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 futures::future::BoxFuture;
13 : use postgres_ffi::v14::xlog_utils::{IsPartialXLogFileName, IsXLogFileName, XLogFromFileName};
14 : use postgres_ffi::{XLogSegNo, PG_TLI};
15 : use remote_storage::RemotePath;
16 : use std::cmp::{max, min};
17 : use std::io::{self, SeekFrom};
18 : use std::path::{Path, PathBuf};
19 : use std::pin::Pin;
20 : use tokio::fs::{self, remove_file, File, OpenOptions};
21 : use tokio::io::{AsyncRead, AsyncWriteExt};
22 : use tokio::io::{AsyncReadExt, AsyncSeekExt};
23 : use tracing::*;
24 :
25 : use crate::metrics::{time_io_closure, WalStorageMetrics, REMOVED_WAL_SEGMENTS};
26 : use crate::safekeeper::SafeKeeperState;
27 : use crate::wal_backup::read_object;
28 : use crate::SafeKeeperConf;
29 : use postgres_ffi::waldecoder::WalStreamDecoder;
30 : use postgres_ffi::XLogFileName;
31 : use postgres_ffi::XLOG_BLCKSZ;
32 : use pq_proto::SystemId;
33 : use utils::{id::TenantTimelineId, lsn::Lsn};
34 :
35 : #[async_trait::async_trait]
36 : pub trait Storage {
37 : /// LSN of last durably stored WAL record.
38 : fn flush_lsn(&self) -> Lsn;
39 :
40 : /// Write piece of WAL from buf to disk, but not necessarily sync it.
41 : async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()>;
42 :
43 : /// Truncate WAL at specified LSN, which must be the end of WAL record.
44 : async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()>;
45 :
46 : /// Durably store WAL on disk, up to the last written WAL record.
47 : async fn flush_wal(&mut self) -> Result<()>;
48 :
49 : /// Remove all segments <= given segno. Returns function doing that as we
50 : /// want to perform it without timeline lock.
51 : fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>>;
52 :
53 : /// Release resources associated with the storage -- technically, close FDs.
54 : /// Currently we don't remove timelines until restart (#3146), so need to
55 : /// spare descriptors. This would be useful for temporary tli detach as
56 : /// well.
57 0 : fn close(&mut self) {}
58 :
59 : /// Get metrics for this timeline.
60 : fn get_metrics(&self) -> WalStorageMetrics;
61 : }
62 :
63 : /// PhysicalStorage is a storage that stores WAL on disk. Writes are separated from flushes
64 : /// for better performance. Storage is initialized in the constructor.
65 : ///
66 : /// WAL is stored in segments, each segment is a file. Last segment has ".partial" suffix in
67 : /// its filename and may be not fully flushed.
68 : ///
69 : /// Relationship of LSNs:
70 : /// `write_lsn` >= `write_record_lsn` >= `flush_record_lsn`
71 : ///
72 : /// When storage is created first time, all LSNs are zeroes and there are no segments on disk.
73 : pub struct PhysicalStorage {
74 : metrics: WalStorageMetrics,
75 : timeline_dir: PathBuf,
76 : conf: SafeKeeperConf,
77 :
78 : /// Size of WAL segment in bytes.
79 : wal_seg_size: usize,
80 :
81 : /// Written to disk, but possibly still in the cache and not fully persisted.
82 : /// Also can be ahead of record_lsn, if happen to be in the middle of a WAL record.
83 : write_lsn: Lsn,
84 :
85 : /// The LSN of the last WAL record written to disk. Still can be not fully flushed.
86 : write_record_lsn: Lsn,
87 :
88 : /// The LSN of the last WAL record flushed to disk.
89 : flush_record_lsn: Lsn,
90 :
91 : /// Decoder is required for detecting boundaries of WAL records.
92 : decoder: WalStreamDecoder,
93 :
94 : /// Cached open file for the last segment.
95 : ///
96 : /// If Some(file) is open, then it always:
97 : /// - has ".partial" suffix
98 : /// - points to write_lsn, so no seek is needed for writing
99 : /// - doesn't point to the end of the segment
100 : file: Option<File>,
101 :
102 : /// When false, we have just initialized storage using the LSN from find_end_of_wal().
