{% extends "_layout.html" %} {% set title = "The run failed" %} {% set status = "Not an HTTP response" %} {% set description = "Never an HTTP response. A type on a job’s or node’s error object." %} {% block main %}

The run itself failed, in a way resubmitting unchanged repeats. The job was accepted and ran, and a node — or the run around the nodes — ended in failure: a tool or flow failure, or a deployment time or memory limit, a queued job’s included, with detail naming the limit.

What to do

  1. It will fail the same way: change something before you resubmit.
  2. Where detail names a time or memory limit, the fix is a smaller run or a larger limit.
  3. Otherwise look at the nodes: the one whose error.type is this, and whose exit_code is non-zero, is where it happened. Its own error.detail says what the server knows, and its log says the rest. Where no node failed, the job’s own log says why the run stopped.

The body

{
  "id":    "01J9...",
  "state": "failed",
  "error": {
    "type":   "https://sc.example.com/server-errors/run-failed",
    "title":  "The run failed",
    "detail": "..."
  }
}

detail is prose and may be reworded at any time. Branch on type — never parse detail.

Where it comes from

GET /v1/jobs/{id} As the job’s error.type, and as the error.type of each failed node, whose detail names the limit it ran into.

Getting help with your specific failure

This type arrives inside a job, not on a request, so there is no trace_id on it. Quote the job’s id when you escalate — it is the one identifier that finds the run on the server, its scheduler job and its logs.

This page documents the error type. It is the same page for everyone who hits it and it cannot see your job — the job id is what connects the two.

What this is not

run-interrupted is the environment ending the run — preemption, a failed compute node, a lost job, an image that could not be pulled — and resubmitting unchanged may work.

{% endblock %}