Line data Source code
1 : //! This module has everything to deal with WAL -- reading and writing to disk.
2 : //!
3 : //! Safekeeper WAL is stored in the timeline directory, in format similar to pg_wal.
4 : //! PG timeline is always 1, so WAL segments are usually have names like this:
5 : //! - 000000010000000000000001
6 : //! - 000000010000000000000002.partial
7 : //!
8 : //! Note that last file has `.partial` suffix, that's different from postgres.
9 :
10 : use anyhow::{bail, Context, Result};
11 : use bytes::Bytes;
12 : use camino::{Utf8Path, Utf8PathBuf};
13 : use futures::future::BoxFuture;
14 : use postgres_ffi::v14::xlog_utils::{IsPartialXLogFileName, IsXLogFileName, XLogFromFileName};
15 : use postgres_ffi::{dispatch_pgversion, XLogSegNo, PG_TLI};
16 : use remote_storage::RemotePath;
17 : use std::cmp::{max, min};
18 : use std::future::Future;
19 : use std::io::{self, SeekFrom};
20 : use std::pin::Pin;
21 : use tokio::fs::{self, remove_file, File, OpenOptions};
22 : use tokio::io::{AsyncRead, AsyncWriteExt};
23 : use tokio::io::{AsyncReadExt, AsyncSeekExt};
24 : use tracing::*;
25 : use utils::crashsafe::durable_rename;
26 :
27 : use crate::metrics::{
28 : time_io_closure, WalStorageMetrics, REMOVED_WAL_SEGMENTS, WAL_STORAGE_OPERATION_SECONDS,
29 : };
30 : use crate::state::TimelinePersistentState;
31 : use crate::wal_backup::{read_object, remote_timeline_path};
32 : use crate::SafeKeeperConf;
33 : use postgres_ffi::waldecoder::WalStreamDecoder;
34 : use postgres_ffi::XLogFileName;
35 : use postgres_ffi::XLOG_BLCKSZ;
36 : use pq_proto::SystemId;
37 : use utils::{id::TenantTimelineId, lsn::Lsn};
38 :
39 : pub trait Storage {
40 : // Last written LSN.
41 : fn write_lsn(&self) -> Lsn;
42 : /// LSN of last durably stored WAL record.
43 : fn flush_lsn(&self) -> Lsn;
44 :
45 : /// Initialize segment by creating proper long header at the beginning of
46 : /// the segment and short header at the page of given LSN. This is only used
47 : /// for timeline initialization because compute will stream data only since
48 : /// init_lsn. Other segment headers are included in compute stream.
49 : fn initialize_first_segment(
50 : &mut self,
51 : init_lsn: Lsn,
52 : ) -> impl Future<Output = Result<()>> + Send;
53 :
54 : /// Write piece of WAL from buf to disk, but not necessarily sync it.
55 : fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> impl Future<Output = Result<()>> + Send;
56 :
57 : /// Truncate WAL at specified LSN, which must be the end of WAL record.
58 : fn truncate_wal(&mut self, end_pos: Lsn) -> impl Future<Output = Result<()>> + Send;
59 :
60 : /// Durably store WAL on disk, up to the last written WAL record.
61 : fn flush_wal(&mut self) -> impl Future<Output = Result<()>> + Send;
62 :
63 : /// Remove all segments <= given segno. Returns function doing that as we
64 : /// want to perform it without timeline lock.
65 : fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>>;
66 :
67 : /// Release resources associated with the storage -- technically, close FDs.
68 : /// Currently we don't remove timelines until restart (#3146), so need to
69 : /// spare descriptors. This would be useful for temporary tli detach as
70 : /// well.
71 0 : fn close(&mut self) {}
72 :
73 : /// Get metrics for this timeline.
74 : fn get_metrics(&self) -> WalStorageMetrics;
75 : }
76 :
77 : /// PhysicalStorage is a storage that stores WAL on disk. Writes are separated from flushes
78 : /// for better performance. Storage is initialized in the constructor.
79 : ///
80 : /// WAL is stored in segments, each segment is a file. Last segment has ".partial" suffix in
81 : /// its filename and may be not fully flushed.
82 : ///
83 : /// Relationship of LSNs:
84 : /// `write_lsn` >= `write_record_lsn` >= `flush_record_lsn`
85 : ///
86 : /// When storage is created first time, all LSNs are zeroes and there are no segments on disk.
