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 super::tenant::{PageReconstructError, Timeline};
10 : use crate::context::RequestContext;
11 : use crate::keyspace::{KeySpace, KeySpaceAccum};
12 : use crate::repository::*;
13 : use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
14 : use crate::walrecord::NeonWalRecord;
15 : use anyhow::{ensure, Context};
16 : use bytes::{Buf, Bytes, BytesMut};
17 : use enum_map::Enum;
18 : use itertools::Itertools;
19 : use pageserver_api::key::{
20 : dbdir_key_range, is_rel_block_key, is_slru_block_key, rel_block_to_key, rel_dir_to_key,
21 : rel_key_range, rel_size_to_key, relmap_file_key, 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 : AUX_FILES_KEY, CHECKPOINT_KEY, CONTROLFILE_KEY, DBDIR_KEY, TWOPHASEDIR_KEY,
24 : };
25 : use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
26 : use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
27 : use postgres_ffi::BLCKSZ;
28 : use postgres_ffi::{Oid, TimestampTz, TransactionId};
29 : use serde::{Deserialize, Serialize};
30 : use std::collections::{hash_map, HashMap, HashSet};
31 : use std::ops::ControlFlow;
32 : use std::ops::Range;
33 : use strum::IntoEnumIterator;
34 : use tokio_util::sync::CancellationToken;
35 : use tracing::{debug, trace, warn};
36 : use utils::bin_ser::DeserializeError;
37 : use utils::{bin_ser::BeSer, lsn::Lsn};
38 :
39 0 : #[derive(Debug)]
40 : pub enum LsnForTimestamp {
41 : /// Found commits both before and after the given timestamp
42 : Present(Lsn),
43 :
44 : /// Found no commits after the given timestamp, this means
45 : /// that the newest data in the branch is older than the given
46 : /// timestamp.
47 : ///
48 : /// All commits <= LSN happened before the given timestamp
49 : Future(Lsn),
50 :
51 : /// The queried timestamp is past our horizon we look back at (PITR)
52 : ///
53 : /// All commits > LSN happened after the given timestamp,
54 : /// but any commits < LSN might have happened before or after
55 : /// the given timestamp. We don't know because no data before
56 : /// the given lsn is available.
57 : Past(Lsn),
58 :
59 : /// We have found no commit with a timestamp,
60 : /// so we can't return anything meaningful.
61 : ///
62 : /// The associated LSN is the lower bound value we can safely
63 : /// create branches on, but no statement is made if it is
64 : /// older or newer than the timestamp.
65 : ///
66 : /// This variant can e.g. be returned right after a
67 : /// cluster import.
68 : NoData(Lsn),
69 : }
70 :
71 0 : #[derive(Debug, thiserror::Error)]
72 : pub enum CalculateLogicalSizeError {
73 : #[error("cancelled")]
74 : Cancelled,
75 : #[error(transparent)]
76 : Other(#[from] anyhow::Error),
77 : }
78 :
79 0 : #[derive(Debug, thiserror::Error)]
80 : pub(crate) enum CollectKeySpaceError {
81 : #[error(transparent)]
82 : Decode(#[from] DeserializeError),
83 : #[error(transparent)]
84 : PageRead(PageReconstructError),
85 : #[error("cancelled")]
86 : Cancelled,
87 : }
88 :
89 : impl From<PageReconstructError> for CollectKeySpaceError {
90 0 : fn from(err: PageReconstructError) -> Self {
91 0 : match err {
92 0 : PageReconstructError::Cancelled => Self::Cancelled,
93 0 : err => Self::PageRead(err),
94 : }
95 0 : }
96 : }
97 :
98 : impl From<PageReconstructError> for CalculateLogicalSizeError {
99 0 : fn from(pre: PageReconstructError) -> Self {
100 0 : match pre {
101 : PageReconstructError::AncestorStopping(_) | PageReconstructError::Cancelled => {
102 0 : Self::Cancelled
103 : }
104 0 : _ => Self::Other(pre.into()),
105 : }
106 0 : }
107 : }
108 :
109 0 : #[derive(Debug, thiserror::Error)]
110 : pub enum RelationError {
111 : #[error("Relation Already Exists")]
112 : AlreadyExists,
113 : #[error("invalid relnode")]
114 : InvalidRelnode,
115 : #[error(transparent)]
116 : Other(#[from] anyhow::Error),
117 : }
118 :
119 : ///
120 : /// This impl provides all the functionality to store PostgreSQL relations, SLRUs,
121 : /// and other special kinds of files, in a versioned key-value store. The
122 : /// Timeline struct provides the key-value store.
123 : ///
124 : /// This is a separate impl, so that we can easily include all these functions in a Timeline
125 : /// implementation, and might be moved into a separate struct later.
126 : impl Timeline {
127 : /// Start ingesting a WAL record, or other atomic modification of
128 : /// the timeline.
129 : ///
130 : /// This provides a transaction-like interface to perform a bunch
131 : /// of modifications atomically.
132 : ///
133 : /// To ingest a WAL record, call begin_modification(lsn) to get a
134 : /// DatadirModification object. Use the functions in the object to
135 : /// modify the repository state, updating all the pages and metadata
136 : /// that the WAL record affects. When you're done, call commit() to
137 : /// commit the changes.
138 : ///
139 : /// Lsn stored in modification is advanced by `ingest_record` and
140 : /// is used by `commit()` to update `last_record_lsn`.
141 : ///
142 : /// Calling commit() will flush all the changes and reset the state,
143 : /// so the `DatadirModification` struct can be reused to perform the next modification.
144 : ///
145 : /// Note that any pending modifications you make through the
146 : /// modification object won't be visible to calls to the 'get' and list
147 : /// functions of the timeline until you finish! And if you update the
148 : /// same page twice, the last update wins.
149 : ///
150 268268 : pub fn begin_modification(&self, lsn: Lsn) -> DatadirModification
151 268268 : where
152 268268 : Self: Sized,
153 268268 : {
154 268268 : DatadirModification {
155 268268 : tline: self,
156 268268 : pending_lsns: Vec::new(),
157 268268 : pending_updates: HashMap::new(),
158 268268 : pending_deletions: Vec::new(),
159 268268 : pending_nblocks: 0,
160 268268 : pending_aux_files: None,
161 268268 : pending_directory_entries: Vec::new(),
162 268268 : lsn,
163 268268 : }
164 268268 : }
165 :
166 : //------------------------------------------------------------------------------
167 : // Public GET functions
168 : //------------------------------------------------------------------------------
169 :
170 : /// Look up given page version.
171 18384 : pub(crate) async fn get_rel_page_at_lsn(
172 18384 : &self,
173 18384 : tag: RelTag,
174 18384 : blknum: BlockNumber,
175 18384 : version: Version<'_>,
176 18384 : latest: bool,
177 18384 : ctx: &RequestContext,
178 18384 : ) -> Result<Bytes, PageReconstructError> {
179 18384 : if tag.relnode == 0 {
180 0 : return Err(PageReconstructError::Other(
181 0 : RelationError::InvalidRelnode.into(),
182 0 : ));
183 18384 : }
184 :
185 18384 : let nblocks = self.get_rel_size(tag, version, latest, ctx).await?;
186 18384 : if blknum >= nblocks {
187 0 : debug!(
188 0 : "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
189 0 : tag,
190 0 : blknum,
191 0 : version.get_lsn(),
192 0 : nblocks
193 0 : );
194 0 : return Ok(ZERO_PAGE.clone());
195 18384 : }
196 18384 :
197 18384 : let key = rel_block_to_key(tag, blknum);
198 18384 : version.get(self, key, ctx).await
199 18384 : }
200 :
201 : // Get size of a database in blocks
202 0 : pub(crate) async fn get_db_size(
203 0 : &self,
204 0 : spcnode: Oid,
205 0 : dbnode: Oid,
206 0 : version: Version<'_>,
207 0 : latest: bool,
208 0 : ctx: &RequestContext,
209 0 : ) -> Result<usize, PageReconstructError> {
210 0 : let mut total_blocks = 0;
211 :
212 0 : let rels = self.list_rels(spcnode, dbnode, version, ctx).await?;
213 :
214 0 : for rel in rels {
215 0 : let n_blocks = self.get_rel_size(rel, version, latest, ctx).await?;
216 0 : total_blocks += n_blocks as usize;
217 : }
218 0 : Ok(total_blocks)
219 0 : }
220 :
221 : /// Get size of a relation file
222 24434 : pub(crate) async fn get_rel_size(
223 24434 : &self,
224 24434 : tag: RelTag,
225 24434 : version: Version<'_>,
226 24434 : latest: bool,
227 24434 : ctx: &RequestContext,
228 24434 : ) -> Result<BlockNumber, PageReconstructError> {
229 24434 : if tag.relnode == 0 {
230 0 : return Err(PageReconstructError::Other(
231 0 : RelationError::InvalidRelnode.into(),
232 0 : ));
233 24434 : }
234 :
235 24434 : if let Some(nblocks) = self.get_cached_rel_size(&tag, version.get_lsn()) {
236 19294 : return Ok(nblocks);
237 5140 : }
238 5140 :
239 5140 : if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
240 0 : && !self.get_rel_exists(tag, version, latest, ctx).await?
241 : {
242 : // FIXME: Postgres sometimes calls smgrcreate() to create
243 : // FSM, and smgrnblocks() on it immediately afterwards,
244 : // without extending it. Tolerate that by claiming that
245 : // any non-existent FSM fork has size 0.
246 0 : return Ok(0);
247 5140 : }
248 5140 :
249 5140 : let key = rel_size_to_key(tag);
250 5140 : let mut buf = version.get(self, key, ctx).await?;
251 5136 : let nblocks = buf.get_u32_le();
252 5136 :
253 5136 : if latest {
254 0 : // Update relation size cache only if "latest" flag is set.
255 0 : // This flag is set by compute when it is working with most recent version of relation.
256 0 : // Typically master compute node always set latest=true.
257 0 : // Please notice, that even if compute node "by mistake" specifies old LSN but set
258 0 : // latest=true, then it can not cause cache corruption, because with latest=true
259 0 : // pageserver choose max(request_lsn, last_written_lsn) and so cached value will be
260 0 : // associated with most recent value of LSN.
261 0 : self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
262 5136 : }
263 5136 : Ok(nblocks)
264 24434 : }
265 :
266 : /// Does relation exist?
