{% extends "_layout.html" %} {% set title = "Limit exceeded" %} {% set status = "429 Too Many Requests" %} {% set description = "A ceiling that refills. limit names which; honour Retry-After." %} {% block main %}

You have reached a published ceiling on how much you may have going at once — jobs, uploads, open log streams. Each of these refills as what you already have finishes.

What to do

  1. Read limit. It is the exact key published in limits — GET /v1’s, or your account’s in GET /v1/me — so you can compare it with what you hold.
  2. Wait for Retry-After, then send the same request again.
  3. On pending_uploads, job_ids names the jobs holding the slots that you may read: cancel one you will not submit instead of waiting. It may be empty, as for a CI credential bound to a project, whose person’s other jobs hold the slots and are cancelled from a session that can see them.

Formerly four separate types, each of which a client may still have in its code: quota-exhausted, pending-upload-limit, concurrent-job-limit and concurrent-stream-limit. They are one type now, and limit says which. max_download_bytes was briefly answered here too, and moved to its own type because it does not refill.

The body

{
  "type":     "https://sc.example.com/server-errors/limit-exceeded",
  "title":    "Limit exceeded",
  "status":   429,
  "detail":   "...",
  "instance": "/v1/jobs/01J9...",
  "limit":    "concurrent_jobs",  // REQUIRED on this type
  "trace_id": "4bf92f3577b34da6a3ce929d0e0e4736"
}
Member
limitREQUIRED The published key that bound: concurrent_jobs, pending_uploads or concurrent_log_streams. The usage keys — compute_seconds, storage_bytes, license_seconds — are registered and not raised today: nothing checks the compute quota at create or at submit.
job_idson pending_uploads The jobs holding the slots, in created or awaiting_input, that you may read. A CI credential bound to a project is named only those in its project, possibly none: an empty list means the slots are held by jobs this session cannot see.
trace_idoptional The request’s W3C trace id. Quote it in a support request.

detail is prose and may be reworded at any time. Branch on type, and read the members above for the specifics — never parse detail.

Where it comes from

POST /v1/jobs pending_uploads: too many jobs created and not yet submitted, with job_ids. concurrent_jobs: too many jobs staging, queued, running or cancelling.
POST /v1/jobs/{id}/submit concurrent_jobs, checked again.
the stream host concurrent_log_streams: too many live logs open at once, after …/logs redirected. A slot is held only while a connection is open.

A null pending_uploads or concurrent_jobs has no enforced ceiling and does not refuse. The device cap is not this type: a login over it is refused access_denied, reason: "devices", at the token endpoint, and a slot frees only when you revoke a device.

{% include "_help.html" %}

What this is not

node-limit-exceeded, upload-too-large and download-too-large are ceilings that do not refill: the request itself is too big, and waiting changes nothing.

rate-limited is about how often you call, not what you hold; it names no limit.

{% endblock %}