87 : pub struct PhysicalStorage {
88 : metrics: WalStorageMetrics,
89 : timeline_dir: Utf8PathBuf,
90 : conf: SafeKeeperConf,
91 :
92 : /// Size of WAL segment in bytes.
93 : wal_seg_size: usize,
94 : pg_version: u32,
95 : system_id: u64,
96 :
97 : /// Written to disk, but possibly still in the cache and not fully persisted.
98 : /// Also can be ahead of record_lsn, if happen to be in the middle of a WAL record.
99 : write_lsn: Lsn,
100 :
101 : /// The LSN of the last WAL record written to disk. Still can be not fully flushed.
102 : write_record_lsn: Lsn,
103 :
104 : /// The LSN of the last WAL record flushed to disk.
105 : flush_record_lsn: Lsn,
106 :
107 : /// Decoder is required for detecting boundaries of WAL records.
108 : decoder: WalStreamDecoder,
109 :
110 : /// Cached open file for the last segment.
111 : ///
112 : /// If Some(file) is open, then it always:
113 : /// - has ".partial" suffix
114 : /// - points to write_lsn, so no seek is needed for writing
115 : /// - doesn't point to the end of the segment
116 : file: Option<File>,
117 :
118 : /// When false, we have just initialized storage using the LSN from find_end_of_wal().
119 : /// In this case, [`write_lsn`] can be less than actually written WAL on disk. In particular,
120 : /// there can be a case with unexpected .partial file.
121 : ///
122 : /// Imagine the following:
123 : /// - 000000010000000000000001
124 : /// - it was fully written, but the last record is split between 2 segments
125 : /// - after restart, `find_end_of_wal()` returned 0/1FFFFF0, which is in the end of this segment
126 : /// - `write_lsn`, `write_record_lsn` and `flush_record_lsn` were initialized to 0/1FFFFF0
127 : /// - 000000010000000000000002.partial
128 : /// - it has only 1 byte written, which is not enough to make a full WAL record
129 : ///
130 : /// Partial segment 002 has no WAL records, and it will be removed by the next truncate_wal().
131 : /// This flag will be set to true after the first truncate_wal() call.
132 : ///
133 : /// [`write_lsn`]: Self::write_lsn
134 : is_truncated_after_restart: bool,
135 : }
136 :
137 : impl PhysicalStorage {
138 : /// Create new storage. If commit_lsn is not zero, flush_lsn is tried to be restored from
139 : /// the disk. Otherwise, all LSNs are set to zero.
140 0 : pub fn new(
141 0 : ttid: &TenantTimelineId,
142 0 : timeline_dir: Utf8PathBuf,
143 0 : conf: &SafeKeeperConf,
144 0 : state: &TimelinePersistentState,
145 0 : ) -> Result<PhysicalStorage> {
146 0 : let wal_seg_size = state.server.wal_seg_size as usize;
147 :
148 : // Find out where stored WAL ends, starting at commit_lsn which is a
149 : // known recent record boundary (unless we don't have WAL at all).
150 : //
151 : // NB: find_end_of_wal MUST be backwards compatible with the previously
152 : // written WAL. If find_end_of_wal fails to read any WAL written by an
153 : // older version of the code, we could lose data forever.
154 0 : let write_lsn = if state.commit_lsn == Lsn(0) {
155 0 : Lsn(0)
156 : } else {
157 0 : let version = state.server.pg_version / 10000;
158 0 :
159 0 : dispatch_pgversion!(
160 0 : version,
161 0 : pgv::xlog_utils::find_end_of_wal(
162 0 : timeline_dir.as_std_path(),
163 0 : wal_seg_size,
164 0 : state.commit_lsn,
165 0 : )?,
166 0 : bail!("unsupported postgres version: {}", version)
167 : )
168 : };
169 :
170 : // TODO: do we really know that write_lsn is fully flushed to disk?
171 : // If not, maybe it's better to call fsync() here to be sure?
