rust/neon
1
0
Fork 0
mirror of https://github.com/neondatabase/neon.git synced 2026-10-21 19:23:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Rename full segment to .partial in WAL truncation if needed.

Browse Source 
c058e1cec2 omitted it if write_lsn is equal to truncation position, but
write_lsn is updated only on full record, so following was possible:
- insert half of record, crossing segment, and thus rename .partial to full;
  write_lsn doesn't change
- next truncate omits rename full back to .partial
- write after truncate errors about write into non-partial segment

Fixes regression looking like

ERROR WAL acceptor{cid=22 ...}:panic{thread=WAL acceptor
19b6c1743666ec02991a7633c57178db/b07db8c88f4c76ea5ed0954c04cc1e74
location=safekeeper/src/wal_storage.rs:230:13}: unexpected write into
non-partial segment file

I'd prefer to drop .partial segments completely, but let's do it separately.

Also fsync last segment on safekeeper restart to make sure we correctly init
flush_lsn.
This commit is contained in:
Arseny Sher
2023-06-11 17:11:02 +04:00
parent 2e687bca5b
commit 003e967a33
1 changed files with 43 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
safekeeper/src/wal_storage.rs
+43 -10
View File
@@ -146,15 +146,7 @@ impl PhysicalStorage {
// If not, maybe it's better to call fsync() here to be sure?
let flush_lsn = write_lsn;
debug!(
"initialized storage for timeline {}, flush_lsn={}, commit_lsn={}, peer_horizon_lsn={}",
ttid.timeline_id, flush_lsn, state.commit_lsn, state.peer_horizon_lsn,
);
if flush_lsn < state.commit_lsn || flush_lsn < state.peer_horizon_lsn {
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);
}
Ok(PhysicalStorage {
let mut storage = PhysicalStorage {
metrics: WalStorageMetrics::default(),
timeline_dir,
conf: conf.clone(),
@@ -164,7 +156,39 @@ impl PhysicalStorage {
flush_record_lsn: flush_lsn,
decoder: WalStreamDecoder::new(write_lsn, state.server.pg_version / 10000),
file: None,
})
};
if write_lsn != Lsn::INVALID {
// fsync last segment as we're setting flush_lsn to write_lsn.
// Previous segments must have been fsynced before starting new one.
let mut last_segno = storage.write_lsn.segment_number(storage.wal_seg_size);
// We could have stopped on the segment boundary; fsync the previous
// segment then, we might have crashed before that.
if storage.write_lsn.segment_offset(storage.wal_seg_size) == 0 {
last_segno -= 1;
}
// 1 is the first segment
if last_segno >= 1 {
// We can create timeline without underlying WAL (and do that in
// s3_wal_reply test) if e.g. it is in s3, so ignore missing
// file.
if let Ok((mut last_file, _)) = storage.open_or_create(last_segno) {
storage
.fsync_file(&mut last_file)
.with_context(|| format!("fsync last segment segno={last_segno}"))?;
}
}
}
debug!(
"initialized storage for timeline {}, flush_lsn={}, commit_lsn={}, peer_horizon_lsn={}",
ttid.timeline_id, flush_lsn, state.commit_lsn, state.peer_horizon_lsn,
);
if flush_lsn < state.commit_lsn || flush_lsn < state.peer_horizon_lsn {
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);
}
Ok(storage)
}
/// Get all known state of the storage.
@@ -382,6 +406,15 @@ impl Storage for PhysicalStorage {
// Quick exit if nothing to do to avoid writing up to 16 MiB of zeros on
// disk (this happens on each connect).
if end_pos == self.write_lsn {
// ... we still need to rename last to partial to confirm asserts.
// It's likely better to remove .partial altogether.
let segno = end_pos.segment_number(self.wal_seg_size);
let (_, is_partial) = self.open_or_create(segno)?;
if !is_partial {
let (wal_file_path, wal_file_partial_path) =
wal_file_paths(&self.timeline_dir, segno, self.wal_seg_size)?;
fs::rename(wal_file_path, wal_file_partial_path)?;
}
return Ok(());
}