Unclear if we are matching functionality for FirstKeyOnlyFilter

[billyjacobson]

Can we add a check to this test that sees if the row data is returned by HBase's FirstKeyOnlyFilter and if so, we should potentially change the Bigtable HBase client's FirstKeyOnlyFilter to act in the same way.

https://github.com/googleapis/java-bigtable-hbase/blob/master/bigtable-client-core-parent/bigtable-hbase-integration-tests-common/src/test/java/com/google/cloud/bigtable/hbase/AbstractTestFilters.java#L1381

[rahulKQL]

Hi @billyjacobson

I think HBase's FirstKeyOnlyFilter should return the first KV for the rows it matches. Our FirstKeyOnlyFilterAdapter has the same filter condition:

java-bigtable-hbase/bigtable-client-core-parent/bigtable-hbase/src/main/java/com/google/cloud/bigtable/hbase/adapters/filters/FirstKeyOnlyFilterAdapter.java

Lines 30 to 33 in 7b2d0df

	public class FirstKeyOnlyFilterAdapter extends TypedFilterAdapterBase<FirstKeyOnlyFilter> {
	private static Filters.Filter LIMIT_ONE =
	FILTERS.chain().filter(FILTERS.limit().cellsPerRow(1)).filter(FILTERS.value().strip());

For more clarity, I have updated the mentioned test case to perform a scan instead of point read. Please let me know if I am missing something.

[billyjacobson]

My comment was that HBase says it returns the first KV pair but ours returns only the first key because we strip out the value. It says that in their documentation here: https://hbase.apache.org/apidocs/org/apache/hadoop/hbase/filter/FirstKeyOnlyFilter.html

I'm not sure if I am misinterpreting the documentation, so I think we should check if there is a value returned (rather than just the key) and make sure we are doing the same thing.

[rahulKQL]

@billyjacobson Thanks for the explanation!! Indeed HBase returns value along with the first column details.
It seems this change was intentionally done via #1989 for performance improvement. I am not sure what should be our priority here.

@igorbernstein2 / @kolea2 Please let us know how should we proceed here.

[igorbernstein2]

I think we need to revert that PR. We should stay consistent with HBase rather then try to optimize it. Sorry for creating this issue

[billyjacobson]

I think we should have a filter that does what it currently does though. It could go by another name, but I think it would be very useful to have when counting rows. We recommend that people use this filter for that and if it is retrieving the data for even a single column, it could send a ton more data.

[igorbernstein2]

Shouldnt we recommend people to use https://hbase.apache.org/apidocs/org/apache/hadoop/hbase/filter/KeyOnlyFilter.html for that?

[billyjacobson]

Oh that could work!