267 6050 : pub(crate) async fn get_rel_exists(
268 6050 : &self,
269 6050 : tag: RelTag,
270 6050 : version: Version<'_>,
271 6050 : _latest: bool,
272 6050 : ctx: &RequestContext,
273 6050 : ) -> Result<bool, PageReconstructError> {
274 6050 : if tag.relnode == 0 {
275 0 : return Err(PageReconstructError::Other(
276 0 : RelationError::InvalidRelnode.into(),
277 0 : ));
278 6050 : }
279 :
280 : // first try to lookup relation in cache
281 6050 : if let Some(_nblocks) = self.get_cached_rel_size(&tag, version.get_lsn()) {
282 6032 : return Ok(true);
283 18 : }
284 18 : // fetch directory listing
285 18 : let key = rel_dir_to_key(tag.spcnode, tag.dbnode);
286 18 : let buf = version.get(self, key, ctx).await?;
287 :
288 18 : match RelDirectory::des(&buf).context("deserialization failure") {
289 18 : Ok(dir) => {
290 18 : let exists = dir.rels.get(&(tag.relnode, tag.forknum)).is_some();
291 18 : Ok(exists)
292 : }
293 0 : Err(e) => Err(PageReconstructError::from(e)),
294 : }
295 6050 : }
296 :
297 : /// Get a list of all existing relations in given tablespace and database.
298 : ///
299 : /// # Cancel-Safety
300 : ///
301 : /// This method is cancellation-safe.
302 0 : pub(crate) async fn list_rels(
303 0 : &self,
304 0 : spcnode: Oid,
305 0 : dbnode: Oid,
306 0 : version: Version<'_>,
307 0 : ctx: &RequestContext,
308 0 : ) -> Result<HashSet<RelTag>, PageReconstructError> {
309 0 : // fetch directory listing
310 0 : let key = rel_dir_to_key(spcnode, dbnode);
311 0 : let buf = version.get(self, key, ctx).await?;
312 :
313 0 : match RelDirectory::des(&buf).context("deserialization failure") {
314 0 : Ok(dir) => {
315 0 : let rels: HashSet<RelTag> =
316 0 : HashSet::from_iter(dir.rels.iter().map(|(relnode, forknum)| RelTag {
317 0 : spcnode,
318 0 : dbnode,
319 0 : relnode: *relnode,
320 0 : forknum: *forknum,
321 0 : }));
322 0 :
323 0 : Ok(rels)
324 : }
325 0 : Err(e) => Err(PageReconstructError::from(e)),
326 : }
327 0 : }
328 :
329 : /// Get the whole SLRU segment
330 0 : pub(crate) async fn get_slru_segment(
331 0 : &self,
332 0 : kind: SlruKind,
333 0 : segno: u32,
334 0 : lsn: Lsn,
335 0 : ctx: &RequestContext,
336 0 : ) -> Result<Bytes, PageReconstructError> {
337 0 : let n_blocks = self
338 0 : .get_slru_segment_size(kind, segno, Version::Lsn(lsn), ctx)
339 0 : .await?;
340 0 : let mut segment = BytesMut::with_capacity(n_blocks as usize * BLCKSZ as usize);
341 0 : for blkno in 0..n_blocks {
342 0 : let block = self
343 0 : .get_slru_page_at_lsn(kind, segno, blkno, lsn, ctx)
344 0 : .await?;
345 0 : segment.extend_from_slice(&block[..BLCKSZ as usize]);
346 : }
347 0 : Ok(segment.freeze())
348 0 : }
349 :
350 : /// Look up given SLRU page version.
351 0 : pub(crate) async fn get_slru_page_at_lsn(
352 0 : &self,
353 0 : kind: SlruKind,
354 0 : segno: u32,
355 0 : blknum: BlockNumber,
356 0 : lsn: Lsn,
357 0 : ctx: &RequestContext,
358 0 : ) -> Result<Bytes, PageReconstructError> {
359 0 : let key = slru_block_to_key(kind, segno, blknum);
360 0 : self.get(key, lsn, ctx).await
361 0 : }
362 :
363 : /// Get size of an SLRU segment
364 0 : pub(crate) async fn get_slru_segment_size(
365 0 : &self,
366 0 : kind: SlruKind,
367 0 : segno: u32,
368 0 : version: Version<'_>,
369 0 : ctx: &RequestContext,
370 0 : ) -> Result<BlockNumber, PageReconstructError> {
371 0 : let key = slru_segment_size_to_key(kind, segno);
372 0 : let mut buf = version.get(self, key, ctx).await?;
373 0 : Ok(buf.get_u32_le())
374 0 : }
375 :
376 : /// Get size of an SLRU segment
377 0 : pub(crate) async fn get_slru_segment_exists(
378 0 : &self,
379 0 : kind: SlruKind,
380 0 : segno: u32,
381 0 : version: Version<'_>,
382 0 : ctx: &RequestContext,
383 0 : ) -> Result<bool, PageReconstructError> {
384 0 : // fetch directory listing
385 0 : let key = slru_dir_to_key(kind);
386 0 : let buf = version.get(self, key, ctx).await?;
387 :
388 0 : match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
389 0 : Ok(dir) => {
390 0 : let exists = dir.segments.get(&segno).is_some();
391 0 : Ok(exists)
392 : }
393 0 : Err(e) => Err(PageReconstructError::from(e)),
394 : }
395 0 : }
396 :
397 : /// Locate LSN, such that all transactions that committed before
398 : /// 'search_timestamp' are visible, but nothing newer is.
399 : ///
400 : /// This is not exact. Commit timestamps are not guaranteed to be ordered,
401 : /// so it's not well defined which LSN you get if there were multiple commits
402 : /// "in flight" at that point in time.
403 : ///
404 0 : pub(crate) async fn find_lsn_for_timestamp(
405 0 : &self,
406 0 : search_timestamp: TimestampTz,
407 0 : cancel: &CancellationToken,
408 0 : ctx: &RequestContext,
409 0 : ) -> Result<LsnForTimestamp, PageReconstructError> {
410 0 : let gc_cutoff_lsn_guard = self.get_latest_gc_cutoff_lsn();
411 0 : // We use this method to figure out the branching LSN for the new branch, but the
412 0 : // GC cutoff could be before the branching point and we cannot create a new branch
413 0 : // with LSN < `ancestor_lsn`. Thus, pick the maximum of these two to be
414 0 : // on the safe side.
415 0 : let min_lsn = std::cmp::max(*gc_cutoff_lsn_guard, self.get_ancestor_lsn());
416 0 : let max_lsn = self.get_last_record_lsn();
417 0 :
418 0 : // LSNs are always 8-byte aligned. low/mid/high represent the
419 0 : // LSN divided by 8.
420 0 : let mut low = min_lsn.0 / 8;
421 0 : let mut high = max_lsn.0 / 8 + 1;
422 0 :
423 0 : let mut found_smaller = false;
424 0 : let mut found_larger = false;
425 0 : while low < high {
426 0 : if cancel.is_cancelled() {
427 0 : return Err(PageReconstructError::Cancelled);
428 0 : }
429 0 : // cannot overflow, high and low are both smaller than u64::MAX / 2
430 0 : let mid = (high + low) / 2;
431 :
432 0 : let cmp = self
433 0 : .is_latest_commit_timestamp_ge_than(
434 0 : search_timestamp,
435 0 : Lsn(mid * 8),
436 0 : &mut found_smaller,
437 0 : &mut found_larger,
438 0 : ctx,
439 0 : )
440 0 : .await?;
441 :
442 0 : if cmp {
443 0 : high = mid;
444 0 : } else {
445 0 : low = mid + 1;
446 0 : }
447 : }
448 : // If `found_smaller == true`, `low = t + 1` where `t` is the target LSN,
449 : // so the LSN of the last commit record before or at `search_timestamp`.
450 : // Remove one from `low` to get `t`.
451 : //
452 : // FIXME: it would be better to get the LSN of the previous commit.
453 : // Otherwise, if you restore to the returned LSN, the database will
454 : // include physical changes from later commits that will be marked
455 : // as aborted, and will need to be vacuumed away.
456 0 : let commit_lsn = Lsn((low - 1) * 8);
457 0 : match (found_smaller, found_larger) {
458 : (false, false) => {
459 : // This can happen if no commit records have been processed yet, e.g.
460 : // just after importing a cluster.
461 0 : Ok(LsnForTimestamp::NoData(min_lsn))
462 : }
463 : (false, true) => {
464 : // Didn't find any commit timestamps smaller than the request
465 0 : Ok(LsnForTimestamp::Past(min_lsn))
466 : }
467 : (true, false) => {
468 : // Only found commits with timestamps smaller than the request.
469 : // It's still a valid case for branch creation, return it.
470 : // And `update_gc_info()` ignores LSN for a `LsnForTimestamp::Future`
471 : // case, anyway.
472 0 : Ok(LsnForTimestamp::Future(commit_lsn))
473 : }
474 0 : (true, true) => Ok(LsnForTimestamp::Present(commit_lsn)),
475 : }
476 0 : }
477 :
478 : /// Subroutine of find_lsn_for_timestamp(). Returns true, if there are any
479 : /// commits that committed after 'search_timestamp', at LSN 'probe_lsn'.
480 : ///
481 : /// Additionally, sets 'found_smaller'/'found_Larger, if encounters any commits
482 : /// with a smaller/larger timestamp.
483 : ///
484 0 : pub(crate) async fn is_latest_commit_timestamp_ge_than(
485 0 : &self,
486 0 : search_timestamp: TimestampTz,
487 0 : probe_lsn: Lsn,
488 0 : found_smaller: &mut bool,
489 0 : found_larger: &mut bool,
490 0 : ctx: &RequestContext,
491 0 : ) -> Result<bool, PageReconstructError> {
492 0 : self.map_all_timestamps(probe_lsn, ctx, |timestamp| {
493 0 : if timestamp >= search_timestamp {
494 0 : *found_larger = true;
495 0 : return ControlFlow::Break(true);
496 0 : } else {
497 0 : *found_smaller = true;
498 0 : }
499 0 : ControlFlow::Continue(())
500 0 : })
501 0 : .await
502 0 : }
503 :
504 : /// Obtain the possible timestamp range for the given lsn.
