Document changes to forcemerge as a breaking change for 6.5

[jpountz]

As of Lucene 7.5 (probably Elasticsearch 6.5), the max_num_segments option will only be a best effort, and will not be honored if it would require creating segments that are larger than the maximum segment size. https://issues.apache.org/jira/browse/LUCENE-7976

Stalled on the upgrade to Lucene 7.5.

[elasticmachine]

Pinging @elastic/es-distributed

[polyfractal]

I think this is unstalled since #32390 landed?

I think the behavior is slightly different than originally intended, since #32291 also landed (e.g. we respect max_num_segments), but I think that means we may also end up with fewer segments than max_num_segments? The docs still state that you'll get exactly max_num_segments

Related: #32323

[jasontedor]

Relates #36616

[danielkasen]

This leads to Force merging down to 1 segment everytime. So even if I say force merge to 20 it'll do the large operation and try to shrink it to 1.

[danielkasen]

Any update on if this is the expected behavior going forward? This is causing serious issues within our cluster due to an inability to perform force merge operations.

[jpountz]

We made the decision that forcemerge should keep honoring the max_num_segments parameter in #32291.