fix(ci): raise AVM check-circuit per-tx timeout from 30s to 120s

[AztecBot]

Problem

The nightly AVM Circuit Inputs Collection and Check workflow failed on next (run 27394504089). The GitHub job avm-check-circuit exited with code 124 (the SSM command surfaced the remote timeout exit code), and parallelize --halt now,fail=1 aborted the whole job on the first failing command.

Root cause

A single check-circuit command timed out — not a correctness failure:

FAILED: bb-avm avm_check_circuit -v --avm-inputs .../e2e_multiple_blobs/avm-circuit-inputs-tx-0x2636082f...bin (34s) (code: 124)

That tx produces a large AVM circuit. From its log:

Generating trace...                                   (mem: 825 MiB)   @ 04:55:22
Checking circuit... (mem: 3912 MiB)                                    @ 04:55:41   (~19s trace gen)
Running check (with skippable) circuit over 700560 rows.
timeout: sending signal TERM to command 'bash'                        @ 04:55:51

With the default ISOLATE resources (2 CPUs), trace generation (~19s) plus the check pass over ~700k rows exceeds the per-command TIMEOUT=30s, so timeout kills it. That 30s value has been in place unchanged since the feature was added (#18747) and was never tuned; the inline comment already anticipated exactly this ("transactions could need more CPU and MEM than we allocate by default ... they might start timing out").

Fix

Raise the per-command timeout from 30s to 120s in avm_check_circuit_cmds. The heaviest observed tx needs ~40s end-to-end at 2 CPUs, so 120s gives comfortable margin as circuits grow while staying under the 5-minute "test too slow" warning threshold.

CPUS is intentionally left at the default 2: parallelize runs num_cpus/2 jobs (96 on the 192-core CI box), so 2 CPUs per command already fully utilizes the machine. Bumping CPUS without changing the job count would oversubscribe the cores and get throttled back by CFS, providing no real speedup — a timeout bump is the correct, scheduler-consistent fix.

Testing

Scripted change to a nightly CI path; behavior can't be exercised in unit CI. The fix is a timeout increase only — the circuit itself passed its checks up to the kill point, it simply needed more wall time than the old budget allowed.

---

Created by claudebox · group: slackbot