Skip to content

ci: add a Blacksmith workflow for the server-log file backfill - #1312

Merged
cquil11 merged 1 commit into
masterfrom
ci/backfill-server-log-files-workflow
Oct 9, 2026
Merged

cquil11 merged 1 commit into
masterfrom
ci/backfill-server-log-files-workflow

Conversation

@cquil11

@cquil11 cquil11 commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Summary

db:backfill-server-log-files has only ever run locally. A local machine needs enough free disk for the largest extracted artifact, and the backfill needed after #1308 covers about 54 GiB of artifacts across 51 runs. This adds a manual-dispatch workflow that runs it on blacksmith-16vcpu-ubuntu-2404, following recompute-agentic-metrics.yml.

Inputs:

  • run-id: one GitHub benchmark run.
  • from-run: every DB run at or above this GitHub run ID (--all --from-run). Leaving both empty uses the script's 90-day default.
  • force: revisit bundles already marked complete; only missing filenames are inserted.
  • neon-branch: an existing Neon child branch. Without it the job writes to production, and only from master.

It uses the same database selection as recompute-agentic-metrics.yml, passes INFX_MAIN_PAT so gh can download InferenceX artifacts, and invalidates the production cache after a production run.


Note

Medium Risk
Manual workflow can write production data and purge the live site cache; guards limit production runs to master and optional Neon branches.

Overview
Adds a manual GitHub Actions workflow so the existing backfill-server-log-files script can run on Blacksmith (blacksmith-16vcpu-ubuntu-2404, 6h timeout) instead of only on a developer machine with enough disk for large benchmark artifacts.

Dispatch inputs map to the CLI: single run-id, range from-run (mutually exclusive), force for incomplete bundles, and optional neon-branch for a Neon child DB. Without a branch, the job writes production and is allowed only from master; with a branch it resolves a write URL via the Neon API (child branch checks mirror the recompute workflow). The backfill step uses INFX_MAIN_PAT for GitHub artifact access and runs packages/db/src/backfill-server-log-files.ts with --yes.

A follow-up invalidate job runs after production backfills and calls admin:cache:invalidate for the live site so cached data reflects new server-log file rows.

Reviewed by Cursor Bugbot for commit ba0bb2c. Bugbot is set up for automated code reviews on this repo. Configure here.

Run db:backfill-server-log-files on a Blacksmith runner by manual dispatch,
for one run or a run-ID range, optionally with --force. Like the agentic
recompute workflow, it targets production only from master or a named Neon
child branch, and invalidates the production cache after a production run.
@cquil11
cquil11 requested a review from adibarra as a code owner October 9, 2026 20:24
@vercel

vercel Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
inferencemax-app Ready Ready Preview Oct 9, 2026 8:25pm UTC

Request Review

@cquil11
cquil11 merged commit c7396aa into master Oct 9, 2026
14 checks passed
@cquil11
cquil11 deleted the ci/backfill-server-log-files-workflow branch October 9, 2026 20:30

This branch was successfully deployed

1 active deployment
Preview — ba0bb2c7 Deployed Oct 9, 2026 by vercel[bot]
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