Support Dynamic Pruning in Cardinality Aggregation

[bowenlan-amzn]

Description

First I want to thanks Rishabh's work on this which I learnt from.

Background

Cardinality aggregation count the number of distinct values of a field.
Some common use cases are counting the visitors of a service, detect anomly of unique IP address or event types and count unique product or brands in e-commerce, etc.

Idea

By defaults, aggregation works on every document that matches the top level query, collects them in an iterative manner.
However, once we collect one document, we don't need to care about the values of this document anymore, since they are already counted as the distinct values in cardinality aggregation results.
So the idea is to stop working on other documents that also have these collected values as the aggregation going on.

Implementation

1. Segment level

To implement this idea, we need to know all the possible values of the field and the documents of each value before the aggregation started.
Lucene provide a dictionary of terms and a list the documents containing these terms.
We can build a disjunction of the documentID iterators for these term values. Every time after we collect values of a doc, we remove the iterators of the collected. So we actually skipped many documents for the next doc iteration and won't see these collected values anymore.

2. For numeric field

Lucene provide a dictionary of terms and the documents that contain these terms.
However, for numeric field, lucene provides a tree structure that let user traverse the numeric values in sorted order. So it would be different to do the same kind of dynamic pruning for numeric field. It's still possible but we can work on it as a follow up considering numeric cardinality is less common comparing to keyword term cardinality.

3. Across segments or Shard level

One shard may contain multiple segments, and by default segments are processed in sequential. If the seen values are saved at shard level, for the subsequential segments, there's no need to build the iterator for these values.
This will be a follow up tasks.

Public Facing Changes

New cluster setting that can dynamically tune the threshold of supported maximum cardinality for doing the pruning
search.dynamic_pruning.cardinality_aggregation_threshold
Set to 0 can disable the feature.
Currently use 100 which should be a safe number.

opensearch-project/documentation-website#7334
opensearch-project/documentation-website#7341

Benchmark

Add cardinality aggregation operations in big5 workload
opensearch-project/opensearch-benchmark-workloads#311

Performance improvements before and after

|                                       50th percentile latency |  cardinality-agg-low |      903463 |     10.3347 |      -903453 |     ms |
|                                       90th percentile latency |  cardinality-agg-low | 1.04617e+06 |     10.8622 | -1.04615e+06 |     ms |
|                                       99th percentile latency |  cardinality-agg-low | 1.07827e+06 |      13.704 | -1.07826e+06 |     ms |
|                                      100th percentile latency |  cardinality-agg-low | 1.08184e+06 |      14.249 | -1.08183e+06 |     ms |
|                                  50th percentile service time |  cardinality-agg-low |     4103.42 |     9.22686 |      -4094.2 |     ms |
|                                  90th percentile service time |  cardinality-agg-low |     4109.98 |     9.54821 |     -4100.43 |     ms |
|                                  99th percentile service time |  cardinality-agg-low |     4112.87 |     13.4495 |     -4099.42 |     ms |
|                                 100th percentile service time |  cardinality-agg-low |     4115.37 |     13.6122 |     -4101.76 |     ms |
|                                                    error rate |  cardinality-agg-low |           0 |           0 |            0 |      % |

|                                       50th percentile latency | cardinality-agg-high |      918502 |      690842 |      -227661 |     ms |
|                                       90th percentile latency | cardinality-agg-high | 1.06362e+06 |      800014 |      -263604 |     ms |
|                                       99th percentile latency | cardinality-agg-high | 1.09628e+06 |      824581 |      -271698 |     ms |
|                                      100th percentile latency | cardinality-agg-high |  1.0999e+06 |      827314 |      -272586 |     ms |
|                                  50th percentile service time | cardinality-agg-high |     4162.98 |     3256.42 |     -906.562 |     ms |
|                                  90th percentile service time | cardinality-agg-high |     4169.53 |     3260.06 |     -909.463 |     ms |
|                                  99th percentile service time | cardinality-agg-high |     4177.45 |     3266.89 |     -910.562 |     ms |
|                                 100th percentile service time | cardinality-agg-high |     4178.73 |     3280.35 |     -898.381 |     ms |
|                                                    error rate | cardinality-agg-high |           0 |           0 |            0 |      % |

Related Issues

Resolves #11959

Check List

• New functionality includes testing.
  • All tests pass
• New functionality has been documented.
  • New functionality has javadoc added
• API changes companion pull request created.
• Failing checks are inspected and point to the corresponding known issue(s) (See: Troubleshooting Failing Builds)
• Commits are signed per the DCO using --signoff
• Commit changes are listed out in CHANGELOG.md file (See: Changelog)
• Public documentation issue/PR created

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[github-actions]

❌ Gradle check result for c2775ac: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for 91f7a04: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for 8fc9bee: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for 85133c4: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❕ Gradle check result for 9d4701c: UNSTABLE

Please review all flaky tests that succeeded after retry and create an issue if one does not already exist to track the flaky failure.

[codecov]

Codecov Report

Attention: Patch coverage is 78.18182% with 24 lines in your changes missing coverage. Please review.

Project coverage is 71.66%. Comparing base (b15cb0c) to head (5f7dad3).
Report is 405 commits behind head on main.

Files	Patch %	Lines
...ch/aggregations/metrics/CardinalityAggregator.java	78.64%	14 Missing and 8 partials ⚠️
...va/org/opensearch/search/DefaultSearchContext.java	80.00%	1 Missing ⚠️
.../org/opensearch/search/internal/SearchContext.java	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main   #13821      +/-   ##
============================================
+ Coverage     71.42%   71.66%   +0.24%     
- Complexity    59978    61795    +1817     
============================================
  Files          4985     5105     +120     
  Lines        282275   290175    +7900     
  Branches      40946    41917     +971     
============================================
+ Hits         201603   207968    +6365     
- Misses        63999    65043    +1044     
- Partials      16673    17164     +491     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

❌ Gradle check result for d53182b: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for 77fceea: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[bowenlan-amzn]

❌ Gradle check result for 77fceea: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

Pretty interesting, seems this failure can be reproduced consistently with the seed

./gradlew ':server:internalClusterTest' --tests "org.opensearch.indices.IndicesRequestCacheIT" -Dtests.method="testDeleteAndCreateSameIndexShardOnSameNode {p0={"search.concurrent_segment_search.enabled":"false"}}" -Dtests.seed=8199B3619B8101C

[github-actions]

❌ Gradle check result for 09f957c: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for 09f957c: null

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[opensearch-trigger-bot]

The backport to 2.x failed:

The process '/usr/bin/git' failed with exit code 128

To backport manually, run these commands in your terminal:

# Navigate to the root of your repository
cd $(git rev-parse --show-toplevel)
# Fetch latest updates from GitHub
git fetch
# Create a new working tree
git worktree add ../.worktrees/OpenSearch/backport-2.x 2.x
# Navigate to the new working tree
pushd ../.worktrees/OpenSearch/backport-2.x
# Create a new branch
git switch --create backport/backport-13821-to-2.x
# Cherry-pick the merged commit of this pull request and resolve the conflicts
git cherry-pick -x --mainline 1 2f8cb079c2c220304b8e6965853a15732c3bf57c
# Push it to GitHub
git push --set-upstream origin backport/backport-13821-to-2.x
# Go back to the original working tree
popd
# Delete the working tree
git worktree remove ../.worktrees/OpenSearch/backport-2.x

Then, create a pull request where the base branch is 2.x and the compare/head branch is backport/backport-13821-to-2.x.