mudhorn/prometheus
1
0
Fork 0
mirror of https://github.com/prometheus/prometheus.git synced 2024-09-19 23:37:31 -07:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

tsdb: Replay m-map chunk only when required

Browse source 
M-map chunks replayed on startup are discarded if there
was no WAL and no snapshot loaded, because there is no
series created in the Head that it can map to. So only
load m-map chunks from disk if there is either a snapshot
loaded or there is WAL on disk.

Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
This commit is contained in:
Ganesh Vernekar 2023-03-13 11:15:45 +05:30
parent 6c008ec56a
commit c9d06f2826
No known key found for this signature in database
GPG key ID: F056451B52F1DC34
1 changed files with 28 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

44
tsdb/head.go
View file

@ -590,6 +590,7 @@ func (h *Head) Init(minValidTime int64) error {
snapIdx, snapOffset := -1, 0
refSeries := make(map[chunks.HeadSeriesRef]*memSeries)
snapshotLoaded := false
if h.opts.EnableMemorySnapshotOnShutdown {
level.Info(h.logger).Log("msg", "Chunk snapshot is enabled, replaying from the snapshot")
// If there are any WAL files, there should be at least one WAL file with an index that is current or newer
@ -619,6 +620,7 @@ func (h *Head) Init(minValidTime int64) error {
var err error
snapIdx, snapOffset, refSeries, err = h.loadChunkSnapshot()
if err == nil {
snapshotLoaded = true
level.Info(h.logger).Log("msg", "Chunk snapshot loading time", "duration", time.Since(start).String())
}
if err != nil {
@ -636,26 +638,36 @@ func (h *Head) Init(minValidTime int64) error {
}
mmapChunkReplayStart := time.Now()
mmappedChunks, oooMmappedChunks, lastMmapRef, err := h.loadMmappedChunks(refSeries)
if err != nil {
// TODO(codesome): clear out all m-map chunks here for refSeries.
level.Error(h.logger).Log("msg", "Loading on-disk chunks failed", "err", err)
if _, ok := errors.Cause(err).(*chunks.CorruptionErr); ok {
h.metrics.mmapChunkCorruptionTotal.Inc()
}
// Discard snapshot data since we need to replay the WAL for the missed m-map chunks data.
snapIdx, snapOffset = -1, 0
// If this fails, data will be recovered from WAL.
// Hence we wont lose any data (given WAL is not corrupt).
mmappedChunks, oooMmappedChunks, lastMmapRef, err = h.removeCorruptedMmappedChunks(err)
var (
mmappedChunks map[chunks.HeadSeriesRef][]*mmappedChunk
oooMmappedChunks map[chunks.HeadSeriesRef][]*mmappedChunk
lastMmapRef chunks.ChunkDiskMapperRef
err error
)
if snapshotLoaded || h.wal != nil {
// If snapshot was not loaded and if there is no WAL, then m-map chunks will be discarded
// anyway. So we only load m-map chunks when it won't be discarded.
mmappedChunks, oooMmappedChunks, lastMmapRef, err = h.loadMmappedChunks(refSeries)
if err != nil {
return err
// TODO(codesome): clear out all m-map chunks here for refSeries.
level.Error(h.logger).Log("msg", "Loading on-disk chunks failed", "err", err)
if _, ok := errors.Cause(err).(*chunks.CorruptionErr); ok {
h.metrics.mmapChunkCorruptionTotal.Inc()
}
// Discard snapshot data since we need to replay the WAL for the missed m-map chunks data.
snapIdx, snapOffset = -1, 0
// If this fails, data will be recovered from WAL.
// Hence we wont lose any data (given WAL is not corrupt).
mmappedChunks, oooMmappedChunks, lastMmapRef, err = h.removeCorruptedMmappedChunks(err)
if err != nil {
return err
}
}
level.Info(h.logger).Log("msg", "On-disk memory mappable chunks replay completed", "duration", time.Since(mmapChunkReplayStart).String())
}
level.Info(h.logger).Log("msg", "On-disk memory mappable chunks replay completed", "duration", time.Since(mmapChunkReplayStart).String())
if h.wal == nil {
level.Info(h.logger).Log("msg", "WAL not found")
return nil