Serverless vLLM: jobs stuck IN_QUEUE, no workers start (A40, balance OK)
Hi, I’m trying to use serverless vLLM and running into an issue where jobs never actually run.
Setup:
Endpoint name: omni-fast-7b
Endpoint type: Serverless vLLM, queue-based
GPU: A40 48GB
Min workers: 0, Max workers: 1
Balance: ~$45 shown in the console
Behavior:
Endpoint URL format: https://api.runpod.ai/v2/<ENDPOINT_ID>/openai
API key is valid (200 OK on a simple health check).
Requests from the dashboard and from my own client are accepted and show status: "IN_QUEUE".
The Workers tab always shows 0/1 running workers, 0 jobs in progress, even after waiting 60+ seconds.
This happens across multiple endpoints and GPUs (A40 / L40S / A100).
So the endpoint exists and accepts requests, but the autoscaler never starts a worker, so nothing ever leaves IN_QUEUE.
Could someone please check if there is any account‑level / serverless enablement / quota issue on my account that would prevent workers from initializing? I don’t need help with my code, just to understand why workers never spawn.
Thanks!
Setup:
Endpoint name: omni-fast-7b
Endpoint type: Serverless vLLM, queue-based
GPU: A40 48GB
Min workers: 0, Max workers: 1
Balance: ~$45 shown in the console
Behavior:
Endpoint URL format: https://api.runpod.ai/v2/<ENDPOINT_ID>/openai
API key is valid (200 OK on a simple health check).
Requests from the dashboard and from my own client are accepted and show status: "IN_QUEUE".
The Workers tab always shows 0/1 running workers, 0 jobs in progress, even after waiting 60+ seconds.
This happens across multiple endpoints and GPUs (A40 / L40S / A100).
So the endpoint exists and accepts requests, but the autoscaler never starts a worker, so nothing ever leaves IN_QUEUE.
Could someone please check if there is any account‑level / serverless enablement / quota issue on my account that would prevent workers from initializing? I don’t need help with my code, just to understand why workers never spawn.
Thanks!
