This pull request's diff hasn't been prepared yet. Log in with GitHub to have us prepare it for review.

Cap vectordb_columnar string chunks at 32KB #162735

Closed
mayya-sharipova opened 4:34pm on October 20, 2026 wanted to merge 25 Ξ” into elastic/elasticsearch main from
vectordb-columnar-chunks-32
Reviewing
6 fewer changed lines
(43% less) vs GitHub
This saves about 75.3 hours per year vs conventional diff tools
mayya-sharipova authored
of work during October 20
Diff Delta:
25
About 35 Diff Delta/hour
Classified as:  General

mayya-sharipova's Description of Work

Follow-up to #160792. Plain ColumNAR string chunks in
vectordb_columnar were bounded at 128KB, so a top-k fetch of one
long text value could decompress up to 128KB. Lower the plain byte
target to 32KB, matching escape chunks, so no single read of a plain
or escaped value decompresses more than 32KB.

Size
columnar, logsdb_columnar
vectordb_columnar

Plain chunks
512KB / 16384 val
32KB / 128 val

Escape chunks
32KB / 16384 val
32KB / 128 val

Compressed ordinal block
2048
512

Building updated commit group diff...
2 total changed files
Loading changes...