I have a docker (version 17.05.0-ce-rc3) container "logger" running on a host in a VPC (for simplicity, call the underlying host "worker") that is unable to communicate with a host in the same VPC. The target host is running kafka, so I'll refer to that as "kafka".
A few points:
1) worker can connect to kafka on port 9092.
2) logger can connect to worker if I open a port on worker
3) when logger attempts to connect to kafka, it appears to be routing it outside the vpc, to the external internet, which is obviously wrong.
traceroute to 172.31.24.83 (172.31.24.83), 30 hops max, 46 byte packets
1 ip-172-18-0-1.us-west-2.compute.internal (172.18.0.1) 0.005 ms 0.004 ms 0.003 ms
2 * * *
route on logger yields:
Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
default ip-172-18-0-1.u 0.0.0.0 UG 0 0 0 eth1
172.18.0.0 * 255.255.0.0 U 0 0 0 eth1
172.31.115.0 * 255.255.255.0 U 0 0 0 eth0
5) Logger is able to resolve kafka via hostname.
6) I believe it isn't a security group setting, as if that were the case, worker wouldn't be able to communicate with kafka either.
docker network ls on worker yields:
docker network ls
NETWORK ID NAME DRIVER SCOPE
ca66dc7f7326 bridge bridge local
f4e66ecf1f5e docker_gwbridge bridge local
29946d1fe867 host host local
npm3fm1gkipp ingress overlay swarm
8d369d123025 none null local
bdprje9fq1nq ops_default overlay swarm
logger is running within the context of the ops_default network above.
8) I'm running all of this in AWS, I'm using Swarm as my orchestration engine.
9) Both worker and kafka are in the same VPC.
Can anyone tell me what I need to do to ensure that logger can connect to kafka? Has anyone else seen something like this? I've been searching and haven't turned up anything.