This message was deleted.
# rke2a
adamant-kite-43734
03/14/2023, 12:38 AMThis message was deleted.
c
creamy-pencil-82913
03/14/2023, 12:39 AMThis sounds like an issue with Rancher, not with RKE2?
creamy-pencil-82913
03/14/2023, 12:41 AMThat said…
this hostname suggest that you are using a private DNS zone for your Rancher server. Can you confirm that the resolv.conf file on your nodes is properly configured to point at that? Check the RKE2 logs for a message about using 8.8.8.8 instead of your private DNS server.Copy codeERROR: <https://rancher75182.senode.dev/ping> is not accessible (Could not resolve host: rancher75182.senode.dev)
creamy-pencil-82913
03/14/2023, 12:42 AMHost resolv.conf includes loopback or multicast nameservers - kubelet will use autogenerated resolv.conf with nameserver 8.8.8.8s
stale-painting-80203
03/14/2023, 12:42 AMyes. I can ping the rancher server from all nodes on the downstream cluster
c
creamy-pencil-82913
03/14/2023, 12:42 AM^^ itll look like that
creamy-pencil-82913
03/14/2023, 12:43 AMmake sure that your resolv.conf doesn’t point at multicast or loopback resolvers
creamy-pencil-82913
03/14/2023, 12:43 AMIf you can’t resolve private hostnames from within pods, RKE2 falling back to 8.8.8.8 is the most likely reason why
s
stale-painting-80203
03/14/2023, 12:46 AMmine only has the nameserver pointing to an ipaddr
stale-painting-80203
03/14/2023, 12:46 AMeverything else is commented out
stale-painting-80203
03/14/2023, 3:14 AMEven rebooting the cluster VMs on a working cluster causes the same issue and the cluster never comes back up again. Same pods go into a CrashLoop
c
creamy-pencil-82913
03/14/2023, 3:22 AMIs the name server not reachable from within pods? What do the coredns pod logs say?
s
stale-painting-80203
03/14/2023, 4:03 AMcoredns pods have the right nameserver and can access from there. Unable to exec omnt the pods which are in a crashloop
stale-painting-80203
03/14/2023, 4:07 AMseems bazar that a working cluster is unable to establish connection with a rancher after the cluster VMs were rebooted
c
creamy-pencil-82913
03/14/2023, 4:54 AMSo just to confirm, the coredns pods can resolve the rancher server address but the rancher agent pods cannot? How did you test that it can be resolved by coredns?
s
stale-painting-80203
03/14/2023, 5:04 AMRecreated a fresh cluster and collected more data. It seems on cluster create the cluster-agent pods successfully come up even though they are unable to ping rancher. Cluster also connects to rancher. One has an error on the certificate and the other cannot reach it. After rebooting the cluster, the cluster-agent pods go in a crashloop
stale-painting-80203
03/14/2023, 5:10 AMalso from my local machine or VMs, I don't see any issues with rancher ping:
Copy code
curl <https://rancher75182.senode.dev/ping>
pongstale-painting-80203
03/14/2023, 5:19 AMAlso to answer your question regarding coredns, Yes. exec into coredns and curl google.com or my rancher instance and it works, but not from within the cluster-agents
stale-painting-80203
03/14/2023, 5:25 AMcoredns IPs are 10.42.247.129 and 10.42.213.131, but the cluster-agent resolv.conf has a nameserver 10.43.0.10. Is that correct?
c
creamy-pencil-82913
03/14/2023, 9:01 PMyes, that is the dns service address. Can you show me how specifically you are testing resolution from within the coredns pods?
s
stale-painting-80203
03/14/2023, 9:02 PMJust doing a curl as shown above.
c
creamy-pencil-82913
03/14/2023, 9:03 PMYou said you were doing that from the local machine or VM, not inside the coredns pod
s
stale-painting-80203
03/14/2023, 9:05 PMBoth. I also created a new cluster on the same server as Rancher and it seems to work fine. I am able to restart the cluster and it comes back up. I suspect it’s some networking issue on the Harvester server on which the cluster fails to come up
stale-painting-80203
03/14/2023, 9:05 PMAfter a restart
c
creamy-pencil-82913
03/14/2023, 9:07 PMhmm
231 Views