Merge quantile sketches as they arrive #11544
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Trivy scan found the following vulnerabilities:
|
jeschkies
commented
Jan 2, 2024
Comment on lines
+298
to
+326
| type quantileSketchAccumulator struct { | ||
| matrix logql.ProbabilisticQuantileMatrix | ||
| } | ||
|
|
||
| func newQuantileSketchAccumulator() *quantileSketchAccumulator { | ||
| return &quantileSketchAccumulator{} | ||
| } | ||
|
|
||
| func (a *quantileSketchAccumulator) Accumulate(res logqlmodel.Result, _ int) error { | ||
| if res.Data.Type() != logql.QuantileSketchMatrixType { | ||
| return fmt.Errorf("unexpected matrix data type: got (%s), want (%s)", res.Data.Type(), logql.QuantileSketchMatrixType) | ||
| } | ||
| data, ok := res.Data.(logql.ProbabilisticQuantileMatrix) | ||
| if !ok { | ||
| return fmt.Errorf("unexpected matrix type: got (%T), want (ProbabilisticQuantileMatrix)", res.Data) | ||
| } | ||
| if a.matrix == nil { | ||
| a.matrix = data | ||
| return nil | ||
| } | ||
|
|
||
| var err error | ||
| a.matrix, err = a.matrix.Merge(data) | ||
| return err | ||
| } | ||
|
|
||
| func (a *quantileSketchAccumulator) Result() []logqlmodel.Result { | ||
| return []logqlmodel.Result{{Data: a.matrix}} | ||
| } |
There was a problem hiding this comment.
Should this be the place where we merge results?
|
I still need to benchmark this. |
that we use to store and check stream hashes Signed-off-by: Callum Styan <callumstyan@gmail.com>
Signed-off-by: Callum Styan <callumstyan@gmail.com>
|
I've pushed two commits to your branch, one with benchmarks for single merge vs accumulator and one for an optimization to the map usage in the quantile matrix/vector merging. First looking at just the single merge as we have it prior to the accumulator change, with and without the pooling: Then the pooled single merge vs accumulator merge: And the final result with my two commits vs the single merge usage: |
jeschkies
commented
Jan 3, 2024
Comment on lines
613
to
639
| func BenchmarkAccumulateMerge(b *testing.B) { | ||
| b.ReportAllocs() | ||
| results := make([]logql.ProbabilisticQuantileVector, 100) | ||
| for i := range results { | ||
| results[i] = make(logql.ProbabilisticQuantileVector, 100) | ||
| for j := range results[i] { | ||
| results[i][j] = logql.ProbabilisticQuantileSample{ | ||
| T: int64(i), | ||
| F: newRandomSketch(), | ||
| Metric: []labels.Label{{Name: "foo", Value: fmt.Sprintf("bar-%d", j)}}, | ||
| } | ||
| } | ||
| } | ||
| res := make([]logqlmodel.Result, 0, 100) | ||
| for _, r := range results { | ||
| res = append(res, logqlmodel.Result{ | ||
| Data: logql.ProbabilisticQuantileMatrix([]logql.ProbabilisticQuantileVector{r}), | ||
| }) | ||
| } | ||
| b.ResetTimer() | ||
| for n := 0; n < b.N; n++ { | ||
| acc := newQuantileSketchAccumulator() | ||
| for _, r := range res { | ||
| require.NoError(b, acc.Accumulate(r, 0)) | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Nice. I was about to push my change and had a conflict. Your code was almost exactly as mine 🎉
…t-a-time' into karsten/merge-one-at-a-time
jeschkies
commented
Jan 3, 2024
| @@ -248,7 +238,8 @@ func newDownstreamAccumulator(params logql.Params, nQueries int) *downstreamAccu | |||
| } | |||
|
|
|||
| func (a *downstreamAccumulator) build(acc logqlmodel.Result) { | |||
There was a problem hiding this comment.
I'm not too fond of this design. We do know whether the results are samples, logs or quantile sketches. So ideally the logql engine would inject the accumulator on the execution plan. Moving this into the logql package would also make the MockQuerier simpler.
There was a problem hiding this comment.
this seems like a good change to me
cstyan
approved these changes
Jan 4, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What this PR does / why we need it:
Currently, the quantile sketch merge logic in the query frontend waits for all shards to complete and merges them then. That puts a lot of pressure on the memory. Luckily the merges can happen out of order as they come in.
I've simulated the later merge.
Checklist
CONTRIBUTING.mdguide (required)CHANGELOG.mdupdatedadd-to-release-noteslabeldocs/sources/setup/upgrade/_index.mdproduction/helm/loki/Chart.yamland updateproduction/helm/loki/CHANGELOG.mdandproduction/helm/loki/README.md. Example PRdeprecated-config.yamlanddeleted-config.yamlfiles respectively in thetools/deprecated-config-checkerdirectory. Example PR