Dispatching multiple workflow instances

[Neosettler]

Greetings,

For context, my current setup generates only one image per workflow queued. Each workflow is identical but with a different seed. Somehow, I was expecting that queued items would be distributed to available GPUs automatically. Am I missing something or that's a no go?

Thank you,

[robertvoy]

Can you screenshot me your workflow?

[Neosettler]

Hi Robert, thank you for your support.

Sure thing, note that sampler batch_size is 1 but the main batch count is 8. Results, Master is processing images sequentially. Denoising goes as fast as usual but it hangs for each image:

"ComfyUI" : [Distributed] Master - Heartbeat timeout: worker=070ba07f-29d9-468c-9a82-a21db8b398e9, elapsed=60.3s
[Distributed] Master - Heartbeat timeout: worker=6ee925bd-5ca4-49f6-92ca-0322efe7b23e, elapsed=60.3s
[Distributed] Master - Heartbeat timeout: worker=c29f0d9f-c49b-42db-b73f-a634ce119334, elapsed=60.3s
[Distributed] Master - Heartbeat timeout. Still waiting for workers: ['070ba07f-29d9-468c-9a82-a21db8b398e9', '6ee925bd-5ca4-49f6-92ca-0322efe7b23e', 'c29f0d9f-c49b-42db-b73f-a634ce119334'] (elapsed=60.3s)

"ComfyUI" : Prompt executed in 60.86 seconds

PS: The 8 images takes about 10 seconds to generate whiteout the distributed nodes. Also, sampler batch_size to 3 will dispatch to the workers as expected so that works but still, it will hang at the Distributed Collector node for 60 seconds.

EDIT: Latest portable comfyUI with the Manager and the Distributed nodes, that's it.

[robertvoy]

What do the logs on the workers say?

it's hanging for 60s, because it's waiting for the results from the workers that never come.

[Neosettler]

It's just a single machine with only one ComfyUI instance running. And the log is just looping the code block above. What am I missing?

[robertvoy]

in the GPU panel you have 3 workers, each of them have a view log button. What is the contents of the log from a worker when u run this workflow?

[Neosettler]

Ah... it spins an instance for each GPU device?

[Distributed] Worker - Failed to send canonical image envelope to master: Cannot connect to host x.x.x.x:8188 ssl:default [The remote computer refused the network connection]
!!! Exception during processing !!! Cannot connect to host x.x.x.x:8188 ssl:default [The remote computer refused the network connection]
Traceback (most recent call last):

...SNIP

  File "comfy\python_embeded\Lib\site-packages\aiohttp\connector.py", line 1321, in _wrap_create_connection
    raise client_error(req.connection_key, exc) from exc
aiohttp.client_exceptions.ClientConnectorError: Cannot connect to host x.x.x.x:8188 ssl:default [The remote computer refused the network connection]

[robertvoy]

Ive pushed an update. It should fix this.