172 0 : let flush_lsn = write_lsn;
173 0 :
174 0 : debug!(
175 0 : "initialized storage for timeline {}, flush_lsn={}, commit_lsn={}, peer_horizon_lsn={}",
176 : ttid.timeline_id, flush_lsn, state.commit_lsn, state.peer_horizon_lsn,
177 : );
178 0 : if flush_lsn < state.commit_lsn || flush_lsn < state.peer_horizon_lsn {
179 0 : warn!("timeline {} potential data loss: flush_lsn by find_end_of_wal is less than either commit_lsn or peer_horizon_lsn from control file", ttid.timeline_id);
180 0 : }
181 :
182 0 : Ok(PhysicalStorage {
183 0 : metrics: WalStorageMetrics::default(),
184 0 : timeline_dir,
185 0 : conf: conf.clone(),
186 0 : wal_seg_size,
187 0 : pg_version: state.server.pg_version,
188 0 : system_id: state.server.system_id,
189 0 : write_lsn,
190 0 : write_record_lsn: write_lsn,
191 0 : flush_record_lsn: flush_lsn,
192 0 : decoder: WalStreamDecoder::new(write_lsn, state.server.pg_version / 10000),
193 0 : file: None,
194 0 : is_truncated_after_restart: false,
195 0 : })
196 0 : }
197 :
198 : /// Get all known state of the storage.
199 0 : pub fn internal_state(&self) -> (Lsn, Lsn, Lsn, bool) {
200 0 : (
201 0 : self.write_lsn,
202 0 : self.write_record_lsn,
203 0 : self.flush_record_lsn,
204 0 : self.file.is_some(),
205 0 : )
206 0 : }
207 :
208 : /// Call fdatasync if config requires so.
209 0 : async fn fdatasync_file(&mut self, file: &File) -> Result<()> {
210 0 : if !self.conf.no_sync {
211 0 : self.metrics
212 0 : .observe_flush_seconds(time_io_closure(file.sync_data()).await?);
213 0 : }
214 0 : Ok(())
215 0 : }
216 :
217 : /// Open or create WAL segment file. Caller must call seek to the wanted position.
218 : /// Returns `file` and `is_partial`.
219 0 : async fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
220 0 : let (wal_file_path, wal_file_partial_path) =
221 0 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size);
222 :
223 : // Try to open already completed segment
224 0 : if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path).await {
225 0 : Ok((file, false))
226 0 : } else if let Ok(file) = OpenOptions::new()
227 0 : .write(true)
228 0 : .open(&wal_file_partial_path)
229 0 : .await
230 : {
231 : // Try to open existing partial file
232 0 : Ok((file, true))
233 : } else {
234 : // Create and fill new partial file
235 : //
236 : // We're using fdatasync during WAL writing, so file size must not
237 : // change; to this end it is filled with zeros here. To avoid using
238 : // half initialized segment, first bake it under tmp filename and
239 : // then rename.
240 0 : let tmp_path = self.timeline_dir.join("waltmp");
241 0 : let mut file = File::create(&tmp_path)
242 0 : .await
243 0 : .with_context(|| format!("Failed to open tmp wal file {:?}", &tmp_path))?;
244 :
245 0 : write_zeroes(&mut file, self.wal_seg_size).await?;
246 :
247 : // Note: this doesn't get into observe_flush_seconds metric. But
248 : // segment init should be separate metric, if any.
249 0 : if let Err(e) =
250 0 : durable_rename(&tmp_path, &wal_file_partial_path, !self.conf.no_sync).await
251 : {
252 : // Probably rename succeeded, but fsync of it failed. Remove
253 : // the file then to avoid using it.
254 0 : remove_file(wal_file_partial_path)
255 0 : .await
256 0 : .or_else(utils::fs_ext::ignore_not_found)?;
257 0 : return Err(e.into());
258 0 : }
259 0 : Ok((file, true))
260 : }
261 0 : }
262 :
263 : /// Write WAL bytes, which are known to be located in a single WAL segment.
264 0 : async fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
265 0 : let mut file = if let Some(file) = self.file.take() {
266 0 : file
267 : } else {
268 0 : let (mut file, is_partial) = self.open_or_create(segno).await?;
269 0 : assert!(is_partial, "unexpected write into non-partial segment file");
270 0 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
271 0 : file
272 : };
273 :
274 0 : file.write_all(buf).await?;
275 : // Note: flush just ensures write above reaches the OS (this is not
276 : // needed in case of sync IO as Write::write there calls directly write
277 : // syscall, but needed in case of async). It does *not* fsyncs the file.
278 0 : file.flush().await?;
279 :
280 0 : if xlogoff + buf.len() == self.wal_seg_size {
281 : // If we reached the end of a WAL segment, flush and close it.
