Fix tsdb codec when doc-values spread in two blocks

[dnhatn]

Currently, loading ordinals multiple times (after advanceExact) for documents with values spread across multiple blocks in the TSDB codec will fail due to the absence of re-seeking for the ordinals block. Doc-values of a document can spread across multiple blocks in two cases: when it has more than 128 values or when it exceeds the remaining space in the current block.

[elasticsearchmachine]

Hi @dnhatn, I've created a changelog YAML for you.

[elasticsearchmachine]

Pinging @elastic/es-storage-engine (Team:StorageEngine)

[benwtrent]

I see this exact pattern in NumericDocValues#longValue. Do we have off-by-one errors there as well?

elasticsearch/server/src/main/java/org/elasticsearch/index/codec/tsdb/ES87TSDBDocValuesProducer.java

Line 1015 in ae69a61

if (blockIndex - 1 > currentBlockIndex) {

elasticsearch/server/src/main/java/org/elasticsearch/index/codec/tsdb/ES87TSDBDocValuesProducer.java

Line 1075 in ae69a61

if (blockIndex - 1 > currentBlockIndex) {

?

[dnhatn]

Thanks Bent. These are single value paths, so they shouldn't be affected by this bug. Anyway, I've extended the tests and made the seeking patterns more consistent in this class.

[dnhatn]

@benwtrent Thanks for review.

[elasticsearchmachine]

💚 Backport successful

Status	Branch	Result
✅	8.14
✅	8.13

[martijnvg]

LGTM2

[jpountz]

Thanks for fixing! I tried to understand why BaseDocValuesFormatTestCase#testSortedSetVariableLengthVsStoredFields did not catch this bug, and this seems to be because it only uses nextDoc(), not advance()/advanceExact()? Could you improve BaseDocValuesFormatTestCase#doTestSortedSetVsStoredFields in Lucene to add more coverage for this?

[dnhatn]

@jpountz +1, I will open a PR in the Lucene repository.

[dnhatn]

I've opened apache/lucene#13346