103 : /// In this case, [`write_lsn`] can be less than actually written WAL on disk. In particular,
104 : /// there can be a case with unexpected .partial file.
105 : ///
106 : /// Imagine the following:
107 : /// - 000000010000000000000001
108 : /// - it was fully written, but the last record is split between 2 segments
109 : /// - after restart, `find_end_of_wal()` returned 0/1FFFFF0, which is in the end of this segment
110 : /// - `write_lsn`, `write_record_lsn` and `flush_record_lsn` were initialized to 0/1FFFFF0
111 : /// - 000000010000000000000002.partial
112 : /// - it has only 1 byte written, which is not enough to make a full WAL record
113 : ///
114 : /// Partial segment 002 has no WAL records, and it will be removed by the next truncate_wal().
115 : /// This flag will be set to true after the first truncate_wal() call.
116 : ///
117 : /// [`write_lsn`]: Self::write_lsn
118 : is_truncated_after_restart: bool,
119 : }
120 :
121 : impl PhysicalStorage {
122 : /// Create new storage. If commit_lsn is not zero, flush_lsn is tried to be restored from
123 : /// the disk. Otherwise, all LSNs are set to zero.
124 605 : pub fn new(
125 605 : ttid: &TenantTimelineId,
126 605 : timeline_dir: PathBuf,
127 605 : conf: &SafeKeeperConf,
128 605 : state: &SafeKeeperState,
129 605 : ) -> Result<PhysicalStorage> {
130 605 : let wal_seg_size = state.server.wal_seg_size as usize;
131 :
132 : // Find out where stored WAL ends, starting at commit_lsn which is a
133 : // known recent record boundary (unless we don't have WAL at all).
134 : //
135 : // NB: find_end_of_wal MUST be backwards compatible with the previously
136 : // written WAL. If find_end_of_wal fails to read any WAL written by an
137 : // older version of the code, we could lose data forever.
138 605 : let write_lsn = if state.commit_lsn == Lsn(0) {
139 514 : Lsn(0)
140 : } else {
141 91 : match state.server.pg_version / 10000 {
142 91 : 14 => postgres_ffi::v14::xlog_utils::find_end_of_wal(
143 91 : &timeline_dir,
144 91 : wal_seg_size,
145 91 : state.commit_lsn,
146 91 : )?,
147 0 : 15 => postgres_ffi::v15::xlog_utils::find_end_of_wal(
148 0 : &timeline_dir,
149 0 : wal_seg_size,
150 0 : state.commit_lsn,
151 0 : )?,
152 0 : _ => bail!("unsupported postgres version: {}", state.server.pg_version),
153 : }
154 : };
155 :
156 : // TODO: do we really know that write_lsn is fully flushed to disk?
157 : // If not, maybe it's better to call fsync() here to be sure?
158 605 : let flush_lsn = write_lsn;
159 605 :
160 605 : debug!(
161 0 : "initialized storage for timeline {}, flush_lsn={}, commit_lsn={}, peer_horizon_lsn={}",
162 0 : ttid.timeline_id, flush_lsn, state.commit_lsn, state.peer_horizon_lsn,
163 0 : );
164 605 : if flush_lsn < state.commit_lsn || flush_lsn < state.peer_horizon_lsn {
165 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);
166 605 : }
167 :
168 605 : Ok(PhysicalStorage {
169 605 : metrics: WalStorageMetrics::default(),
170 605 : timeline_dir,
171 605 : conf: conf.clone(),
172 605 : wal_seg_size,
173 605 : write_lsn,
174 605 : write_record_lsn: write_lsn,
175 605 : flush_record_lsn: flush_lsn,
176 605 : decoder: WalStreamDecoder::new(write_lsn, state.server.pg_version / 10000),
177 605 : file: None,
178 605 : is_truncated_after_restart: false,
179 605 : })
180 605 : }
181 :
182 : /// Get all known state of the storage.
