Services are not unregistering properly in docker container

[xetys]

Whenever you run a JHipster microservice application without docker container it is unregistering itself immediatly when is stopped, so it disappears in eureka server and is no longer offered for load balancer in zuul....

this is not happening when using docker. You can easily reproduce it by start a jhipster microservice setup with docker-compose up -d and then stop and start it several times. you will quickly get several dead instances inside of zuul....

I am not sure, but this was working before....and it is still working without docker...

[jdubois]

It depends on how the instance is killed, but probably if you do a docker-compose down then JHipster doesn't have the time to send a notification, so indeed the registry doesn't know it was removed.
I don't think there is anything we can do about it.
It works differently if you kill the app "nicely", as there is a shutdown hook that is triggered, and JHipster can send the notification before dying. This is what happens "outside Docker", probably.
If you wait a little bit, dead instances should disappear from the Registry, after a timeout: there's a heartbeat that should handle that issue after some time. It could be configured to be more aggressive, but we keep the default Netflix Eureka configuration, as usual.

[pascalgrimaud]

@xetys / @jdubois : for your information, I have the same thing on my current project.
There is an interesting ticket opened here: see spring-cloud/spring-cloud-netflix#1322

So I tested this config on my registry:

eureka:
    server:
        enable-self-preservation: false

Here the log I have:

2016-09-08 11:00:17.875  INFO 12828 --- [a-EvictionTimer] c.n.e.registry.AbstractInstanceRegistry  : Running the evict task with compensationTime 0ms
2016-09-08 11:01:17.875  INFO 12828 --- [a-EvictionTimer] c.n.e.registry.AbstractInstanceRegistry  : Running the evict task with compensationTime 0ms
2016-09-08 11:01:17.876  INFO 12828 --- [a-EvictionTimer] c.n.e.registry.AbstractInstanceRegistry  : Evicting 1 items (expired=1, evictionLimit=1)
2016-09-08 11:01:17.876  WARN 12828 --- [a-EvictionTimer] c.n.e.registry.AbstractInstanceRegistry  : DS: Registry: expired lease for MSXXXX/msxxxx:edaf71c06bb773379b435e1c39f61e50
2016-09-08 11:01:17.878  INFO 12828 --- [a-EvictionTimer] c.n.e.registry.AbstractInstanceRegistry  : Cancelled instance MSXXXX/msxxxx:edaf71c06bb773379b435e1c39f61e50 (replication=false)

And the instance is cleared on the home page of the registry

[ruddell]

I had the same thing happening with my registry (see screenshot). It caused an issue where requests would fail because they are trying to request from a non-existing microservice. Following @pascalgrimaud's tip fixed my issue. Should this be included in the registry by default (or at least document)?

[jdubois]

OK I'm re-opening this, let's try this out, and fix it if needed.

[xetys]

I am also trying self-preservation mode to false, but still got zombie applications in eureka...maybe I am still missing something?

[xetys]

i tried to play with different attributes like enablePreservationMode, renewalThresholdUpdateIntervalMs, and evictionIntervalTimerInMs...

after about 2 minutes after service kill my registry is recognizing the dead service....for frequent updates this is really uggly...

[jdubois]

Closing this as I merged @pascalgrimaud's PR on the registry (note to self: I need to merge this in the master branch and do a new release of the Registry).