Re-ran a UGC variant bulk run and got the old queue: why

If you re-run a Sume bulk request with the same Idempotency-Key and the same { concurrency, items } body, you get 202 and the existing queue back, not a new batch of UGC variants. With the same key and a different body, you get 409 idempotency_conflict, and details.queue_id names the original. To make a new batch, mint a new key.

This is stated in Sume's Bulk runs docs under Idempotency. Keys are scoped to one Format. The replay stays 202 for bulk, unlike a single run, and the queue object has no idempotency_hit field, so you cannot tell a replay from a fresh create by the status code.

What happens for each replay?

Sume's docs give three cases.

Why does it cause surprise?

A script that hard-codes the key, or derives it from the Format name, will quietly return the first queue forever. You edit the hook lines, run again, and still see last week's queue, because the body only changes if the items changed. If the items are identical you get the old queue; if they differ you get a 409. Either way no new variants render until you change the key.

How should I key a batch?

Generate one key per intended batch and store it next to the batch, so a network retry resends the same key and a deliberate re-run gets a new one. The docs also accept idempotency_key in the body, and the header wins when both are present. The sample uses the catalog Format sume-close-camera-ugc called at the reserved sume handle, which any key with formats:write may call.

import json, os, urllib.request, uuid

KEY = os.environ.get("SUME_API_KEY", "")
if not KEY:
    raise SystemExit("SUME_API_KEY is empty")

def create_queue(items, idem_key):
    body = json.dumps({"concurrency": 3, "items": items}).encode()
    req = urllib.request.Request(
        "https://api.sume.com/v1/formats/sume/sume-close-camera-ugc/bulk-runs",
        data=body,
        method="POST",
        headers={
            "Authorization": f"Bearer {KEY}",
            "Content-Type": "application/json",
            "Idempotency-Key": idem_key,
        },
    )
    with urllib.request.urlopen(req) as r:
        return json.load(r)["data"]["id"]

batch_key = str(uuid.uuid4())  # new key per batch, saved with the batch

What if the retry came from a timeout?

Resend the same body with the same key. You get the queue you already created, so there is no second charge for the same items. Then follow status_url and read counts; completed does not mean all succeeded.