282 0 : self.fdatasync_file(&file).await?;
283 :
284 : // Rename partial file to completed file
285 0 : let (wal_file_path, wal_file_partial_path) =
286 0 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size);
287 0 : fs::rename(wal_file_partial_path, wal_file_path).await?;
288 0 : } else {
289 0 : // otherwise, file can be reused later
290 0 : self.file = Some(file);
291 0 : }
292 :
293 0 : Ok(())
294 0 : }
295 :
296 : /// Writes WAL to the segment files, until everything is writed. If some segments
297 : /// are fully written, they are flushed to disk. The last (partial) segment can
298 : /// be flushed separately later.
299 : ///
300 : /// Updates `write_lsn`.
301 0 : async fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
302 0 : if self.write_lsn != pos {
303 : // need to flush the file before discarding it
304 0 : if let Some(file) = self.file.take() {
305 0 : self.fdatasync_file(&file).await?;
306 0 : }
307 :
308 0 : self.write_lsn = pos;
309 0 : }
310 :
311 0 : while !buf.is_empty() {
312 : // Extract WAL location for this block
313 0 : let xlogoff = self.write_lsn.segment_offset(self.wal_seg_size);
314 0 : let segno = self.write_lsn.segment_number(self.wal_seg_size);
315 :
316 : // If crossing a WAL boundary, only write up until we reach wal segment size.
317 0 : let bytes_write = if xlogoff + buf.len() > self.wal_seg_size {
318 0 : self.wal_seg_size - xlogoff
319 : } else {
320 0 : buf.len()
321 : };
322 :
323 0 : self.write_in_segment(segno, xlogoff, &buf[..bytes_write])
324 0 : .await?;
325 0 : self.write_lsn += bytes_write as u64;
326 0 : buf = &buf[bytes_write..];
327 : }
328 :
329 0 : Ok(())
330 0 : }
331 : }
332 :
333 : impl Storage for PhysicalStorage {
334 : // Last written LSN.
335 0 : fn write_lsn(&self) -> Lsn {
336 0 : self.write_lsn
337 0 : }
338 : /// flush_lsn returns LSN of last durably stored WAL record.
339 0 : fn flush_lsn(&self) -> Lsn {
340 0 : self.flush_record_lsn
341 0 : }
342 :
343 0 : async fn initialize_first_segment(&mut self, init_lsn: Lsn) -> Result<()> {
344 0 : let _timer = WAL_STORAGE_OPERATION_SECONDS
345 0 : .with_label_values(&["initialize_first_segment"])
346 0 : .start_timer();
347 0 :
348 0 : let segno = init_lsn.segment_number(self.wal_seg_size);
349 0 : let (mut file, _) = self.open_or_create(segno).await?;
350 0 : let major_pg_version = self.pg_version / 10000;
351 0 : let wal_seg =
352 0 : postgres_ffi::generate_wal_segment(segno, self.system_id, major_pg_version, init_lsn)?;
353 0 : file.seek(SeekFrom::Start(0)).await?;
354 0 : file.write_all(&wal_seg).await?;
355 0 : file.flush().await?;
356 0 : info!("initialized segno {} at lsn {}", segno, init_lsn);
357 : // note: file is *not* fsynced
358 0 : Ok(())
359 0 : }
360 :
361 : /// Write WAL to disk.
362 0 : async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
363 0 : // Disallow any non-sequential writes, which can result in gaps or overwrites.
364 0 : // If we need to move the pointer, use truncate_wal() instead.
