Nodes with duplicate hostnames fail to join kong cluster when using default cluster_listen

[nguilford]

Summary:
Nodes with identical hostnames using the default cluster_listen ipv4:port fail to join a cluster because serf expects node names to be unique, and the code for get_node_name does not generate unique names in these conditions:

https://github.com/Mashape/kong/blob/997561b42c27d5917f261065d9b1f5e99079e337/kong/tools/cluster.lua#L6

Why this is a problem:
I use docker to spin up multiple instances of kong containers that live behind a load balancer. Those instances are brought up as members of the same pool and, as such, are given the same hostname; /etc/hosts on each instance is configured to resolve that hostname to localhost. ipv4 addresses are not assigned to these instances until runtime, so I wish to use kong's default cluster_listen ip:port (0.0.0.0:7946). Unfortunately, in this case get_node_name will return the same name for each node in my cluster, resulting in all but one node failing to join the cluster.

At the moment, my container grabs the node's local ipv4 address at runtime and appends it to kong.yml file for cluster_listen before starting up kong. But it would be nice to have this working out of the box.

Version/Platform:
I am running Kong version 0.8.1 in Docker.

Suggested fix:
Just as the first local non-loopback address is used to advertise a node's presence to other nodes when the cluster_listen ipv4 address is 0.0.0.0, first local non-loopback address could be used to calculate the node's name. There is no reason two nodes in the same cluster should have the same non-loopback ipv4 address, so this would be unique.

[subnetmarco]

@nguilford could you try to manually apply the patch at #1281 and see if that fixes your problem?

[nguilford]

Yes, #1281 fixes this issue. Thank you!

[Tieske]

released in 0.8.3. closing this.