I was working with autoscaler to do some tests and i had the crazy idea to try the GPU for large datasets.
The thing is, to do that i must change the image to
opendronemap/nodeodm:gpu and my cloudProvider (scaleway) is asking for another machine image (ubuntu GPU) instead of my generic instance image.
So i was thinking, the simplest way to do that in the autoscaler is to allow optional parameters in
imageSizeMapping such as
ami (for aws) and
dockerImage to tune with specific
Then just modify the function
getCreateArgs in each provider to handle this configuration. It is a small task and i would be willing to take it.
To handle GPUs, another optional parameter could be needed to add
--gpus all to the
docker run command
Would you be interested by such a feature?