183 5 : pub fn internal_state(&self) -> (Lsn, Lsn, Lsn, bool) {
184 5 : (
185 5 : self.write_lsn,
186 5 : self.write_record_lsn,
187 5 : self.flush_record_lsn,
188 5 : self.file.is_some(),
189 5 : )
190 5 : }
191 :
192 : /// Call fdatasync if config requires so.
193 1187301 : async fn fdatasync_file(&mut self, file: &mut File) -> Result<()> {
194 1187301 : if !self.conf.no_sync {
195 : self.metrics
196 0 : .observe_flush_seconds(time_io_closure(file.sync_data()).await?);
197 1187301 : }
198 1187301 : Ok(())
199 1187301 : }
200 :
201 : /// Call fsync if config requires so.
202 1331 : async fn fsync_file(&mut self, file: &mut File) -> Result<()> {
203 1331 : if !self.conf.no_sync {
204 : self.metrics
205 0 : .observe_flush_seconds(time_io_closure(file.sync_all()).await?);
206 1331 : }
207 1331 : Ok(())
208 1331 : }
209 :
210 : /// Open or create WAL segment file. Caller must call seek to the wanted position.
211 : /// Returns `file` and `is_partial`.
212 1981 : async fn open_or_create(&mut self, segno: XLogSegNo) -> Result<(File, bool)> {
213 1981 : let (wal_file_path, wal_file_partial_path) =
214 1981 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
215 :
216 : // Try to open already completed segment
217 1981 : if let Ok(file) = OpenOptions::new().write(true).open(&wal_file_path).await {
218 0 : Ok((file, false))
219 1981 : } else if let Ok(file) = OpenOptions::new()
220 1981 : .write(true)
221 1981 : .open(&wal_file_partial_path)
222 1957 : .await
223 : {
224 : // Try to open existing partial file
225 650 : Ok((file, true))
226 : } else {
227 : // Create and fill new partial file
228 1331 : let mut file = OpenOptions::new()
229 1331 : .create(true)
230 1331 : .write(true)
231 1331 : .open(&wal_file_partial_path)
232 1317 : .await
233 1331 : .with_context(|| format!("Failed to open log file {:?}", &wal_file_path))?;
234 :
235 2632805 : write_zeroes(&mut file, self.wal_seg_size).await?;
236 1331 : self.fsync_file(&mut file).await?;
237 1331 : Ok((file, true))
238 : }
239 1981 : }
240 :
241 : /// Write WAL bytes, which are known to be located in a single WAL segment.
242 1461159 : async fn write_in_segment(&mut self, segno: u64, xlogoff: usize, buf: &[u8]) -> Result<()> {
243 1461159 : let mut file = if let Some(file) = self.file.take() {
244 1459781 : file
245 : } else {
246 1598117 : let (mut file, is_partial) = self.open_or_create(segno).await?;
247 1378 : assert!(is_partial, "unexpected write into non-partial segment file");
248 1378 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
249 1378 : file
250 : };
251 :
252 1461159 : file.write_all(buf).await?;
253 : // Note: flush just ensures write above reaches the OS (this is not
254 : // needed in case of sync IO as Write::write there calls directly write
255 : // syscall, but needed in case of async). It does *not* fsyncs the file.
256 1461159 : file.flush().await?;
257 :
258 1461159 : if xlogoff + buf.len() == self.wal_seg_size {
259 : // If we reached the end of a WAL segment, flush and close it.
260 818 : self.fdatasync_file(&mut file).await?;
261 :
262 : // Rename partial file to completed file
263 818 : let (wal_file_path, wal_file_partial_path) =
264 818 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
265 818 : fs::rename(wal_file_partial_path, wal_file_path).await?;
266 1460341 : } else {
267 1460341 : // otherwise, file can be reused later
268 1460341 : self.file = Some(file);
269 1460341 : }
270 :
271 1461159 : Ok(())
272 1461159 : }
273 :
274 : /// Writes WAL to the segment files, until everything is writed. If some segments
275 : /// are fully written, they are flushed to disk. The last (partial) segment can
276 : /// be flushed separately later.
277 : ///
278 : /// Updates `write_lsn`.