505 : ///
506 : /// If the lsn has no timestamps, returns None. returns `(min, max, median)` if it has timestamps.
507 0 : pub(crate) async fn get_timestamp_for_lsn(
508 0 : &self,
509 0 : probe_lsn: Lsn,
510 0 : ctx: &RequestContext,
511 0 : ) -> Result<Option<TimestampTz>, PageReconstructError> {
512 0 : let mut max: Option<TimestampTz> = None;
513 0 : self.map_all_timestamps(probe_lsn, ctx, |timestamp| {
514 0 : if let Some(max_prev) = max {
515 0 : max = Some(max_prev.max(timestamp));
516 0 : } else {
517 0 : max = Some(timestamp);
518 0 : }
519 0 : ControlFlow::Continue(())
520 0 : })
521 0 : .await?;
522 :
523 0 : Ok(max)
524 0 : }
525 :
526 : /// Runs the given function on all the timestamps for a given lsn
527 : ///
528 : /// The return value is either given by the closure, or set to the `Default`
529 : /// impl's output.
530 0 : async fn map_all_timestamps<T: Default>(
531 0 : &self,
532 0 : probe_lsn: Lsn,
533 0 : ctx: &RequestContext,
534 0 : mut f: impl FnMut(TimestampTz) -> ControlFlow<T>,
535 0 : ) -> Result<T, PageReconstructError> {
536 0 : for segno in self
537 0 : .list_slru_segments(SlruKind::Clog, Version::Lsn(probe_lsn), ctx)
538 0 : .await?
539 : {
540 0 : let nblocks = self
541 0 : .get_slru_segment_size(SlruKind::Clog, segno, Version::Lsn(probe_lsn), ctx)
542 0 : .await?;
543 0 : for blknum in (0..nblocks).rev() {
544 0 : let clog_page = self
545 0 : .get_slru_page_at_lsn(SlruKind::Clog, segno, blknum, probe_lsn, ctx)
546 0 : .await?;
547 :
548 0 : if clog_page.len() == BLCKSZ as usize + 8 {
549 0 : let mut timestamp_bytes = [0u8; 8];
550 0 : timestamp_bytes.copy_from_slice(&clog_page[BLCKSZ as usize..]);
551 0 : let timestamp = TimestampTz::from_be_bytes(timestamp_bytes);
552 0 :
553 0 : match f(timestamp) {
554 0 : ControlFlow::Break(b) => return Ok(b),
555 0 : ControlFlow::Continue(()) => (),
556 : }
557 0 : }
558 : }
559 : }
560 0 : Ok(Default::default())
561 0 : }
562 :
563 0 : pub(crate) async fn get_slru_keyspace(
564 0 : &self,
565 0 : version: Version<'_>,
566 0 : ctx: &RequestContext,
567 0 : ) -> Result<KeySpace, PageReconstructError> {
568 0 : let mut accum = KeySpaceAccum::new();
569 :
570 0 : for kind in SlruKind::iter() {
571 0 : let mut segments: Vec<u32> = self
572 0 : .list_slru_segments(kind, version, ctx)
573 0 : .await?
574 0 : .into_iter()
575 0 : .collect();
576 0 : segments.sort_unstable();
577 :
578 0 : for seg in segments {
579 0 : let block_count = self.get_slru_segment_size(kind, seg, version, ctx).await?;
580 :
581 0 : accum.add_range(
582 0 : slru_block_to_key(kind, seg, 0)..slru_block_to_key(kind, seg, block_count),
583 0 : );
584 : }
585 : }
586 :
587 0 : Ok(accum.to_keyspace())
588 0 : }
589 :
590 : /// Get a list of SLRU segments
591 0 : pub(crate) async fn list_slru_segments(
592 0 : &self,
593 0 : kind: SlruKind,
594 0 : version: Version<'_>,
595 0 : ctx: &RequestContext,
596 0 : ) -> Result<HashSet<u32>, PageReconstructError> {
597 0 : // fetch directory entry
598 0 : let key = slru_dir_to_key(kind);
599 :
600 0 : let buf = version.get(self, key, ctx).await?;
601 0 : match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
602 0 : Ok(dir) => Ok(dir.segments),
603 0 : Err(e) => Err(PageReconstructError::from(e)),
604 : }
605 0 : }
606 :
607 0 : pub(crate) async fn get_relmap_file(
608 0 : &self,
609 0 : spcnode: Oid,
610 0 : dbnode: Oid,
611 0 : version: Version<'_>,
612 0 : ctx: &RequestContext,
613 0 : ) -> Result<Bytes, PageReconstructError> {
614 0 : let key = relmap_file_key(spcnode, dbnode);
615 :
616 0 : let buf = version.get(self, key, ctx).await?;
617 0 : Ok(buf)
618 0 : }
619 :
620 0 : pub(crate) async fn list_dbdirs(
621 0 : &self,
622 0 : lsn: Lsn,
623 0 : ctx: &RequestContext,
624 0 : ) -> Result<HashMap<(Oid, Oid), bool>, PageReconstructError> {
625 : // fetch directory entry
626 0 : let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
627 :
628 0 : match DbDirectory::des(&buf).context("deserialization failure") {
629 0 : Ok(dir) => Ok(dir.dbdirs),
630 0 : Err(e) => Err(PageReconstructError::from(e)),
631 : }
632 0 : }
633 :
634 0 : pub(crate) async fn get_twophase_file(
635 0 : &self,
636 0 : xid: TransactionId,
637 0 : lsn: Lsn,
638 0 : ctx: &RequestContext,
639 0 : ) -> Result<Bytes, PageReconstructError> {
640 0 : let key = twophase_file_key(xid);
641 0 : let buf = self.get(key, lsn, ctx).await?;
642 0 : Ok(buf)
643 0 : }
644 :
645 0 : pub(crate) async fn list_twophase_files(
646 0 : &self,
647 0 : lsn: Lsn,
648 0 : ctx: &RequestContext,
649 0 : ) -> Result<HashSet<TransactionId>, PageReconstructError> {
650 : // fetch directory entry
651 0 : let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
652 :
653 0 : match TwoPhaseDirectory::des(&buf).context("deserialization failure") {
654 0 : Ok(dir) => Ok(dir.xids),
655 0 : Err(e) => Err(PageReconstructError::from(e)),
656 : }
657 0 : }
658 :
659 0 : pub(crate) async fn get_control_file(
660 0 : &self,
661 0 : lsn: Lsn,
662 0 : ctx: &RequestContext,
663 0 : ) -> Result<Bytes, PageReconstructError> {
664 0 : self.get(CONTROLFILE_KEY, lsn, ctx).await
665 0 : }
666 :
667 12 : pub(crate) async fn get_checkpoint(
668 12 : &self,
669 12 : lsn: Lsn,
670 12 : ctx: &RequestContext,
671 12 : ) -> Result<Bytes, PageReconstructError> {
672 12 : self.get(CHECKPOINT_KEY, lsn, ctx).await
673 12 : }
674 :
675 6 : pub(crate) async fn list_aux_files(
676 6 : &self,
677 6 : lsn: Lsn,
678 6 : ctx: &RequestContext,
679 6 : ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
680 6 : match self.get(AUX_FILES_KEY, lsn, ctx).await {
681 6 : Ok(buf) => match AuxFilesDirectory::des(&buf).context("deserialization failure") {
682 6 : Ok(dir) => Ok(dir.files),
683 0 : Err(e) => Err(PageReconstructError::from(e)),
684 : },
685 0 : Err(e) => {
686 : // This is expected: historical databases do not have the key.
687 0 : debug!("Failed to get info about AUX files: {}", e);
688 0 : Ok(HashMap::new())
689 : }
690 : }
691 6 : }
692 :
693 : /// Does the same as get_current_logical_size but counted on demand.
694 : /// Used to initialize the logical size tracking on startup.
695 : ///
696 : /// Only relation blocks are counted currently. That excludes metadata,
697 : /// SLRUs, twophase files etc.
698 : ///
699 : /// # Cancel-Safety
700 : ///
701 : /// This method is cancellation-safe.
702 0 : pub async fn get_current_logical_size_non_incremental(
703 0 : &self,
704 0 : lsn: Lsn,
705 0 : ctx: &RequestContext,
706 0 : ) -> Result<u64, CalculateLogicalSizeError> {
707 0 : debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id();
708 :
709 : // Fetch list of database dirs and iterate them
710 0 : let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
711 0 : let dbdir = DbDirectory::des(&buf).context("deserialize db directory")?;
712 :
713 0 : let mut total_size: u64 = 0;
714 0 : for (spcnode, dbnode) in dbdir.dbdirs.keys() {
715 0 : for rel in self
716 0 : .list_rels(*spcnode, *dbnode, Version::Lsn(lsn), ctx)
717 0 : .await?
718 : {
719 0 : if self.cancel.is_cancelled() {
720 0 : return Err(CalculateLogicalSizeError::Cancelled);
721 0 : }
722 0 : let relsize_key = rel_size_to_key(rel);
723 0 : let mut buf = self.get(relsize_key, lsn, ctx).await?;
724 0 : let relsize = buf.get_u32_le();
725 0 :
726 0 : total_size += relsize as u64;
727 : }
728 : }
729 0 : Ok(total_size * BLCKSZ as u64)
730 0 : }
731 :
732 : ///
733 : /// Get a KeySpace that covers all the Keys that are in use at the given LSN.
734 : /// Anything that's not listed maybe removed from the underlying storage (from
735 : /// that LSN forwards).
736 170 : pub(crate) async fn collect_keyspace(
737 170 : &self,
738 170 : lsn: Lsn,
739 170 : ctx: &RequestContext,
740 170 : ) -> Result<KeySpace, CollectKeySpaceError> {
741 170 : // Iterate through key ranges, greedily packing them into partitions
742 170 : let mut result = KeySpaceAccum::new();
743 170 :
744 170 : // The dbdir metadata always exists
745 170 : result.add_key(DBDIR_KEY);
746 :
747 : // Fetch list of database dirs and iterate them
748 2160 : let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
749 170 : let dbdir = DbDirectory::des(&buf)?;
750 :
751 170 : let mut dbs: Vec<(Oid, Oid)> = dbdir.dbdirs.keys().cloned().collect();
752 170 : dbs.sort_unstable();
753 170 : for (spcnode, dbnode) in dbs {
754 0 : result.add_key(relmap_file_key(spcnode, dbnode));
755 0 : result.add_key(rel_dir_to_key(spcnode, dbnode));
756 :
757 0 : let mut rels: Vec<RelTag> = self
758 0 : .list_rels(spcnode, dbnode, Version::Lsn(lsn), ctx)
759 0 : .await?
