Repository navigation
fix(db): ingest server logs from srt-slurm single-node archives - #1308
Merged
Merged
Conversation
Native srt-slurm single-node runs upload the job's logs directory, including the server worker log <host>_agg_w0.out, inside srt-single-node-logs.tar.gz in the server_logs_* artifact. Ingest only unpacked the multinode archive, so those runs stored just benchmark.log and srt-setup.log. Extract the single-node archive in place so its .log/.out files are stored alongside the root files. Artifacts without it, and multinode artifacts, read as before; the primary log choice is unchanged. Wire the existing --force flag in the server-log backfill so bundles already marked complete can pick up the newly visible files.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Since native srt-slurm single-node runs landed (fixed-sequence in SemiAnalysisAI/InferenceX#3352 on 2026-09-24, AgentX in SemiAnalysisAI/InferenceX#3428 on 2026-09-26), the
server_logs_*artifact no longer containsresults/server.log. The job's wholelogs/directory, including the server worker log<host>_agg_w0.out, is uploaded insidesrt-single-node-logs.tar.gzinstead.serverLogArtifactRootonly unpackedmultinode_server_logs.tar.gz, so for these runs ingest stored justresults/benchmark.logandsrt-setup.log. Every single-node point since then has no server log in the DB.Change
server_logs_*artifacts that containsrt-single-node-logs.tar.gzare extracted in place tosrt-single-node-logs/. Their.log/.outfiles are stored next to the root files with artifact-relative paths, for examplesrt-single-node-logs/logs/<host>_agg_w0.out.results/benchmark.logfor new artifacts, the same as today. KV-cache derivation is untouched.readServerLogArtifact, so both pick this up.db:backfill-server-log-filesnow honors the already-parsed--forceflag. Affected bundles are markedfiles_complete, so they were skipped; with--forcethey are revisited, and only filenames the bundle does not already hold are inserted.docs/data-pipeline.mddocuments the archive and--force.Backfill after merge
The first affected run in the DB is 36270056341 (AgentX, 2026-09-26); fixed-sequence rows start at 36507377618 (2026-09-29).
Note
Medium Risk
Touches ETL server-log expansion for all new single-node artifacts and backfill recovery; changes are backward compatible but require a forced backfill to repair already-ingested incomplete bundles.
Overview
Native srt-slurm single-node
server_logs_*artifacts now unpacksrt-single-node-logs.tar.gzin place so worker.log/.outfiles (e.g.<host>_agg_w0.out) are ingested undersrt-single-node-logs/next to root client logs. Legacy layouts without the archive and multinodemultinode_server_logs.tar.gzbehavior stay the same, with extraction refactored through a sharedextractArchiveOncehelper.db:backfill-server-log-filesfinally applies the existing--forceflag: bundles already markedfiles_completeare revisited and only missing filenames are inserted, so historical runs can pick up archive contents without re-downloading everything blindly.docs/data-pipeline.mddocuments the archive layout and--force.Reviewed by Cursor Bugbot for commit 275de85. Bugbot is set up for automated code reviews on this repo. Configure here.