问题
I have a swarm composed of three nodes:
$ sudo docker node ls
ID HOSTNAME STATUS AVAILABILITY MANAGER STATUS
i12s3zxsn4vu1c98bv3i5idr8 node03 Ready Active
i2ckxvsju4tmommxim3dbfq7l node02 Ready Active
wak4isl46dn7pbo39drrhphju * node01 Ready Active Leader
Then I run 1 replica of nginx on that swarm and map his port to 8080:
$ sudo docker service create --replicas 1 --publish 8080:80 --name nginx nginx
$ sudo docker service ls
ID NAME MODE REPLICAS IMAGE PORTS
neahnb9mvi1i nginx replicated 1/1 nginx:latest *:8080->80/tcp
From there, i can reach nginx on http://node01:8080
Next, I scale nginx instances to 6:
$ sudo docker service scale nginx=6
$ sudo docker service ls
ID NAME MODE REPLICAS IMAGE PORTS
neahnb9mvi1i nginx replicated 6/6 nginx:latest *:8080->80/tcp
From there, i'm still able to reach nginx on http://node01:8080.
However, if docker swarm expose several node as a unique host, how does he manage the port during such a scaling operation as all my nginx services are mapped on the same 8080 port? Is there a round robin load balancing between all services instances done by swarm internally and returning the answer on 8080?
回答1:
I believe the requests to the hosts get assigned in a round-robin type assignment.
Found this handy article about it http://blog.scottlogic.com/2016/08/30/docker-1-12-swarm-mode-round-robin.html . Checkout the part titled 'INGRESS AND ROUND ROBIN LOAD BALANCING'.
回答2:
You can reach any service (deployed with a stack) which exposes its port by using the IP of any machine in the swarm and that port. Docker swarm will then forward that request to one of the replicas. Which replica is decided by the healthstate (unhealthy services are ignored) and the load balancing option that is configured which by default is round robin.
来源：https://stackoverflow.com/questions/46407531/docker-swarm-scaling-behaviour-on-port-mapping