279 1460402 : async fn write_exact(&mut self, pos: Lsn, mut buf: &[u8]) -> Result<()> {
280 1460402 : if self.write_lsn != pos {
281 : // need to flush the file before discarding it
282 0 : if let Some(mut file) = self.file.take() {
283 0 : self.fdatasync_file(&mut file).await?;
284 0 : }
285 :
286 0 : self.write_lsn = pos;
287 1460402 : }
288 :
289 2921561 : while !buf.is_empty() {
290 : // Extract WAL location for this block
291 1461159 : let xlogoff = self.write_lsn.segment_offset(self.wal_seg_size);
292 1461159 : let segno = self.write_lsn.segment_number(self.wal_seg_size);
293 :
294 : // If crossing a WAL boundary, only write up until we reach wal segment size.
295 1461159 : let bytes_write = if xlogoff + buf.len() > self.wal_seg_size {
296 757 : self.wal_seg_size - xlogoff
297 : } else {
298 1460402 : buf.len()
299 : };
300 :
301 1461159 : self.write_in_segment(segno, xlogoff, &buf[..bytes_write])
302 3041998 : .await?;
303 1461159 : self.write_lsn += bytes_write as u64;
304 1461159 : buf = &buf[bytes_write..];
305 : }
306 :
307 1460402 : Ok(())
308 1460402 : }
309 : }
310 :
311 : #[async_trait::async_trait]
312 : impl Storage for PhysicalStorage {
313 : /// flush_lsn returns LSN of last durably stored WAL record.
314 9726306 : fn flush_lsn(&self) -> Lsn {
315 9726306 : self.flush_record_lsn
316 9726306 : }
317 :
318 : /// Write WAL to disk.
319 1460402 : async fn write_wal(&mut self, startpos: Lsn, buf: &[u8]) -> Result<()> {
320 : // Disallow any non-sequential writes, which can result in gaps or overwrites.
321 : // If we need to move the pointer, use truncate_wal() instead.
322 1460402 : if self.write_lsn > startpos {
323 0 : bail!(
324 0 : "write_wal rewrites WAL written before, write_lsn={}, startpos={}",
325 0 : self.write_lsn,
326 0 : startpos
327 0 : );
328 1460402 : }
329 1460402 : if self.write_lsn < startpos && self.write_lsn != Lsn(0) {
330 0 : bail!(
331 0 : "write_wal creates gap in written WAL, write_lsn={}, startpos={}",
332 0 : self.write_lsn,
333 0 : startpos
334 0 : );
335 1460402 : }
336 :
337 3041998 : let write_seconds = time_io_closure(self.write_exact(startpos, buf)).await?;
338 : // WAL is written, updating write metrics
339 1460402 : self.metrics.observe_write_seconds(write_seconds);
340 1460402 : self.metrics.observe_write_bytes(buf.len());
341 1460402 :
342 1460402 : // figure out last record's end lsn for reporting (if we got the
343 1460402 : // whole record)
344 1460402 : if self.decoder.available() != startpos {
345 492 : info!(
346 492 : "restart decoder from {} to {}",
347 492 : self.decoder.available(),
348 492 : startpos,
349 492 : );
350 492 : let pg_version = self.decoder.pg_version;
351 492 : self.decoder = WalStreamDecoder::new(startpos, pg_version);
352 1459910 : }
353 1460402 : self.decoder.feed_bytes(buf);
354 : loop {
355 103016069 : match self.decoder.poll_decode()? {
356 1460402 : None => break, // no full record yet
357 101555667 : Some((lsn, _rec)) => {
358 101555667 : self.write_record_lsn = lsn;
359 101555667 : }
360 : }
361 : }
362 :
363 1460402 : Ok(())
364 2920804 : }
365 :
366 2202314 : async fn flush_wal(&mut self) -> Result<()> {
367 2202314 : if self.flush_record_lsn == self.write_record_lsn {
368 : // no need to do extra flush
369 1016424 : return Ok(());
370 1185890 : }
371 :
372 1185890 : if let Some(mut unflushed_file) = self.file.take() {
373 1185871 : self.fdatasync_file(&mut unflushed_file).await?;
374 1185871 : self.file = Some(unflushed_file);
375 : } else {
376 : // We have unflushed data (write_lsn != flush_lsn), but no file.
