Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fork tdigest library #95903

Closed
martijnvg opened this issue May 8, 2023 · 5 comments
Closed

Fork tdigest library #95903

martijnvg opened this issue May 8, 2023 · 5 comments
Assignees
@kkrik-es
Labels
:Analytics/Aggregations Aggregations :Delivery/Build Build or test infrastructure >refactoring Team:Analytics Meta label for analytical engine team (ESQL/Aggs/Geo) Team:Delivery Meta label for Delivery team

Comments

@martijnvg
Copy link
Member

martijnvg commented May 8, 2023 •
edited
Loading

We plan to for the tdigest library.

There are two main reasons behind this choice:

  1. We would like to control semantic version and backward compatibility according to our definition. Right now, for instance, TDigest does not match our usage of semantic versioning when changing the library code and that makes upgrading quite challenging because exposes us to backward compatibility issues.
  2. We would like to change those libraries to use some specific Elasticsearch libraries/tools/frameworks such as BigArrays. Right now when running some aggregations (percentiles, boxplot,...) we experience OOMs due to large memory usage. Using BigArrays, for instance, would allow us to deal with OOMs using Circuit Breakers.

The immediate goal of this issue is to fork the library and then at a later stage enhance to forked library to make use of the big arrays infrastructure.

Currently t-digest version 3.2 is used. The current version is 3.3 We have been locked to the 3.2 version because of at least one breaking change (how p50 is computed). The plan is to fork from the latest commit and change the forked library such that the results it produces are similar to that of version 3.2.
@martijnvg martijnvg added :Delivery/Build Build or test infrastructure :Analytics/Aggregations Aggregations >refactoring labels May 8, 2023
@elasticsearchmachine elasticsearchmachine added Team:Analytics Meta label for analytical engine team (ESQL/Aggs/Geo) Team:Delivery Meta label for Delivery team labels May 8, 2023
@elasticsearchmachine
Copy link
Collaborator

Pinging @elastic/es-analytics-geo (Team:Analytics)

@elasticsearchmachine
Copy link
Collaborator

Pinging @elastic/es-delivery (Team:Delivery)

@martijnvg martijnvg mentioned this issue May 8, 2023
Closed
@kkrik-es kkrik-es self-assigned this May 8, 2023
@martijnvg martijnvg mentioned this issue May 9, 2023
Closed
7 tasks
@mark-vieira
Copy link
Contributor

@martijnvg @wchaparro what's the priority on this. Is the upgrade to 3.3 blocking something that is targeting a particular Elasticsearch release?

@martijnvg
Copy link
Member Author

@mark-vieira We're not blocked by other items. The forking has a high priority for us and we like to get this in with the 8.9 release. We plan to fork it as a library under the top-level libs directory, similar to how h3 was forked.

The integration with big arrays is currently not a blocker. We do like to adopt big arrays in the forked library, but that will be done at a later point of time and has a lower priority than the forking itself.

@wchaparro wchaparro mentioned this issue Jun 7, 2023
Closed
This was referenced Jun 13, 2023
Merged
Closed
Merged
Merged
Closed
Closed
Merged
Merged
elasticsearchmachine pushed a commit that referenced this issue Jun 19, 2023
@kkrik-es
Remove invalid asserts in MergingDigest. (#96928)
2f93cf5 
The asserts were misfiring in this test:

```
REPRODUCE WITH: ./gradlew ':server:test' --tests "org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviationTests.testReduceRandom" -Dtests.seed=AA1D81AD056870F0 -Dtests.locale=en-CA -Dtests.timezone=US/Eastern -Druntime.java=20

org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviationTests > testReduceRandom FAILED
    java.lang.AssertionError
        at __randomizedtesting.SeedInfo.seed([AA1D81AD056870F0:6A202DBC2335E9A6]:0)
        at org.elasticsearch.tdigest.MergingDigest.merge(MergingDigest.java:316)
        at org.elasticsearch.tdigest.MergingDigest.mergeNewValues(MergingDigest.java:298)
        at org.elasticsearch.tdigest.MergingDigest.mergeNewValues(MergingDigest.java:288)
        at org.elasticsearch.tdigest.MergingDigest.quantile(MergingDigest.java:485)
        at org.elasticsearch.tdigest.HybridDigest.quantile(HybridDigest.java:141)
        at org.elasticsearch.search.aggregations.metrics.TDigestState.quantile(TDigestState.java:247)
        at org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviation.computeMedianAbsoluteDeviation(InternalMedianAbsoluteDeviation.java:38)
        at org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviation.<init>(InternalMedianAbsoluteDeviation.java:48)
        at org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviationTests.createTestInstance(InternalMedianAbsoluteDeviationTests.java:33)
        at org.elasticsearch.search.aggregations.metrics.InternalMedianAbsoluteDeviationTests.createTestInstance(InternalMedianAbsoluteDeviationTests.java:23)
```

They should have been removed earlier, it's possible for the first and
the last centroid to have weight > 1.

Related to #95903
@kkrik-es kkrik-es mentioned this issue Jun 20, 2023
Merged
kkrik-es added a commit to kkrik-es/elasticsearch that referenced this issue Jun 20, 2023
@kkrik-es
Restore V_8_500_014 executionHint serialization.
c662442 
This was replaced by optional serialization in version
V_8_500_018 (elastic#96943), breaking backwards compatibility for serverless.

Related to elastic#95903
@kkrik-es kkrik-es mentioned this issue Jun 20, 2023
Merged
kkrik-es added a commit that referenced this issue Jun 20, 2023
@kkrik-es
Restore V_8_500_014 executionHint serialization. (#96965)
d6ff43b 
This was replaced by optional serialization in version
V_8_500_018 (#96943), breaking backwards compatibility for serverless.

Related to #95903
elasticsearchmachine pushed a commit that referenced this issue Jun 21, 2023
@kkrik-es
Add tests for execution_hint in aggregations with percentiles. (#96927)
bd29706 
Setting `experiment_hint` to `high_accuracy` leads to using a slower but
more accurate implementation for TDigest.

Add YAML tests to cover for settings this param properly and for parsing
errors.

Related to #95903
@kkrik-es
Copy link
Contributor

This is now complete.

@sorenlouv sorenlouv mentioned this issue Jul 2, 2023
Closed
@mark-vieira mark-vieira mentioned this issue Oct 4, 2023
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@kkrik-es kkrik-es

Labels
:Analytics/Aggregations Aggregations :Delivery/Build Build or test infrastructure >refactoring Team:Analytics Meta label for analytical engine team (ESQL/Aggs/Geo) Team:Delivery Meta label for Delivery team
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@martijnvg @mark-vieira @elasticsearchmachine @kkrik-es