No available workers

[flat sedge]

All of my workers have been throttled. Is this expected?
That only 1 left sometimes back to throttled and then back to initialization stage and downloading the image, back and force.

[wanton scroll]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[analog mist]

No, just add more workers if you want

means theres a high demand for the gpu in that region

[flat sedge]

I think it's the algorithm has changed, to more aggressively make the worker throttled. I already changed the region to be almost all, and I see it's high supply for my machine

[flat sedge]

And I am not sure if machines switch the state between idle and initializing is an efficient way of doing the scheduling, isn't downloading and remove the image a resource consuming thing to do?

[analog mist]

Yes maybe it's consuming but maybe they need the resources for other to use using the other's docker image

[analog mist]

I see, are you experiencing any problems with using the endpoint because of this?

[flat sedge]

At first, since I was only using 6 machines, there was only 1 idle/available machine, which made me nervous.
I also noticed that the request delay increased when I sent more requests at the same time, since the machine comes from throttled to idle and can be used.
But now, I see that 9 out of my 10 machines are idle. Not sure if runpod adjusts the algo during the time, but this does look much better.

[analog mist]

I see

How's the usage pattern on your endpoint? Is it often full? Perhaps if it gets more usage it will be throttled less