A new stat for bytes read off the disk #1702
Conversation
Thejas-bhat
left a comment
Thejas-bhat
left a comment
There was a problem hiding this comment.
Thanks sreekanth, I've included your suggestions
index/scorch/merge.go
Outdated
| } | ||
|
|
||
| func cumulateBytesRead(sbs []segment.Segment) uint64 { | ||
| rv := uint64(0) |
index/scorch/scorch.go
Outdated
| analysisResults[itemsDeQueued] = result | ||
| itemsDeQueued++ | ||
| result.VisitFields(func(f index.Field) { | ||
| atomic.AddUint64(&s.stats.TotIndexedAnalysisBytes, |
There was a problem hiding this comment.
better stats wordings
TotBytesIndexedAfterAnalysis?
TotBytesReadDuringQueryTime Or TotBytesReadAtQueryTime
so that both stats share the same prefix.
index/scorch/snapshot_index.go
Outdated
| for index, segment := range i.segment { | ||
| go func(index int, segment *SegmentSnapshot) { | ||
| var prevBytesRead uint64 | ||
| if seg, ok := segment.segment.(diskStatsReporter); ok { |
There was a problem hiding this comment.
same interface checks were repeated at 152 and 159.
Can't we make that to 1?
index/scorch/snapshot_index.go
Outdated
| postings.BytesRead()-prevBytesReadPL) | ||
| } | ||
|
|
||
| if itr, ok := rv.iterators[i].(diskStatsReporter); ok && |
There was a problem hiding this comment.
avoid repeated interface checks for the same vars..
There was a problem hiding this comment.
there is a re-initialisation of the var rv.iterators[i], so i can't reuse the same variable over here, since the bytes read as part of the iterator initialisation won't be part of the old variable.
| analysisResults := make([]index.Document, int(numUpdates)) | ||
| var itemsDeQueued uint64 | ||
| var totalAnalysisSize int | ||
| analysisBytes := func(tokMap index.TokenFrequencies) (rv uint64) { |
There was a problem hiding this comment.
Should we add a flag here to calculate this only when the flag is set?
Feel like this is something we don't always necessarily need to do - hope none of this calculation shows up in CPU profiles?
There was a problem hiding this comment.
Not sure if this is enough, we've to record copies - stored values and doc values. Also term vectors. I don't see you doing this in your zap PR.
There was a problem hiding this comment.
Just to close the thread, these two changes are going to be included in upcoming PRs.
| analysisResults := make([]index.Document, int(numUpdates)) | ||
| var itemsDeQueued uint64 | ||
| var totalAnalysisSize int | ||
| analysisBytes := func(tokMap index.TokenFrequencies) (rv uint64) { |
There was a problem hiding this comment.
Not sure if this is enough, we've to record copies - stored values and doc values. Also term vectors. I don't see you doing this in your zap PR.
| } | ||
|
|
||
| switch segI := seg.(type) { | ||
| case segment.DiskStatsReporter: |
There was a problem hiding this comment.
Per my latest conversation with Jon, we shouldn't be accounting for content written to disk by the merger. More conversation needed here I suppose.
There was a problem hiding this comment.
That is the case here as well, we are only accounting for the BytesRead from disk for the new merge created segment. And this would anyway be needed/accounted for the next query on this newly formed segment.
There was a problem hiding this comment.
We need to carry further the stats from the older segments that got merged into the newly formed segment so that the accounting of stats is cumulative. And hence the need for the SetBytesRead() api.
If we fail to consider the stats from the pre-merge segments, then the bytes accounting go wrong.
Have posted alternative thought in zapx PR.
index/scorch/snapshot_index.go
Outdated
|
|
||
| rvd := document.NewDocument(id) | ||
| var prevBytesRead uint64 | ||
| seg, ok := i.segment[segmentIndex].segment.(segment.DiskStatsReporter) |
There was a problem hiding this comment.
Rename ok to say diskStatsAvailable so in line 472 below we know why things are ok.
index/scorch/snapshot_index.go
Outdated
| rv.dicts = make([]segment.TermDictionary, len(is.segment)) | ||
| for i, segment := range is.segment { | ||
| var prevBytesRead uint64 | ||
| segP, ok := segment.segment.(diskStatsReporter) |
There was a problem hiding this comment.
same here, rename ok to diskStatsAvailable.
index/scorch/snapshot_index.go
Outdated
| if ssvOk && ssv != nil && len(vFields) > 0 { | ||
| var prevBytesRead uint64 | ||
| ssvp, ok := ssv.(segment.DiskStatsReporter) | ||
| if ok { |
index/scorch/snapshot_index_tfr.go
Outdated
| // find the next hit | ||
| for i.segmentOffset < len(i.iterators) { | ||
| prevBytesRead := uint64(0) | ||
| itr, ok := i.iterators[i.segmentOffset].(segment.DiskStatsReporter) |
3 participants
No description provided.