_lsn(),
676 0 : ctx,
677 0 : io_concurrency,
678 0 : )
679 0 : .await?;
680 0 : let mut rels = rels_v1;
681 0 : for (key, val) in results {
682 0 : let val = RelDirExists::decode(&val?)
683 0 : .map_err(|_| PageReconstructError::Other(anyhow::anyhow!("invalid reldir key")))?;
684 0 : assert_eq!(key.field6, 1);
685 0 : assert_eq!(key.field2, spcnode);
686 0 : assert_eq!(key.field3, dbnode);
687 0 : let tag = RelTag {
688 0 : spcnode,
689 0 : dbnode,
690 0 : relnode: key.field4,
691 0 : forknum: key.field5,
692 0 : };
693 0 : if val == RelDirExists::Removed {
694 0 : debug_assert!(!rels.contains(&tag), "removed reltag in v2");
695 0 : continue;
696 0 : }
697 0 : let did_not_contain = rels.insert(tag);
698 0 : debug_assert!(did_not_contain, "duplicate reltag in v2");
699 : }
700 0 : Ok(rels)
701 0 : }
702 :
703 : /// Get the whole SLRU segment
704 0 : pub(crate) async fn get_slru_segment(
705 0 : &self,
706 0 : kind: SlruKind,
707 0 : segno: u32,
708 0 : lsn: Lsn,
709 0 : ctx: &RequestContext,
710 0 : ) -> Result<Bytes, PageReconstructError> {
711 0 : assert!(self.tenant_shard_id.is_shard_zero());
712 0 : let n_blocks = self
713 0 : .get_slru_segment_size(kind, segno, Version::at(lsn), ctx)
714 0 : .await?;
715 :
716 0 : let keyspace = KeySpace::single(
717 0 : slru_block_to_key(kind, segno, 0)..slru_block_to_key(kind, segno, n_blocks),
718 0 : );
719 0 :
720 0 : let batches = keyspace.partition(
721 0 : self.get_shard_identity(),
722 0 : self.conf.max_get_vectored_keys.get() as u64 * BLCKSZ as u64,
723 0 : );
724 :
725 0 : let io_concurrency = IoConcurrency::spawn_from_conf(
726 0 : self.conf.get_vectored_concurrent_io,
727 0 : self.gate
728 0 : .enter()
729 0 : .map_err(|_| PageReconstructError::Cancelled)?,
730 : );
731 :
732 0 : let mut segment = BytesMut::with_capacity(n_blocks as usize * BLCKSZ as usize);
733 0 : for batch in batches.parts {
734 0 : let query = VersionedKeySpaceQuery::uniform(batch, lsn);
735 0 : let blocks = self
736 0 : .get_vectored(query, io_concurrency.clone(), ctx)
737 0 : .await?;
738 :
739 0 : for (_key, block) in blocks {
740 0 : let block = block?;
741 0 : segment.extend_from_slice(&block[..BLCKSZ as usize]);
742 : }
743 : }
744 :
745 0 : Ok(segment.freeze())
746 0 : }
747 :
748 : /// Get size of an SLRU segment
749 0 : pub(crate) async fn get_slru_segment_size(
750 0 : &self,
751 0 : kind: SlruKind,
752 0 : segno: u32,
753 0 : version: Version<'_>,
754 0 : ctx: &RequestContext,
755 0 : ) -> Result<BlockNumber, PageReconstructError> {
756 0 : assert!(self.tenant_shard_id.is_shard_zero());
757 0 : let key = slru_segment_size_to_key(kind, segno);
758 0 : let mut buf = version.get(self, key, ctx).await?;
759 0 : Ok(buf.get_u32_le())
760 0 : }
761 :
762 : /// Does the slru segment exist?
763 0 : pub(crate) async fn get_slru_segment_exists(
764 0 : &self,
765 0 : kind: SlruKind,
766 0 : segno: u32,
767 0 : version: Version<'_>,
768 0 : ctx: &RequestContext,
769 0 : ) -> Result<bool, PageReconstructError> {
770 0 : assert!(self.tenant_shard_id.is_shard_zero());
771 : // fetch directory listing
772 0 : let key = slru_dir_to_key(kind);
773 0 : let buf = version.get(self, key, ctx).await?;
774 :
775 0 : let dir = SlruSegmentDirectory::des(&buf)?;
776 0 : Ok(dir.segments.contains(&segno))
777 0 : }
778 :
779 : /// Locate LSN, such that all transactions that committed before
780 : /// 'search_timestamp' are visible, but nothing newer is.
781 : ///
782 : /// This is not exact. Commit timestamps are not guaranteed to be ordered,
783 : /// so it's not well defined which LSN you get if there were multiple commits
784 : /// "in flight" at that point in time.
785 : ///
786 0 : pub(crate) async fn find_lsn_for_timestamp(
787 0 : &self,
788 0 : search_timestamp: TimestampTz,
789 0 : cancel: &CancellationToken,
790 0 : ctx: &RequestContext,
791 0 : ) -> Result<LsnForTimestamp, PageReconstructError> {
792 0 : pausable_failpoint!("find-lsn-for-timestamp-pausable");
793 :
794 0 : let gc_cutoff_lsn_guard = self.get_applied_gc_cutoff_lsn();
795 0 : let gc_cutoff_planned = {
796 0 : let gc_info = self.gc_info.read().unwrap();
797 0 : gc_info.min_cutoff()
798 0 : };
799 0 : // Usually the planned cutoff is newer than the cutoff of the last gc run,
800 0 : // but let's be defensive.
801 0 : let gc_cutoff = gc_cutoff_planned.max(*gc_cutoff_lsn_guard);
802 0 : // We use this method to figure out the branching LSN for the new branch, but the
803 0 : // GC cutoff could be before the branching point and we cannot create a new branch
804 0 : // with LSN < `ancestor_lsn`. Thus, pick the maximum of these two to be
805 0 : // on the safe side.
806 0 : let min_lsn = std::cmp::max(gc_cutoff, self.get_ancestor_lsn());
807 0 : let max_lsn = self.get_last_record_lsn();
808 0 :
809 0 : // LSNs are always 8-byte aligned. low/mid/high represent the
810 0 : // LSN divided by 8.
811 0 : let mut low = min_lsn.0 / 8;
812 0 : let mut high = max_lsn.0 / 8 + 1;
813 0 :
814 0 : let mut found_smaller = false;
815 0 : let mut found_larger = false;
816 :
817 0 : while low < high {
818 0 : if cancel.is_cancelled() {
819 0 : return Err(PageReconstructError::Cancelled);
820 0 : }
821 0 : // cannot overflow, high and low are both smaller than u64::MAX / 2
822 0 : let mid = (high + low) / 2;
823 :
824 0 : let cmp = match self
825 0 : .is_latest_commit_timestamp_ge_than(
826 0 : search_timestamp,
827 0 : Lsn(mid * 8),
828 0 : &mut found_smaller,
829 0 : &mut found_larger,
830 0 : ctx,
831 0 : )
832 0 : .await
833 : {
834 0 : Ok(res) => res,
835 0 : Err(PageReconstructError::MissingKey(e)) => {
836 0 : warn!(
837 0 : "Missing key while find_lsn_for_timestamp. Either we might have already garbage-collected that data or the key is really missing. Last error: {:#}",
838 : e
839 : );
840 : // Return that we didn't find any requests smaller than the LSN, and logging the error.
841 0 : return Ok(LsnForTimestamp::Past(min_lsn));
842 : }
843 0 : Err(e) => return Err(e),
844 : };
845 :
846 0 : if cmp {
847 0 : high = mid;
848 0 : } else {
849 0 : low = mid + 1;
850 0 : }
851 : }
852 :
853 : // If `found_smaller == true`, `low = t + 1` where `t` is the target LSN,
854 : // so the LSN of the last commit record before or at `search_timestamp`.
855 : // Remove one from `low` to get `t`.
856 : //
857 : // FIXME: it would be better to get the LSN of the previous commit.
858 : // Otherwise, if you restore to the returned LSN, the database will
859 : // include physical changes from later commits that will be marked
860 : // as aborted, and will need to be vacuumed away.
861 0 : let commit_lsn = Lsn((low - 1) * 8);
862 0 : match (found_smaller, found_larger) {
863 : (false, false) => {
864 : // This can happen if no commit records have been processed yet, e.g.
865 : // just after importing a cluster.
866 0 : Ok(LsnForTimestamp::NoData(min_lsn))
867 : }
868 : (false, true) => {
869 : // Didn't find any commit timestamps smaller than the request
870 0 : Ok(LsnForTimestamp::Past(min_lsn))
871 : }
872 0 : (true, _) if commit_lsn < min_lsn => {
873 0 : // the search above did set found_smaller to true but it never increased the lsn.
874 0 : // Then, low is still the old min_lsn, and the subtraction above gave a value
875 0 : // below the min_lsn. We should never do that.
876 0 : Ok(LsnForTimestamp::Past(min_lsn))
877 : }
878 : (true, false) => {
879 : // Only found commits with timestamps smaller than the request.
880 : // It's still a valid case for branch creation, return it.
881 : // And `update_gc_info()` ignores LSN for a `LsnForTimestamp::Future`
882 : // case, anyway.
883 0 : Ok(LsnForTimestamp::Future(commit_lsn))
884 : }
885 0 : (true, true) => Ok(LsnForTimestamp::Present(commit_lsn)),
886 : }
887 0 : }
888 :
889 : /// Subroutine of find_lsn_for_timestamp(). Returns true, if there are any
890 : /// commits that committed after 'search_timestamp', at LSN 'probe_lsn'.
891 : ///
892 : /// Additionally, sets 'found_smaller'/'found_Larger, if encounters any commits
893 : /// with a smaller/larger timestamp.
894 : ///
895 0 : pub(crate) async fn is_latest_commit_timestamp_ge_than(
896 0 : &self,
897 0 : search_timestamp: TimestampTz,
898 0 : probe_lsn: Lsn,
899 0 : found_smaller: &mut bool,
900 0 : found_larger: &mut bool,
901 0 : ctx: &RequestContext,
902 0 : ) -> Result<bool, PageReconstructError> {
903 0 : self.map_all_timestamps(probe_lsn, ctx, |timestamp| {
904 0 : if timestamp >= search_timestamp {
905 0 : *found_larger = true;
906 0 : return ControlFlow::Break(true);
907 0 : } else {
908 0 : *found_smaller = true;
909 0 : }
910 0 : ControlFlow::Continue(())
911 0 : })
912 0 : .await
913 0 : }
914 :
915 : /// Obtain the timestamp for the given lsn.
916 : ///
917 : /// If the lsn has no timestamps (e.g. no commits), returns None.
918 0 : pub(crate) async fn get_timestamp_for_lsn(
919 0 : &self,
920 0 : probe_lsn: Lsn,
921 0 : ctx: &RequestContext,
922 0 : ) -> Result<Option<TimestampTz>, PageReconstructError> {
923 0 : let mut max: Option<TimestampTz> = None;
924 0 : self.map_all_timestamps::<()>(probe_lsn, ctx, |timestamp| {
925 0 : if let Some(max_prev) = max {
926 0 : max = Some(max_prev.max(timestamp));
927 0 : } else {
928 0 : max = Some(timestamp);
929 0 : }
930 0 : ControlFlow::Continue(())
931 0 : })
932 0 : .await?;
933 :
934 0 : Ok(max)
935 0 : }
936 :
937 : /// Runs the given function on all the timestamps for a given lsn
938 : ///
939 : /// The return value is either given by the closure, or set to the `Default`
940 : /// impl's output.
941 0 : async fn map_all_timestamps<T: Default>(
942 0 : &self,
943 0 : probe_lsn: Lsn,
944 0 : ctx: &RequestContext,
945 0 : mut f: impl FnMut(TimestampTz) -> ControlFlow<T>,
946 0 : ) -> Result<T, PageReconstructError> {
947 0 : for segno in self
948 0 : .list_slru_segments(SlruKind::Clog, Version::at(probe_lsn), ctx)
949 0 : .await?
950 : {
951 0 : let nblocks = self
952 0 : .get_slru_segment_size(SlruKind::Clog, segno, Version::at(probe_lsn), ctx)
953 0 : .await?;
954 :
955 0 : let keyspace = KeySpace::single(
956 0 : slru_block_to_key(SlruKind::Clog, segno, 0)
957 0 : ..slru_block_to_key(SlruKind::Clog, segno, nblocks),
958 0 : );
959 0 :
960 0 : let batches = keyspace.partition(
961 0 : self.get_shard_identity(),
962 0 : self.conf.max_get_vectored_keys.get() as u64 * BLCKSZ as u64,
963 0 : );
964 :
965 0 : let io_concurrency = IoConcurrency::spawn_from_conf(
966 0 : self.conf.get_vectored_concurrent_io,
967 0 : self.gate
968 0 : .enter()
969 0 : .map_err(|_| PageReconstructError::Cancelled)?,
970 : );
971 :
972 0 : for batch in batches.parts.into_iter().rev() {
973 0 : let query = VersionedKeySpaceQuery::uniform(batch, probe_lsn);
974 0 : let blocks = self
975 0 : .get_vectored(query, io_concurrency.clone(), ctx)
976 0 : .await?;
977 :
978 0 : for (_key, clog_page) in blocks.into_iter().rev() {
979 0 : let clog_page = clog_page?;
980 :
981 0 : if clog_page.len() == BLCKSZ as usize + 8 {
982 0 : let mut timestamp_bytes = [0u8; 8];
983 0 : timestamp_bytes.copy_from_slice(&clog_page[BLCKSZ as usize..]);
984 0 : let timestamp = TimestampTz::from_be_bytes(timestamp_bytes);
985 0 :
986 0 : match f(timestamp) {
987 0 : ControlFlow::Break(b) => return Ok(b),
988 0 : ControlFlow::Continue(()) => (),
989 : }
990 0 : }
991 : }
992 : }
993 : }
994 0 : Ok(Default::default())
995 0 : }
996 :
997 0 : pub(crate) async fn get_slru_keyspace(
998 0 : &self,
999 0 : version: Version<'_>,
1000 0 : ctx: &RequestContext,
1001 0 : ) -> Result<KeySpace, PageReconstructError> {
1002 0 : let mut accum = KeySpaceAccum::new();
1003 :
1004 0 : for kind in SlruKind::iter() {
1005 0 : let mut segments: Vec<u32> = self
1006 0 : .list_slru_segments(kind, version, ctx)
1007 0 : .await?
