tree 007a67e30499fcceaf7a2f59a1878107483d861b
parent 0547b5dce1faad6f6f26952fd1aaed42974cbe3d
author Prudhvi Akhil Alahari <prudhvi.alahari@linaro.org> 1665486449 +0530
committer Prudhvi Akhil Alahari <prudhvi.alahari@linaro.org> 1667233878 +0530

Add HasCardinality interface which helps in defining a cardinality

Before this change, cardinality of the IndexSource was always set to
10. This makes AndSource to choose IndexSource over other source almost
always.

Add a HasCardinality interface which can be used by all the predicates
which implement IndexPredicate to set a cardinality. This helps
AndSource to choose the right datasource.

Unfortunately there are no operators which are datasources other
than Index in the Core as of now. We at Qualcomm have a datasource
operator in a plugin which takes a bug number and queries the bug
tracker's DB for changes related to that bug number. Suppose this
operator (say issue:123) returns around 15 changes. On a 3.4 test site
against LUCENE which contains ~20K open changes, ~2.8M merged changes,
the performance of,

query "issue:123 status:open"
  when status:open is datasource: 2.1s, 2.1s, 2.1s
  when issue:123 is datasource: 0.07s, 0.07s, 0.07s

query "issue:123 status:merged"
  when status:merged is datasource: 6m49s, 6m38s, 6m34s
  when issue:123 is datasource: 0.07s, 0.07s, 0.07s

The cardinality of the predicates are restored from their previous
values [1] and [2].

[1] https://gerrit-review.googlesource.com/c/gerrit/+/52078/8/gerrit-server/src/main/java/com/google/gerrit/server/query/change/SqlRewriterImpl.java#b576
[2] https://gerrit-review.googlesource.com/c/gerrit/+/63214

Release-Notes: AndSource chooses right datasource more often
Change-Id: I802ab592d121b01dab9557da535f99411a1c7bc0