openclaw - 💡(How to fix) Fix [Bug] Cron isolated session first LLM call times out in 2026.4.2 [1 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Cron job with sessionTarget=isolated fails with LLM request timeout on first call, preventing all isolated cron tasks from executing

Steps to reproduce

1. Create a cron job with sessionTarget: "isolated" (e.g., openclaw cron add with default isolated session target)
2. Set payload.kind to agentTurn with a simple task (e.g., run a Python script and send a message)
3. Schedule to trigger at a specific time (e.g., 0 9 * * 1-5)
4. Wait for cron to fire

Observed timeline on 2026-04-09:

• 09:00:00 — cron triggered
• 09:00:05 — first LLM call issued
• 09:01:05 — exactly 60s later, timeout: "Profile minimax:cn timed out. Trying next account..."
• 09:02:05 — second attempt also times out
• Final error: "FailoverError: LLM request timed out."

Expected behavior

The isolated session should complete bootstrap and execute the cron payload within a reasonable time. The first LLM call should not time out on a lightweight task with the same model that works normally in webchat/feishu sessions.

Actual behavior

Every isolated cron job times out on the first LLM call with "FailoverError: LLM request timed out." The bootstrap phase takes ~65 seconds before the first call is even issued, then the call itself hits the 60s timeout and fails. This has been consistently reproducible across multiple dates (2026-03-30, 2026-04-06, 2026-04-08, 2026-04-09).

Key log evidence:

2026-04-09T09:01:05.534 [agent/embedded] Profile minimax:cn timed out. Trying next account...
2026-04-09T09:01:05.642 [agent/embedded] embedded run failover decision: ... reason=timeout provider=minimax/MiniMax-M2.7-highspeed
2026-04-09T09:02:05.922 [diagnostic] lane task error: lane=nested durationMs=125759 error="FailoverError: LLM request timed out."

OpenClaw version

2026.4.2 (d74a122)

Operating system

macOS 24.6.0 (Darwin, x64)

Install method

npm global (npm install -g openclaw)

Model

minimax/MiniMax-M2.7-highspeed

Provider / routing chain

minimax direct API (auth profile: minimax:cn)

Additional provider/model setup details

Cron job configured with:

• sessionTarget: "isolated"
• agentId: "main"
• payload.kind: "agentTurn"
• timeoutSeconds: 300

The same minimax/MiniMax-M2.7-highspeed model works fine in webchat and feishu sessions on the same machine at the same time. The issue is specific to isolated session bootstrap. auth-profiles.json shows two minimax profiles (minimax:cn and minimax:default), both timing out during the isolated session first call.

Logs, screenshots, and evidence

Gateway error log excerpts:

2026-04-09T09:01:05.534+08:00 [agent/embedded] Profile minimax:cn timed out. Trying next account...
2026-04-09T09:01:05.642+08:00 [agent/embedded] embedded run failover decision: runId=5cf8ef8b-4ac9-46d9-8f3a-924996033fb7 stage=assistant decision=rotate_profile reason=timeout provider=minimax/MiniMax-M2.7-highspeed profile=sha256:c38c74a5066a
2026-04-09T09:02:05.915+08:00 [agent/embedded] Profile minimax:default timed out. Trying next account...
2026-04-09T09:02:05.919+08:00 [agent/embedded] embedded run failover decision: runId=5cf8ef8b-4ac9-46d9-8f3a-924996033fb7 stage=assistant decision=fallback_model reason=timeout provider=minimax/MiniMax-M2.7-highspeed profile=sha256:b58aa0cc713a
2026-04-09T09:02:05.922+08:00 [diagnostic] lane task error: lane=nested durationMs=125759 error="FailoverError: LLM request timed out."
2026-04-09T09:02:05.924+08:00 [diagnostic] lane task error: lane=session:agent:main:cron:8474672a-4354-4a95-aef9-9a4977bf2142 durationMs=125762 error="FailoverError: LLM request timed out."

Previous occurrences (same pattern):
- 2026-03-30T09:28: isolated cron timeout, durationMs=946218
- 2026-04-06T09:43: isolated cron timeout, durationMs=1161425

Impact and severity

Affected: All cron jobs configured with sessionTarget="isolated" (background scheduled tasks) Severity: High — completely blocks the cron feature, making it unusable Frequency: 100% — every isolated cron attempt fails on first call Consequence: Scheduled background tasks (monitoring, reports, alerts) never execute. Users must manually run tasks that should be automated.

Additional information

This is a regression: isolated cron sessions worked (or at least did not timeout) in earlier versions. The bootstrap phase appears to have an abnormally long overhead (~65 seconds before first LLM call is even made) in the 2026.4.2 isolated session environment.

Root cause hypothesis: Isolated session bootstrap in 2026.4.2 has excessive fixed overhead (workspace file loading, skills snapshot processing) that causes the first LLM call to exceed the 60s timeout before it even gets to send the request.

Note: The cron job has deleteAfterRun:true, so failed sessions leave zombie entries in sessions.json that are never cleaned up.