1008 0 : .into_iter()
1009 0 : .collect();
1010 0 : segments.sort_unstable();
1011 :
1012 0 : for seg in segments {
1013 0 : let block_count = self.get_slru_segment_size(kind, seg, version, ctx).await?;
1014 :
1015 0 : accum.add_range(
1016 0 : slru_block_to_key(kind, seg, 0)..slru_block_to_key(kind, seg, block_count),
1017 0 : );
1018 : }
1019 : }
1020 :
1021 0 : Ok(accum.to_keyspace())
1022 0 : }
1023 :
1024 : /// Get a list of SLRU segments
1025 0 : pub(crate) async fn list_slru_segments(
1026 0 : &self,
1027 0 : kind: SlruKind,
1028 0 : version: Version<'_>,
1029 0 : ctx: &RequestContext,
1030 0 : ) -> Result<HashSet<u32>, PageReconstructError> {
1031 0 : // fetch directory entry
1032 0 : let key = slru_dir_to_key(kind);
1033 :
1034 0 : let buf = version.get(self, key, ctx).await?;
1035 0 : Ok(SlruSegmentDirectory::des(&buf)?.segments)
1036 0 : }
1037 :
1038 0 : pub(crate) async fn get_relmap_file(
1039 0 : &self,
1040 0 : spcnode: Oid,
1041 0 : dbnode: Oid,
1042 0 : version: Version<'_>,
1043 0 : ctx: &RequestContext,
1044 0 : ) -> Result<Bytes, PageReconstructError> {
1045 0 : let key = relmap_file_key(spcnode, dbnode);
1046 :
1047 0 : let buf = version.get(self, key, ctx).await?;
1048 0 : Ok(buf)
1049 0 : }
1050 :
1051 168 : pub(crate) async fn list_dbdirs(
1052 168 : &self,
1053 168 : lsn: Lsn,
1054 168 : ctx: &RequestContext,
1055 168 : ) -> Result<HashMap<(Oid, Oid), bool>, PageReconstructError> {
1056 : // fetch directory entry
1057 168 : let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
1058 :
1059 168 : Ok(DbDirectory::des(&buf)?.dbdirs)
1060 168 : }
1061 :
1062 0 : pub(crate) async fn get_twophase_file(
1063 0 : &self,
1064 0 : xid: u64,
1065 0 : lsn: Lsn,
1066 0 : ctx: &RequestContext,
1067 0 : ) -> Result<Bytes, PageReconstructError> {
1068 0 : let key = twophase_file_key(xid);
1069 0 : let buf = self.get(key, lsn, ctx).await?;
1070 0 : Ok(buf)
1071 0 : }
1072 :
1073 169 : pub(crate) async fn list_twophase_files(
1074 169 : &self,
1075 169 : lsn: Lsn,
1076 169 : ctx: &RequestContext,
1077 169 : ) -> Result<HashSet<u64>, PageReconstructError> {
1078 : // fetch directory entry
1079 169 : let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
1080 :
1081 169 : if self.pg_version >= 17 {
1082 156 : Ok(TwoPhaseDirectoryV17::des(&buf)?.xids)
1083 : } else {
1084 13 : Ok(TwoPhaseDirectory::des(&buf)?
1085 : .xids
1086 13 : .iter()
1087 13 : .map(|x| u64::from(*x))
1088 13 : .collect())
1089 : }
1090 169 : }
1091 :
1092 0 : pub(crate) async fn get_control_file(
1093 0 : &self,
1094 0 : lsn: Lsn,
1095 0 : ctx: &RequestContext,
1096 0 : ) -> Result<Bytes, PageReconstructError> {
1097 0 : self.get(CONTROLFILE_KEY, lsn, ctx).await
1098 0 : }
1099 :
1100 6 : pub(crate) async fn get_checkpoint(
1101 6 : &self,
1102 6 : lsn: Lsn,
1103 6 : ctx: &RequestContext,
1104 6 : ) -> Result<Bytes, PageReconstructError> {
1105 6 : self.get(CHECKPOINT_KEY, lsn, ctx).await
1106 6 : }
1107 :
1108 6 : async fn list_aux_files_v2(
1109 6 : &self,
1110 6 : lsn: Lsn,
1111 6 : ctx: &RequestContext,
1112 6 : io_concurrency: IoConcurrency,
1113 6 : ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
1114 6 : let kv = self
1115 6 : .scan(
1116 6 : KeySpace::single(Key::metadata_aux_key_range()),
1117 6 : lsn,
1118 6 : ctx,
1119 6 : io_concurrency,
1120 6 : )
1121 6 : .await?;
1122 6 : let mut result = HashMap::new();
1123 6 : let mut sz = 0;
1124 15 : for (_, v) in kv {
1125 9 : let v = v?;
1126 9 : let v = aux_file::decode_file_value_bytes(&v)
1127 9 : .context("value decode")
1128 9 : .map_err(PageReconstructError::Other)?;
1129 17 : for (fname, content) in v {
1130 8 : sz += fname.len();
1131 8 : sz += content.len();
1132 8 : result.insert(fname, content);
1133 8 : }
1134 : }
1135 6 : self.aux_file_size_estimator.on_initial(sz);
1136 6 : Ok(result)
1137 6 : }
1138 :
1139 0 : pub(crate) async fn trigger_aux_file_size_computation(
1140 0 : &self,
1141 0 : lsn: Lsn,
1142 0 : ctx: &RequestContext,
1143 0 : io_concurrency: IoConcurrency,
1144 0 : ) -> Result<(), PageReconstructError> {
1145 0 : self.list_aux_files_v2(lsn, ctx, io_concurrency).await?;
1146 0 : Ok(())
1147 0 : }
1148 :
1149 6 : pub(crate) async fn list_aux_files(
1150 6 : &self,
1151 6 : lsn: Lsn,
1152 6 : ctx: &RequestContext,
1153 6 : io_concurrency: IoConcurrency,
1154 6 : ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
1155 6 : self.list_aux_files_v2(lsn, ctx, io_concurrency).await
1156 6 : }
1157 :
1158 2 : pub(crate) async fn get_replorigins(
1159 2 : &self,
1160 2 : lsn: Lsn,
1161 2 : ctx: &RequestContext,
1162 2 : io_concurrency: IoConcurrency,
1163 2 : ) -> Result<HashMap<RepOriginId, Lsn>, PageReconstructError> {
1164 2 : let kv = self
1165 2 : .scan(
1166 2 : KeySpace::single(repl_origin_key_range()),
1167 2 : lsn,
1168 2 : ctx,
1169 2 : io_concurrency,
1170 2 : )
1171 2 : .await?;
1172 2 : let mut result = HashMap::new();
1173 6 : for (k, v) in kv {
1174 5 : let v = v?;
1175 5 : if v.is_empty() {
1176 : // This is a tombstone -- we can skip it.
1177 : // Originally, the replorigin code uses `Lsn::INVALID` to represent a tombstone. However, as it part of
1178 : // the sparse keyspace and the sparse keyspace uses an empty image to universally represent a tombstone,
1179 : // we also need to consider that. Such tombstones might be written on the detach ancestor code path to
1180 : // avoid the value going into the child branch. (See [`crate::tenant::timeline::detach_ancestor::generate_tombstone_image_layer`] for more details.)
1181 2 : continue;
1182 3 : }
1183 3 : let origin_id = k.field6 as RepOriginId;
1184 3 : let origin_lsn = Lsn::des(&v)
1185 3 : .with_context(|| format!("decode replorigin value for {}: {v:?}", origin_id))?;
1186 2 : if origin_lsn != Lsn::INVALID {
1187 2 : result.insert(origin_id, origin_lsn);
1188 2 : }
1189 : }
1190 1 : Ok(result)
1191 2 : }
1192 :
1193 : /// Does the same as get_current_logical_size but counted on demand.
1194 : /// Used to initialize the logical size tracking on startup.
1195 : ///
1196 : /// Only relation blocks are counted currently. That excludes metadata,
1197 : /// SLRUs, twophase files etc.
1198 : ///
1199 : /// # Cancel-Safety
1200 : ///
1201 : /// This method is cancellation-safe.
1202 7 : pub(crate) async fn get_current_logical_size_non_incremental(
1203 7 : &self,
1204 7 : lsn: Lsn,
1205 7 : ctx: &RequestContext,
1206 7 : ) -> Result<u64, CalculateLogicalSizeError> {
1207 7 : debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id();
1208 7 :
1209 7 : fail::fail_point!("skip-logical-size-calculation", |_| { Ok(0) });
1210 :
1211 : // Fetch list of database dirs and iterate them
1212 7 : let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
1213 7 : let dbdir = DbDirectory::des(&buf)?;
1214 :
1215 7 : let mut total_size: u64 = 0;
1216 7 : for (spcnode, dbnode) in dbdir.dbdirs.keys() {
1217 0 : for rel in self
1218 0 : .list_rels(*spcnode, *dbnode, Version::at(lsn), ctx)
1219 0 : .await?
1220 : {
1221 0 : if self.cancel.is_cancelled() {
1222 0 : return Err(CalculateLogicalSizeError::Cancelled);
1223 0 : }
1224 0 : let relsize_key = rel_size_to_key(rel);
1225 0 : let mut buf = self.get(relsize_key, lsn, ctx).await?;
1226 0 : let relsize = buf.get_u32_le();
1227 0 :
1228 0 : total_size += relsize as u64;
1229 : }
1230 : }
1231 7 : Ok(total_size * BLCKSZ as u64)
1232 7 : }
1233 :
1234 : /// Get a KeySpace that covers all the Keys that are in use at AND below the given LSN. This is only used
1235 : /// for gc-compaction.
1236 : ///
1237 : /// gc-compaction cannot use the same `collect_keyspace` function as the legacy compaction because it
1238 : /// processes data at multiple LSNs and needs to be aware of the fact that some key ranges might need to
1239 : /// be kept only for a specific range of LSN.
1240 : ///
1241 : /// Consider the case that the user created branches at LSN 10 and 20, where the user created a table A at
1242 : /// LSN 10 and dropped that table at LSN 20. `collect_keyspace` at LSN 10 will return the key range
1243 : /// corresponding to that table, while LSN 20 won't. The keyspace info at a single LSN is not enough to
1244 : /// determine which keys to retain/drop for gc-compaction.
1245 : ///
1246 : /// For now, it only drops AUX-v1 keys. But in the future, the function will be extended to return the keyspace
1247 : /// to be retained for each of the branch LSN.
1248 : ///
1249 : /// The return value is (dense keyspace, sparse keyspace).
1250 27 : pub(crate) async fn collect_gc_compaction_keyspace(
1251 27 : &self,
1252 27 : ) -> Result<(KeySpace, SparseKeySpace), CollectKeySpaceError> {
1253 27 : let metadata_key_begin = Key::metadata_key_range().start;
1254 27 : let aux_v1_key = AUX_FILES_KEY;
1255 27 : let dense_keyspace = KeySpace {
1256 27 : ranges: vec![Key::MIN..aux_v1_key, aux_v1_key.next()..metadata_key_begin],
1257 27 : };
1258 27 : Ok((
1259 27 : dense_keyspace,
1260 27 : SparseKeySpace(KeySpace::single(Key::metadata_key_range())),
1261 27 : ))
1262 27 : }
1263 :
1264 : ///
1265 : /// Get a KeySpace that covers all the Keys that are in use at the given LSN.
1266 : /// Anything that's not listed maybe removed from the underlying storage (from
1267 : /// that LSN forwards).
1268 : ///
1269 : /// The return value is (dense keyspace, sparse keyspace).
1270 168 : pub(crate) async fn collect_keyspace(
1271 168 : &self,
1272 168 : lsn: Lsn,
1273 168 : ctx: &RequestContext,
1274 168 : ) -> Result<(KeySpace, SparseKeySpace), CollectKeySpaceError> {
1275 168 : // Iterate through key ranges, greedily packing them into partitions
1276 168 : let mut result = KeySpaceAccum::new();
1277 168 :
1278 168 : // The dbdir metadata always exists
1279 168 : result.add_key(DBDIR_KEY);
1280 :
1281 : // Fetch list of database dirs and iterate them
1282 168 : let dbdir = self.list_dbdirs(lsn, ctx).await?;
1283 168 : let mut dbs: Vec<((Oid, Oid), bool)> = dbdir.into_iter().collect();
1284 168 :
1285 168 : dbs.sort_unstable_by(|(k_a, _), (k_b, _)| k_a.cmp(k_b));
1286 168 : for ((spcnode, dbnode), has_relmap_file) in dbs {
1287 0 : if has_relmap_file {
1288 0 : result.add_key(relmap_file_key(spcnode, dbnode));
1289 0 : }
1290 0 : result.add_key(rel_dir_to_key(spcnode, dbnode));
1291 :
1292 0 : let mut rels: Vec<RelTag> = self
1293 0 : .list_rels(spcnode, dbnode, Version::at(lsn), ctx)
1294 0 : .await?
1295 0 : .into_iter()
1296 0 : .collect();
1297 0 : rels.sort_unstable();
1298 0 : for rel in rels {
1299 0 : let relsize_key = rel_size_to_key(rel);
1300 0 : let mut buf = self.get(relsize_key, lsn, ctx).await?;
1301 0 : let relsize = buf.get_u32_le();
1302 0 :
1303 0 : result.add_range(rel_block_to_key(rel, 0)..rel_block_to_key(rel, relsize));
1304 0 : result.add_key(relsize_key);
1305 : }
1306 : }
1307 :
1308 : // Iterate SLRUs next
1309 168 : if self.tenant_shard_id.is_shard_zero() {
1310 495 : for kind in [
1311 165 : SlruKind::Clog,
1312 165 : SlruKind::MultiXactMembers,
1313 165 : SlruKind::MultiXactOffsets,
1314 : ] {
1315 495 : let slrudir_key = slru_dir_to_key(kind);
1316 495 : result.add_key(slrudir_key);
1317 495 : let buf = self.get(slrudir_key, lsn, ctx).await?;
1318 495 : let dir = SlruSegmentDirectory::des(&buf)?;
1319 495 : let mut segments: Vec<u32> = dir.segments.iter().cloned().collect();
1320 495 : segments.sort_unstable();
1321 495 : for segno in segments {
1322 0 : let segsize_key = slru_segment_size_to_key(kind, segno);
1323 0 : let mut buf = self.get(segsize_key, lsn, ctx).await?;
1324 0 : let segsize = buf.get_u32_le();
1325 0 :
1326 0 : result.add_range(
1327 0 : slru_block_to_key(kind, segno, 0)..slru_block_to_key(kind, segno, segsize),
1328 0 : );
1329 0 : result.add_key(segsize_key);
1330 : }
1331 : }
1332 3 : }
1333 :
1334 : // Then pg_twophase
1335 168 : result.add_key(TWOPHASEDIR_KEY);
1336 :
1337 168 : let mut xids: Vec<u64> = self
1338 168 : .list_twophase_files(lsn, ctx)
1339 168 : .await?