377 : // This should only happen if last file was fully written and flushed,
378 : // but haven't updated flush_lsn yet.
379 19 : if self.write_lsn.segment_offset(self.wal_seg_size) != 0 {
380 0 : bail!(
381 0 : "unexpected unflushed data with no open file, write_lsn={}, flush_lsn={}",
382 0 : self.write_lsn,
383 0 : self.flush_record_lsn
384 0 : );
385 19 : }
386 : }
387 :
388 : // everything is flushed now, let's update flush_lsn
389 1185890 : self.flush_record_lsn = self.write_record_lsn;
390 1185890 : Ok(())
391 4404628 : }
392 :
393 : /// Truncate written WAL by removing all WAL segments after the given LSN.
394 : /// end_pos must point to the end of the WAL record.
395 971 : async fn truncate_wal(&mut self, end_pos: Lsn) -> Result<()> {
396 : // Streaming must not create a hole, so truncate cannot be called on non-written lsn
397 971 : if self.write_lsn != Lsn(0) && end_pos > self.write_lsn {
398 0 : bail!(
399 0 : "truncate_wal called on non-written WAL, write_lsn={}, end_pos={}",
400 0 : self.write_lsn,
401 0 : end_pos
402 0 : );
403 971 : }
404 971 :
405 971 : // Quick exit if nothing to do to avoid writing up to 16 MiB of zeros on
406 971 : // disk (this happens on each connect).
407 971 : if self.is_truncated_after_restart
408 377 : && end_pos == self.write_lsn
409 368 : && end_pos == self.flush_record_lsn
410 : {
411 368 : return Ok(());
412 603 : }
413 :
414 : // Close previously opened file, if any
415 603 : if let Some(mut unflushed_file) = self.file.take() {
416 9 : self.fdatasync_file(&mut unflushed_file).await?;
417 594 : }
418 :
419 603 : let xlogoff = end_pos.segment_offset(self.wal_seg_size);
420 603 : let segno = end_pos.segment_number(self.wal_seg_size);
421 603 :
422 603 : // Remove all segments after the given LSN.
423 1204 : remove_segments_from_disk(&self.timeline_dir, self.wal_seg_size, |x| x > segno).await?;
424 :
425 1039915 : let (mut file, is_partial) = self.open_or_create(segno).await?;
426 :
427 : // Fill end with zeroes
428 603 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
429 692101 : write_zeroes(&mut file, self.wal_seg_size - xlogoff).await?;
430 603 : self.fdatasync_file(&mut file).await?;
431 :
432 603 : if !is_partial {
433 : // Make segment partial once again
434 0 : let (wal_file_path, wal_file_partial_path) =
435 0 : wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
436 0 : fs::rename(wal_file_path, wal_file_partial_path).await?;
437 603 : }
438 :
439 : // Update LSNs
440 603 : self.write_lsn = end_pos;
441 603 : self.write_record_lsn = end_pos;
442 603 : self.flush_record_lsn = end_pos;
443 603 : self.is_truncated_after_restart = true;
444 603 : Ok(())
445 1942 : }
446 :
447 14 : fn remove_up_to(&self, segno_up_to: XLogSegNo) -> BoxFuture<'static, anyhow::Result<()>> {
448 14 : let timeline_dir = self.timeline_dir.clone();
449 14 : let wal_seg_size = self.wal_seg_size;
450 14 : Box::pin(async move {
451 54 : remove_segments_from_disk(&timeline_dir, wal_seg_size, |x| x <= segno_up_to).await
452 14 : })
453 14 : }
454 :
455 31 : fn close(&mut self) {
456 31 : // close happens in destructor
457 31 : let _open_file = self.file.take();
458 31 : }
459 :
460 51 : fn get_metrics(&self) -> WalStorageMetrics {
461 51 : self.metrics.clone()
462 51 : }
463 : }
464 :
465 : /// Remove all WAL segments in timeline_dir that match the given predicate.
