K8s Operater deploy port conflict.

[feloxx]

We followed the example in operater to deploy in a k8s.

Because it's a test, there are fewer nodes.

The following problems arise:

0/2 nodes are available: 2 node(s) didn't have free ports for the requested pod ports.

At the source code, the Coordinater and shuffle server use the same port by default, and conflicts do arise if they are deployed together.

---

We did the following:

1. Coordinater use hostNetwork: true, Shuffle use hostNetwork: false, is ok.
2. Yaml specifies rpcPort, httpPort, is error.
3. Yaml add env COORDINATOR_RPC_PORT, COORDINATOR_HTTP_PORT, is error.

---

How to resolve the conflict?

or

If there is a actual example of operator, such as uniffle.apache.org_remoteshuffleservices.yaml is just a reference, but the specific how much this configuration, we need to keep trying.

[advancedxy]

@wangao1236 ao if you have time, would you like to answer this question in more details?

By default, coordinator and shuffle servers use host network, so that it can serve apps outside K8S cluster.

For coordinator and shuffle server, it both requires two ports: 19996 & 19997, and the count of coordinators is at least 2.

So, if you only have two nodes, you need to change ShuffleServer's rpcPort and httpPort to other values with coordinators.

[wangao1236]

Yes, as @advancedxy says, usually both components should use the host network.

[wangao1236]

@feloxx Could you please paste the whole yaml of rss object and pods of shuffle server and coordinator?

[jerqi]

@feloxx Your problem should be solved by #627