760 0 : .into_iter()
761 0 : .collect();
762 0 : rels.sort_unstable();
763 0 : for rel in rels {
764 0 : let relsize_key = rel_size_to_key(rel);
765 0 : let mut buf = self.get(relsize_key, lsn, ctx).await?;
766 0 : let relsize = buf.get_u32_le();
767 0 :
768 0 : result.add_range(rel_block_to_key(rel, 0)..rel_block_to_key(rel, relsize));
769 0 : result.add_key(relsize_key);
770 : }
771 : }
772 :
773 : // Iterate SLRUs next
774 510 : for kind in [
775 170 : SlruKind::Clog,
776 170 : SlruKind::MultiXactMembers,
777 170 : SlruKind::MultiXactOffsets,
778 : ] {
779 510 : let slrudir_key = slru_dir_to_key(kind);
780 510 : result.add_key(slrudir_key);
781 7958 : let buf = self.get(slrudir_key, lsn, ctx).await?;
782 510 : let dir = SlruSegmentDirectory::des(&buf)?;
783 510 : let mut segments: Vec<u32> = dir.segments.iter().cloned().collect();
784 510 : segments.sort_unstable();
785 510 : for segno in segments {
786 0 : let segsize_key = slru_segment_size_to_key(kind, segno);
787 0 : let mut buf = self.get(segsize_key, lsn, ctx).await?;
788 0 : let segsize = buf.get_u32_le();
789 0 :
790 0 : result.add_range(
791 0 : slru_block_to_key(kind, segno, 0)..slru_block_to_key(kind, segno, segsize),
792 0 : );
793 0 : result.add_key(segsize_key);
794 : }
795 : }
796 :
797 : // Then pg_twophase
798 170 : result.add_key(TWOPHASEDIR_KEY);
799 3054 : let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
800 170 : let twophase_dir = TwoPhaseDirectory::des(&buf)?;
801 170 : let mut xids: Vec<TransactionId> = twophase_dir.xids.iter().cloned().collect();
802 170 : xids.sort_unstable();
803 170 : for xid in xids {
804 0 : result.add_key(twophase_file_key(xid));
805 0 : }
806 :
807 170 : result.add_key(CONTROLFILE_KEY);
808 170 : result.add_key(CHECKPOINT_KEY);
809 170 : if self.get(AUX_FILES_KEY, lsn, ctx).await.is_ok() {
810 70 : result.add_key(AUX_FILES_KEY);
811 100 : }
812 170 : Ok(result.to_keyspace())
813 170 : }
814 :
815 : /// Get cached size of relation if it not updated after specified LSN
816 448540 : pub fn get_cached_rel_size(&self, tag: &RelTag, lsn: Lsn) -> Option<BlockNumber> {
817 448540 : let rel_size_cache = self.rel_size_cache.read().unwrap();
818 448540 : if let Some((cached_lsn, nblocks)) = rel_size_cache.get(tag) {
819 448518 : if lsn >= *cached_lsn {
820 443372 : return Some(*nblocks);
821 5146 : }
822 22 : }
823 5168 : None
824 448540 : }
825 :
826 : /// Update cached relation size if there is no more recent update
827 0 : pub fn update_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
828 0 : let mut rel_size_cache = self.rel_size_cache.write().unwrap();
829 0 : match rel_size_cache.entry(tag) {
830 0 : hash_map::Entry::Occupied(mut entry) => {
831 0 : let cached_lsn = entry.get_mut();
832 0 : if lsn >= cached_lsn.0 {
833 0 : *cached_lsn = (lsn, nblocks);
834 0 : }
835 : }
836 0 : hash_map::Entry::Vacant(entry) => {
837 0 : entry.insert((lsn, nblocks));
838 0 : }
839 : }
840 0 : }
841 :
842 : /// Store cached relation size
843 288732 : pub fn set_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
844 288732 : let mut rel_size_cache = self.rel_size_cache.write().unwrap();
845 288732 : rel_size_cache.insert(tag, (lsn, nblocks));
846 288732 : }
847 :
848 : /// Remove cached relation size
849 2 : pub fn remove_cached_rel_size(&self, tag: &RelTag) {
850 2 : let mut rel_size_cache = self.rel_size_cache.write().unwrap();
851 2 : rel_size_cache.remove(tag);
852 2 : }
853 : }
854 :
855 : /// DatadirModification represents an operation to ingest an atomic set of
856 : /// updates to the repository. It is created by the 'begin_record'
857 : /// function. It is called for each WAL record, so that all the modifications
858 : /// by a one WAL record appear atomic.
859 : pub struct DatadirModification<'a> {
860 : /// The timeline this modification applies to. You can access this to
861 : /// read the state, but note that any pending updates are *not* reflected
862 : /// in the state in 'tline' yet.
863 : pub tline: &'a Timeline,
864 :
865 : /// Current LSN of the modification
866 : lsn: Lsn,
867 :
868 : // The modifications are not applied directly to the underlying key-value store.
869 : // The put-functions add the modifications here, and they are flushed to the
870 : // underlying key-value store by the 'finish' function.
871 : pending_lsns: Vec<Lsn>,
872 : pending_updates: HashMap<Key, Vec<(Lsn, Value)>>,
873 : pending_deletions: Vec<(Range<Key>, Lsn)>,
874 : pending_nblocks: i64,
875 :
876 : // If we already wrote any aux file changes in this modification, stash the latest dir. If set,
877 : // [`Self::put_file`] may assume that it is safe to emit a delta rather than checking
878 : // if AUX_FILES_KEY is already set.
879 : pending_aux_files: Option<AuxFilesDirectory>,
880 :
881 : /// For special "directory" keys that store key-value maps, track the size of the map
882 : /// if it was updated in this modification.
883 : pending_directory_entries: Vec<(DirectoryKind, usize)>,
884 : }
885 :
886 : impl<'a> DatadirModification<'a> {
887 : /// Get the current lsn
888 418056 : pub(crate) fn get_lsn(&self) -> Lsn {
889 418056 : self.lsn
890 418056 : }
891 :
892 : /// Set the current lsn
893 145856 : pub(crate) fn set_lsn(&mut self, lsn: Lsn) -> anyhow::Result<()> {
894 145856 : ensure!(
895 145856 : lsn >= self.lsn,
896 0 : "setting an older lsn {} than {} is not allowed",
897 : lsn,
898 : self.lsn
899 : );
900 145856 : if lsn > self.lsn {
901 145856 : self.pending_lsns.push(self.lsn);
902 145856 : self.lsn = lsn;
903 145856 : }
904 145856 : Ok(())
905 145856 : }
906 :
907 : /// Initialize a completely new repository.
908 : ///
909 : /// This inserts the directory metadata entries that are assumed to
910 : /// always exist.
911 72 : pub fn init_empty(&mut self) -> anyhow::Result<()> {
912 72 : let buf = DbDirectory::ser(&DbDirectory {
913 72 : dbdirs: HashMap::new(),
914 72 : })?;
915 72 : self.pending_directory_entries.push((DirectoryKind::Db, 0));
916 72 : self.put(DBDIR_KEY, Value::Image(buf.into()));
917 72 :
918 72 : // Create AuxFilesDirectory
919 72 : self.init_aux_dir()?;
920 :
921 72 : let buf = TwoPhaseDirectory::ser(&TwoPhaseDirectory {
922 72 : xids: HashSet::new(),
923 72 : })?;
924 72 : self.pending_directory_entries
925 72 : .push((DirectoryKind::TwoPhase, 0));
926 72 : self.put(TWOPHASEDIR_KEY, Value::Image(buf.into()));
927 :
928 72 : let buf: Bytes = SlruSegmentDirectory::ser(&SlruSegmentDirectory::default())?.into();
929 72 : let empty_dir = Value::Image(buf);
930 72 : self.put(slru_dir_to_key(SlruKind::Clog), empty_dir.clone());
931 72 : self.pending_directory_entries
932 72 : .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
933 72 : self.put(
934 72 : slru_dir_to_key(SlruKind::MultiXactMembers),
935 72 : empty_dir.clone(),
936 72 : );
937 72 : self.pending_directory_entries
938 72 : .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
939 72 : self.put(slru_dir_to_key(SlruKind::MultiXactOffsets), empty_dir);
940 72 : self.pending_directory_entries
941 72 : .push((DirectoryKind::SlruSegment(SlruKind::MultiXactOffsets), 0));
942 72 :
943 72 : Ok(())
944 72 : }
945 :
946 : #[cfg(test)]
947 70 : pub fn init_empty_test_timeline(&mut self) -> anyhow::Result<()> {
948 70 : self.init_empty()?;
949 70 : self.put_control_file(bytes::Bytes::from_static(
950 70 : b"control_file contents do not matter",
951 70 : ))
952 70 : .context("put_control_file")?;
953 70 : self.put_checkpoint(bytes::Bytes::from_static(
954 70 : b"checkpoint_file contents do not matter",
955 70 : ))
956 70 : .context("put_checkpoint_file")?;
957 70 : Ok(())
958 70 : }
959 :
960 : /// Put a new page version that can be constructed from a WAL record
961 : ///
962 : /// NOTE: this will *not* implicitly extend the relation, if the page is beyond the
963 : /// current end-of-file. It's up to the caller to check that the relation size
964 : /// matches the blocks inserted!
965 145630 : pub fn put_rel_wal_record(
966 145630 : &mut self,
967 145630 : rel: RelTag,
968 145630 : blknum: BlockNumber,
969 145630 : rec: NeonWalRecord,
970 145630 : ) -> anyhow::Result<()> {
971 145630 : anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
972 145630 : self.put(rel_block_to_key(rel, blknum), Value::WalRecord(rec));
973 145630 : Ok(())
974 145630 : }
975 :
976 : // Same, but for an SLRU.