466 617 : async fn remove_segments_from_disk(
467 617 : timeline_dir: &Path,
468 617 : wal_seg_size: usize,
469 617 : remove_predicate: impl Fn(XLogSegNo) -> bool,
470 617 : ) -> Result<()> {
471 617 : let mut n_removed = 0;
472 617 : let mut min_removed = u64::MAX;
473 617 : let mut max_removed = u64::MIN;
474 :
475 617 : let mut entries = fs::read_dir(timeline_dir).await?;
476 1466 : while let Some(entry) = entries.next_entry().await? {
477 849 : let entry_path = entry.path();
478 849 : let fname = entry_path.file_name().unwrap();
479 :
480 849 : if let Some(fname_str) = fname.to_str() {
481 : /* Ignore files that are not XLOG segments */
482 849 : if !IsXLogFileName(fname_str) && !IsPartialXLogFileName(fname_str) {
483 617 : continue;
484 232 : }
485 232 : let (segno, _) = XLogFromFileName(fname_str, wal_seg_size);
486 232 : if remove_predicate(segno) {
487 26 : remove_file(entry_path).await?;
488 26 : n_removed += 1;
489 26 : min_removed = min(min_removed, segno);
490 26 : max_removed = max(max_removed, segno);
491 26 : REMOVED_WAL_SEGMENTS.inc();
492 206 : }
493 0 : }
494 : }
495 :
496 617 : if n_removed > 0 {
497 13 : info!(
498 13 : "removed {} WAL segments [{}; {}]",
499 13 : n_removed, min_removed, max_removed
500 13 : );
501 604 : }
502 617 : Ok(())
503 617 : }
504 :
505 : pub struct WalReader {
506 : workdir: PathBuf,
507 : timeline_dir: PathBuf,
508 : wal_seg_size: usize,
509 : pos: Lsn,
510 : wal_segment: Option<Pin<Box<dyn AsyncRead + Send + Sync>>>,
511 :
512 : // S3 will be used to read WAL if LSN is not available locally
513 : enable_remote_read: bool,
514 :
515 : // We don't have WAL locally if LSN is less than local_start_lsn
516 : local_start_lsn: Lsn,
517 : // We will respond with zero-ed bytes before this Lsn as long as
518 : // pos is in the same segment as timeline_start_lsn.
519 : timeline_start_lsn: Lsn,
520 : // integer version number of PostgreSQL, e.g. 14; 15; 16
521 : pg_version: u32,
522 : system_id: SystemId,
523 : timeline_start_segment: Option<Bytes>,
524 : }
525 :
526 : impl WalReader {
527 830 : pub fn new(
528 830 : workdir: PathBuf,
529 830 : timeline_dir: PathBuf,
530 830 : state: &SafeKeeperState,
531 830 : start_pos: Lsn,
532 830 : enable_remote_read: bool,
533 830 : ) -> Result<Self> {
534 830 : if state.server.wal_seg_size == 0 || state.local_start_lsn == Lsn(0) {
535 0 : bail!("state uninitialized, no data to read");
536 830 : }
537 830 :
538 830 : // TODO: Upgrade to bail!() once we know this couldn't possibly happen
539 830 : if state.timeline_start_lsn == Lsn(0) {
540 8 : warn!("timeline_start_lsn uninitialized before initializing wal reader");
541 822 : }
542 :
543 830 : if start_pos
544 830 : < state
545 830 : .timeline_start_lsn
546 830 : .segment_lsn(state.server.wal_seg_size as usize)
547 : {
548 0 : bail!(
549 0 : "Requested streaming from {}, which is before the start of the timeline {}, and also doesn't start at the first segment of that timeline",
550 0 : start_pos,
551 0 : state.timeline_start_lsn
552 0 : );
553 830 : }
554 830 :
555 830 : Ok(Self {
556 830 : workdir,
557 830 : timeline_dir,
558 830 : wal_seg_size: state.server.wal_seg_size as usize,
559 830 : pos: start_pos,
560 830 : wal_segment: None,
561 830 : enable_remote_read,
562 830 : local_start_lsn: state.local_start_lsn,
563 830 : timeline_start_lsn: state.timeline_start_lsn,
564 830 : pg_version: state.server.pg_version / 10000,
565 830 : system_id: state.server.system_id,
566 830 : timeline_start_segment: None,
567 830 : })
568 830 : }
569 :
570 740438 : pub async fn read(&mut self, buf: &mut [u8]) -> Result<usize> {
571 740438 : // If this timeline is new, we may not have a full segment yet, so
572 740438 : // we pad the first bytes of the timeline's first WAL segment with 0s
573 740438 : if self.pos < self.timeline_start_lsn {
574 : debug_assert_eq!(
575 53 : self.pos.segment_number(self.wal_seg_size),
576 53 : self.timeline_start_lsn.segment_number(self.wal_seg_size)
577 : );
578 :
579 : // All bytes after timeline_start_lsn are in WAL, but those before
580 : // are not, so we manually construct an empty segment for the bytes
581 : // not available in this timeline.