1340 168 : .iter()
1341 168 : .cloned()
1342 168 : .collect();
1343 168 : xids.sort_unstable();
1344 168 : for xid in xids {
1345 0 : result.add_key(twophase_file_key(xid));
1346 0 : }
1347 :
1348 168 : result.add_key(CONTROLFILE_KEY);
1349 168 : result.add_key(CHECKPOINT_KEY);
1350 168 :
1351 168 : // Add extra keyspaces in the test cases. Some test cases write keys into the storage without
1352 168 : // creating directory keys. These test cases will add such keyspaces into `extra_test_dense_keyspace`
1353 168 : // and the keys will not be garbage-colllected.
1354 168 : #[cfg(test)]
1355 168 : {
1356 168 : let guard = self.extra_test_dense_keyspace.load();
1357 168 : for kr in &guard.ranges {
1358 0 : result.add_range(kr.clone());
1359 0 : }
1360 0 : }
1361 0 :
1362 168 : let dense_keyspace = result.to_keyspace();
1363 168 : let sparse_keyspace = SparseKeySpace(KeySpace {
1364 168 : ranges: vec![
1365 168 : Key::metadata_aux_key_range(),
1366 168 : repl_origin_key_range(),
1367 168 : Key::rel_dir_sparse_key_range(),
1368 168 : ],
1369 168 : });
1370 168 :
1371 168 : if cfg!(debug_assertions) {
1372 : // Verify if the sparse keyspaces are ordered and non-overlapping.
1373 :
1374 : // We do not use KeySpaceAccum for sparse_keyspace because we want to ensure each
1375 : // category of sparse keys are split into their own image/delta files. If there
1376 : // are overlapping keyspaces, they will be automatically merged by keyspace accum,
1377 : // and we want the developer to keep the keyspaces separated.
1378 :
1379 168 : let ranges = &sparse_keyspace.0.ranges;
1380 :
1381 : // TODO: use a single overlaps_with across the codebase
1382 504 : fn overlaps_with<T: Ord>(a: &Range<T>, b: &Range<T>) -> bool {
1383 504 : !(a.end <= b.start || b.end <= a.start)
1384 504 : }
1385 504 : for i in 0..ranges.len() {
1386 504 : for j in 0..i {
1387 504 : if overlaps_with(&ranges[i], &ranges[j]) {
1388 0 : panic!(
1389 0 : "overlapping sparse keyspace: {}..{} and {}..{}",
1390 0 : ranges[i].start, ranges[i].end, ranges[j].start, ranges[j].end
1391 0 : );
1392 504 : }
1393 : }
1394 : }
1395 336 : for i in 1..ranges.len() {
1396 336 : assert!(
1397 336 : ranges[i - 1].end <= ranges[i].start,
1398 0 : "unordered sparse keyspace: {}..{} and {}..{}",
1399 0 : ranges[i - 1].start,
1400 0 : ranges[i - 1].end,
1401 0 : ranges[i].start,
1402 0 : ranges[i].end
1403 : );
1404 : }
1405 0 : }
1406 :
1407 168 : Ok((dense_keyspace, sparse_keyspace))
1408 168 : }
1409 :
1410 : /// Get cached size of relation. There are two caches: one for primary updates, it captures the latest state of
1411 : /// of the timeline and snapshot cache, which key includes LSN and so can be used by replicas to get relation size
1412 : /// at the particular LSN (snapshot).
1413 224270 : pub fn get_cached_rel_size(&self, tag: &RelTag, version: Version<'_>) -> Option<BlockNumber> {
1414 224270 : let lsn = version.get_lsn();
1415 224270 : {
1416 224270 : let rel_size_cache = self.rel_size_latest_cache.read().unwrap();
1417 224270 : if let Some((cached_lsn, nblocks)) = rel_size_cache.get(tag) {
1418 224259 : if lsn >= *cached_lsn {
1419 221686 : RELSIZE_LATEST_CACHE_HITS.inc();
1420 221686 : return Some(*nblocks);
1421 2573 : }
1422 2573 : RELSIZE_CACHE_MISSES_OLD.inc();
1423 11 : }
1424 : }
1425 : {
1426 2584 : let mut rel_size_cache = self.rel_size_snapshot_cache.lock().unwrap();
1427 2584 : if let Some(nblock) = rel_size_cache.get(&(lsn, *tag)) {
1428 2563 : RELSIZE_SNAPSHOT_CACHE_HITS.inc();
1429 2563 : return Some(*nblock);
1430 21 : }
1431 21 : }
1432 21 : if version.is_latest() {
1433 10 : RELSIZE_LATEST_CACHE_MISSES.inc();
1434 11 : } else {
1435 11 : RELSIZE_SNAPSHOT_CACHE_MISSES.inc();
1436 11 : }
1437 21 : None
1438 224270 : }
1439 :
1440 : /// Update cached relation size if there is no more recent update
1441 5 : pub fn update_cached_rel_size(&self, tag: RelTag, version: Version<'_>, nblocks: BlockNumber) {
1442 5 : let lsn = version.get_lsn();
1443 5 : if version.is_latest() {
1444 0 : let mut rel_size_cache = self.rel_size_latest_cache.write().unwrap();
1445 0 : match rel_size_cache.entry(tag) {
1446 0 : hash_map::Entry::Occupied(mut entry) => {
1447 0 : let cached_lsn = entry.get_mut();
1448 0 : if lsn >= cached_lsn.0 {
1449 0 : *cached_lsn = (lsn, nblocks);
1450 0 : }
1451 : }
1452 0 : hash_map::Entry::Vacant(entry) => {
1453 0 : entry.insert((lsn, nblocks));
1454 0 : RELSIZE_LATEST_CACHE_ENTRIES.inc();
1455 0 : }
1456 : }
1457 : } else {
1458 5 : let mut rel_size_cache = self.rel_size_snapshot_cache.lock().unwrap();
1459 5 : if rel_size_cache.capacity() != 0 {
1460 5 : rel_size_cache.insert((lsn, tag), nblocks);
1461 5 : RELSIZE_SNAPSHOT_CACHE_ENTRIES.set(rel_size_cache.len() as u64);
1462 5 : }
1463 : }
1464 5 : }
1465 :
1466 : /// Store cached relation size
1467 141360 : pub fn set_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
1468 141360 : let mut rel_size_cache = self.rel_size_latest_cache.write().unwrap();
1469 141360 : if rel_size_cache.insert(tag, (lsn, nblocks)).is_none() {
1470 960 : RELSIZE_LATEST_CACHE_ENTRIES.inc();
1471 140400 : }
1472 141360 : }
1473 :
1474 : /// Remove cached relation size
1475 1 : pub fn remove_cached_rel_size(&self, tag: &RelTag) {
1476 1 : let mut rel_size_cache = self.rel_size_latest_cache.write().unwrap();
1477 1 : if rel_size_cache.remove(tag).is_some() {
1478 1 : RELSIZE_LATEST_CACHE_ENTRIES.dec();
1479 1 : }
1480 1 : }
1481 : }
1482 :
1483 : /// DatadirModification represents an operation to ingest an atomic set of
1484 : /// updates to the repository.
1485 : ///
1486 : /// It is created by the 'begin_record' function. It is called for each WAL
1487 : /// record, so that all the modifications by a one WAL record appear atomic.
1488 : pub struct DatadirModification<'a> {
1489 : /// The timeline this modification applies to. You can access this to
1490 : /// read the state, but note that any pending updates are *not* reflected
1491 : /// in the state in 'tline' yet.
1492 : pub tline: &'a Timeline,
1493 :
1494 : /// Current LSN of the modification
1495 : lsn: Lsn,
1496 :
1497 : // The modifications are not applied directly to the underlying key-value store.
1498 : // The put-functions add the modifications here, and they are flushed to the
1499 : // underlying key-value store by the 'finish' function.
1500 : pending_lsns: Vec<Lsn>,
1501 : pending_deletions: Vec<(Range<Key>, Lsn)>,
1502 : pending_nblocks: i64,
1503 :
1504 : /// Metadata writes, indexed by key so that they can be read from not-yet-committed modifications
1505 : /// while ingesting subsequent records. See [`Self::is_data_key`] for the definition of 'metadata'.
1506 : pending_metadata_pages: HashMap<CompactKey, Vec<(Lsn, usize, Value)>>,
1507 :
1508 : /// Data writes, ready to be flushed into an ephemeral layer. See [`Self::is_data_key`] for
1509 : /// which keys are stored here.
1510 : pending_data_batch: Option<SerializedValueBatch>,
1511 :
1512 : /// For special "directory" keys that store key-value maps, track the size of the map
1513 : /// if it was updated in this modification.
1514 : pending_directory_entries: Vec<(DirectoryKind, MetricsUpdate)>,
1515 :
1516 : /// An **approximation** of how many metadata bytes will be written to the EphemeralFile.
1517 : pending_metadata_bytes: usize,
1518 : }
1519 :
1520 : #[derive(Debug, Clone, Copy, PartialEq, Eq)]
1521 : pub enum MetricsUpdate {
1522 : /// Set the metrics to this value
1523 : Set(u64),
1524 : /// Increment the metrics by this value
1525 : Add(u64),
1526 : /// Decrement the metrics by this value
1527 : Sub(u64),
1528 : }
1529 :
1530 : impl DatadirModification<'_> {
1531 : // When a DatadirModification is committed, we do a monolithic serialization of all its contents. WAL records can
1532 : // contain multiple pages, so the pageserver's record-based batch size isn't sufficient to bound this allocation: we
1533 : // additionally specify a limit on how much payload a DatadirModification may contain before it should be committed.
1534 : pub(crate) const MAX_PENDING_BYTES: usize = 8 * 1024 * 1024;
1535 :
1536 : /// Get the current lsn
1537 1 : pub(crate) fn get_lsn(&self) -> Lsn {
1538 1 : self.lsn
1539 1 : }
1540 :
1541 0 : pub(crate) fn approx_pending_bytes(&self) -> usize {
1542 0 : self.pending_data_batch
1543 0 : .as_ref()
1544 0 : .map_or(0, |b| b.buffer_size())
1545 0 : + self.pending_metadata_bytes
1546 0 : }
1547 :
1548 0 : pub(crate) fn has_dirty_data(&self) -> bool {
1549 0 : self.pending_data_batch
1550 0 : .as_ref()
1551 0 : .is_some_and(|b| b.has_data())
1552 0 : }
1553 :
1554 : /// Returns statistics about the currently pending modifications.
1555 0 : pub(crate) fn stats(&self) -> DatadirModificationStats {
1556 0 : let mut stats = DatadirModificationStats::default();
1557 0 : for (_, _, value) in self.pending_metadata_pages.values().flatten() {
1558 0 : match value {
1559 0 : Value::Image(_) => stats.metadata_images += 1,
1560 0 : Value::WalRecord(r) if r.will_init() => stats.metadata_images += 1,
1561 0 : Value::WalRecord(_) => stats.metadata_deltas += 1,
1562 : }
1563 : }
1564 0 : for valuemeta in self.pending_data_batch.iter().flat_map(|b| &b.metadata) {
1565 0 : match valuemeta {
1566 0 : ValueMeta::Serialized(s) if s.will_init => stats.data_images += 1,
1567 0 : ValueMeta::Serialized(_) => stats.data_deltas += 1,
1568 0 : ValueMeta::Observed(_) => {}
1569 : }
1570 : }
1571 0 : stats
1572 0 : }
1573 :
1574 : /// Set the current lsn
1575 72929 : pub(crate) fn set_lsn(&mut self, lsn: Lsn) -> Result<(), WalIngestError> {
1576 72929 : ensure_walingest!(
1577 72929 : lsn >= self.lsn,
1578 72929 : "setting an older lsn {} than {} is not allowed",
1579 72929 : lsn,
1580 72929 : self.lsn
1581 72929 : );
1582 :
1583 72929 : if lsn > self.lsn {
1584 72929 : self.pending_lsns.push(self.lsn);
1585 72929 : self.lsn = lsn;
1586 72929 : }
1587 72929 : Ok(())
1588 72929 : }
1589 :
1590 : /// In this context, 'metadata' means keys that are only read by the pageserver internally, and 'data' means
1591 : /// keys that represent literal blocks that postgres can read. So data includes relation blocks and
1592 : /// SLRU blocks, which are read directly by postgres, and everything else is considered metadata.
1593 : ///
1594 : /// The distinction is important because data keys are handled on a fast path where dirty writes are
1595 : /// not readable until this modification is committed, whereas metadata keys are visible for read
1596 : /// via [`Self::get`] as soon as their record has been ingested.
1597 425364 : fn is_data_key(key: &Key) -> bool {
1598 425364 : key.is_rel_block_key() || key.is_slru_block_key()
1599 425364 : }
1600 :
1601 : /// Initialize a completely new repository.
1602 : ///
1603 : /// This inserts the directory metadata entries that are assumed to
1604 : /// always exist.