977 8 : pub fn put_slru_wal_record(
978 8 : &mut self,
979 8 : kind: SlruKind,
980 8 : segno: u32,
981 8 : blknum: BlockNumber,
982 8 : rec: NeonWalRecord,
983 8 : ) -> anyhow::Result<()> {
984 8 : self.put(
985 8 : slru_block_to_key(kind, segno, blknum),
986 8 : Value::WalRecord(rec),
987 8 : );
988 8 : Ok(())
989 8 : }
990 :
991 : /// Like put_wal_record, but with ready-made image of the page.
992 280864 : pub fn put_rel_page_image(
993 280864 : &mut self,
994 280864 : rel: RelTag,
995 280864 : blknum: BlockNumber,
996 280864 : img: Bytes,
997 280864 : ) -> anyhow::Result<()> {
998 280864 : anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
999 280864 : self.put(rel_block_to_key(rel, blknum), Value::Image(img));
1000 280864 : Ok(())
1001 280864 : }
1002 :
1003 6 : pub fn put_slru_page_image(
1004 6 : &mut self,
1005 6 : kind: SlruKind,
1006 6 : segno: u32,
1007 6 : blknum: BlockNumber,
1008 6 : img: Bytes,
1009 6 : ) -> anyhow::Result<()> {
1010 6 : self.put(slru_block_to_key(kind, segno, blknum), Value::Image(img));
1011 6 : Ok(())
1012 6 : }
1013 :
1014 : /// Store a relmapper file (pg_filenode.map) in the repository
1015 16 : pub async fn put_relmap_file(
1016 16 : &mut self,
1017 16 : spcnode: Oid,
1018 16 : dbnode: Oid,
1019 16 : img: Bytes,
1020 16 : ctx: &RequestContext,
1021 16 : ) -> anyhow::Result<()> {
1022 : // Add it to the directory (if it doesn't exist already)
1023 16 : let buf = self.get(DBDIR_KEY, ctx).await?;
1024 16 : let mut dbdir = DbDirectory::des(&buf)?;
1025 :
1026 16 : let r = dbdir.dbdirs.insert((spcnode, dbnode), true);
1027 16 : if r.is_none() || r == Some(false) {
1028 : // The dbdir entry didn't exist, or it contained a
1029 : // 'false'. The 'insert' call already updated it with
1030 : // 'true', now write the updated 'dbdirs' map back.
1031 16 : let buf = DbDirectory::ser(&dbdir)?;
1032 16 : self.put(DBDIR_KEY, Value::Image(buf.into()));
1033 16 :
1034 16 : // Create AuxFilesDirectory as well
1035 16 : self.init_aux_dir()?;
1036 0 : }
1037 16 : if r.is_none() {
1038 8 : // Create RelDirectory
1039 8 : let buf = RelDirectory::ser(&RelDirectory {
1040 8 : rels: HashSet::new(),
1041 8 : })?;
1042 8 : self.pending_directory_entries.push((DirectoryKind::Rel, 0));
1043 8 : self.put(
1044 8 : rel_dir_to_key(spcnode, dbnode),
1045 8 : Value::Image(Bytes::from(buf)),
1046 8 : );
1047 8 : }
1048 :
1049 16 : self.put(relmap_file_key(spcnode, dbnode), Value::Image(img));
1050 16 : Ok(())
1051 16 : }
1052 :
1053 0 : pub async fn put_twophase_file(
1054 0 : &mut self,
1055 0 : xid: TransactionId,
1056 0 : img: Bytes,
1057 0 : ctx: &RequestContext,
1058 0 : ) -> anyhow::Result<()> {
1059 : // Add it to the directory entry
1060 0 : let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
1061 0 : let mut dir = TwoPhaseDirectory::des(&buf)?;
1062 0 : if !dir.xids.insert(xid) {
1063 0 : anyhow::bail!("twophase file for xid {} already exists", xid);
1064 0 : }
1065 0 : self.pending_directory_entries
1066 0 : .push((DirectoryKind::TwoPhase, dir.xids.len()));
1067 0 : self.put(
1068 0 : TWOPHASEDIR_KEY,
1069 0 : Value::Image(Bytes::from(TwoPhaseDirectory::ser(&dir)?)),
1070 : );
1071 :
1072 0 : self.put(twophase_file_key(xid), Value::Image(img));
1073 0 : Ok(())
1074 0 : }
1075 :
1076 72 : pub fn put_control_file(&mut self, img: Bytes) -> anyhow::Result<()> {
1077 72 : self.put(CONTROLFILE_KEY, Value::Image(img));
1078 72 : Ok(())
1079 72 : }
1080 :
1081 86 : pub fn put_checkpoint(&mut self, img: Bytes) -> anyhow::Result<()> {
1082 86 : self.put(CHECKPOINT_KEY, Value::Image(img));
1083 86 : Ok(())
1084 86 : }
1085 :
1086 0 : pub async fn drop_dbdir(
1087 0 : &mut self,
1088 0 : spcnode: Oid,
1089 0 : dbnode: Oid,
1090 0 : ctx: &RequestContext,
1091 0 : ) -> anyhow::Result<()> {
1092 0 : let total_blocks = self
1093 0 : .tline
1094 0 : .get_db_size(spcnode, dbnode, Version::Modified(self), true, ctx)
1095 0 : .await?;
1096 :
1097 : // Remove entry from dbdir
1098 0 : let buf = self.get(DBDIR_KEY, ctx).await?;
1099 0 : let mut dir = DbDirectory::des(&buf)?;
1100 0 : if dir.dbdirs.remove(&(spcnode, dbnode)).is_some() {
1101 0 : let buf = DbDirectory::ser(&dir)?;
1102 0 : self.pending_directory_entries
1103 0 : .push((DirectoryKind::Db, dir.dbdirs.len()));
1104 0 : self.put(DBDIR_KEY, Value::Image(buf.into()));
1105 : } else {
1106 0 : warn!(
1107 0 : "dropped dbdir for spcnode {} dbnode {} did not exist in db directory",
1108 0 : spcnode, dbnode
1109 0 : );
1110 : }
1111 :
1112 : // Update logical database size.
1113 0 : self.pending_nblocks -= total_blocks as i64;
1114 0 :
1115 0 : // Delete all relations and metadata files for the spcnode/dnode
1116 0 : self.delete(dbdir_key_range(spcnode, dbnode));
1117 0 : Ok(())
1118 0 : }
1119 :
1120 : /// Create a relation fork.
1121 : ///
1122 : /// 'nblocks' is the initial size.
1123 1920 : pub async fn put_rel_creation(
1124 1920 : &mut self,
1125 1920 : rel: RelTag,
1126 1920 : nblocks: BlockNumber,
1127 1920 : ctx: &RequestContext,
1128 1920 : ) -> Result<(), RelationError> {
1129 1920 : if rel.relnode == 0 {
1130 0 : return Err(RelationError::InvalidRelnode);
1131 1920 : }
1132 : // It's possible that this is the first rel for this db in this
1133 : // tablespace. Create the reldir entry for it if so.
1134 1920 : let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await.context("read db")?)
1135 1920 : .context("deserialize db")?;
1136 1920 : let rel_dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
1137 1920 : let mut rel_dir = if dbdir.dbdirs.get(&(rel.spcnode, rel.dbnode)).is_none() {
1138 : // Didn't exist. Update dbdir
1139 8 : dbdir.dbdirs.insert((rel.spcnode, rel.dbnode), false);
1140 8 : let buf = DbDirectory::ser(&dbdir).context("serialize db")?;
1141 8 : self.pending_directory_entries
1142 8 : .push((DirectoryKind::Db, dbdir.dbdirs.len()));
1143 8 : self.put(DBDIR_KEY, Value::Image(buf.into()));
1144 8 :
1145 8 : // and create the RelDirectory
1146 8 : RelDirectory::default()
1147 : } else {
1148 : // reldir already exists, fetch it
1149 1912 : RelDirectory::des(&self.get(rel_dir_key, ctx).await.context("read db")?)
1150 1912 : .context("deserialize db")?
1151 : };
1152 :
1153 : // Add the new relation to the rel directory entry, and write it back
1154 1920 : if !rel_dir.rels.insert((rel.relnode, rel.forknum)) {
1155 0 : return Err(RelationError::AlreadyExists);
1156 1920 : }
1157 1920 :
1158 1920 : self.pending_directory_entries
1159 1920 : .push((DirectoryKind::Rel, rel_dir.rels.len()));
1160 1920 :
1161 1920 : self.put(
1162 1920 : rel_dir_key,
1163 1920 : Value::Image(Bytes::from(
1164 1920 : RelDirectory::ser(&rel_dir).context("serialize")?,
1165 : )),
1166 : );
1167 :
1168 : // Put size
1169 1920 : let size_key = rel_size_to_key(rel);
1170 1920 : let buf = nblocks.to_le_bytes();
1171 1920 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
1172 1920 :
1173 1920 : self.pending_nblocks += nblocks as i64;
1174 1920 :
1175 1920 : // Update relation size cache
1176 1920 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
1177 1920 :
1178 1920 : // Even if nblocks > 0, we don't insert any actual blocks here. That's up to the
1179 1920 : // caller.
1180 1920 : Ok(())
1181 1920 : }
1182 :
1183 : /// Truncate relation
1184 6012 : pub async fn put_rel_truncation(
1185 6012 : &mut self,
1186 6012 : rel: RelTag,
1187 6012 : nblocks: BlockNumber,
1188 6012 : ctx: &RequestContext,
1189 6012 : ) -> anyhow::Result<()> {
1190 6012 : anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
1191 6012 : if self
1192 6012 : .tline
1193 6012 : .get_rel_exists(rel, Version::Modified(self), true, ctx)
1194 0 : .await?
1195 : {
1196 6012 : let size_key = rel_size_to_key(rel);
1197 : // Fetch the old size first
1198 6012 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
1199 6012 :
1200 6012 : // Update the entry with the new size.
1201 6012 : let buf = nblocks.to_le_bytes();
1202 6012 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
1203 6012 :
1204 6012 : // Update relation size cache
1205 6012 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
1206 6012 :
1207 6012 : // Update relation size cache
1208 6012 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
1209 6012 :
1210 6012 : // Update logical database size.
1211 6012 : self.pending_nblocks -= old_size as i64 - nblocks as i64;
1212 0 : }
1213 6012 : Ok(())
1214 6012 : }
1215 :
1216 : /// Extend relation
1217 : /// If new size is smaller, do nothing.
