error in judging status of mxjob about mxnet-operator HOT 5 CLOSED

@jokerwenxiao ok, we'd like to keep same behavior with tf-operator, we will handle it later.

from mxnet-operator.

@jokerwenxiao mxnet-operator is designed for distributed training for the moment. In other words, mxnet-operator doesn't support the configuration of "scheduler replicas:0 , server replicas:0 and worker replicas:1". The behavior of what you watched is a bug and have to be fixed later. Why don't run your container instance just as a pod for your case?

from mxnet-operator.

@suleisl2000 just like tf-operator, i can use worker-0 to train no-distributed job. I wonder if mxnet-operator can do this in the future. thank you!

from mxnet-operator.

Did you modify the crd of mxnet-operator? I can't create mxjob with the same settings, its crd has set the minimum of replica to 1, and tf-operator does things like it, too.

from mxnet-operator.

@KingOnTheStar
After deploying kubeflow, I haven't modified the crd of mxnet-operator.
for tf-operator, I deleted ps replicaSpec, leave only worker replicaSpec.

from mxnet-operator.