I'm using Flyte-demo-sandbox. I have build the wor...
# ask-the-community
r
I'm using Flyte-demo-sandbox. I have build the workflow which training with FasterRCNN_ResNet50_FPN. I just want to testing with CPU pod first. But i catch the error on Kubenetes-dashboard that the pod which exec my wf had been killed with status ( OOMKilled), and i saw the resource of each pod is only 2Gb.I have 2 question: My first question is What is limit resource of @task, cuz i have set up @task(limits=Resources(mem="20000Mi")) ? Thus i, think that if i set up like this, the limit memory for my pod should be 20000Mi. And the second question is how to up the limit of resource of the pod.
m
You probably need to increase the cluster resource limits, see this thread
y
you should also just set requests. on the backend flyte will set requests=limits. this is how most containers are run at least.
if you want to set both you of course can.
just keep in mind when requests are not equal to limits, k8s uses a different eviction policy for the pod.
r
Ths, I have clearly understand about it