1218 276680 : pub async fn put_rel_extend(
1219 276680 : &mut self,
1220 276680 : rel: RelTag,
1221 276680 : nblocks: BlockNumber,
1222 276680 : ctx: &RequestContext,
1223 276680 : ) -> anyhow::Result<()> {
1224 276680 : anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
1225 :
1226 : // Put size
1227 276680 : let size_key = rel_size_to_key(rel);
1228 276680 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
1229 276680 :
1230 276680 : // only extend relation here. never decrease the size
1231 276680 : if nblocks > old_size {
1232 274788 : let buf = nblocks.to_le_bytes();
1233 274788 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
1234 274788 :
1235 274788 : // Update relation size cache
1236 274788 : self.tline.set_cached_rel_size(rel, self.lsn, nblocks);
1237 274788 :
1238 274788 : self.pending_nblocks += nblocks as i64 - old_size as i64;
1239 274788 : }
1240 276680 : Ok(())
1241 276680 : }
1242 :
1243 : /// Drop a relation.
1244 2 : pub async fn put_rel_drop(&mut self, rel: RelTag, ctx: &RequestContext) -> anyhow::Result<()> {
1245 2 : anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
1246 :
1247 : // Remove it from the directory entry
1248 2 : let dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
1249 2 : let buf = self.get(dir_key, ctx).await?;
1250 2 : let mut dir = RelDirectory::des(&buf)?;
1251 :
1252 2 : self.pending_directory_entries
1253 2 : .push((DirectoryKind::Rel, dir.rels.len()));
1254 2 :
1255 2 : if dir.rels.remove(&(rel.relnode, rel.forknum)) {
1256 2 : self.put(dir_key, Value::Image(Bytes::from(RelDirectory::ser(&dir)?)));
1257 : } else {
1258 0 : warn!("dropped rel {} did not exist in rel directory", rel);
1259 : }
1260 :
1261 : // update logical size
1262 2 : let size_key = rel_size_to_key(rel);
1263 2 : let old_size = self.get(size_key, ctx).await?.get_u32_le();
1264 2 : self.pending_nblocks -= old_size as i64;
1265 2 :
1266 2 : // Remove enty from relation size cache
1267 2 : self.tline.remove_cached_rel_size(&rel);
1268 2 :
1269 2 : // Delete size entry, as well as all blocks
1270 2 : self.delete(rel_key_range(rel));
1271 2 :
1272 2 : Ok(())
1273 2 : }
1274 :
1275 6 : pub async fn put_slru_segment_creation(
1276 6 : &mut self,
1277 6 : kind: SlruKind,
1278 6 : segno: u32,
1279 6 : nblocks: BlockNumber,
1280 6 : ctx: &RequestContext,
1281 6 : ) -> anyhow::Result<()> {
1282 6 : // Add it to the directory entry
1283 6 : let dir_key = slru_dir_to_key(kind);
1284 6 : let buf = self.get(dir_key, ctx).await?;
1285 6 : let mut dir = SlruSegmentDirectory::des(&buf)?;
1286 :
1287 6 : if !dir.segments.insert(segno) {
1288 0 : anyhow::bail!("slru segment {kind:?}/{segno} already exists");
1289 6 : }
1290 6 : self.pending_directory_entries
1291 6 : .push((DirectoryKind::SlruSegment(kind), dir.segments.len()));
1292 6 : self.put(
1293 6 : dir_key,
1294 6 : Value::Image(Bytes::from(SlruSegmentDirectory::ser(&dir)?)),
1295 : );
1296 :
1297 : // Put size
1298 6 : let size_key = slru_segment_size_to_key(kind, segno);
1299 6 : let buf = nblocks.to_le_bytes();
1300 6 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
1301 6 :
1302 6 : // even if nblocks > 0, we don't insert any actual blocks here
1303 6 :
1304 6 : Ok(())
1305 6 : }
1306 :
1307 : /// Extend SLRU segment
1308 0 : pub fn put_slru_extend(
1309 0 : &mut self,
1310 0 : kind: SlruKind,
1311 0 : segno: u32,
1312 0 : nblocks: BlockNumber,
1313 0 : ) -> anyhow::Result<()> {
1314 0 : // Put size
1315 0 : let size_key = slru_segment_size_to_key(kind, segno);
1316 0 : let buf = nblocks.to_le_bytes();
1317 0 : self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
1318 0 : Ok(())
1319 0 : }
1320 :
1321 : /// This method is used for marking truncated SLRU files
1322 0 : pub async fn drop_slru_segment(
1323 0 : &mut self,
1324 0 : kind: SlruKind,
1325 0 : segno: u32,
1326 0 : ctx: &RequestContext,
1327 0 : ) -> anyhow::Result<()> {
1328 0 : // Remove it from the directory entry
1329 0 : let dir_key = slru_dir_to_key(kind);
1330 0 : let buf = self.get(dir_key, ctx).await?;
1331 0 : let mut dir = SlruSegmentDirectory::des(&buf)?;
1332 :
1333 0 : if !dir.segments.remove(&segno) {
1334 0 : warn!("slru segment {:?}/{} does not exist", kind, segno);
1335 0 : }
1336 0 : self.pending_directory_entries
1337 0 : .push((DirectoryKind::SlruSegment(kind), dir.segments.len()));
1338 0 : self.put(
1339 0 : dir_key,
1340 0 : Value::Image(Bytes::from(SlruSegmentDirectory::ser(&dir)?)),
1341 : );
1342 :
1343 : // Delete size entry, as well as all blocks
1344 0 : self.delete(slru_segment_key_range(kind, segno));
1345 0 :
1346 0 : Ok(())
1347 0 : }
1348 :
1349 : /// Drop a relmapper file (pg_filenode.map)
1350 0 : pub fn drop_relmap_file(&mut self, _spcnode: Oid, _dbnode: Oid) -> anyhow::Result<()> {
1351 0 : // TODO
1352 0 : Ok(())
1353 0 : }
1354 :
1355 : /// This method is used for marking truncated SLRU files
1356 0 : pub async fn drop_twophase_file(
1357 0 : &mut self,
1358 0 : xid: TransactionId,
1359 0 : ctx: &RequestContext,
1360 0 : ) -> anyhow::Result<()> {
1361 : // Remove it from the directory entry
1362 0 : let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
1363 0 : let mut dir = TwoPhaseDirectory::des(&buf)?;
1364 :
1365 0 : if !dir.xids.remove(&xid) {
1366 0 : warn!("twophase file for xid {} does not exist", xid);
1367 0 : }
1368 0 : self.pending_directory_entries
1369 0 : .push((DirectoryKind::TwoPhase, dir.xids.len()));
1370 0 : self.put(
1371 0 : TWOPHASEDIR_KEY,
1372 0 : Value::Image(Bytes::from(TwoPhaseDirectory::ser(&dir)?)),
1373 : );
1374 :
1375 : // Delete it
1376 0 : self.delete(twophase_key_range(xid));
1377 0 :
1378 0 : Ok(())
1379 0 : }
1380 :
1381 88 : pub fn init_aux_dir(&mut self) -> anyhow::Result<()> {
1382 88 : let buf = AuxFilesDirectory::ser(&AuxFilesDirectory {
1383 88 : files: HashMap::new(),
1384 88 : })?;
1385 88 : self.pending_directory_entries
1386 88 : .push((DirectoryKind::AuxFiles, 0));
1387 88 : self.put(AUX_FILES_KEY, Value::Image(Bytes::from(buf)));
1388 88 : Ok(())
1389 88 : }
1390 :
1391 8 : pub async fn put_file(
1392 8 : &mut self,
1393 8 : path: &str,
1394 8 : content: &[u8],
1395 8 : ctx: &RequestContext,
1396 8 : ) -> anyhow::Result<()> {
1397 8 : let file_path = path.to_string();
1398 8 : let content = if content.is_empty() {
1399 2 : None
1400 : } else {
1401 6 : Some(Bytes::copy_from_slice(content))
1402 : };
1403 :
1404 8 : let dir = if let Some(mut dir) = self.pending_aux_files.take() {
1405 : // We already updated aux files in `self`: emit a delta and update our latest value
1406 :
1407 4 : self.put(
1408 4 : AUX_FILES_KEY,
1409 4 : Value::WalRecord(NeonWalRecord::AuxFile {
1410 4 : file_path: file_path.clone(),
1411 4 : content: content.clone(),
1412 4 : }),
1413 4 : );
1414 4 :
1415 4 : dir.upsert(file_path, content);
1416 4 : dir
1417 : } else {
1418 : // Check if the AUX_FILES_KEY is initialized
1419 4 : match self.get(AUX_FILES_KEY, ctx).await {
1420 2 : Ok(dir_bytes) => {
1421 2 : let mut dir = AuxFilesDirectory::des(&dir_bytes)?;
1422 : // Key is already set, we may append a delta
1423 2 : self.put(
1424 2 : AUX_FILES_KEY,
1425 2 : Value::WalRecord(NeonWalRecord::AuxFile {
1426 2 : file_path: file_path.clone(),
1427 2 : content: content.clone(),
1428 2 : }),
1429 2 : );
1430 2 : dir.upsert(file_path, content);
1431 2 : dir
1432 : }
1433 : Err(
1434 0 : e @ (PageReconstructError::AncestorStopping(_)
1435 : | PageReconstructError::Cancelled
1436 : | PageReconstructError::AncestorLsnTimeout(_)),
1437 : ) => {
1438 : // Important that we do not interpret a shutdown error as "not found" and thereby
1439 : // reset the map.
1440 0 : return Err(e.into());
1441 : }
1442 : // FIXME: PageReconstructError doesn't have an explicit variant for key-not-found, so
1443 : // we are assuming that all _other_ possible errors represents a missing key. If some
1444 : // other error occurs, we may incorrectly reset the map of aux files.
1445 : Err(PageReconstructError::Other(_) | PageReconstructError::WalRedo(_)) => {
1446 : // Key is missing, we must insert an image as the basis for subsequent deltas.
