Repository navigation
Tune falsifier discovery-batch memory envelope to prevent OOM at width 4 - #6396
Merged
Merged
Conversation
gunbai-bot
Bot
changed the title
Contributor
Author
|
Verified both dashboard review findings against claude (36426) / cursor (36431): Confirmed in tree:
Merge readiness @ — sent from nimble-newt-334 |
briansrls
added a commit
that referenced
this pull request
Jul 9, 2026
…tants #6396 landed falsifier-specific per-shard samples (max 6.44GB) and its own width authority; this branch re-measured executor base overhead to 8.91GB. Merged semantics: the memory formula at the live 24GiB cap now derives 2 (not #6396's 3, computed against the old 1.5GiB base) - renamed/updated witness_falsifier_derived_width_at_live_cap_is_three -> _is_two; the shipped falsifier width stays 1 (cpu_bound pins shard_count: 1), so the runtime witnesses hold unchanged. Note text updated to the post-merge numbers. Verified by execution: measurement, plan, budget-tree, and ci_spec witness files all PASS on the merged tree. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
The affected-set falsifier's first post-#6393 run completed plan setup but OOM-killed (exit 137) ~16min into the cold full-corpus discovery batch on srv1 (
spawn_width=4, live runner cap ~25.7GiB, run 28993330753). Floor gate per-shard samples (~4.25GiB) underestimated cold scan-dir discovery load.This adds a falsifier-specific per-shard authority (
gunbc_falsifier_per_shard_peak_rss_bytes, max of local width-5 calibration and CI cap÷width floor from the OOM run) and routesgunbc_falsifier_batches+gunbc_falsifier_plan_spawn_widththroughgunbc_falsifier_memory_derived_width. At the live 24GiB cap that backs spawn width off from 4 → 3 while keeping committed demand inside the 17/20 usable margin (~19.5GiB committed vs ~20.4GiB margin).Test plan
claim_batch --entry dag/test/claim/ci_floor_measurement_per_shard_test.dag --function ci_floor_measurement_per_shard_rss_witnesses— PASSclaim_batch --entry src/v2/test/claim/ci_floor_plan_witness_test.dag --function falsifier_plan_spawn_width_witnesses— PASS