This repository was archived by the owner on Jul 17, 2026. It is now read-only.
feat(tools): gc_dead_runs — GC early-crashed runs' checkpoints#961
Open
ocg-goodfire wants to merge 1 commit into
Open
feat(tools): gc_dead_runs — GC early-crashed runs' checkpoints#961ocg-goodfire wants to merge 1 commit into
ocg-goodfire wants to merge 1 commit into
Conversation
Finds runs under $PARAM_DECOMP_OUT_DIR/{runs,jax_runs} that are safely
deletable: owned by the invoking user, have checkpoints, max ckpt step
below --max-step (crashed early, ckpts useless for analysis or resume),
not referenced by any live SLURM job's stdout under slurm_logs/, and
quiet for --quiet-minutes. Dry-run by default; --delete removes.
First runs reclaimed 19.28 TiB (44 runs, 2026-07-03) and 4.02 TiB
(3 runs, 2026-07-07).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G7BRWT94QzDbqV43BtTFUE
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
param_decomp/tools/gc_dead_runs.py: finds and (with--delete) removes early-crashed runs under$PARAM_DECOMP_OUT_DIR/{runs,jax_runs}— the dead runs whose leftover orbax checkpoints are the dominant storage waste (checkpoint retention on live runs already works; the waste is crashed runs nobody cleaned up).A run is deletable iff ALL of:
st_uidcheck — never touches teammates' runs)ckpts/<step>entries) — that's where the bytes are--max-step(default 5000): crashed early, so the ckpts are useless for analysis or resumptionslurm_logs/(a healthy run passes through low steps early — this guard is what makes "early" watertight)--quiet-minutes(default 60; race guard between submission and first stdout)Dry-run by default, prints a sized candidate list.
Usage
Track record
Known sharp edge (from lore
2026-07-03--task--solve-resumption-oom): it GC'dp-05101e15, a crashed run that was deliberately kept as a resume-OOM diagnostic artifact. Runs referenced only by lore/investigations (not by live jobs) are not protected — possible follow-up.🤖 Generated with Claude Code
https://claude.ai/code/session_01G7BRWT94QzDbqV43BtTFUE