1447 :
1448 2 : let mut dir = AuxFilesDirectory {
1449 2 : files: HashMap::new(),
1450 2 : };
1451 2 : dir.upsert(file_path, content);
1452 2 : self.put(
1453 2 : AUX_FILES_KEY,
1454 2 : Value::Image(Bytes::from(
1455 2 : AuxFilesDirectory::ser(&dir).context("serialize")?,
1456 : )),
1457 : );
1458 2 : dir
1459 : }
1460 : }
1461 : };
1462 :
1463 8 : self.pending_directory_entries
1464 8 : .push((DirectoryKind::AuxFiles, dir.files.len()));
1465 8 : self.pending_aux_files = Some(dir);
1466 8 :
1467 8 : Ok(())
1468 8 : }
1469 :
1470 : ///
1471 : /// Flush changes accumulated so far to the underlying repository.
1472 : ///
1473 : /// Usually, changes made in DatadirModification are atomic, but this allows
1474 : /// you to flush them to the underlying repository before the final `commit`.
1475 : /// That allows to free up the memory used to hold the pending changes.
1476 : ///
1477 : /// Currently only used during bulk import of a data directory. In that
1478 : /// context, breaking the atomicity is OK. If the import is interrupted, the
1479 : /// whole import fails and the timeline will be deleted anyway.
1480 : /// (Or to be precise, it will be left behind for debugging purposes and
1481 : /// ignored, see <https://github.com/neondatabase/neon/pull/1809>)
1482 : ///
1483 : /// Note: A consequence of flushing the pending operations is that they
1484 : /// won't be visible to subsequent operations until `commit`. The function
1485 : /// retains all the metadata, but data pages are flushed. That's again OK
1486 : /// for bulk import, where you are just loading data pages and won't try to
1487 : /// modify the same pages twice.
1488 1930 : pub async fn flush(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
1489 1930 : // Unless we have accumulated a decent amount of changes, it's not worth it
1490 1930 : // to scan through the pending_updates list.
1491 1930 : let pending_nblocks = self.pending_nblocks;
1492 1930 : if pending_nblocks < 10000 {
1493 1930 : return Ok(());
1494 0 : }
1495 :
1496 0 : let mut writer = self.tline.writer().await;
1497 :
1498 : // Flush relation and SLRU data blocks, keep metadata.
1499 0 : let mut retained_pending_updates = HashMap::<_, Vec<_>>::new();
1500 0 : for (key, values) in self.pending_updates.drain() {
1501 0 : for (lsn, value) in values {
1502 0 : if is_rel_block_key(&key) || is_slru_block_key(key) {
1503 : // This bails out on first error without modifying pending_updates.
1504 : // That's Ok, cf this function's doc comment.
1505 0 : writer.put(key, lsn, &value, ctx).await?;
1506 0 : } else {
1507 0 : retained_pending_updates
1508 0 : .entry(key)
1509 0 : .or_default()
1510 0 : .push((lsn, value));
1511 0 : }
1512 : }
1513 : }
1514 :
1515 0 : self.pending_updates = retained_pending_updates;
1516 0 :
1517 0 : if pending_nblocks != 0 {
1518 0 : writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
1519 0 : self.pending_nblocks = 0;
1520 0 : }
1521 :
1522 0 : for (kind, count) in std::mem::take(&mut self.pending_directory_entries) {
1523 0 : writer.update_directory_entries_count(kind, count as u64);
1524 0 : }
1525 :
1526 0 : Ok(())
1527 1930 : }
1528 :
1529 : ///
1530 : /// Finish this atomic update, writing all the updated keys to the
1531 : /// underlying timeline.
1532 : /// All the modifications in this atomic update are stamped by the specified LSN.
1533 : ///
1534 742952 : pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
1535 742952 : let mut writer = self.tline.writer().await;
1536 :
1537 742952 : let pending_nblocks = self.pending_nblocks;
1538 742952 : self.pending_nblocks = 0;
1539 742952 :
1540 742952 : if !self.pending_updates.is_empty() {
1541 413936 : let prev_pending_updates = std::mem::take(&mut self.pending_updates);
1542 413936 :
1543 413936 : // The put_batch call below expects expects the inputs to be sorted by Lsn,
1544 413936 : // so we do that first.
1545 413936 : let lsn_ordered_batch: Vec<(Key, Lsn, Value)> = prev_pending_updates
1546 413936 : .into_iter()
1547 699664 : .map(|(key, vals)| vals.into_iter().map(move |(lsn, val)| (key, lsn, val)))
1548 413936 : .kmerge_by(|lhs, rhs| lhs.1 .0 < rhs.1 .0)
1549 413936 : .collect();
1550 413936 :
1551 413936 : writer.put_batch(lsn_ordered_batch, ctx).await?;
1552 413936 : self.pending_updates.clear();
1553 329016 : }
1554 :
1555 742952 : if !self.pending_deletions.is_empty() {
1556 2 : writer.delete_batch(&self.pending_deletions).await?;
1557 2 : self.pending_deletions.clear();
1558 742950 : }
1559 :
1560 742952 : self.pending_lsns.push(self.lsn);
1561 888808 : for pending_lsn in self.pending_lsns.drain(..) {
1562 888808 : // Ideally, we should be able to call writer.finish_write() only once
1563 888808 : // with the highest LSN. However, the last_record_lsn variable in the
1564 888808 : // timeline keeps track of the latest LSN and the immediate previous LSN
1565 888808 : // so we need to record every LSN to not leave a gap between them.
1566 888808 : writer.finish_write(pending_lsn);
1567 888808 : }
1568 :
1569 742952 : if pending_nblocks != 0 {
1570 270570 : writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
1571 472382 : }
1572 :
1573 742952 : for (kind, count) in std::mem::take(&mut self.pending_directory_entries) {
1574 2400 : writer.update_directory_entries_count(kind, count as u64);
1575 2400 : }
1576 :
1577 742952 : Ok(())
1578 742952 : }
1579 :
1580 291704 : pub(crate) fn len(&self) -> usize {
1581 291704 : self.pending_updates.len() + self.pending_deletions.len()
1582 291704 : }
1583 :
1584 : // Internal helper functions to batch the modifications
1585 :
1586 286564 : async fn get(&self, key: Key, ctx: &RequestContext) -> Result<Bytes, PageReconstructError> {
1587 : // Have we already updated the same key? Read the latest pending updated
1588 : // version in that case.
1589 : //
1590 : // Note: we don't check pending_deletions. It is an error to request a
1591 : // value that has been removed, deletion only avoids leaking storage.
1592 286564 : if let Some(values) = self.pending_updates.get(&key) {
1593 15928 : if let Some((_, value)) = values.last() {
1594 15928 : return if let Value::Image(img) = value {
1595 15928 : Ok(img.clone())
1596 : } else {
1597 : // Currently, we never need to read back a WAL record that we
1598 : // inserted in the same "transaction". All the metadata updates
1599 : // work directly with Images, and we never need to read actual
1600 : // data pages. We could handle this if we had to, by calling
1601 : // the walredo manager, but let's keep it simple for now.
1602 0 : Err(PageReconstructError::from(anyhow::anyhow!(
1603 0 : "unexpected pending WAL record"
1604 0 : )))
1605 : };
1606 0 : }
1607 270636 : }
1608 270636 : let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
1609 270636 : self.tline.get(key, lsn, ctx).await
1610 286564 : }
1611 :
1612 711824 : fn put(&mut self, key: Key, val: Value) {
1613 711824 : let values = self.pending_updates.entry(key).or_default();
1614 : // Replace the previous value if it exists at the same lsn
1615 711824 : if let Some((last_lsn, last_value)) = values.last_mut() {
1616 12164 : if *last_lsn == self.lsn {
1617 12160 : *last_value = val;
1618 12160 : return;
1619 4 : }
1620 699660 : }
1621 699664 : values.push((self.lsn, val));
1622 711824 : }
1623 :
1624 2 : fn delete(&mut self, key_range: Range<Key>) {
1625 2 : trace!("DELETE {}-{}", key_range.start, key_range.end);
1626 2 : self.pending_deletions.push((key_range, self.lsn));
1627 2 : }
1628 : }
1629 :
1630 : /// This struct facilitates accessing either a committed key from the timeline at a
1631 : /// specific LSN, or the latest uncommitted key from a pending modification.
1632 : /// During WAL ingestion, the records from multiple LSNs may be batched in the same
1633 : /// modification before being flushed to the timeline. Hence, the routines in WalIngest
1634 : /// need to look up the keys in the modification first before looking them up in the
1635 : /// timeline to not miss the latest updates.
1636 0 : #[derive(Clone, Copy)]
1637 : pub enum Version<'a> {
1638 : Lsn(Lsn),
1639 : Modified(&'a DatadirModification<'a>),
1640 : }
1641 :
1642 : impl<'a> Version<'a> {
1643 23542 : async fn get(
1644 23542 : &self,
1645 23542 : timeline: &Timeline,
1646 23542 : key: Key,
1647 23542 : ctx: &RequestContext,
1648 23542 : ) -> Result<Bytes, PageReconstructError> {
1649 23542 : match self {
1650 23532 : Version::Lsn(lsn) => timeline.get(key, *lsn, ctx).await,
1651 10 : Version::Modified(modification) => modification.get(key, ctx).await,
1652 : }
1653 23542 : }
1654 :
1655 30484 : fn get_lsn(&self) -> Lsn {
1656 30484 : match self {
1657 24438 : Version::Lsn(lsn) => *lsn,
1658 6046 : Version::Modified(modification) => modification.lsn,
1659 : }
1660 30484 : }
1661 : }
1662 :
1663 : //--- Metadata structs stored in key-value pairs in the repository.
1664 :
1665 2106 : #[derive(Debug, Serialize, Deserialize)]
1666 : struct DbDirectory {
1667 : // (spcnode, dbnode) -> (do relmapper and PG_VERSION files exist)
1668 : dbdirs: HashMap<(Oid, Oid), bool>,
1669 : }
1670 :
1671 170 : #[derive(Debug, Serialize, Deserialize)]
1672 : struct TwoPhaseDirectory {
1673 : xids: HashSet<TransactionId>,
1674 : }
1675 :
1676 3860 : #[derive(Debug, Serialize, Deserialize, Default)]
1677 : struct RelDirectory {
1678 : // Set of relations that exist. (relfilenode, forknum)
1679 : //
1680 : // TODO: Store it as a btree or radix tree or something else that spans multiple
1681 : // key-value pairs, if you have a lot of relations
1682 : rels: HashSet<(Oid, u8)>,
1683 : }
1684 :
1685 202 : #[derive(Debug, Serialize, Deserialize, Default)]
1686 : pub(crate) struct AuxFilesDirectory {
1687 : pub(crate) files: HashMap<String, Bytes>,
1688 : }
1689 :
1690 : impl AuxFilesDirectory {
1691 26 : pub(crate) fn upsert(&mut self, key: String, value: Option<Bytes>) {
1692 26 : if let Some(value) = value {
1693 20 : self.files.insert(key, value);
1694 20 : } else {
1695 6 : self.files.remove(&key);
1696 6 : }
1697 26 : }
1698 : }
1699 :
1700 0 : #[derive(Debug, Serialize, Deserialize)]
1701 : struct RelSizeEntry {
1702 : nblocks: u32,
1703 : }
1704 :
1705 516 : #[derive(Debug, Serialize, Deserialize, Default)]
1706 : struct SlruSegmentDirectory {
1707 : // Set of SLRU segments that exist.