1605 111 : pub fn init_empty(&mut self) -> anyhow::Result<()> {
1606 111 : let buf = DbDirectory::ser(&DbDirectory {
1607 111 : dbdirs: HashMap::new(),
1608 111 : })?;
1609 111 : self.pending_directory_entries
1610 111 : .push((DirectoryKind::Db, MetricsUpdate::Set(0)));
1611 111 : self.put(DBDIR_KEY, Value::Image(buf.into()));
1612 :
1613 111 : let buf = if self.tline.pg_version >= 17 {
1614 98 : TwoPhaseDirectoryV17::ser(&TwoPhaseDirectoryV17 {
1615 98 : xids: HashSet::new(),
1616 98 : })
1617 : } else {
1618 13 : TwoPhaseDirectory::ser(&TwoPhaseDirectory {
1619 13 : xids: HashSet::new(),
1620 13 : })
1621 0 : }?;
1622 111 : self.pending_directory_entries
1623 111 : .push((DirectoryKind::TwoPhase, MetricsUpdate::Set(0)));
1624 111 : self.put(TWOPHASEDIR_KEY, Value::Image(buf.into()));
1625 :
1626 111 : let buf: Bytes = SlruSegmentDirectory::ser(&SlruSegmentDirectory::default())?.into();
1627 111 : let empty_dir = Value::Image(buf);
1628 111 :
1629 111 : // Initialize SLRUs on shard 0 only: creating these on other shards would be
1630 111 : // harmless but they'd just be dropped on later compaction.
1631 111 : if self.tline.tenant_shard_id.is_shard_zero() {
1632 108 : self.put(slru_dir_to_key(SlruKind::Clog), empty_dir.clone());
1633 108 : self.pending_directory_entries.push((
1634 108 : DirectoryKind::SlruSegment(SlruKind::Clog),
1635 108 : MetricsUpdate::Set(0),
1636 108 : ));
1637 108 : self.put(
1638 108 : slru_dir_to_key(SlruKind::MultiXactMembers),
1639 108 : empty_dir.clone(),
1640 108 : );
1641 108 : self.pending_directory_entries.push((
1642 108 : DirectoryKind::SlruSegment(SlruKind::Clog),
1643 108 : MetricsUpdate::Set(0),
1644 108 : ));
1645 108 : self.put(slru_dir_to_key(SlruKind::MultiXactOffsets), empty_dir);
1646 108 : self.pending_directory_entries.push((
1647 108 : DirectoryKind::SlruSegment(SlruKind::MultiXactOffsets),
1648 108 : MetricsUpdate::Set(0),
1649 108 : ));
1650 108 : }
1651 :
1652 111 : Ok(())
1653 111 : }
1654 :
1655 : #[cfg(test)]
1656 110 : pub fn init_empty_test_timeline(&mut self) -> anyhow::Result<()> {
1657 110 : self.init_empty()?;
1658 110 : self.put_control_file(bytes::Bytes::from_static(
1659 110 : b"control_file contents do not matter",
1660 110 : ))
1661 110 : .context("put_control_file")?;
1662 110 : self.put_checkpoint(bytes::Bytes::from_static(
1663 110 : b"checkpoint_file contents do not matter",
1664 110 : ))
1665 110 : .context("put_checkpoint_file")?;
1666 110 : Ok(())
1667 110 : }
1668 :
1669 : /// Creates a relation if it is not already present.
1670 : /// Returns the current size of the relation
1671 209028 : pub(crate) async fn create_relation_if_required(
1672 209028 : &mut self,
1673 209028 : rel: RelTag,
1674 209028 : ctx: &RequestContext,
1675 209028 : ) -> Result<u32, WalIngestError> {
1676 : // Get current size and put rel creation if rel doesn't exist
1677 : //
1678 : // NOTE: we check the cache first even though get_rel_exists and get_rel_size would
1679 : // check the cache too. This is because eagerly checking the cache results in
1680 : // less work overall and 10% better performance. It's more work on cache miss
1681 : // but cache miss is rare.
1682 209028 : if let Some(nblocks) = self
1683 209028 : .tline
1684 209028 : .get_cached_rel_size(&rel, Version::Modified(self))
1685 : {
1686 209023 : Ok(nblocks)
1687 5 : } else if !self
1688 5 : .tline
1689 5 : .get_rel_exists(rel, Version::Modified(self), ctx)
1690 5 : .await?
1691 : {
1692 : // create it with 0 size initially, the logic below will extend it
1693 5 : self.put_rel_creation(rel, 0, ctx).await?;
1694 5 : Ok(0)
1695 : } else {
1696 0 : Ok(self
1697 0 : .tline
1698 0 : .get_rel_size(rel, Version::Modified(self), ctx)
1699 0 : .await?)
1700 : }
1701 209028 : }
1702 :
1703 : /// Given a block number for a relation (which represents a newly written block),
1704 : /// the previous block count of the relation, and the shard info, find the gaps
1705 : /// that were created by the newly written block if any.
1706 72835 : fn find_gaps(
1707 72835 : rel: RelTag,
1708 72835 : blkno: u32,
1709 72835 : previous_nblocks: u32,
1710 72835 : shard: &ShardIdentity,
1711 72835 : ) -> Option<KeySpace> {
1712 72835 : let mut key = rel_block_to_key(rel, blkno);
1713 72835 : let mut gap_accum = None;
1714 :
1715 72835 : for gap_blkno in previous_nblocks..blkno {
1716 16 : key.field6 = gap_blkno;
1717 16 :
1718 16 : if shard.get_shard_number(&key) != shard.number {
1719 4 : continue;
1720 12 : }
1721 12 :
1722 12 : gap_accum
1723 12 : .get_or_insert_with(KeySpaceAccum::new)
1724 12 : .add_key(key);
1725 : }
1726 :
1727 72835 : gap_accum.map(|accum| accum.to_keyspace())
1728 72835 : }
1729 :
1730 72926 : pub async fn ingest_batch(
1731 72926 : &mut self,
1732 72926 : mut batch: SerializedValueBatch,
1733 72926 : // TODO(vlad): remove this argument and replace the shard check with is_key_local
1734 72926 : shard: &ShardIdentity,
1735 72926 : ctx: &RequestContext,
1736 72926 : ) -> Result<(), WalIngestError> {
1737 72926 : let mut gaps_at_lsns = Vec::default();
1738 :
1739 72926 : for meta in batch.metadata.iter() {
1740 72821 : let key = Key::from_compact(meta.key());
1741 72821 : let (rel, blkno) = key
1742 72821 : .to_rel_block()
1743 72821 : .map_err(|_| WalIngestErrorKind::InvalidKey(key, meta.lsn()))?;
1744 72821 : let new_nblocks = blkno + 1;
1745 :
1746 72821 : let old_nblocks = self.create_relation_if_required(rel, ctx).await?;
1747 72821 : if new_nblocks > old_nblocks {
1748 1195 : self.put_rel_extend(rel, new_nblocks, ctx).await?;
1749 71626 : }
1750 :
1751 72821 : if let Some(gaps) = Self::find_gaps(rel, blkno, old_nblocks, shard) {
1752 0 : gaps_at_lsns.push((gaps, meta.lsn()));
1753 72821 : }
1754 : }
1755 :
1756 72926 : if !gaps_at_lsns.is_empty() {
1757 0 : batch.zero_gaps(gaps_at_lsns);
1758 72926 : }
1759 :
1760 72926 : match self.pending_data_batch.as_mut() {
1761 10 : Some(pending_batch) => {
1762 10 : pending_batch.extend(batch);
1763 10 : }
1764 72916 : None if batch.has_data() => {
1765 72815 : self.pending_data_batch = Some(batch);
1766 72815 : }
1767 101 : None => {
1768 101 : // Nothing to initialize the batch with
1769 101 : }
1770 : }
1771 :
1772 72926 : Ok(())
1773 72926 : }
1774 :
1775 : /// Put a new page version that can be constructed from a WAL record
1776 : ///
1777 : /// NOTE: this will *not* implicitly extend the relation, if the page is beyond the
1778 : /// current end-of-file. It's up to the caller to check that the relation size
1779 : /// matches the blocks inserted!
1780 6 : pub fn put_rel_wal_record(
1781 6 : &mut self,
1782 6 : rel: RelTag,
1783 6 : blknum: BlockNumber,
1784 6 : rec: NeonWalRecord,
1785 6 : ) -> Result<(), WalIngestError> {
1786 6 : ensure_walingest!(rel.relnode != 0, RelationError::InvalidRelnode);
1787 6 : self.put(rel_block_to_key(rel, blknum), Value::WalRecord(rec));
1788 6 : Ok(())
1789 6 : }
1790 :
1791 : // Same, but for an SLRU.
1792 4 : pub fn put_slru_wal_record(
1793 4 : &mut self,
1794 4 : kind: SlruKind,
1795 4 : segno: u32,
1796 4 : blknum: BlockNumber,
1797 4 : rec: NeonWalRecord,
1798 4 : ) -> Result<(), WalIngestError> {
1799 4 : if !self.tline.tenant_shard_id.is_shard_zero() {
1800 0 : return Ok(());
1801 4 : }
1802 4 :
1803 4 : self.put(
1804 4 : slru_block_to_key(kind, segno, blknum),
1805 4 : Value::WalRecord(rec),
1806 4 : );
1807 4 : Ok(())
1808 4 : }
1809 :
1810 : /// Like put_wal_record, but with ready-made image of the page.
1811 138921 : pub fn put_rel_page_image(
1812 138921 : &mut self,
1813 138921 : rel: RelTag,
1814 138921 : blknum: BlockNumber,
1815 138921 : img: Bytes,
1816 138921 : ) -> Result<(), WalIngestError> {
1817 138921 : ensure_walingest!(rel.relnode != 0, RelationError::InvalidRelnode);
1818 138921 : let key = rel_block_to_key(rel, blknum);
1819 138921 : if !key.is_valid_key_on_write_path() {
1820 0 : Err(WalIngestErrorKind::InvalidKey(key, self.lsn))?;
1821 138921 : }
1822 138921 : self.put(rel_block_to_key(rel, blknum), Value::Image(img));
1823 138921 : Ok(())
1824 138921 : }
1825 :
1826 3 : pub fn put_slru_page_image(
1827 3 : &mut self,
1828 3 : kind: SlruKind,
1829 3 : segno: u32,
1830 3 : blknum: BlockNumber,
1831 3 : img: Bytes,
1832 3 : ) -> Result<(), WalIngestError> {
1833 3 : assert!(self.tline.tenant_shard_id.is_shard_zero());
1834 :
1835 3 : let key = slru_block_to_key(kind, segno, blknum);
1836 3 : if !key.is_valid_key_on_write_path() {
1837 0 : Err(WalIngestErrorKind::InvalidKey(key, self.lsn))?;
1838 3 : }
1839 3 : self.put(key, Value::Image(img));
1840 3 : Ok(())
1841 3 : }
1842 :
1843 1499 : pub(crate) fn put_rel_page_image_zero(
1844 1499 : &mut self,
1845 1499 : rel: RelTag,
1846 1499 : blknum: BlockNumber,
1847 1499 : ) -> Result<(), WalIngestError> {
1848 1499 : ensure_walingest!(rel.relnode != 0, RelationError::InvalidRelnode);
1849 1499 : let key = rel_block_to_key(rel, blknum);
1850 1499 : if !key.is_valid_key_on_write_path() {
1851 0 : Err(WalIngestErrorKind::InvalidKey(key, self.lsn))?;
1852 1499 : }
1853 :
1854 1499 : let batch = self
1855 1499 : .pending_data_batch
1856 1499 : .get_or_insert_with(SerializedValueBatch::default);
1857 1499 :
1858 1499 : batch.put(key.to_compact(), Value::Image(ZERO_PAGE.clone()), self.lsn);
1859 1499 :
1860 1499 : Ok(())
1861 1499 : }
1862 :
1863 0 : pub(crate) fn put_slru_page_image_zero(
1864 0 : &mut self,
1865 0 : kind: SlruKind,
1866 0 : segno: u32,
1867 0 : blknum: BlockNumber,
1868 0 : ) -> Result<(), WalIngestError> {
1869 0 : assert!(self.tline.tenant_shard_id.is_shard_zero());
1870 0 : let key = slru_block_to_key(kind, segno, blknum);
1871 0 : if !key.is_valid_key_on_write_path() {
1872 0 : Err(WalIngestErrorKind::InvalidKey(key, self.lsn))?;
1873 0 : }
1874 :
1875 0 : let batch = self
1876 0 : .pending_data_batch
1877 0 : .get_or_insert_with(SerializedValueBatch::default);
1878 0 :
1879 0 : batch.put(key.to_compact(), Value::Image(ZERO_PAGE.clone()), self.lsn);
1880 0 :
1881 0 : Ok(())
1882 0 : }
1883 :
1884 : /// Returns `true` if the rel_size_v2 write path is enabled. If it is the first time that
1885 : /// we enable it, we also need to persist it in `index_part.json`.
1886 973 : pub fn maybe_enable_rel_size_v2(&mut self) -> anyhow::Result<bool> {
1887 973 : let status = self.tline.get_rel_size_v2_status();
1888 973 : let config = self.tline.get_rel_size_v2_enabled();
1889 973 : match (config, status) {
1890 : (false, RelSizeMigration::Legacy) => {
1891 : // tenant config didn't enable it and we didn't write any reldir_v2 key yet
1892 973 : Ok(false)
1893 : }
1894 : (false, RelSizeMigration::Migrating | RelSizeMigration::Migrated) => {
1895 : // index_part already persisted that the timeline has enabled rel_size_v2
1896 0 : Ok(true)
1897 : }
1898 : (true, RelSizeMigration::Legacy) => {
1899 : // The first time we enable it, we need to persist it in `index_part.json`
1900 0 : self.tline
1901 0 : .update_rel_size_v2_status(RelSizeMigration::Migrating)?;
1902 0 : tracing::info!("enabled rel_size_v2");
1903 0 : Ok(true)
1904 : }
1905 : (true, RelSizeMigration::Migrating | RelSizeMigration::Migrated) => {
1906 : // index_part already persisted that the timeline has enabled rel_size_v2
1907 : // and we don't need to do anything
1908 0 : Ok(true)
1909 : }
1910 : }
1911 973 : }
1912 :
1913 : /// Store a relmapper file (pg_filenode.map) in the repository
1914 8 : pub async fn put_relmap_file(
1915 8 : &mut self,
1916 8 : spcnode: Oid,
1917 8 : dbnode: Oid,
1918 8 : img: Bytes,
1919 8 : ctx: &RequestContext,
1920 8 : ) -> Result<(), WalIngestError> {
1921 8 : let v2_enabled = self
1922 8 : .maybe_enable_rel_size_v2()
1923 8 : .map_err(WalIngestErrorKind::MaybeRelSizeV2Error)?;
1924 :
1925 : // Add it to the directory (if it doesn't exist already)
1926 8 : let buf = self.get(DBDIR_KEY, ctx).await?;
1927 8 : let mut dbdir = DbDirectory::des(&buf)?;
1928 :
1929 8 : let r = dbdir.dbdirs.insert((spcnode, dbnode), true);
1930 8 : if r.is_none() || r == Some(false) {
1931 : // The dbdir entry didn't exist, or it contained a
1932 : // 'false'. The 'insert' call already updated it with
1933 : // 'true', now write the updated 'dbdirs' map back.
