Skip to content

bench: prod baselines for 0.15.0, and JSON vs Parquet uploads - #75

Merged
amrit110 merged 1 commit into
mainfrom
bench/0.15.0-baselines
Oct 21, 2026
Merged

amrit110 merged 1 commit into
mainfrom
bench/0.15.0-baselines

Conversation

@amrit110

Copy link
Copy Markdown
Member
  • benchmarks/baselines/prod-bench-0.15.0.json: prod_bench on 0.15.0. tabctx's own work is unchanged from 0.14.1. The worst calls, 45–170 s, are all in the hop to tabctx; tabctx's work was 14 s or less. Those stalls came more often than in the overnight 0.14.1 run.

  • benchmarks/bench_upload_formats.py and baselines/upload-formats-0.15.0.json: the same tables as JSON and as Parquet, interleaved, 12 fits each.

    Format Mean size Time in transit, median Worst Fits over 10 s
    JSON 14.0 MiB 2.7 s 113 s 3/12
    Parquet 8.5 MiB 1.2 s 42 s 1/12

    Smaller bodies spend less time in transit, but Parquet doesn't remove the stalls. The evidence is on aieng-proxy-gateway#472.

🤖 Generated with Claude Code

…eway

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@amrit110
amrit110 force-pushed the bench/0.15.0-baselines branch from 85957e5 to bdb9d73 Compare October 21, 2026 14:36
@amrit110
amrit110 merged commit 329dabc into main Oct 21, 2026
3 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.

1 participant