Skip to content

enhance: pipeline sealed BM25 stats materialization - #30

Merged
sunby merged 1 commit into
sunby:codex/load-1m-segments-pr-stack-rebased-qv-workfrom
sijie-ni-0214:idf-materialization-pipeline
Sep 10, 2026
Merged

enhance: pipeline sealed BM25 stats materialization#30
sunby merged 1 commit into
sunby:codex/load-1m-segments-pr-stack-rebased-qv-workfrom
sijie-ni-0214:idf-materialization-pipeline

Conversation

@sijie-ni-0214

Copy link
Copy Markdown

What this PR does

  • Merge each sealed BM25 contribution as soon as its load completes, overlapping aggregation with the remaining object storage reads during lazy IDF materialization.
  • Reuse immutable cache-owned BM25 statistics instead of cloning the full segment statistics on every acquire.
  • Adopt the first deserialized statistics value per field and merge only subsequent files.
  • Cover immediate pipelined consumption, limiter cancellation, lease cleanup, and shared cache statistics.

Verification

  • go test -race -tags dynamic,test -gcflags="all=-N -l" -count=1 ./internal/streamingnode/server/wal/vchannel/idf
  • go test -tags dynamic,test -gcflags="all=-N -l" -run "^TestAcquireSealedContributionsCancellationWhileWaitingForLimit$" -count=1000 ./internal/streamingnode/server/wal/vchannel/idf
  • make build-go
  • git diff --check

Merge sealed segment statistics as individual loads complete so lazy IDF materialization overlaps object storage reads with aggregation. Reuse immutable cache-owned statistics and adopt the first deserialized field value to avoid redundant full copies.

Add coverage for pipelined consumption, limiter cancellation, lease cleanup, and shared cache statistics.

Signed-off-by: sijie-ni-0214 <sijie.ni@zilliz.com>
@sunby
sunby merged commit 00da51e into sunby:codex/load-1m-segments-pr-stack-rebased-qv-work Sep 10, 2026
6 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants