[SPARK-49846][SS] Add numUpdatedStateRows and numRemovedStateRows metrics for use with transformWithState operator

[anishshri-db]

What changes were proposed in this pull request?

Add numUpdatedStateRows and numRemovedStateRows metrics for use with transformWithState operator

Why are the changes needed?

Without this change, metrics around these operations are not available in the query progress metrics

Does this PR introduce any user-facing change?

No

Metrics updated as part of the streaming query progress

    "operatorName" : "transformWithStateExec",
    "numRowsTotal" : 1,
    "numRowsUpdated" : 1,
    "numRowsRemoved" : 1,

How was this patch tested?

Added unit tests

[info] Run completed in 25 seconds, 697 milliseconds.
[info] Total number of tests run: 2
[info] Suites: completed 1, aborted 0
[info] Tests: succeeded 2, failed 0, canceled 0, ignored 0, pending 0
[info] All tests passed.

Was this patch authored or co-authored using generative AI tooling?

No

[anishshri-db]

cc - @neilramaswamy @HeartSaVioR - PTAL, thx !

[neilramaswamy]

Probably also want a suite for TTL (making sure we count the removal due to TTL in metrics), but the other comments are probably more important.

[neilramaswamy]

Hm, every time we append a value we add one to the numUpdatedStateRows metric, even though we're not actually adding a row to RocksDB, we're merging something into an existing row.

But over here, we're counting removing an entire row (which may have multiple values all merged together) as one removal. So then if you do k calls to appendValue and then a clear(), numUpdatedRows != numRemovedStateRows.

I don't have a good solution for you (without iterating through everything in clear()), but this does seem trickier than it might have initially seemed.

[anishshri-db]

@neilramaswamy - yes thats correct. we have that inherent limitation here since we either need to iterate on all the values on clear just to get the count or if we want to count only once - we need to do a get explicitly each time we call appendValue. Also, we don't really expect users to tally these numbers across batches since these are cumulative across all state operations on all variables. If you still think its too confusing for the users, I think its prob better to just mark this as TODO for listState and we will circle back to this issue

[neilramaswamy]

Yeah, seems like my earlier comment is getting at exactly this. It's 1 here, even though it should probably be 6? To the user, the fact that we use merge for lists is an implementation detail.