365 0 : if self.write_lsn > startpos {
366 0 : bail!(
367 0 : "write_wal rewrites WAL written before, write_lsn={}, startpos={}",
368 0 : self.write_lsn,
369 0 : startpos
370 0 : );
371 0 : }
372 0 : if self.write_lsn < startpos && self.write_lsn != Lsn(0) {
373 0 : bail!(
374 0 : "write_wal creates gap in written WAL, write_lsn={}, startpos={}",
375 0 : self.write_lsn,
376 0 : startpos
377 0 : );
378 0 : }
379 :
380 0 : let write_seconds = time_io_closure(self.write_exact(startpos, buf)).await?;
381 : // WAL is written, updating write metrics
382 0 : self.metrics.observe_write_seconds(write_seconds);
383 0 : self.metrics.observe_write_bytes(buf.len());
384 0 :
385 0 : // figure out last record's end lsn for reporting (if we got the
386 0 : // whole record)
387 0 : if self.decoder.available() != startpos {
388 0 : info!(
389 0 : "restart decoder from {} to {}",
390 0 : self.decoder.available(),
391 : startpos,
392 : );
393 0 : let pg_version = self.decoder.pg_version;
394 0 : self.decoder = WalStreamDecoder::new(startpos, pg_version);
395 0 : }
396 0 : self.decoder.feed_bytes(buf);
397 : loop {
398 0 : match self.decoder.poll_decode()? {
399 0 : None => break, // no full record yet
400 0 : Some((lsn, _rec)) => {
401 0 : self.write_record_lsn = lsn;
402 0 : }
403 : }
404 : }
405 :
406 0 : Ok(())
407 0 : }
408 :
409 0 : async fn flush_wal(&mut self) -> Result<()> {
410 0 : if self.flush_record_lsn == self.write_record_lsn {
411 : // no need to do extra flush
412 0 : return Ok(());
413 0 : }
414 :
415 0 : if let Some(unflushed_file) = self.file.take() {
416 0 : self.fdatasync_file(&unflushed_file).await?;
417 0 : self.file = Some(unflushed_file);
418 : } else {
419 : // We have unflushed data (write_lsn != flush_lsn), but no file.
420 : // This should only happen if last file was fully written and flushed,
421 : // but haven't updated flush_lsn yet.
422 0 : if self.write_lsn.segment_offset(self.wal_seg_size) != 0 {
423 0 : bail!(
424 0 : "unexpected unflushed data with no open file, write_lsn={}, flush_lsn={}",
425 0 : self.write_lsn,
426 0 : self.flush_record_lsn
427 0 : );
428 0 : }
429 : }
430 :
431 : // everything is flushed now, let's update flush_lsn
432 0 : self.flush_record_lsn = self.write_record_lsn;
433 0 : Ok(())
434 0 : }
435 :
436 : /// Truncate written WAL by removing all WAL segments after the given LSN.
437 : /// end_pos must point to the end of the WAL record.
438 0 : async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
439 0 : let _timer = WAL_STORAGE_OPERATION_SECONDS
440 0 : .with_label_values(&["truncate_wal"])
441 0 : .start_timer();
442 0 :
443 0 : // Streaming must not create a hole, so truncate cannot be called on non-written lsn
444 0 : if self.write_lsn != Lsn(0) && end_pos > self.write_lsn {
445 0 : bail!(
446 0 : "truncate_wal called on non-written WAL, write_lsn={}, end_pos={}",
447 0 : self.write_lsn,
448 0 : end_pos
449 0 : );
450 0 : }
451 0 :
452 0 : // Quick exit if nothing to do to avoid writing up to 16 MiB of zeros on
453 0 : // disk (this happens on each connect).
454 0 : if self.is_truncated_after_restart
455 0 : && end_pos == self.write_lsn
456 0 : && end_pos == self.flush_record_lsn
457 : {
458 0 : return Ok(());
459 0 : }
460 :
461 : // Close previously opened file, if any
462 0 : if let Some(unflushed_file) = self.file.take() {
463 0 : self.fdatasync_file(&unflushed_file).await?;
464 0 : }
465 :
466 0 : let xlogoff = end_pos.segment_offset(self.wal_seg_size);
467 0 : let segno = end_pos.segment_number(self.wal_seg_size);
468 0 :
469 0 : // Remove all segments after the given LSN.
470 0 : remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno).await?;
471 :
472 0 : let (mut file, is_partial) = self.open_or_create(segno).await?;
473 :
474 : // Fill end with zeroes
475 0 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
476 0 : write_zeroes(&mut file, self.wal_seg_size - xlogoff).await?;
477 0 : self.fdatasync_file(&file).await?;
478 :
479 0 : if !is_partial {
480 : // Make segment partial once again
481 0 : let (wal_file_path, wal_file_partial_path) =
482 0 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size);
483 0 : fs::rename(wal_file_path, wal_file_partial_path).await?;
484 0 : }
485 :
486 : // Update LSNs
487 0 : self.write_lsn = end_pos;
488 0 : self.write_record_lsn = end_pos;
489 0 : self.flush_record_lsn = end_pos;
490 0 : self.is_truncated_after_restart = true;
491 0 : Ok(())
492 0 : }
493 :
494 0 : fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
495 0 : let timeline_dir = self.timeline_dir.clone();
496 0 : let wal_seg_size = self.wal_seg_size;
497 0 : Box::pin(async move {
498 0 : remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to).await
499 0 : })
500 0 : }
501 :
502 0 : fn close(&mut self) {
503 0 : // close happens in destructor
504 0 : let _open_file = self.file.take();
505 0 : }
506 :
507 0 : fn get_metrics(&self) -> WalStorageMetrics {
508 0 : self.metrics.clone()
509 0 : }
510 : }
511 :
512 : /// Remove all WAL segments in timeline_dir that match the given predicate.
