fix(ci3): don't abort CI when redis is unavailable (redis_setexz broken pipe)

[AztecBot]

Root cause

The nightly barretenberg debug build failed instantly (~50ms, before any build work) with:

--- Run barretenberg-debug CI ---

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

(failing run: AztecProtocol/aztec-claude run 28313796517)

ci.sh barretenberg-debug calls bootstrap_ec2, whose very first action logs a heartbeat to redis:

# ci3/bootstrap_ec2:24
echo "CI booting..." | redis_setexz "$CI_LOG_ID" 300

redis_setexz was:

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

When redis is unavailable, redis_cli is a documented no-op ("Noop when redis is unavailable") that never reads stdin. So gzip compresses, tries to write to the closed pipe, takes a SIGPIPE/stdout: Broken pipe, and exits non-zero. Under set -o pipefail + set -e (from ci3/source_options) that non-zero pipeline status aborts the entire CI run at the first log write — long before reaching EC2. The &>/dev/null only covers redis_cli, so gzip's stderr still leaks the confusing message.

This bit the aztec-claude run because its next mirror is stale and predates the job guard now present in the aztec-packages workflow (if: github.event_name != 'schedule' || github.repository == 'AztecProtocol/aztec-packages'), so the schedule fired in an environment with no AWS creds and no reachable redis. But it is a genuine latent landmine for the real aztec-packages nightly too: if the redis tunnel ever drops, the whole run would abort here with a cryptic broken-pipe error instead of degrading gracefully.

Fix

Make redis_setexz honor the same "noop when redis is unavailable" contract as redis_cli, while still draining stdin so upstream writers in a pipeline don't get SIGPIPE:

function redis_setexz {
  if [ "$CI_REDIS_AVAILABLE" -eq 1 ]; then
    gzip | redis_cli -x SETEX $1 $2 &>/dev/null
  else
    cat >/dev/null
  fi
}

This centrally protects all callers of redis_setexz (bootstrap_ec2, cache_log, denoise, run_test_cmd) — CI log/cache writes now degrade gracefully when redis is down instead of aborting the run.

Verification

Reproduced the original abort and confirmed the fix, under set -euo pipefail:

• Before: echo ... | redis_setexz → pipeline exits 141 (SIGPIPE) / gzip non-zero → script aborts (matches CI).
• After: redis-unavailable path returns cleanly (rc=0, stdin drained); redis-available path still pipes through gzip unchanged.

Note on recurring duplicates

This nightly auto-dispatches a claudebox investigation on every failure, and prior daily runs left ~15 cb/*redis-setexz* branches on the remote with no associated PRs (the fix never actually landed). This PR is the consolidated fix; the stale branches can be pruned.

---

Created by claudebox · group: slackbot