---
name: jobqueue-namespace-pitfall
description: "Start a jobqueue scheduler only with JOBQUEUE_NAMESPACE inline on the same command; an unset namespace runs the DEFAULT queue (other users' jobs)"
metadata:
  node_type: memory
  type: feedback
  originSessionId: d4dac82a-f3e4-4196-a274-a5c105cb0541
  modified: 2026-11-11T01:11:36.956Z
---

When starting a scheduler, put the namespace on the SAME line: `JOBQUEUE_NAMESPACE=x nohup ../../.venv/bin/python runners/jobqueue.py run ...`, never `nohup ... &` followed by `export JOBQUEUE_NAMESPACE=x`. Then check `logs/jobqueue_<x>.log` (not `logs/jobqueue.log`) shows "scheduler started".

**Why:** 2026-11-11 01:08 UTC a scheduler started without the namespace ran the default queue: it "requeued after scheduler restart" two stale running jobs of another user (78100/78110 lpftcbramod) and started 7900_stimpt_USTC. Killed after ~2 min by PID with all children; the jobs only re-scored cached folds, but their log files (`> logs/...out`) were overwritten and the original start times lost. Queue state restored by hand with a note in each job file and in logs/jobqueue.log.

**How to apply:** any `jobqueue.py run` (and `add`/`status` that must target a namespace) gets the variable inline; a new scheduler also requeues whatever is in its namespace's running/ folder, so look at that folder first. Related: [[ram-is-the-limit]], [[no-blocking-waits]].