513 0 : async fn remove_segments_from_disk(
514 0 : timeline_dir: &Utf8Path,
515 0 : wal_seg_size: usize,
516 0 : remove_predicate: impl Fn(XLogSegNo) -> bool,
517 0 : ) -> Result<()> {
518 0 : let _timer = WAL_STORAGE_OPERATION_SECONDS
519 0 : .with_label_values(&["remove_segments_from_disk"])
520 0 : .start_timer();
521 0 :
522 0 : let mut n_removed = 0;
523 0 : let mut min_removed = u64::MAX;
524 0 : let mut max_removed = u64::MIN;
525 :
526 0 : let mut entries = fs::read_dir(timeline_dir).await?;
527 0 : while let Some(entry) = entries.next_entry().await? {
528 0 : let entry_path = entry.path();
529 0 : let fname = entry_path.file_name().unwrap();
530 :
531 0 : if let Some(fname_str) = fname.to_str() {
532 : /* Ignore files that are not XLOG segments */
533 0 : if !IsXLogFileName(fname_str) && !IsPartialXLogFileName(fname_str) {
534 0 : continue;
535 0 : }
536 0 : let (segno, _) = XLogFromFileName(fname_str, wal_seg_size);
537 0 : if remove_predicate(segno) {
538 0 : remove_file(entry_path).await?;
539 0 : n_removed += 1;
540 0 : min_removed = min(min_removed, segno);
541 0 : max_removed = max(max_removed, segno);
542 0 : REMOVED_WAL_SEGMENTS.inc();
543 0 : }
544 0 : }
545 : }
546 :
547 0 : if n_removed > 0 {
548 0 : info!(
549 0 : "removed {} WAL segments [{}; {}]",
550 : n_removed, min_removed, max_removed
551 : );
552 0 : }
553 0 : Ok(())
554 0 : }
555 :
556 : pub struct WalReader {
557 : remote_path: RemotePath,
558 : timeline_dir: Utf8PathBuf,
559 : wal_seg_size: usize,
560 : pos: Lsn,
561 : wal_segment: Option<Pin<Box<dyn AsyncRead + Send + Sync>>>,
562 :
563 : // S3 will be used to read WAL if LSN is not available locally
564 : enable_remote_read: bool,
565 :
566 : // We don't have WAL locally if LSN is less than local_start_lsn
567 : local_start_lsn: Lsn,
568 : // We will respond with zero-ed bytes before this Lsn as long as
569 : // pos is in the same segment as timeline_start_lsn.
570 : timeline_start_lsn: Lsn,
571 : // integer version number of PostgreSQL, e.g. 14; 15; 16
572 : pg_version: u32,
573 : system_id: SystemId,
574 : timeline_start_segment: Option<Bytes>,
575 : }
576 :
577 : impl WalReader {
578 0 : pub fn new(
579 0 : ttid: &TenantTimelineId,
580 0 : timeline_dir: Utf8PathBuf,
581 0 : state: &TimelinePersistentState,
582 0 : start_pos: Lsn,
583 0 : enable_remote_read: bool,
584 0 : ) -> Result<Self> {
585 0 : if state.server.wal_seg_size == 0 || state.local_start_lsn == Lsn(0) {
586 0 : bail!("state uninitialized, no data to read");
587 0 : }
588 0 :
589 0 : // TODO: Upgrade to bail!() once we know this couldn't possibly happen
590 0 : if state.timeline_start_lsn == Lsn(0) {
591 0 : warn!("timeline_start_lsn uninitialized before initializing wal reader");
592 0 : }
593 :
594 0 : if start_pos
595 0 : < state
596 0 : .timeline_start_lsn
597 0 : .segment_lsn(state.server.wal_seg_size as usize)
598 : {
599 0 : bail!(
600 0 : "Requested streaming from {}, which is before the start of the timeline {}, and also doesn't start at the first segment of that timeline",
601 0 : start_pos,
602 0 : state.timeline_start_lsn
603 0 : );
604 0 : }
605 0 :
606 0 : Ok(Self {
607 0 : remote_path: remote_timeline_path(ttid)?,
608 0 : timeline_dir,
609 0 : wal_seg_size: state.server.wal_seg_size as usize,
610 0 : pos: start_pos,
611 0 : wal_segment: None,
612 0 : enable_remote_read,
613 0 : local_start_lsn: state.local_start_lsn,
614 0 : timeline_start_lsn: state.timeline_start_lsn,
615 0 : pg_version: state.server.pg_version / 10000,
616 0 : system_id: state.server.system_id,
617 0 : timeline_start_segment: None,
618 : })
619 0 : }
620 :
621 : /// Read WAL at current position into provided buf, returns number of bytes
622 : /// read. It can be smaller than buf size only if segment boundary is
623 : /// reached.
