cyrildiagne / kuda

Serverless APIs on remote GPUs

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Increase stable-window

cyrildiagne opened this issue · comments

The default 30s grace period of Knative is too short in comparison to the time most CUDA based services take to boot.
A few minutes would make more sense: https://knative.dev/docs/serving/configuring-autoscaling/