...
https://hadoop.apache.org/docs/r3.1.1/hadoop-yarn/hadoop-yarn-site/PlacementConstraints.html
...
Kubernetes
My code only supports Yarn integration. But as far as I know, we could use nodeaffinity or podaffinity to achieve the same goal with Yarn PlacementConstraint in K8s integration.
For Kubernetes, we could the node selector to control where to assign the pod. However, one difference is that the K8s Apiserver could not help us to persistent this information. We need to store it by Flink internally.
Even we could add some label for specified node that we do not want to schedule for a Flink cluster. I do not suggest to go in this direction. Since it will pollute the node labels and may be residual when the Flink cluster destroyed.
Mesos
According to
Jira | ||||||
---|---|---|---|---|---|---|
|
...