He tries to get a token, but since the service has not yet started, he gets a connection refused message. I checked the ports, 8030 and the others are available on the leader, but not on this follower.

It doesn't start, which means the service isn't sending traffic to it. There's no endpoint.

Until the main service is up, it won't route traffic or open ports. I even tried deleting the probes, but the result is the same.

I ran nginx on the same node, and when it started, traffic started flowing. It turns out the k8s node itself doesn't open the port until the pod starts.

The probes are running on the same port as the web server.

alter system add follower ...

2025-12-08 13:32:43.696Z INFO (main|1) [NodeMgr.getFeNodeTypeAndNameFromHelpers():553] get fe node type FOLLOWER, name 172.21.0.101_9010_1765200762950 from 172.21.0.100:8030
2025-12-08 13:32:43.705Z INFO (main|1) [NodeMgr.getVersionFileFromHelper():701] Downloading version file from <http://172.21.0.100:8030/version>
2025-12-08 13:32:43.763Z INFO (main|1) [NodeMgr.getNewImageOnStartup():743] skip download image for /opt/starrocks/fe/meta/image, current version 0 >= version 0 from 172.21.0.100:9010
2025-12-08 13:32:43.763Z INFO (main|1) [NodeMgr.getClusterIdAndRoleOnStartup():499] Current run_mode is shared_nothing
2025-12-08 13:32:43.764Z INFO (main|1) [NodeMgr.getClusterIdAndRoleOnStartup():504] Got role: FOLLOWER, node name: 172.21.0.101_9010_1765200762950 and run_mode: shared_nothing
2025-12-08 13:32:43.765Z INFO (main|1) [BDBEnvironment.ensureHelperInLocal():351] start to check if local replica environment from /opt/starrocks/fe/meta/bdb contains 172.21.0.100:9010

sr-port-service

experimented for a couple of hours.
What assortment would you recommend?

A client connected via a load balancer makes a request from the Kubernetes node, not through the service....

The K8S load balancer has a static public IP address, and the airflow request goes over the internet. Using virtual cables, you get two computers from different networks connected via a single cable. This makes them accessible via OSI Layer 2 over MAC. I'll test this theory today.

For some reason, a request is coming from the node's IP address?

2025-12-10 16:00:00.286+03:00 WARN (starrocks-mysql-nio-pool-3|190) [StmtExecutor.execute():800] execute Exception, sql SHOW COMPUTE NODES

`org.apache.thrift.TApplicationException: Internal error processing forward`

How is this pool assembled? Can I look at the code?

We did this, and it handles single requests fine. But when we launch Airflow, we get this error. As far as I understand, some requests are coming from the Kubernetes node's IP and are being rejected.
I've contacted Cloud support, and maybe they can explain.

requests once per minute, request with Internal error processing forward, once every three minutes.

Can you tell me what problem you're seeing? Again?
Cloud support said everything's fine, check your Starrocks settings.
I've read a lot of issues and seen that detailed problems are resolved by initing the container that parses DNS names, and then configuring fe.conf based on that.
Something like this.
      - bash
      - '-c'
      - |
      set -ex
      hostname=`hostname`
      ip=`ifconfig eth0|grep inet|awk '{print $2}'`
      echo "$(sed 's/^'"$ip"'.*/'"$ip"'        '"$hostname"'.starrocks-shared-data-fe-service.***.svc.dev-blz1.hefei '"$hostname"'/g' /etc/hosts)" > /etc/hosts
      [[ $hostname =~ -([0-9]+)$ ]] || exit 1
      ordinal=${BASH_REMATCH[1]}
       if [[ $ordinal -eq 0 ]]; then
           #first pod as leader default, FQDN host type
           /opt/starrocks/fe/bin/start_fe.sh --host_type FQDN
       else
           #other pods as followers,helper is the first pod's service, FQDN host type 
           /opt/starrocks/fe/bin/start_fe.sh --helper starrocks-shared-data-fe-0.starrocks-shared-data-fe-service.***.svc.dev-blz1.***i:9010 --host_type FQDN 
       fi

Of course, you can try and then look at the service manifests; there might be some option that will solve everything.

It would be interesting to understand and solve the problem)

You can say, "I need manifests for this and that," I'll generate them and send them to your email. Take a look.