From what I observe in my testing, and what I see in the code, nodes use akka.remote.netty.tcp.bind-hostname (via akka.cluster.Cluster#selfAddress, which calls akka.remote.RemoteTransport#defaultAddress), not akka.remote.netty.tcp.hostname, as self-address in coordination backend.
This means, nodes running in Docker can't form the cluster, since they try to join the internal address of the other nodes, which is obviously inaccessible from outside.
Is this understanding correct? If yes, maybe we can somehow accommodate for this (quite popular) deployment scenario?
I've tested more carefully, and found this issue invalid. ConstructR uses the correct hostname from the cluster, and everything works fine in that sense.
From what I observe in my testing, and what I see in the code, nodes use
akka.remote.netty.tcp.bind-hostname
(viaakka.cluster.Cluster#selfAddress
, which callsakka.remote.RemoteTransport#defaultAddress
), notakka.remote.netty.tcp.hostname
, as self-address in coordination backend.This means, nodes running in Docker can't form the cluster, since they try to join the internal address of the other nodes, which is obviously inaccessible from outside.
Is this understanding correct? If yes, maybe we can somehow accommodate for this (quite popular) deployment scenario?