582 53 : if self.timeline_start_segment.is_none() {
583 1 : let it = postgres_ffi::generate_wal_segment(
584 1 : self.timeline_start_lsn.segment_number(self.wal_seg_size),
585 1 : self.system_id,
586 1 : self.pg_version,
587 1 : self.timeline_start_lsn,
588 1 : )?;
589 1 : self.timeline_start_segment = Some(it);
590 52 : }
591 :
592 53 : assert!(self.timeline_start_segment.is_some());
593 53 : let segment = self.timeline_start_segment.take().unwrap();
594 53 :
595 53 : let seg_bytes = &segment[..];
596 53 :
597 53 : // How much of the current segment have we already consumed?
598 53 : let pos_seg_offset = self.pos.segment_offset(self.wal_seg_size);
599 53 :
600 53 : // How many bytes may we consume in total?
601 53 : let tl_start_seg_offset = self.timeline_start_lsn.segment_offset(self.wal_seg_size);
602 :
603 53 : debug_assert!(seg_bytes.len() > pos_seg_offset);
604 53 : debug_assert!(seg_bytes.len() > tl_start_seg_offset);
605 :
606 : // Copy as many bytes as possible into the buffer
607 53 : let len = (tl_start_seg_offset - pos_seg_offset).min(buf.len());
608 53 : buf[0..len].copy_from_slice(&seg_bytes[pos_seg_offset..pos_seg_offset + len]);
609 53 :
610 53 : self.pos += len as u64;
611 53 :
612 53 : // If we're done with the segment, we can release it's memory.
613 53 : // However, if we're not yet done, store it so that we don't have to
614 53 : // construct the segment the next time this function is called.
615 53 : if self.pos < self.timeline_start_lsn {
616 52 : self.timeline_start_segment = Some(segment);
617 52 : }
618 :
619 53 : return Ok(len);
620 740385 : }
621 :
622 740385 : let mut wal_segment = match self.wal_segment.take() {
623 739038 : Some(reader) => reader,
624 2893 : None => self.open_segment().await?,
625 : };
626 :
627 : // How much to read and send in message? We cannot cross the WAL file
628 : // boundary, and we don't want send more than provided buffer.
629 740385 : let xlogoff = self.pos.segment_offset(self.wal_seg_size);
630 740385 : let send_size = min(buf.len(), self.wal_seg_size - xlogoff);
631 740385 :
632 740385 : // Read some data from the file.
633 740385 : let buf = &mut buf[0..send_size];
634 740385 : let send_size = wal_segment.read_exact(buf).await?;
635 740384 : self.pos += send_size as u64;
636 740384 :
637 740384 : // Decide whether to reuse this file. If we don't set wal_segment here
638 740384 : // a new reader will be opened next time.
639 740384 : if self.pos.segment_offset(self.wal_seg_size) != 0 {
640 739652 : self.wal_segment = Some(wal_segment);
641 739652 : }
642 :
643 740384 : Ok(send_size)
644 740437 : }
645 :
646 : /// Open WAL segment at the current position of the reader.