1934 8 : let buf = DbDirectory::ser(&dbdir)?;
1935 8 : self.put(DBDIR_KEY, Value::Image(buf.into()));
1936 0 : }
1937 8 : if r.is_none() {
1938 : // Create RelDirectory
1939 : // TODO: if we have fully migrated to v2, no need to create this directory
1940 4 : let buf = RelDirectory::ser(&RelDirectory {
1941 4 : rels: HashSet::new(),
1942 4 : })?;
1943 4 : self.pending_directory_entries
1944 4 : .push((DirectoryKind::Rel, MetricsUpdate::Set(0)));
1945 4 : if v2_enabled {
1946 0 : self.pending_directory_entries
1947 0 : .push((DirectoryKind::RelV2, MetricsUpdate::Set(0)));
1948 4 : }
1949 4 : self.put(
1950 4 : rel_dir_to_key(spcnode, dbnode),
1951 4 : Value::Image(Bytes::from(buf)),
1952 4 : );
1953 4 : }
1954 :
1955 8 : self.put(relmap_file_key(spcnode, dbnode), Value::Image(img));
1956 8 : Ok(())
1957 8 : }
1958 :
1959 0 : pub async fn put_twophase_file(
1960 0 : &mut self,
1961 0 : xid: u64,
1962 0 : img: Bytes,
1963 0 : ctx: &RequestContext,
1964 0 : ) -> Result<(), WalIngestError> {
1965 : // Add it to the directory entry
1966 0 : let dirbuf = self.get(TWOPHASEDIR_KEY, ctx).await?;
1967 0 : let newdirbuf = if self.tline.pg_version >= 17 {
1968 0 : let mut dir = TwoPhaseDirectoryV17::des(&dirbuf)?;
1969 0 : if !dir.xids.insert(xid) {
1970 0 : Err(WalIngestErrorKind::FileAlreadyExists(xid))?;
1971 0 : }
1972 0 : self.pending_directory_entries.push((
1973 0 : DirectoryKind::TwoPhase,
1974 0 : MetricsUpdate::Set(dir.xids.len() as u64),
1975 0 : ));
1976 0 : Bytes::from(TwoPhaseDirectoryV17::ser(&dir)?)
1977 : } else {
1978 0 : let xid = xid as u32;
1979 0 : let mut dir = TwoPhaseDirectory::des(&dirbuf)?;
1980 0 : if !dir.xids.insert(xid) {
1981 0 : Err(WalIngestErrorKind::FileAlreadyExists(xid.into()))?;
1982 0 : }
1983 0 : self.pending_directory_entries.push((
1984 0 : DirectoryKind::TwoPhase,
1985 0 : MetricsUpdate::Set(dir.xids.len() as u64),
1986 0 : ));
1987 0 : Bytes::from(TwoPhaseDirectory::ser(&dir)?)
1988 : };
1989 0 : self.put(TWOPHASEDIR_KEY, Value::Image(newdirbuf));
1990 0 :
1991 0 : self.put(twophase_file_key(xid), Value::Image(img));
1992 0 : Ok(())
1993 0 : }
1994 :
1995 1 : pub async fn set_replorigin(
1996 1 : &mut self,
1997 1 : origin_id: RepOriginId,
1998 1 : origin_lsn: Lsn,
1999 1 : ) -> Result<(), WalIngestError> {
2000 1 : let key = repl_origin_key(origin_id);
2001 1 : self.put(key, Value::Image(origin_lsn.ser().unwrap().into()));
2002 1 : Ok(())
2003 1 : }
2004 :
2005 0 : pub async fn drop_replorigin(&mut self, origin_id: RepOriginId) -> Result<(), WalIngestError> {
2006 0 : self.set_replorigin(origin_id, Lsn::INVALID).await
2007 0 : }
2008 :
2009 111 : pub fn put_control_file(&mut self, img: Bytes) -> Result<(), WalIngestError> {
2010 111 : self.put(CONTROLFILE_KEY, Value::Image(img));
2011 111 : Ok(())
2012 111 : }
2013 :
2014 118 : pub fn put_checkpoint(&mut self, img: Bytes) -> Result<(), WalIngestError> {
2015 118 : self.put(CHECKPOINT_KEY, Value::Image(img));
2016 118 : Ok(())
2017 118 : }
2018 :
2019 0 : pub async fn drop_dbdir(
2020 0 : &mut self,
2021 0 : spcnode: Oid,
2022 0 : dbnode: Oid,
2023 0 : ctx: &RequestContext,
2024 0 : ) -> Result<(), WalIngestError> {
2025 0 : let total_blocks = self
2026 0 : .tline
2027 0 : .get_db_size(spcnode, dbnode, Version::Modified(self), ctx)
2028 0 : .await?;
2029 :
2030 : // Remove entry from dbdir
2031 0 : let buf = self.get(DBDIR_KEY, ctx).await?;
2032 0 : let mut dir = DbDirectory::des(&buf)?;
2033 0 : if dir.dbdirs.remove(&(spcnode, dbnode)).is_some() {
2034 0 : let buf = DbDirectory::ser(&dir)?;
2035 0 : self.pending_directory_entries.push((
2036 0 : DirectoryKind::Db,
2037 0 : MetricsUpdate::Set(dir.dbdirs.len() as u64),
2038 0 : ));
2039 0 : self.put(DBDIR_KEY, Value::Image(buf.into()));
2040 : } else {
2041 0 : warn!(
2042 0 : "dropped dbdir for spcnode {} dbnode {} did not exist in db directory",
2043 : spcnode, dbnode
2044 : );
2045 : }
2046 :
2047 : // Update logical database size.
2048 0 : self.pending_nblocks -= total_blocks as i64;
2049 0 :
2050 0 : // Delete all relations and metadata files for the spcnode/dnode
2051 0 : self.delete(dbdir_key_range(spcnode, dbnode));
2052 0 : Ok(())
2053 0 : }
2054 :
2055 : /// Create a relation fork.
2056 : ///
2057 : /// 'nblocks' is the initial size.
2058 960 : pub async fn put_rel_creation(
2059 960 : &mut self,
2060 960 : rel: RelTag,
2061 960 : nblocks: BlockNumber,
2062 960 : ctx: &RequestContext,
2063 960 : ) -> Result<(), WalIngestError> {
2064 960 : if rel.relnode == 0 {
2065 0 : Err(WalIngestErrorKind::LogicalError(anyhow::anyhow!(
2066 0 : "invalid relnode"
2067 0 : )))?;
2068 960 : }
2069 : // It's possible that this is the first rel for this db in this
2070 : // tablespace. Create the reldir entry for it if so.
2071 960 : let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await?)?;
2072 :
2073 960 : let dbdir_exists =
2074 960 : if let hash_map::Entry::Vacant(e) = dbdir.dbdirs.entry((rel.spcnode, rel.dbnode)) {
2075 : // Didn't exist. Update dbdir
2076 4 : e.insert(false);
2077 4 : let buf = DbDirectory::ser(&dbdir)?;
2078 4 : self.pending_directory_entries.push((
2079 4 : DirectoryKind::Db,
2080 4 : MetricsUpdate::Set(dbdir.dbdirs.len() as u64),
2081 4 : ));
2082 4 : self.put(DBDIR_KEY, Value::Image(buf.into()));
2083 4 : false
2084 : } else {
2085 956 : true
2086 : };
2087 :
2088 960 : let rel_dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
2089 960 : let mut rel_dir = if !dbdir_exists {
2090 : // Create the RelDirectory
2091 4 : RelDirectory::default()
2092 : } else {
2093 : // reldir already exists, fetch it
2094 956 : RelDirectory::des(&self.get(rel_dir_key, ctx).await?)?
2095 : };
2096 :
2097 960 : let v2_enabled = self
2098 960 : .maybe_enable_rel_size_v2()
2099 960 : .map_err(WalIngestErrorKind::MaybeRelSizeV2Error)?;
2100 :
2101 960 : if v2_enabled {
2102 0 : if rel_dir.rels.contains(&(rel.relnode, rel.forknum)) {
2103 0 : Err(WalIngestErrorKind::RelationAlreadyExists(rel))?;
2104 0 : }
2105 0 : let sparse_rel_dir_key =
2106 0 : rel_tag_sparse_key(rel.spcnode, rel.dbnode, rel.relnode, rel.forknum);
2107 : // check if the rel_dir_key exists in v2
2108 0 : let val = self.sparse_get(sparse_rel_dir_key, ctx).await?;
2109 0 : let val = RelDirExists::decode_option(val)
2110 0 : .map_err(|_| WalIngestErrorKind::InvalidRelDirKey(sparse_rel_dir_key))?;
2111 0 : if val == RelDirExists::Exists {
2112 0 : Err(WalIngestErrorKind::RelationAlreadyExists(rel))?;
2113 0 : }
2114 0 : self.put(
2115 0 : sparse_rel_dir_key,
2116 0 : Value::Image(RelDirExists::Exists.encode()),
2117 0 : );
2118 0 : if !dbdir_exists {
2119 0 : self.pending_directory_entries
2120 0 : .push((DirectoryKind::Rel, MetricsUpdate::Set(0)));
2121 0 : self.pending_directory_entries
2122 0 : .push((DirectoryKind::RelV2, MetricsUpdate::Set(0)));
2123 0 : // We don't write `rel_dir_key -> rel_dir.rels` back to the storage in the v2 path unless it's the initial creation.
2124 0 : // TODO: if we have fully migrated to v2, no need to create this directory. Otherwise, there
2125 0 : // will be key not found errors if we don't create an empty one for rel_size_v2.
2126 0 : self.put(
2127 0 : rel_dir_key,
2128 0 : Value::Image(Bytes::from(RelDirectory::ser(&RelDirectory::default())?)),
2129 : );
2130 0 : }
2131 0 : self.pending_directory_entries
2132 0 : .push((DirectoryKind::RelV2, MetricsUpdate::Add(1)));
2133 : } else {
2134 : // Add the new relation to the rel directory entry, and write it back
2135 960 : if !rel_dir.rels.insert((rel.relnode, rel.forknum)) {
2136 0 : Err(WalIngestErrorKind::RelationAlreadyExists(rel))?;
2137 960 : }
2138 960 : if !dbdir_exists {
2139 4 : self.pending_directory_entries
2140 4 : .push((DirectoryKind::Rel, MetricsUpdate::Set(0)))
2141 956 : }
2142 960 : self.pending_directory_entries
2143 960 : .push((DirectoryKind::Rel, MetricsUpdate::Add(1)));
2144 960 : self.put(
2145 960 : rel_dir_key,
2146 960 : Value::Image(Bytes::from(RelDirectory::ser(&rel_dir)?)),
2147 : );
2148 : }
2149 :
2150 : // Put size
2151 960 : let size_key = rel_size_to_key(rel);
2152 960 : let buf = nblocks.to_le_bytes();
2153 960 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
2154 960 :
2155 960 : self.pending_nblocks += nblocks as i64;
2156 960 :
2157 960 : // Update relation size cache
2158 960 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
2159 960 :
2160 960 : // Even if nblocks > 0, we don't insert any actual blocks here. That's up to the
2161 960 : // caller.
2162 960 : Ok(())
2163 960 : }
2164 :
2165 : /// Truncate relation
2166 3006 : pub async fn put_rel_truncation(
2167 3006 : &mut self,
2168 3006 : rel: RelTag,
2169 3006 : nblocks: BlockNumber,
2170 3006 : ctx: &RequestContext,
2171 3006 : ) -> Result<(), WalIngestError> {
2172 3006 : ensure_walingest!(rel.relnode != 0, RelationError::InvalidRelnode);
2173 3006 : if self
2174 3006 : .tline
2175 3006 : .get_rel_exists(rel, Version::Modified(self), ctx)
2176 3006 : .await?
2177 : {
2178 3006 : let size_key = rel_size_to_key(rel);
2179 : // Fetch the old size first
2180 3006 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
2181 3006 :
2182 3006 : // Update the entry with the new size.
2183 3006 : let buf = nblocks.to_le_bytes();
2184 3006 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
2185 3006 :
2186 3006 : // Update relation size cache
2187 3006 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
2188 3006 :
2189 3006 : // Update logical database size.
2190 3006 : self.pending_nblocks -= old_size as i64 - nblocks as i64;
2191 0 : }
2192 3006 : Ok(())
2193 3006 : }
2194 :
2195 : /// Extend relation
2196 : /// If new size is smaller, do nothing.
