Skip to content

Optimize aligned row and tablet memory estimation - #18426

Merged
jt2594838 merged 1 commit into
apache:dev/1.3from
jt2594838:allocate_primitive_array_lazily_1.3
Aug 10, 2026
Merged

Optimize aligned row and tablet memory estimation#18426
jt2594838 merged 1 commit into
apache:dev/1.3from
jt2594838:allocate_primitive_array_lazily_1.3

Conversation

@jt2594838

Copy link
Copy Markdown
Contributor

Summary

  • Add aligned single-row and tablet incremental memory estimators for lazily materialized primitive arrays.
  • Reuse per-device estimators across multi-row inserts to reduce temporary Map/Set allocations.
  • Add focused unit tests and a branch-comparison benchmark for row and tablet paths.

Performance

  • Multi-row InsertRows time improved by about 13.6%-25.1% versus the local dev/1.3 baseline.
  • Large-batch allocations reduced by about 37.7%-91.2%.
  • Existing-column single-row and tablet paths show the largest gains; new-column single-row paths remain a known tradeoff.

Tests

  • AlignedTVListTest: 18/18
  • TsFileProcessorTest: 18/18
  • AlignedRowsBranchComparisonBenchmarkTest: 1/1
  • git diff --check

@jt2594838
jt2594838 merged commit 251cf34 into apache:dev/1.3 Aug 10, 2026
16 checks passed
@jt2594838
jt2594838 deleted the allocate_primitive_array_lazily_1.3 branch August 10, 2026 08:00
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.

1 participant