1708 : segments: HashSet<u32>,
1709 : }
1710 :
1711 4800 : #[derive(Copy, Clone, PartialEq, Eq, Debug, enum_map::Enum)]
1712 : #[repr(u8)]
1713 : pub(crate) enum DirectoryKind {
1714 : Db,
1715 : TwoPhase,
1716 : Rel,
1717 : AuxFiles,
1718 : SlruSegment(SlruKind),
1719 : }
1720 :
1721 : impl DirectoryKind {
1722 : pub(crate) const KINDS_NUM: usize = <DirectoryKind as Enum>::LENGTH;
1723 4800 : pub(crate) fn offset(&self) -> usize {
1724 4800 : self.into_usize()
1725 4800 : }
1726 : }
1727 :
1728 : static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; BLCKSZ as usize]);
1729 :
1730 : #[allow(clippy::bool_assert_comparison)]
1731 : #[cfg(test)]
1732 : mod tests {
1733 : use hex_literal::hex;
1734 : use utils::id::TimelineId;
1735 :
1736 : use super::*;
1737 :
1738 : use crate::{tenant::harness::TenantHarness, DEFAULT_PG_VERSION};
1739 :
1740 : /// Test a round trip of aux file updates, from DatadirModification to reading back from the Timeline
1741 2 : #[tokio::test]
1742 2 : async fn aux_files_round_trip() -> anyhow::Result<()> {
1743 2 : let name = "aux_files_round_trip";
1744 2 : let harness = TenantHarness::create(name)?;
1745 2 :
1746 2 : pub const TIMELINE_ID: TimelineId =
1747 2 : TimelineId::from_array(hex!("11223344556677881122334455667788"));
1748 2 :
1749 2 : let (tenant, ctx) = harness.load().await;
1750 2 : let tline = tenant
1751 2 : .create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)
1752 3 : .await?;
1753 2 : let tline = tline.raw_timeline().unwrap();
1754 2 :
1755 2 : // First modification: insert two keys
1756 2 : let mut modification = tline.begin_modification(Lsn(0x1000));
1757 2 : modification.put_file("foo/bar1", b"content1", &ctx).await?;
1758 2 : modification.set_lsn(Lsn(0x1008))?;
1759 2 : modification.put_file("foo/bar2", b"content2", &ctx).await?;
1760 2 : modification.commit(&ctx).await?;
1761 2 : let expect_1008 = HashMap::from([
1762 2 : ("foo/bar1".to_string(), Bytes::from_static(b"content1")),
1763 2 : ("foo/bar2".to_string(), Bytes::from_static(b"content2")),
1764 2 : ]);
1765 2 :
1766 2 : let readback = tline.list_aux_files(Lsn(0x1008), &ctx).await?;
1767 2 : assert_eq!(readback, expect_1008);
1768 2 :
1769 2 : // Second modification: update one key, remove the other
1770 2 : let mut modification = tline.begin_modification(Lsn(0x2000));
1771 2 : modification.put_file("foo/bar1", b"content3", &ctx).await?;
1772 2 : modification.set_lsn(Lsn(0x2008))?;
1773 2 : modification.put_file("foo/bar2", b"", &ctx).await?;
1774 2 : modification.commit(&ctx).await?;
1775 2 : let expect_2008 =
1776 2 : HashMap::from([("foo/bar1".to_string(), Bytes::from_static(b"content3"))]);
1777 2 :
1778 2 : let readback = tline.list_aux_files(Lsn(0x2008), &ctx).await?;
1779 2 : assert_eq!(readback, expect_2008);
1780 2 :
1781 2 : // Reading back in time works
1782 2 : let readback = tline.list_aux_files(Lsn(0x1008), &ctx).await?;
1783 2 : assert_eq!(readback, expect_1008);
1784 2 :
1785 2 : Ok(())
1786 2 : }
1787 :
1788 : /*
1789 : fn assert_current_logical_size<R: Repository>(timeline: &DatadirTimeline<R>, lsn: Lsn) {
1790 : let incremental = timeline.get_current_logical_size();
1791 : let non_incremental = timeline
1792 : .get_current_logical_size_non_incremental(lsn)
1793 : .unwrap();
1794 : assert_eq!(incremental, non_incremental);
1795 : }
1796 : */
1797 :
1798 : /*
1799 : ///
1800 : /// Test list_rels() function, with branches and dropped relations
1801 : ///
1802 : #[test]
1803 : fn test_list_rels_drop() -> Result<()> {
1804 : let repo = RepoHarness::create("test_list_rels_drop")?.load();
1805 : let tline = create_empty_timeline(repo, TIMELINE_ID)?;
1806 : const TESTDB: u32 = 111;
1807 :
1808 : // Import initial dummy checkpoint record, otherwise the get_timeline() call
1809 : // after branching fails below
1810 : let mut writer = tline.begin_record(Lsn(0x10));
1811 : writer.put_checkpoint(ZERO_CHECKPOINT.clone())?;
1812 : writer.finish()?;
1813 :
1814 : // Create a relation on the timeline
1815 : let mut writer = tline.begin_record(Lsn(0x20));
1816 : writer.put_rel_page_image(TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"))?;
1817 : writer.finish()?;
1818 :
1819 : let writer = tline.begin_record(Lsn(0x00));
1820 : writer.finish()?;
1821 :
1822 : // Check that list_rels() lists it after LSN 2, but no before it
1823 : assert!(!tline.list_rels(0, TESTDB, Lsn(0x10))?.contains(&TESTREL_A));
1824 : assert!(tline.list_rels(0, TESTDB, Lsn(0x20))?.contains(&TESTREL_A));
1825 : assert!(tline.list_rels(0, TESTDB, Lsn(0x30))?.contains(&TESTREL_A));
1826 :
1827 : // Create a branch, check that the relation is visible there
1828 : repo.branch_timeline(&tline, NEW_TIMELINE_ID, Lsn(0x30))?;
1829 : let newtline = match repo.get_timeline(NEW_TIMELINE_ID)?.local_timeline() {
1830 : Some(timeline) => timeline,
1831 : None => panic!("Should have a local timeline"),
1832 : };
1833 : let newtline = DatadirTimelineImpl::new(newtline);
1834 : assert!(newtline
1835 : .list_rels(0, TESTDB, Lsn(0x30))?
1836 : .contains(&TESTREL_A));
1837 :
1838 : // Drop it on the branch
1839 : let mut new_writer = newtline.begin_record(Lsn(0x40));
1840 : new_writer.drop_relation(TESTREL_A)?;
1841 : new_writer.finish()?;
1842 :
1843 : // Check that it's no longer listed on the branch after the point where it was dropped
1844 : assert!(newtline
1845 : .list_rels(0, TESTDB, Lsn(0x30))?
1846 : .contains(&TESTREL_A));
1847 : assert!(!newtline
1848 : .list_rels(0, TESTDB, Lsn(0x40))?
1849 : .contains(&TESTREL_A));
1850 :
1851 : // Run checkpoint and garbage collection and check that it's still not visible
1852 : newtline.checkpoint(CheckpointConfig::Forced)?;
1853 : repo.gc_iteration(Some(NEW_TIMELINE_ID), 0, true)?;
1854 :
1855 : assert!(!newtline
1856 : .list_rels(0, TESTDB, Lsn(0x40))?
1857 : .contains(&TESTREL_A));
1858 :
1859 : Ok(())
1860 : }
1861 : */
1862 :
1863 : /*
1864 : #[test]
1865 : fn test_read_beyond_eof() -> Result<()> {
1866 : let repo = RepoHarness::create("test_read_beyond_eof")?.load();
1867 : let tline = create_test_timeline(repo, TIMELINE_ID)?;
1868 :
1869 : make_some_layers(&tline, Lsn(0x20))?;
1870 : let mut writer = tline.begin_record(Lsn(0x60));
1871 : walingest.put_rel_page_image(
1872 : &mut writer,
1873 : TESTREL_A,
1874 : 0,
1875 : TEST_IMG(&format!("foo blk 0 at {}", Lsn(0x60))),
1876 : )?;
1877 : writer.finish()?;
1878 :
1879 : // Test read before rel creation. Should error out.
1880 : assert!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x10), false).is_err());
1881 :
1882 : // Read block beyond end of relation at different points in time.
1883 : // These reads should fall into different delta, image, and in-memory layers.
1884 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x20), false)?, ZERO_PAGE);
1885 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x25), false)?, ZERO_PAGE);
1886 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x30), false)?, ZERO_PAGE);
1887 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x35), false)?, ZERO_PAGE);
1888 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x40), false)?, ZERO_PAGE);
1889 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x45), false)?, ZERO_PAGE);
1890 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x50), false)?, ZERO_PAGE);
1891 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x55), false)?, ZERO_PAGE);
1892 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_A, 1, Lsn(0x60), false)?, ZERO_PAGE);
1893 :
1894 : // Test on an in-memory layer with no preceding layer
1895 : let mut writer = tline.begin_record(Lsn(0x70));
1896 : walingest.put_rel_page_image(
1897 : &mut writer,
1898 : TESTREL_B,
1899 : 0,
1900 : TEST_IMG(&format!("foo blk 0 at {}", Lsn(0x70))),
1901 : )?;
1902 : writer.finish()?;
1903 :
1904 : assert_eq!(tline.get_rel_page_at_lsn(TESTREL_B, 1, Lsn(0x70), false)?6, ZERO_PAGE);
1905 :
1906 : Ok(())
1907 : }
1908 : */
1909 : }
|