2197 138340 : pub async fn put_rel_extend(
2198 138340 : &mut self,
2199 138340 : rel: RelTag,
2200 138340 : nblocks: BlockNumber,
2201 138340 : ctx: &RequestContext,
2202 138340 : ) -> Result<(), WalIngestError> {
2203 138340 : ensure_walingest!(rel.relnode != 0, RelationError::InvalidRelnode);
2204 :
2205 : // Put size
2206 138340 : let size_key = rel_size_to_key(rel);
2207 138340 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
2208 138340 :
2209 138340 : // only extend relation here. never decrease the size
2210 138340 : if nblocks > old_size {
2211 137394 : let buf = nblocks.to_le_bytes();
2212 137394 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
2213 137394 :
2214 137394 : // Update relation size cache
2215 137394 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
2216 137394 :
2217 137394 : self.pending_nblocks += nblocks as i64 - old_size as i64;
2218 137394 : }
2219 138340 : Ok(())
2220 138340 : }
2221 :
2222 : /// Drop some relations
2223 5 : pub(crate) async fn put_rel_drops(
2224 5 : &mut self,
2225 5 : drop_relations: HashMap<(u32, u32), Vec<RelTag>>,
2226 5 : ctx: &RequestContext,
2227 5 : ) -> Result<(), WalIngestError> {
2228 5 : let v2_enabled = self
2229 5 : .maybe_enable_rel_size_v2()
2230 5 : .map_err(WalIngestErrorKind::MaybeRelSizeV2Error)?;
2231 6 : for ((spc_node, db_node), rel_tags) in drop_relations {
2232 1 : let dir_key = rel_dir_to_key(spc_node, db_node);
2233 1 : let buf = self.get(dir_key, ctx).await?;
2234 1 : let mut dir = RelDirectory::des(&buf)?;
2235 :
2236 1 : let mut dirty = false;
2237 2 : for rel_tag in rel_tags {
2238 1 : let found = if dir.rels.remove(&(rel_tag.relnode, rel_tag.forknum)) {
2239 1 : self.pending_directory_entries
2240 1 : .push((DirectoryKind::Rel, MetricsUpdate::Sub(1)));
2241 1 : dirty = true;
2242 1 : true
2243 0 : } else if v2_enabled {
2244 : // The rel is not found in the old reldir key, so we need to check the new sparse keyspace.
2245 : // Note that a relation can only exist in one of the two keyspaces (guaranteed by the ingestion
2246 : // logic).
2247 0 : let key =
2248 0 : rel_tag_sparse_key(spc_node, db_node, rel_tag.relnode, rel_tag.forknum);
2249 0 : let val = RelDirExists::decode_option(self.sparse_get(key, ctx).await?)
2250 0 : .map_err(|_| WalIngestErrorKind::InvalidKey(key, self.lsn))?;
2251 0 : if val == RelDirExists::Exists {
2252 0 : self.pending_directory_entries
2253 0 : .push((DirectoryKind::RelV2, MetricsUpdate::Sub(1)));
2254 0 : // put tombstone
2255 0 : self.put(key, Value::Image(RelDirExists::Removed.encode()));
2256 0 : // no need to set dirty to true
2257 0 : true
2258 : } else {
2259 0 : false
2260 : }
2261 : } else {
2262 0 : false
2263 : };
2264 :
2265 1 : if found {
2266 : // update logical size
2267 1 : let size_key = rel_size_to_key(rel_tag);
2268 1 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
2269 1 : self.pending_nblocks -= old_size as i64;
2270 1 :
2271 1 : // Remove entry from relation size cache
2272 1 : self.tline.remove_cached_rel_size(&rel_tag);
2273 1 :
2274 1 : // Delete size entry, as well as all blocks; this is currently a no-op because we haven't implemented tombstones in storage.
2275 1 : self.delete(rel_key_range(rel_tag));
2276 0 : }
2277 : }
2278 :
2279 1 : if dirty {
2280 1 : self.put(dir_key, Value::Image(Bytes::from(RelDirectory::ser(&dir)?)));
2281 0 : }
2282 : }
2283 :
2284 5 : Ok(())
2285 5 : }
2286 :
2287 3 : pub async fn put_slru_segment_creation(
2288 3 : &mut self,
2289 3 : kind: SlruKind,
2290 3 : segno: u32,
2291 3 : nblocks: BlockNumber,
2292 3 : ctx: &RequestContext,
2293 3 : ) -> Result<(), WalIngestError> {
2294 3 : assert!(self.tline.tenant_shard_id.is_shard_zero());
2295 :
2296 : // Add it to the directory entry
2297 3 : let dir_key = slru_dir_to_key(kind);
2298 3 : let buf = self.get(dir_key, ctx).await?;
2299 3 : let mut dir = SlruSegmentDirectory::des(&buf)?;
2300 :
2301 3 : if !dir.segments.insert(segno) {
2302 0 : Err(WalIngestErrorKind::SlruAlreadyExists(kind, segno))?;
2303 3 : }
2304 3 : self.pending_directory_entries.push((
2305 3 : DirectoryKind::SlruSegment(kind),
2306 3 : MetricsUpdate::Set(dir.segments.len() as u64),
2307 3 : ));
2308 3 : self.put(
2309 3 : dir_key,
2310 3 : Value::Image(Bytes::from(SlruSegmentDirectory::ser(&dir)?)),
2311 : );
2312 :
2313 : // Put size
2314 3 : let size_key = slru_segment_size_to_key(kind, segno);
2315 3 : let buf = nblocks.to_le_bytes();
2316 3 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
2317 3 :
2318 3 : // even if nblocks > 0, we don't insert any actual blocks here
2319 3 :
2320 3 : Ok(())
2321 3 : }
2322 :
2323 : /// Extend SLRU segment
2324 0 : pub fn put_slru_extend(
2325 0 : &mut self,
2326 0 : kind: SlruKind,
2327 0 : segno: u32,
2328 0 : nblocks: BlockNumber,
2329 0 : ) -> Result<(), WalIngestError> {
2330 0 : assert!(self.tline.tenant_shard_id.is_shard_zero());
2331 :
2332 : // Put size
2333 0 : let size_key = slru_segment_size_to_key(kind, segno);
2334 0 : let buf = nblocks.to_le_bytes();
2335 0 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
2336 0 : Ok(())
2337 0 : }
2338 :
2339 : /// This method is used for marking truncated SLRU files
2340 0 : pub async fn drop_slru_segment(
2341 0 : &mut self,
2342 0 : kind: SlruKind,
2343 0 : segno: u32,
2344 0 : ctx: &RequestContext,
2345 0 : ) -> Result<(), WalIngestError> {
2346 0 : // Remove it from the directory entry
2347 0 : let dir_key = slru_dir_to_key(kind);
2348 0 : let buf = self.get(dir_key, ctx).await?;
2349 0 : let mut dir = SlruSegmentDirectory::des(&buf)?;
2350 :
2351 0 : if !dir.segments.remove(&segno) {
2352 0 : warn!("slru segment {:?}/{} does not exist", kind, segno);
2353 0 : }
2354 0 : self.pending_directory_entries.push((
2355 0 : DirectoryKind::SlruSegment(kind),
2356 0 : MetricsUpdate::Set(dir.segments.len() as u64),
2357 0 : ));
2358 0 : self.put(
2359 0 : dir_key,
2360 0 : Value::Image(Bytes::from(SlruSegmentDirectory::ser(&dir)?)),
2361 : );
2362 :
2363 : // Delete size entry, as well as all blocks
2364 0 : self.delete(slru_segment_key_range(kind, segno));
2365 0 :
2366 0 : Ok(())
2367 0 : }
2368 :
2369 : /// Drop a relmapper file (pg_filenode.map)
2370 0 : pub fn drop_relmap_file(&mut self, _spcnode: Oid, _dbnode: Oid) -> Result<(), WalIngestError> {
2371 0 : // TODO
2372 0 : Ok(())
2373 0 : }
2374 :
2375 : /// This method is used for marking truncated SLRU files
2376 0 : pub async fn drop_twophase_file(
2377 0 : &mut self,
2378 0 : xid: u64,
2379 0 : ctx: &RequestContext,
2380 0 : ) -> Result<(), WalIngestError> {
2381 : // Remove it from the directory entry
2382 0 : let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
2383 0 : let newdirbuf = if self.tline.pg_version >= 17 {
2384 0 : let mut dir = TwoPhaseDirectoryV17::des(&buf)?;
2385 :
2386 0 : if !dir.xids.remove(&xid) {
2387 0 : warn!("twophase file for xid {} does not exist", xid);
2388 0 : }
2389 0 : self.pending_directory_entries.push((
2390 0 : DirectoryKind::TwoPhase,
2391 0 : MetricsUpdate::Set(dir.xids.len() as u64),
2392 0 : ));
2393 0 : Bytes::from(TwoPhaseDirectoryV17::ser(&dir)?)
2394 : } else {
2395 0 : let xid: u32 = u32::try_from(xid)
2396 0 : .map_err(|e| WalIngestErrorKind::LogicalError(anyhow::Error::from(e)))?;
2397 0 : let mut dir = TwoPhaseDirectory::des(&buf)?;
2398 :
2399 0 : if !dir.xids.remove(&xid) {
2400 0 : warn!("twophase file for xid {} does not exist", xid);
2401 0 : }
2402 0 : self.pending_directory_entries.push((
2403 0 : DirectoryKind::TwoPhase,
2404 0 : MetricsUpdate::Set(dir.xids.len() as u64),
2405 0 : ));
2406 0 : Bytes::from(TwoPhaseDirectory::ser(&dir)?)
2407 : };
2408 0 : self.put(TWOPHASEDIR_KEY, Value::Image(newdirbuf));
2409 0 :
2410 0 : // Delete it
2411 0 : self.delete(twophase_key_range(xid));
2412 0 :
2413 0 : Ok(())
2414 0 : }
2415 :
2416 8 : pub async fn put_file(
2417 8 : &mut self,
2418 8 : path: &str,
2419 8 : content: &[u8],
2420 8 : ctx: &RequestContext,
2421 8 : ) -> Result<(), WalIngestError> {
2422 8 : let key = aux_file::encode_aux_file_key(path);
2423 : // retrieve the key from the engine
2424 8 : let old_val = match self.get(key, ctx).await {
2425 2 : Ok(val) => Some(val),
2426 6 : Err(PageReconstructError::MissingKey(_)) => None,
2427 0 : Err(e) => return Err(e.into()),
2428 : };
2429 8 : let files: Vec<(&str, &[u8])> = if let Some(ref old_val) = old_val {
2430 2 : aux_file::decode_file_value(old_val).map_err(WalIngestErrorKind::EncodeAuxFileError)?
2431 : } else {
2432 6 : Vec::new()
2433 : };
2434 8 : let mut other_files = Vec::with_capacity(files.len());
2435 8 : let mut modifying_file = None;
2436 10 : for file @ (p, content) in files {
2437 2 : if path == p {
2438 2 : assert!(
2439 2 : modifying_file.is_none(),
2440 0 : "duplicated entries found for {}",
2441 : path
2442 : );
2443 2 : modifying_file = Some(content);
2444 0 : } else {
2445 0 : other_files.push(file);
2446 0 : }
2447 : }
2448 8 : let mut new_files = other_files;
2449 8 : match (modifying_file, content.is_empty()) {
2450 1 : (Some(old_content), false) => {
2451 1 : self.tline
2452 1 : .aux_file_size_estimator
2453 1 : .on_update(old_content.len(), content.len());
2454 1 : new_files.push((path, content));
2455 1 : }
2456 1 : (Some(old_content), true) => {
2457 1 : self.tline
2458 1 : .aux_file_size_estimator
2459 1 : .on_remove(old_content.len());
2460 1 : // not adding the file key to the final `new_files` vec.
2461 1 : }
2462 6 : (None, false) => {
2463 6 : self.tline.aux_file_size_estimator.on_add(content.len());
2464 6 : new_files.push((path, content));
2465 6 : }
2466 : // Compute may request delete of old version of pgstat AUX file if new one exceeds size limit.
2467 : // Compute doesn't know if previous version of this file exists or not, so
2468 : // attempt to delete non-existing file can cause this message.
2469 : // To avoid false alarms, log it as info rather than warning.
2470 0 : (None, true) if path.starts_with("pg_stat/") => {
2471 0 : info!("removing non-existing pg_stat file: {}", path)
2472 : }
2473 0 : (None, true) => warn!("removing non-existing aux file: {}", path),
2474 : }
2475 8 : let new_val = aux_file::encode_file_value(&new_files)
2476 8 : .map_err(WalIngestErrorKind::EncodeAuxFileError)?;
2477 8 : self.put(key, Value::Image(new_val.into()));
2478 8 :
2479 8 : Ok(())
2480 8 : }
2481 :
2482 : ///
2483 : /// Flush changes accumulated so far to the underlying repository.
2484 : ///
2485 : /// Usually, changes made in DatadirModification are atomic, but this allows
2486 : /// you to flush them to the underlying repository before the final `commit`.
2487 : /// That allows to free up the memory used to hold the pending changes.
2488 : ///
2489 : /// Currently only used during bulk import of a data directory. In that
2490 : /// context, breaking the atomicity is OK. If the import is interrupted, the
2491 : /// whole import fails and the timeline will be deleted anyway.
2492 : /// (Or to be precise, it will be left behind for debugging purposes and
2493 : /// ignored, see <https://github.com/neondatabase/neon/pull/1809>)
2494 : ///
2495 : /// Note: A consequence of flushing the pending operations is that they
2496 : /// won't be visible to subsequent operations until `commit`. The function
2497 : /// retains all the metadata, but data pages are flushed. That's again OK
2498 : /// for bulk import, where you are just loading data pages and won't try to
2499 : /// modify the same pages twice.
2500 965 : pub(crate) async fn flush(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
2501 965 : // Unless we have accumulated a decent amount of changes, it's not worth it
2502 965 : // to scan through the pending_updates list.
2503 965 : let pending_nblocks = self.pending_nblocks;
2504 965 : if pending_nblocks < 10000 {
2505 965 : return Ok(());
2506 0 : }
2507 :
2508 0 : let mut writer = self.tline.writer().await;
2509 :
2510 : // Flush relation and SLRU data blocks, keep metadata.
2511 0 : if let Some(batch) = self.pending_data_batch.take() {
2512 0 : tracing::debug!(
2513 0 : "Flushing batch with max_lsn={}. Last record LSN is {}",
2514 0 : batch.max_lsn,
2515 0 : self.tline.get_last_record_lsn()
2516 : );
2517 :
2518 : // This bails out on first error without modifying pending_updates.
2519 : // That's Ok, cf this function's doc comment.
2520 0 : writer.put_batch(batch, ctx).await?;
2521 0 : }
2522 :
2523 0 : if pending_nblocks != 0 {
2524 0 : writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
2525 0 : self.pending_nblocks = 0;
2526 0 : }
2527 :
2528 0 : for (kind, count) in std::mem::take(&mut self.pending_directory_entries) {
2529 0 : writer.update_directory_entries_count(kind, count);
2530 0 : }
2531 :
2532 0 : Ok(())
2533 965 : }
2534 :
2535 : ///
2536 : /// Finish this atomic update, writing all the updated keys to the
2537 : /// underlying timeline.