624 0 : pub async fn read(&mut self, buf: &mut [u8]) -> Result<usize> {
625 0 : // If this timeline is new, we may not have a full segment yet, so
626 0 : // we pad the first bytes of the timeline's first WAL segment with 0s
627 0 : if self.pos < self.timeline_start_lsn {
628 0 : debug_assert_eq!(
629 0 : self.pos.segment_number(self.wal_seg_size),
630 0 : self.timeline_start_lsn.segment_number(self.wal_seg_size)
631 : );
632 :
633 : // All bytes after timeline_start_lsn are in WAL, but those before
634 : // are not, so we manually construct an empty segment for the bytes
635 : // not available in this timeline.
636 0 : if self.timeline_start_segment.is_none() {
637 0 : let it = postgres_ffi::generate_wal_segment(
638 0 : self.timeline_start_lsn.segment_number(self.wal_seg_size),
639 0 : self.system_id,
640 0 : self.pg_version,
641 0 : self.timeline_start_lsn,
642 0 : )?;
643 0 : self.timeline_start_segment = Some(it);
644 0 : }
645 :
646 0 : assert!(self.timeline_start_segment.is_some());
647 0 : let segment = self.timeline_start_segment.take().unwrap();
648 0 :
649 0 : let seg_bytes = &segment[..];
650 0 :
651 0 : // How much of the current segment have we already consumed?
652 0 : let pos_seg_offset = self.pos.segment_offset(self.wal_seg_size);
653 0 :
654 0 : // How many bytes may we consume in total?
655 0 : let tl_start_seg_offset = self.timeline_start_lsn.segment_offset(self.wal_seg_size);
656 0 :
657 0 : debug_assert!(seg_bytes.len() > pos_seg_offset);
658 0 : debug_assert!(seg_bytes.len() > tl_start_seg_offset);
659 :
660 : // Copy as many bytes as possible into the buffer
661 0 : let len = (tl_start_seg_offset - pos_seg_offset).min(buf.len());
662 0 : buf[0..len].copy_from_slice(&seg_bytes[pos_seg_offset..pos_seg_offset + len]);
663 0 :
664 0 : self.pos += len as u64;
665 0 :
666 0 : // If we're done with the segment, we can release it's memory.
667 0 : // However, if we're not yet done, store it so that we don't have to
668 0 : // construct the segment the next time this function is called.
669 0 : if self.pos < self.timeline_start_lsn {
670 0 : self.timeline_start_segment = Some(segment);
671 0 : }
672 :
673 0 : return Ok(len);
674 0 : }
675 :
676 0 : let mut wal_segment = match self.wal_segment.take() {
677 0 : Some(reader) => reader,
678 0 : None => self.open_segment().await?,
679 : };
680 :
681 : // How much to read and send in message? We cannot cross the WAL file
682 : // boundary, and we don't want send more than provided buffer.
683 0 : let xlogoff = self.pos.segment_offset(self.wal_seg_size);
684 0 : let send_size = min(buf.len(), self.wal_seg_size - xlogoff);
685 0 :
686 0 : // Read some data from the file.
687 0 : let buf = &mut buf[0..send_size];
688 0 : let send_size = wal_segment.read_exact(buf).await?;
689 0 : self.pos += send_size as u64;
690 0 :
691 0 : // Decide whether to reuse this file. If we don't set wal_segment here
692 0 : // a new reader will be opened next time.
693 0 : if self.pos.segment_offset(self.wal_seg_size) != 0 {
694 0 : self.wal_segment = Some(wal_segment);
695 0 : }
696 :
697 0 : Ok(send_size)
698 0 : }
699 :
700 : /// Open WAL segment at the current position of the reader.
