Multi-instance gateways with same HOME share cron store and duplicate-fire every job

Summary

When two openclaw-gateway instances run on the same host with the same HOME env var, they both load ~/.openclaw/cron/jobs.json and both fire every scheduled cron tick. OPENCLAW_CONFIG_PATH is honored for the main config file, but resolveConfigDir() derives the cron/state directory from path.dirname(configPath), which collapses to the same ~/.openclaw for both instances.

Result: every cron job fires twice, once per gateway. Each fires with a different agent/bot context, so one run may succeed while the other fails (or both fail), producing the classic "sometimes works, sometimes doesn't" symptom.

Environment

• macOS (darwin 24.6.0)
• openclaw installed globally at ~/.npm-global/lib/node_modules/openclaw
• Two launchd services running side by side: ai.openclaw.gateway (port 18789) and ai.openclaw.sugi (port 18790)
• Both plists set HOME=/Users/robcolvin
• Sugi's plist sets OPENCLAW_CONFIG_PATH=/Users/robcolvin/.openclaw/sugi.json

Repro

1. Start two gateway instances on the same host with the same HOME and distinct OPENCLAW_CONFIG_PATH values pointing to two separate config files in the same parent dir.
2. Create any cron job on one of them — the job is written to ~/.openclaw/cron/jobs.json.
3. Wait for the next tick.

Expected

Exactly one finished entry per scheduled tick in ~/.openclaw/cron/runs/<jobId>.jsonl, fired by the instance that created (or owns) the job.

Actual

Two finished entries per tick with the same runAtMs but different sessionIds, delivered by whichever bot/agent each gateway has configured:

awk -F'"runAtMs":' '{print $2}' ~/.openclaw/cron/runs/c4f6321b-*.jsonl \
  | awk -F',' '{print $1}' | sort | uniq -c | sort -rn | head
   2 1776773208280
   2 1776600000016
   2 1776340800007
   2 1776168000006
   2 1775908800011

The two runs often produce inconsistent outcomes — e.g., one delivery succeeds (the correct bot for the target chat), the other hits Telegram send failed: chat not found (chat_id=...) because the other gateway's bot isn't in that chat.

Root cause (traced in dist/)

• utils-Supr9KvU.js: resolveConfigDir(env, homedir) resolves to path.dirname(env.OPENCLAW_CONFIG_PATH) when that env var is set, otherwise path.join(homeDir, ".openclaw").
• store-B5g6FEBF.js:9-13:

  function resolveDefaultCronDir() {
      return path.join(resolveConfigDir(), "cron");
  }
  function resolveDefaultCronStorePath() {
      return path.join(resolveDefaultCronDir(), "jobs.json");
  }

Since both instances' OPENCLAW_CONFIG_PATH files live in ~/.openclaw/, path.dirname() collapses them to the same directory and both instances share ~/.openclaw/cron/jobs.json.

Workaround

Add "cron": { "enabled": false } to the subordinate instance's config file. server.impl-BxLfE9ri.js:9066 gates buildGatewayCronService on this flag at boot, which prevents duplicate firing. Downside: the subordinate instance can no longer fire its own crons either.

Suggested fix

Either:

1. Introduce OPENCLAW_STATE_DIR (or similar) that is independent of OPENCLAW_CONFIG_PATH, so each instance can have its own cron/, memory/, workspace/, sessions/, etc. Document that running multiple gateways on one host requires setting it.
2. Namespace the cron store path by instance — e.g., derive from OPENCLAW_LAUNCHD_LABEL or the config filename stem (<dir>/cron/<instance>/jobs.json).
3. At minimum, warn on startup if two gateway processes are detected using the same cron store path.

Happy to test a patch.