2538 : /// All the modifications in this atomic update are stamped by the specified LSN.
2539 : ///
2540 371556 : pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
2541 371556 : let mut writer = self.tline.writer().await;
2542 :
2543 371556 : let pending_nblocks = self.pending_nblocks;
2544 371556 : self.pending_nblocks = 0;
2545 :
2546 : // Ordering: the items in this batch do not need to be in any global order, but values for
2547 : // a particular Key must be in Lsn order relative to one another. InMemoryLayer relies on
2548 : // this to do efficient updates to its index. See [`wal_decoder::serialized_batch`] for
2549 : // more details.
2550 :
2551 371556 : let metadata_batch = {
2552 371556 : let pending_meta = self
2553 371556 : .pending_metadata_pages
2554 371556 : .drain()
2555 371556 : .flat_map(|(key, values)| {
2556 137061 : values
2557 137061 : .into_iter()
2558 137061 : .map(move |(lsn, value_size, value)| (key, lsn, value_size, value))
2559 371556 : })
2560 371556 : .collect::<Vec<_>>();
2561 371556 :
2562 371556 : if pending_meta.is_empty() {
2563 236139 : None
2564 : } else {
2565 135417 : Some(SerializedValueBatch::from_values(pending_meta))
2566 : }
2567 : };
2568 :
2569 371556 : let data_batch = self.pending_data_batch.take();
2570 :
2571 371556 : let maybe_batch = match (data_batch, metadata_batch) {
2572 132278 : (Some(mut data), Some(metadata)) => {
2573 132278 : data.extend(metadata);
2574 132278 : Some(data)
2575 : }
2576 71631 : (Some(data), None) => Some(data),
2577 3139 : (None, Some(metadata)) => Some(metadata),
2578 164508 : (None, None) => None,
2579 : };
2580 :
2581 371556 : if let Some(batch) = maybe_batch {
2582 207048 : tracing::debug!(
2583 0 : "Flushing batch with max_lsn={}. Last record LSN is {}",
2584 0 : batch.max_lsn,
2585 0 : self.tline.get_last_record_lsn()
2586 : );
2587 :
2588 : // This bails out on first error without modifying pending_updates.
2589 : // That's Ok, cf this function's doc comment.
2590 207048 : writer.put_batch(batch, ctx).await?;
2591 164508 : }
2592 :
2593 371556 : if !self.pending_deletions.is_empty() {
2594 1 : writer.delete_batch(&self.pending_deletions, ctx).await?;
2595 1 : self.pending_deletions.clear();
2596 371555 : }
2597 :
2598 371556 : self.pending_lsns.push(self.lsn);
2599 444485 : for pending_lsn in self.pending_lsns.drain(..) {
2600 444485 : // TODO(vlad): pretty sure the comment below is not valid anymore
2601 444485 : // and we can call finish write with the latest LSN
2602 444485 : //
2603 444485 : // Ideally, we should be able to call writer.finish_write() only once
2604 444485 : // with the highest LSN. However, the last_record_lsn variable in the
2605 444485 : // timeline keeps track of the latest LSN and the immediate previous LSN
2606 444485 : // so we need to record every LSN to not leave a gap between them.
2607 444485 : writer.finish_write(pending_lsn);
2608 444485 : }
2609 :
2610 371556 : if pending_nblocks != 0 {
2611 135285 : writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
2612 236271 : }
2613 :
2614 371556 : for (kind, count) in std::mem::take(&mut self.pending_directory_entries) {
2615 1522 : writer.update_directory_entries_count(kind, count);
2616 1522 : }
2617 :
2618 371556 : self.pending_metadata_bytes = 0;
2619 371556 :
2620 371556 : Ok(())
2621 371556 : }
2622 :
2623 145852 : pub(crate) fn len(&self) -> usize {
2624 145852 : self.pending_metadata_pages.len()
2625 145852 : + self.pending_data_batch.as_ref().map_or(0, |b| b.len())
2626 145852 : + self.pending_deletions.len()
2627 145852 : }
2628 :
2629 : /// Read a page from the Timeline we are writing to. For metadata pages, this passes through
2630 : /// a cache in Self, which makes writes earlier in this modification visible to WAL records later
2631 : /// in the modification.
2632 : ///
2633 : /// For data pages, reads pass directly to the owning Timeline: any ingest code which reads a data
2634 : /// page must ensure that the pages they read are already committed in Timeline, for example
2635 : /// DB create operations are always preceded by a call to commit(). This is special cased because
2636 : /// it's rare: all the 'normal' WAL operations will only read metadata pages such as relation sizes,
2637 : /// and not data pages.
2638 143293 : async fn get(&self, key: Key, ctx: &RequestContext) -> Result<Bytes, PageReconstructError> {
2639 143293 : if !Self::is_data_key(&key) {
2640 : // Have we already updated the same key? Read the latest pending updated
2641 : // version in that case.
2642 : //
2643 : // Note: we don't check pending_deletions. It is an error to request a
2644 : // value that has been removed, deletion only avoids leaking storage.
2645 143293 : if let Some(values) = self.pending_metadata_pages.get(&key.to_compact()) {
2646 7964 : if let Some((_, _, value)) = values.last() {
2647 7964 : return if let Value::Image(img) = value {
2648 7964 : Ok(img.clone())
2649 : } else {
2650 : // Currently, we never need to read back a WAL record that we
2651 : // inserted in the same "transaction". All the metadata updates
2652 : // work directly with Images, and we never need to read actual
2653 : // data pages. We could handle this if we had to, by calling
2654 : // the walredo manager, but let's keep it simple for now.
2655 0 : Err(PageReconstructError::Other(anyhow::anyhow!(
2656 0 : "unexpected pending WAL record"
2657 0 : )))
2658 : };
2659 0 : }
2660 135329 : }
2661 : } else {
2662 : // This is an expensive check, so we only do it in debug mode. If reading a data key,
2663 : // this key should never be present in pending_data_pages. We ensure this by committing
2664 : // modifications before ingesting DB create operations, which are the only kind that reads
2665 : // data pages during ingest.
2666 0 : if cfg!(debug_assertions) {
2667 0 : assert!(
2668 0 : !self
2669 0 : .pending_data_batch
2670 0 : .as_ref()
2671 0 : .is_some_and(|b| b.updates_key(&key))
2672 0 : );
2673 0 : }
2674 : }
2675 :
2676 : // Metadata page cache miss, or we're reading a data page.
2677 135329 : let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
2678 135329 : self.tline.get(key, lsn, ctx).await
2679 143293 : }
2680 :
2681 : /// Get a key from the sparse keyspace. Automatically converts the missing key error
2682 : /// and the empty value into None.
2683 0 : async fn sparse_get(
2684 0 : &self,
2685 0 : key: Key,
2686 0 : ctx: &RequestContext,
2687 0 : ) -> Result<Option<Bytes>, PageReconstructError> {
2688 0 : let val = self.get(key, ctx).await;
2689 0 : match val {
2690 0 : Ok(val) if val.is_empty() => Ok(None),
2691 0 : Ok(val) => Ok(Some(val)),
2692 0 : Err(PageReconstructError::MissingKey(_)) => Ok(None),
2693 0 : Err(e) => Err(e),
2694 : }
2695 0 : }
2696 :
2697 : #[cfg(test)]
2698 2 : pub fn put_for_unit_test(&mut self, key: Key, val: Value) {
2699 2 : self.put(key, val);
2700 2 : }
2701 :
2702 282071 : fn put(&mut self, key: Key, val: Value) {
2703 282071 : if Self::is_data_key(&key) {
2704 138934 : self.put_data(key.to_compact(), val)
2705 : } else {
2706 143137 : self.put_metadata(key.to_compact(), val)
2707 : }
2708 282071 : }
2709 :
2710 138934 : fn put_data(&mut self, key: CompactKey, val: Value) {
2711 138934 : let batch = self
2712 138934 : .pending_data_batch
2713 138934 : .get_or_insert_with(SerializedValueBatch::default);
2714 138934 : batch.put(key, val, self.lsn);
2715 138934 : }
2716 :
2717 143137 : fn put_metadata(&mut self, key: CompactKey, val: Value) {
2718 143137 : let values = self.pending_metadata_pages.entry(key).or_default();
2719 : // Replace the previous value if it exists at the same lsn
2720 143137 : if let Some((last_lsn, last_value_ser_size, last_value)) = values.last_mut() {
2721 6076 : if *last_lsn == self.lsn {
2722 : // Update the pending_metadata_bytes contribution from this entry, and update the serialized size in place
2723 6076 : self.pending_metadata_bytes -= *last_value_ser_size;
2724 6076 : *last_value_ser_size = val.serialized_size().unwrap() as usize;
2725 6076 : self.pending_metadata_bytes += *last_value_ser_size;
2726 6076 :
2727 6076 : // Use the latest value, this replaces any earlier write to the same (key,lsn), such as much
2728 6076 : // have been generated by synthesized zero page writes prior to the first real write to a page.
2729 6076 : *last_value = val;
2730 6076 : return;
2731 0 : }
2732 137061 : }
2733 :
2734 137061 : let val_serialized_size = val.serialized_size().unwrap() as usize;
2735 137061 : self.pending_metadata_bytes += val_serialized_size;
2736 137061 : values.push((self.lsn, val_serialized_size, val));
2737 137061 :
2738 137061 : if key == CHECKPOINT_KEY.to_compact() {
2739 118 : tracing::debug!("Checkpoint key added to pending with size {val_serialized_size}");
2740 136943 : }
2741 143137 : }
2742 :
2743 1 : fn delete(&mut self, key_range: Range<Key>) {
2744 1 : trace!("DELETE {}-{}", key_range.start, key_range.end);
2745 1 : self.pending_deletions.push((key_range, self.lsn));
2746 1 : }
2747 : }
2748 :
2749 : /// Statistics for a DatadirModification.
2750 : #[derive(Default)]
2751 : pub struct DatadirModificationStats {
2752 : pub metadata_images: u64,
2753 : pub metadata_deltas: u64,
2754 : pub data_images: u64,
2755 : pub data_deltas: u64,
2756 : }
2757 :
2758 : /// This struct facilitates accessing either a committed key from the timeline at a
2759 : /// specific LSN, or the latest uncommitted key from a pending modification.
2760 : ///
2761 : /// During WAL ingestion, the records from multiple LSNs may be batched in the same
2762 : /// modification before being flushed to the timeline. Hence, the routines in WalIngest
2763 : /// need to look up the keys in the modification first before looking them up in the
2764 : /// timeline to not miss the latest updates.
2765 : #[derive(Clone, Copy)]
2766 : pub enum Version<'a> {
2767 : LsnRange(LsnRange),
2768 : Modified(&'a DatadirModification<'a>),
2769 : }
2770 :
2771 : impl Version<'_> {
2772 25 : async fn get(
2773 25 : &self,
2774 25 : timeline: &Timeline,
2775 25 : key: Key,
2776 25 : ctx: &RequestContext,
2777 25 : ) -> Result<Bytes, PageReconstructError> {
2778 25 : match self {
2779 15 : Version::LsnRange(lsns) => timeline.get(key, lsns.effective_lsn, ctx).await,
2780 10 : Version::Modified(modification) => modification.get(key, ctx).await,
2781 : }
2782 25 : }
2783 :
2784 : /// Get a key from the sparse keyspace. Automatically converts the missing key error
2785 : /// and the empty value into None.
2786 0 : async fn sparse_get(
2787 0 : &self,
2788 0 : timeline: &Timeline,
2789 0 : key: Key,
2790 0 : ctx: &RequestContext,
2791 0 : ) -> Result<Option<Bytes>, PageReconstructError> {
2792 0 : let val = self.get(timeline, key, ctx).await;
2793 0 : match val {
2794 0 : Ok(val) if val.is_empty() => Ok(None),
2795 0 : Ok(val) => Ok(Some(val)),
2796 0 : Err(PageReconstructError::MissingKey(_)) => Ok(None),
2797 0 : Err(e) => Err(e),
2798 : }
2799 0 : }
2800 :
2801 26 : pub fn is_latest(&self) -> bool {
2802 26 : match self {
2803 16 : Version::LsnRange(lsns) => lsns.is_latest(),
2804 10 : Version::Modified(_) => true,
2805 : }
2806 26 : }
2807 :
2808 224275 : pub fn get_lsn(&self) -> Lsn {
2809 224275 : match self {
2810 12224 : Version::LsnRange(lsns) => lsns.effective_lsn,
2811 212051 : Version::Modified(modification) => modification.lsn,
2812 : }
2813 224275 : }
2814 :
2815 12219 : pub fn at(lsn: Lsn) -> Self {
2816 12219 : Version::LsnRange(LsnRange {
2817 12219 : effective_lsn: lsn,
2818 12219 : request_lsn: lsn,
2819 12219 : })
2820 12219 : }
2821 : }
2822 :
2823 : //--- Metadata structs stored in key-value pairs in the repository.
2824 :
2825 0 : #[derive(Debug, Serialize, Deserialize)]
2826 : pub(crate) struct DbDirectory {
2827 : // (spcnode, dbnode) -> (do relmapper and PG_VERSION files exist)
2828 : pub(crate) dbdirs: HashMap<(Oid, Oid), bool>,
2829 : }
2830 :
2831 : // The format of TwoPhaseDirectory changed in PostgreSQL v17, because the filenames of
2832 : // pg_twophase files was expanded from 32-bit XIDs to 64-bit XIDs. Previously, the files
2833 : // were named like "pg_twophase/000002E5", now they're like
2834 : // "pg_twophsae/0000000A000002E4".