701 0 : async fn open_segment(&self) -> Result<Pin<Box<dyn AsyncRead + Send + Sync>>> {
702 0 : let xlogoff = self.pos.segment_offset(self.wal_seg_size);
703 0 : let segno = self.pos.segment_number(self.wal_seg_size);
704 0 : let wal_file_name = XLogFileName(PG_TLI, segno, self.wal_seg_size);
705 0 :
706 0 : // Try to open local file, if we may have WAL locally
707 0 : if self.pos >= self.local_start_lsn {
708 0 : let res = open_wal_file(&self.timeline_dir, segno, self.wal_seg_size).await;
709 0 : match res {
710 0 : Ok((mut file, _)) => {
711 0 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
712 0 : return Ok(Box::pin(file));
713 : }
714 0 : Err(e) => {
715 0 : let is_not_found = e.chain().any(|e| {
716 0 : if let Some(e) = e.downcast_ref::<io::Error>() {
717 0 : e.kind() == io::ErrorKind::NotFound
718 : } else {
719 0 : false
720 : }
721 0 : });
722 0 : if !is_not_found {
723 0 : return Err(e);
724 0 : }
725 : // NotFound is expected, fall through to remote read
726 : }
727 : };
728 0 : }
729 :
730 : // Try to open remote file, if remote reads are enabled
731 0 : if self.enable_remote_read {
732 0 : let remote_wal_file_path = self.remote_path.join(&wal_file_name);
733 0 : return read_object(&remote_wal_file_path, xlogoff as u64).await;
734 0 : }
735 0 :
736 0 : bail!("WAL segment is not found")
737 0 : }
738 : }
739 :
740 : /// Zero block for filling created WAL segments.
741 : const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
742 :
743 : /// Helper for filling file with zeroes.
744 0 : async fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
745 0 : fail::fail_point!("sk-write-zeroes", |_| {
746 0 : info!("write_zeroes hit failpoint");
747 0 : Err(anyhow::anyhow!("failpoint: sk-write-zeroes"))
748 0 : });
749 :
750 0 : while count >= XLOG_BLCKSZ {
751 0 : file.write_all(ZERO_BLOCK).await?;
752 0 : count -= XLOG_BLCKSZ;
753 : }
754 0 : file.write_all(&ZERO_BLOCK[0..count]).await?;
755 0 : file.flush().await?;
756 0 : Ok(())
757 0 : }
758 :
759 : /// Helper function for opening WAL segment `segno` in `dir`. Returns file and
760 : /// whether it is .partial.
761 0 : pub(crate) async fn open_wal_file(
762 0 : timeline_dir: &Utf8Path,
763 0 : segno: XLogSegNo,
764 0 : wal_seg_size: usize,
765 0 : ) -> Result<(tokio::fs::File, bool)> {
766 0 : let (wal_file_path, wal_file_partial_path) = wal_file_paths(timeline_dir, segno, wal_seg_size);
767 0 :
768 0 : // First try to open the .partial file.
769 0 : let mut partial_path = wal_file_path.to_owned();
770 0 : partial_path.set_extension("partial");
771 0 : if let Ok(opened_file) = tokio::fs::File::open(&wal_file_partial_path).await {
772 0 : return Ok((opened_file, true));
773 0 : }
774 :
775 : // If that failed, try it without the .partial extension.
776 0 : let pf = tokio::fs::File::open(&wal_file_path)
777 0 : .await
778 0 : .with_context(|| format!("failed to open WAL file {:#}", wal_file_path))
779 0 : .map_err(|e| {
780 0 : warn!("{}", e);
781 0 : e
782 0 : })?;
783 :
784 0 : Ok((pf, false))
785 0 : }
786 :
787 : /// Helper returning full path to WAL segment file and its .partial brother.
788 0 : pub fn wal_file_paths(
789 0 : timeline_dir: &Utf8Path,
790 0 : segno: XLogSegNo,
791 0 : wal_seg_size: usize,
792 0 : ) -> (Utf8PathBuf, Utf8PathBuf) {
793 0 : let wal_file_name = XLogFileName(PG_TLI, segno, wal_seg_size);
794 0 : let wal_file_path = timeline_dir.join(wal_file_name.clone());
795 0 : let wal_file_partial_path = timeline_dir.join(wal_file_name + ".partial");
796 0 : (wal_file_path, wal_file_partial_path)
797 0 : }
|
