Skip to content

fix: increase SPAWN_TIMEOUT_MS from 30s to 60s#538

Open
homelab-coder wants to merge 1 commit into
the-open-engine:mainfrom
homelab-coder:fix/increase-spawn-timeout-to-60s
Open

fix: increase SPAWN_TIMEOUT_MS from 30s to 60s#538
homelab-coder wants to merge 1 commit into
the-open-engine:mainfrom
homelab-coder:fix/increase-spawn-timeout-to-60s

Conversation

@homelab-coder

Copy link
Copy Markdown

Problem

The hardcoded SPAWN_TIMEOUT_MS = 30000 (30 seconds) in agent-task-executor.js causes validator crashes with the opencode provider when LM Studio model priming takes longer than 30s. This is a known issue on consumer hardware where model priming (warming up KV cache / compute graph) can take 10-15s, pushing total CLI startup to 35-40s.

When the timeout fires:

  1. The validator agent crashes with Spawn timeout after 30s - provider CLI hung
  2. After 3 retry attempts, the validator is marked as crashed and rejects validation
  3. The entire cluster enters an infinite reject→retry→crash loop

Fix

Increase SPAWN_TIMEOUT_MS from 30000 to 60000 in both spawn task process functions (non-isolated and isolated modes). This accommodates slow model initialization without significantly impacting detection of genuinely hung CLIs.

Changes

  • src/agent/agent-task-executor.js: SPAWN_TIMEOUT_MS = 3000060000 (2 occurrences, lines ~830 and ~1486)

Testing

Verified by running zeroshot run 20 --pr --provider opencode -d against the latent-space-dance-off repo (Issue #20: format_duration inconsistency). The previous attempt with 30s timeout produced 725 messages over ~20 minutes in a reject→retry loop. With 60s timeout, validators should complete successfully within their spawn window.

The 30-second spawn timeout causes validator crashes with opencode provider
when LM Studio model priming takes >30s. Increase to 60s to accommodate
slow model initialization on consumer hardware.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant