fix(ci3): don't crash CI when redis is unavailable in redis_setexz

[AztecBot]

What

redis_setexz unconditionally pipes gzip into redis_cli:

function redis_setexz {
  gzip | redis_cli -x SETEX $1 $2 &>/dev/null
}

When redis is unavailable (CI_REDIS_AVAILABLE=0), redis_cli is a no-op that returns immediately without reading stdin. The gzip on the left then gets SIGPIPE, prints gzip: stdout: Broken pipe, and exits non-zero. Because ci3/source → source_options sets set -euo pipefail, that non-zero pipeline status aborts the entire run.

This fires at the very first redis_setexz call in bootstrap_ec2:

echo "CI booting..." | redis_setexz "$CI_LOG_ID" 300

so CI dies at bootstrap, before any build work. All four callers (bootstrap_ec2, cache_log, run_test_cmd, denoise) pipe data into redis_setexz, so every one of them is exposed to the same crash in a redis-less environment.

The fix guards on CI_REDIS_AVAILABLE (mirroring redis_cli) and drains stdin when there is nowhere to store it, so producers never see a broken pipe. When redis is available, behaviour is unchanged.

How this surfaced

The aztec-claude nightly barretenberg debug build (run 29141841707) fails instantly with exactly this error:

--- Run barretenberg-debug CI ---

gzip: stdout: Broken pipe
##[error]Process completed with exit code 1.

That job runs on a GitHub-hosted ubuntu-latest runner where redis is not reachable (no bastion SSH key), so CI_REDIS_AVAILABLE=0 and the first redis_setexz call aborts the run. The identical nightly in aztec-packages is green because its runner has the CI secrets that open the redis tunnel, so this path is never exercised there — this change hardens ci3 for any redis-less environment.

Note: this hardens the ci3 framework and removes the proximate crash. The aztec-claude mirror's nightly has a separate, non-code blocker — that repo lacks the AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / BUILD_INSTANCE_SSH_KEY secrets the EC2 build needs — which is an operator/infra action (provision the secrets in aztec-claude, or drop the mirrored workflow) and cannot be fixed by this PR.

Test

Reproduced the SIGPIPE abort with CI_REDIS_AVAILABLE=0 under set -euo pipefail (run aborts before the next statement), then confirmed the patched function drains stdin and the pipeline exits 0. bash -n ci3/source_redis passes.

---

Created by claudebox · group: slackbot

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.