What

Raise the per-transaction timeout in avm_check_circuit_cmds from 30s to 120s (yarn-project/end-to-end/bootstrap.sh).

Root cause

The avm-check-circuit job of the AVM Circuit Inputs Collection and Check workflow failed on next (run 29287694175, exit code 124).

ci-avm-check-circuit downloads the cached AVM inputs and runs bb-avm avm_check_circuit on every dumped tx via avm_check_circuit_cmds | parallelize 16. Each command runs under TIMEOUT=30s. In this run 643 of 644 checks passed in 1–4s each, but one heavy transaction tripped the timeout:

FAILED (…): bb-avm avm_check_circuit -v --avm-inputs .../e2e_multiple_blobs/avm-circuit-inputs-tx-0x127eb600….bin (31s) (code: 124)
parallel: This job failed:
run_test_cmd 'eda031bed1f51704:ISOLATE=1:TIMEOUT=30s:NAME=avm_cc_e2e_multiple_blobs_0x127eb600 …'
+ local code=124

The e2e_multiple_blobs tx has a large AVM trace and ran 31s — just over the 30s cap. timeout -v killed it (exit 124), and because parallelize runs parallel --halt now,fail=1, that single timeout failed the entire job.

This is not caused by the head commit (a docs-only release PR, #24664); the check runs on the tree-hashed cached inputs and is unrelated to docs. It's a pre-existing borderline-heavy check running under a too-tight timeout — exactly the risk the code comment already flagged ("transactions could need more CPU and MEM than we allocate by default … they might start timing out"). A previous attempt to relieve this reduced parallelism (#24458, "It was timing out"); with the heaviest tx now sitting right at the boundary it regressed again.

Fix

Give the check generous headroom over the slowest observed tx (31s) so a single slow run under 16-way CPU contention no longer fails the whole job. avm_check_circuit is a correctness gate — wall-time doesn't matter as long as it's bounded, and a 120s cap still catches genuine hangs (normal txs finish in seconds). Updated the stale comment to match.

Note

This is a recurring failure that ClaudeBox has been auto-dispatched on many times (dozens of stale cb/avm-check-circuit-timeout* / cb/avm-cc-timeout* branches exist, none merged — next still carries 30s). Worth actually landing a headroom bump so the auto-dispatch stops firing on it.

Testing

No local red/green repro is feasible: the check runs bb-avm avm_check_circuit on an EC2 instance against the S3-cached e2e-avm-circuit-inputs-<tree-hash>.tar.gz with the freshly built AVM binary — none of which is reproducible in this container. The change is a one-line timeout bump validated against the failing job's own log (the timed-out tx took 31s, well within the new 120s).

Created by claudebox · group: slackbot