2835 :
2836 0 : #[derive(Debug, Serialize, Deserialize)]
2837 : pub(crate) struct TwoPhaseDirectory {
2838 : pub(crate) xids: HashSet<TransactionId>,
2839 : }
2840 :
2841 0 : #[derive(Debug, Serialize, Deserialize)]
2842 : struct TwoPhaseDirectoryV17 {
2843 : xids: HashSet<u64>,
2844 : }
2845 :
2846 0 : #[derive(Debug, Serialize, Deserialize, Default)]
2847 : pub(crate) struct RelDirectory {
2848 : // Set of relations that exist. (relfilenode, forknum)
2849 : //
2850 : // TODO: Store it as a btree or radix tree or something else that spans multiple
2851 : // key-value pairs, if you have a lot of relations
2852 : pub(crate) rels: HashSet<(Oid, u8)>,
2853 : }
2854 :
2855 0 : #[derive(Debug, Serialize, Deserialize)]
2856 : struct RelSizeEntry {
2857 : nblocks: u32,
2858 : }
2859 :
2860 0 : #[derive(Debug, Serialize, Deserialize, Default)]
2861 : pub(crate) struct SlruSegmentDirectory {
2862 : // Set of SLRU segments that exist.
2863 : pub(crate) segments: HashSet<u32>,
2864 : }
2865 :
2866 : #[derive(Copy, Clone, PartialEq, Eq, Debug, enum_map::Enum)]
2867 : #[repr(u8)]
2868 : pub(crate) enum DirectoryKind {
2869 : Db,
2870 : TwoPhase,
2871 : Rel,
2872 : AuxFiles,
2873 : SlruSegment(SlruKind),
2874 : RelV2,
2875 : }
2876 :
2877 : impl DirectoryKind {
2878 : pub(crate) const KINDS_NUM: usize = <DirectoryKind as Enum>::LENGTH;
2879 4567 : pub(crate) fn offset(&self) -> usize {
2880 4567 : self.into_usize()
2881 4567 : }
2882 : }
2883 :
2884 : static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; BLCKSZ as usize]);
2885 :
2886 : #[allow(clippy::bool_assert_comparison)]
2887 : #[cfg(test)]
2888 : mod tests {
2889 : use hex_literal::hex;
2890 : use pageserver_api::models::ShardParameters;
2891 : use pageserver_api::shard::ShardStripeSize;
2892 : use utils::id::TimelineId;
2893 : use utils::shard::{ShardCount, ShardNumber};
2894 :
2895 : use super::*;
2896 : use crate::DEFAULT_PG_VERSION;
2897 : use crate::tenant::harness::TenantHarness;
2898 :
2899 : /// Test a round trip of aux file updates, from DatadirModification to reading back from the Timeline
2900 : #[tokio::test]
2901 1 : async fn aux_files_round_trip() -> anyhow::Result<()> {
2902 1 : let name = "aux_files_round_trip";
2903 1 : let harness = TenantHarness::create(name).await?;
2904 1 :
2905 1 : pub const TIMELINE_ID: TimelineId =
2906 1 : TimelineId::from_array(hex!("11223344556677881122334455667788"));
2907 1 :
2908 1 : let (tenant, ctx) = harness.load().await;
2909 1 : let (tline, ctx) = tenant
2910 1 : .create_empty_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
2911 1 : .await?;
2912 1 : let tline = tline.raw_timeline().unwrap();
2913 1 :
2914 1 : // First modification: insert two keys
2915 1 : let mut modification = tline.begin_modification(Lsn(0x1000));
2916 1 : modification.put_file("foo/bar1", b"content1", &ctx).await?;
2917 1 : modification.set_lsn(Lsn(0x1008))?;
2918 1 : modification.put_file("foo/bar2", b"content2", &ctx).await?;
2919 1 : modification.commit(&ctx).await?;
2920 1 : let expect_1008 = HashMap::from([
2921 1 : ("foo/bar1".to_string(), Bytes::from_static(b"content1")),
2922 1 : ("foo/bar2".to_string(), Bytes::from_static(b"content2")),
2923 1 : ]);
2924 1 :
2925 1 : let io_concurrency = IoConcurrency::spawn_for_test();
2926 1 :
2927 1 : let readback = tline
2928 1 : .list_aux_files(Lsn(0x1008), &ctx, io_concurrency.clone())
2929 1 : .await?;
2930 1 : assert_eq!(readback, expect_1008);
2931 1 :
2932 1 : // Second modification: update one key, remove the other
2933 1 : let mut modification = tline.begin_modification(Lsn(0x2000));
2934 1 : modification.put_file("foo/bar1", b"content3", &ctx).await?;
2935 1 : modification.set_lsn(Lsn(0x2008))?;
2936 1 : modification.put_file("foo/bar2", b"", &ctx).await?;
2937 1 : modification.commit(&ctx).await?;
2938 1 : let expect_2008 =
2939 1 : HashMap::from([("foo/bar1".to_string(), Bytes::from_static(b"content3"))]);
2940 1 :
2941 1 : let readback = tline
2942 1 : .list_aux_files(Lsn(0x2008), &ctx, io_concurrency.clone())
2943 1 : .await?;
2944 1 : assert_eq!(readback, expect_2008);
2945 1 :
2946 1 : // Reading back in time works
2947 1 : let readback = tline
2948 1 : .list_aux_files(Lsn(0x1008), &ctx, io_concurrency.clone())
2949 1 : .await?;
2950 1 : assert_eq!(readback, expect_1008);
2951 1 :
2952 1 : Ok(())
2953 1 : }
2954 :
2955 : #[test]
2956 1 : fn gap_finding() {
2957 1 : let rel = RelTag {
2958 1 : spcnode: 1663,
2959 1 : dbnode: 208101,
2960 1 : relnode: 2620,
2961 1 : forknum: 0,
2962 1 : };
2963 1 : let base_blkno = 1;
2964 1 :
2965 1 : let base_key = rel_block_to_key(rel, base_blkno);
2966 1 : let before_base_key = rel_block_to_key(rel, base_blkno - 1);
2967 1 :
2968 1 : let shard = ShardIdentity::unsharded();
2969 1 :
2970 1 : let mut previous_nblocks = 0;
2971 11 : for i in 0..10 {
2972 10 : let crnt_blkno = base_blkno + i;
2973 10 : let gaps = DatadirModification::find_gaps(rel, crnt_blkno, previous_nblocks, &shard);
2974 10 :
2975 10 : previous_nblocks = crnt_blkno + 1;
2976 10 :
2977 10 : if i == 0 {
2978 : // The first block we write is 1, so we should find the gap.
2979 1 : assert_eq!(gaps.unwrap(), KeySpace::single(before_base_key..base_key));
2980 : } else {
2981 9 : assert!(gaps.is_none());
2982 : }
2983 : }
2984 :
2985 : // This is an update to an already existing block. No gaps here.
2986 1 : let update_blkno = 5;
2987 1 : let gaps = DatadirModification::find_gaps(rel, update_blkno, previous_nblocks, &shard);
2988 1 : assert!(gaps.is_none());
2989 :
2990 : // This is an update past the current end block.
2991 1 : let after_gap_blkno = 20;
2992 1 : let gaps = DatadirModification::find_gaps(rel, after_gap_blkno, previous_nblocks, &shard);
2993 1 :
2994 1 : let gap_start_key = rel_block_to_key(rel, previous_nblocks);
2995 1 : let after_gap_key = rel_block_to_key(rel, after_gap_blkno);
2996 1 : assert_eq!(
2997 1 : gaps.unwrap(),
2998 1 : KeySpace::single(gap_start_key..after_gap_key)
2999 1 : );
3000 1 : }
3001 :
3002 : #[test]
3003 1 : fn sharded_gap_finding() {
3004 1 : let rel = RelTag {
3005 1 : spcnode: 1663,
3006 1 : dbnode: 208101,
3007 1 : relnode: 2620,
3008 1 : forknum: 0,
3009 1 : };
3010 1 :
3011 1 : let first_blkno = 6;
3012 1 :
3013 1 : // This shard will get the even blocks
3014 1 : let shard = ShardIdentity::from_params(
3015 1 : ShardNumber(0),
3016 1 : &ShardParameters {
3017 1 : count: ShardCount(2),
3018 1 : stripe_size: ShardStripeSize(1),
3019 1 : },
3020 1 : );
3021 1 :
3022 1 : // Only keys belonging to this shard are considered as gaps.
3023 1 : let mut previous_nblocks = 0;
3024 1 : let gaps =
3025 1 : DatadirModification::find_gaps(rel, first_blkno, previous_nblocks, &shard).unwrap();
3026 1 : assert!(!gaps.ranges.is_empty());
3027 3 : for gap_range in gaps.ranges {
3028 2 : let mut k = gap_range.start;
3029 4 : while k != gap_range.end {
3030 2 : assert_eq!(shard.get_shard_number(&k), shard.number);
3031 2 : k = k.next();
3032 : }
3033 : }
3034 :
3035 1 : previous_nblocks = first_blkno;
3036 1 :
3037 1 : let update_blkno = 2;
3038 1 : let gaps = DatadirModification::find_gaps(rel, update_blkno, previous_nblocks, &shard);
3039 1 : assert!(gaps.is_none());
3040 1 : }
3041 :
3042 : /*
3043 : fn assert_current_logical_size<R: Repository>(timeline: &DatadirTimeline<R>, lsn: Lsn) {
3044 : let incremental = timeline.get_current_logical_size();
3045 : let non_incremental = timeline
3046 : .get_current_logical_size_non_incremental(lsn)
3047 : .unwrap();
3048 : assert_eq!(incremental, non_incremental);
3049 : }
3050 : */
3051 :
3052 : /*
3053 : ///
3054 : /// Test list_rels() function, with branches and dropped relations
3055 : ///
3056 : #[test]
3057 : fn test_list_rels_drop() -> Result<()> {
3058 : let repo = RepoHarness::create("test_list_rels_drop")?.load();
3059 : let tline = create_empty_timeline(repo, TIMELINE_ID)?;
3060 : const TESTDB: u32 = 111;
3061 :
3062 : // Import initial dummy checkpoint record, otherwise the get_timeline() call
3063 : // after branching fails below
3064 : let mut writer = tline.begin_record(Lsn(0x10));
3065 : writer.put_checkpoint(ZERO_CHECKPOINT.clone())?;
3066 : writer.finish()?;
3067 :
3068 : // Create a relation on the timeline
3069 : let mut writer = tline.begin_record(Lsn(0x20));
3070 : writer.put_rel_page_image(TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"))?;
3071 : writer.finish()?;
3072 :
3073 : let writer = tline.begin_record(Lsn(0x00));
3074 : writer.finish()?;
3075 :
3076 : // Check that list_rels() lists it after LSN 2, but no before it
3077 : assert!(!tline.list_rels(0, TESTDB, Lsn(0x10))?.contains(&TESTREL_A));
3078 : assert!(tline.list_rels(0, TESTDB, Lsn(0x20))?.contains(&TESTREL_A));
3079 : assert!(tline.list_rels(0, TESTDB, Lsn(0x30))?.contains(&TESTREL_A));
3080 :
3081 : // Create a branch, check that the relation is visible there
3082 : repo.branch_timeline(&tline, NEW_TIMELINE_ID, Lsn(0x30))?;
3083 : let newtline = match repo.get_timeline(NEW_TIMELINE_ID)?.local_timeline() {
3084 : Some(timeline) => timeline,
3085 : None => panic!("Should have a local timeline"),
3086 : };
3087 : let newtline = DatadirTimelineImpl::new(newtline);
3088 : assert!(newtline
3089 : .list_rels(0, TESTDB, Lsn(0x30))?
3090 : .contains(&TESTREL_A));
3091 :
3092 : // Drop it on the branch
3093 : let mut new_writer = newtline.begin_record(Lsn(0x40));
3094 : new_writer.drop_relation(TESTREL_A)?;
3095 : new_writer.finish()?;
3096 :
3097 : // Check that it's no longer listed on the branch after the point where it was dropped
3098 : assert!(newtline
3099 : .list_rels(0, TESTDB, Lsn(0x30))?
3100 : .contains(&TESTREL_A));
3101 : assert!(!newtline
3102 : .list_rels(0, TESTDB, Lsn(0x40))?
3103 : .contains(&TESTREL_A));
3104 :
3105 : // Run checkpoint and garbage collection and check that it's still not visible
3106 : newtline.checkpoint(CheckpointConfig::Forced)?;
3107 : repo.gc_iteration(Some(NEW_TIMELINE_ID), 0, true)?;
3108 :
3109 : assert!(!newtline
3110 : .list_rels(0, TESTDB, Lsn(0x40))?
3111 : .contains(&TESTREL_A));
3112 :
3113 : Ok(())
3114 : }
3115 : */
3116 :
3117 : /*
3118 : #[test]
3119 : fn test_read_beyond_eof() -> Result<()> {
3120 : let repo = RepoHarness::create("test_read_beyond_eof")?.load();
3121 : let tline = create_test_timeline(repo, TIMELINE_ID)?;
3122 :
3123 : make_some_layers(&tline, Lsn(0x20))?;
3124 : let mut writer = tline.begin_record(Lsn(0x60));
3125 : walingest.put_rel_page_image(
3126 : &mut writer,
3127 : TESTREL_A,
3128 : 0,
3129 : TEST_IMG(&format!("foo blk 0 at {}", Lsn(0x60))),
3130 : )?;
3131 : writer.finish()?;
3132 :
3133 : // Test read before rel creation. Should error out.
3134 : assert!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x10), false).is_err());
3135 :
3136 : // Read block beyond end of relation at different points in time.
3137 : // These reads should fall into different delta, image, and in-memory layers.
3138 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x20), false)?, ZERO_PAGE);
3139 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x25), false)?, ZERO_PAGE);
3140 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x30), false)?, ZERO_PAGE);
3141 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x35), false)?, ZERO_PAGE);
3142 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x40), false)?, ZERO_PAGE);
3143 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x45), false)?, ZERO_PAGE);
3144 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x50), false)?, ZERO_PAGE);
3145 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x55), false)?, ZERO_PAGE);
3146 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x60), false)?, ZERO_PAGE);
3147 :
3148 : // Test on an in-memory layer with no preceding layer
3149 : let mut writer = tline.begin_record(Lsn(0x70));
3150 : walingest.put_rel_page_image(
3151 : &mut writer,
3152 : TESTREL_B,
3153 : 0,
3154 : TEST_IMG(&format!("foo blk 0 at {}", Lsn(0x70))),
3155 : )?;
3156 : writer.finish()?;
3157 :
3158 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_B, 1, Lsn(0x70), false)?6, ZERO_PAGE);
3159 :
3160 : Ok(())
3161 : }
3162 : */
3163 : }
|