647 1347 : async fn open_segment(&self) -> Result<Pin<Box<dyn AsyncRead + Send + Sync>>> {
648 1347 : let xlogoff = self.pos.segment_offset(self.wal_seg_size);
649 1347 : let segno = self.pos.segment_number(self.wal_seg_size);
650 1347 : let wal_file_name = XLogFileName(PG_TLI, segno, self.wal_seg_size);
651 1347 : let wal_file_path = self.timeline_dir.join(wal_file_name);
652 1347 :
653 1347 : // Try to open local file, if we may have WAL locally
654 1347 : if self.pos >= self.local_start_lsn {
655 1566 : let res = Self::open_wal_file(&wal_file_path).await;
656 1341 : match res {
657 1341 : Ok(mut file) => {
658 1341 : file.seek(SeekFrom::Start(xlogoff as u64)).await?;
659 1341 : return Ok(Box::pin(file));
660 : }
661 0 : Err(e) => {
662 0 : let is_not_found = e.chain().any(|e| {
663 0 : if let Some(e) = e.downcast_ref::<io::Error>() {
664 0 : e.kind() == io::ErrorKind::NotFound
665 : } else {
666 0 : false
667 : }
668 0 : });
669 0 : if !is_not_found {
670 0 : return Err(e);
671 0 : }
672 : // NotFound is expected, fall through to remote read
673 : }
674 : };
675 6 : }
676 :
677 : // Try to open remote file, if remote reads are enabled
678 6 : if self.enable_remote_read {
679 6 : let remote_wal_file_path = wal_file_path
680 6 : .strip_prefix(&self.workdir)
681 6 : .context("Failed to strip workdir prefix")
682 6 : .and_then(RemotePath::new)
683 6 : .with_context(|| {
684 0 : format!(
685 0 : "Failed to resolve remote part of path {:?} for base {:?}",
686 0 : wal_file_path, self.workdir,
687 0 : )
688 6 : })?;
689 21 : return read_object(&remote_wal_file_path, xlogoff as u64).await;
690 0 : }
691 0 :
692 0 : bail!("WAL segment is not found")
693 1347 : }
694 :
695 : /// Helper function for opening a wal file.
696 1341 : async fn open_wal_file(wal_file_path: &Path) -> Result<tokio::fs::File> {
697 1341 : // First try to open the .partial file.
698 1341 : let mut partial_path = wal_file_path.to_owned();
699 1341 : partial_path.set_extension("partial");
700 1341 : if let Ok(opened_file) = tokio::fs::File::open(&partial_path).await {
701 1101 : return Ok(opened_file);
702 240 : }
703 240 :
704 240 : // If that failed, try it without the .partial extension.
705 240 : tokio::fs::File::open(&wal_file_path)
706 241 : .await
707 240 : .with_context(|| format!("Failed to open WAL file {:?}", wal_file_path))
708 240 : .map_err(|e| {
709 0 : warn!("{}", e);
710 0 : e
711 240 : })
712 1341 : }
713 : }
714 :
715 : /// Zero block for filling created WAL segments.
716 : const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
717 :
718 : /// Helper for filling file with zeroes.
719 1934 : async fn write_zeroes(file: &mut File, mut count: usize) -> Result<()> {
720 3433682 : while count >= XLOG_BLCKSZ {
721 3431748 : file.write_all(ZERO_BLOCK).await?;
722 3431748 : count -= XLOG_BLCKSZ;
723 : }
724 1934 : file.write_all(&ZERO_BLOCK[0..count]).await?;
725 1934 : file.flush().await?;
726 1934 : Ok(())
727 1934 : }
728 :
729 : /// Helper returning full path to WAL segment file and its .partial brother.
730 2799 : fn wal_file_paths(
731 2799 : timeline_dir: &Path,
732 2799 : segno: XLogSegNo,
733 2799 : wal_seg_size: usize,
734 2799 : ) -> Result<(PathBuf, PathBuf)> {
735 2799 : let wal_file_name = XLogFileName(PG_TLI, segno, wal_seg_size);
736 2799 : let wal_file_path = timeline_dir.join(wal_file_name.clone());
737 2799 : let wal_file_partial_path = timeline_dir.join(wal_file_name + ".partial");
738 2799 : Ok((wal_file_path, wal_file_partial_path))
739 2799 : }
|
