Filter shards for sliced search at coordinator

[msfroh]

Description

Prior to this commit, a sliced search would fan out to every shard, then apply a MatchNoDocsQuery filter on shards that don't correspond to the current slice. This still creates a (useless) search context on each shard for every slice, though. For a long-running sliced scroll, this can quickly exhaust the number of available scroll contexts.

This change avoids fanning out to all the shards by checking at the coordinator if a shard is matched by the current slice. This should reduce the number of open scroll contexts to max(numShards, numSlices) instead of numShards * numSlices.

Related Issues

Related to #16289

Check List

• Functionality includes testing.
• API changes companion pull request created, if applicable.
• Public documentation issue/PR created, if applicable.

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 2d2fd05: 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 541979e: 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 b4aaa2f: 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 1680b9b: 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 8842514: 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 eadaabd: 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 eadaabd: SUCCESS

[codecov]

Codecov Report

Attention: Patch coverage is 73.33333% with 12 lines in your changes missing coverage. Please review.

Project coverage is 72.21%. Comparing base (4a53ff2) to head (41d0f33).
Report is 17 commits behind head on main.

Files with missing lines	Patch %	Lines
...min/cluster/shards/ClusterSearchShardsRequest.java	58.33%	3 Missing and 2 partials ⚠️
...n/admin/cluster/RestClusterSearchShardsAction.java	0.00%	5 Missing ⚠️
...pensearch/action/search/TransportSearchAction.java	75.00%	1 Missing ⚠️
...java/org/opensearch/search/slice/SliceBuilder.java	88.88%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main   #16771      +/-   ##
============================================
- Coverage     72.32%   72.21%   -0.11%     
+ Complexity    65310    65284      -26     
============================================
  Files          5299     5299              
  Lines        303534   303561      +27     
  Branches      43941    43954      +13     
============================================
- Hits         219527   219225     -302     
- Misses        66021    66383     +362     
+ Partials      17986    17953      -33     

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

[github-actions]

❌ Gradle check result for c7abf62: 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 f29e111: 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 6a2de32: 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 f29e111: 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 6a2de32: SUCCESS

[github-actions]

✅ Gradle check result for a5bc4ca: SUCCESS

[reta]

I think you don't need to check for (slice != null) for every iteration, may be something along these lines:

        if (slice != null) {
            for (List<ShardIterator> indexIterators : shardIterators.values()) {
                // Filter the returned shards for the given slice
                CollectionUtil.timSort(indexIterators);
                // We use the ordinal of the iterator in the group (after sorting) rather than the shard id, because
                // computeTargetedShards may return a subset of shards for an index, if a routing parameter was
                // specified. In that case, the set of routable shards is considered the full universe of available
                // shards for each index, when mapping shards to slices. If no routing parameter was specified,
                // then ordinals and shard IDs are the same. This mimics the logic in
                // org.opensearch.search.slice.SliceBuilder.toFilter.
                for (int i = 0; i < indexIterators.size(); i++) {
                    if (slice.shardMatches(i, indexIterators.size())) {
                        allShardIterators.add(indexIterators.get(i));
                    }
                }
            }
        } else {
            shardIterators.values().forEach(allShardIterators::addAll);
        }

[msfroh]

With this change, we avoid creating unnecessary long-lived search contexts on every shard, but you're complaining about unnecessary (extremely cheap) null checks in a tight loop?!

This is what makes you awesome, @reta! I love how you say, "Yeah, this is cool, but couldn't it be better?"

[github-actions]

✅ Gradle check result for 41d0f33: SUCCESS