Upgrading ELK stack to 5.3 and use official images

[ygokirmak]

I couldn't manage to configure elasticsearch to run clustered on my prod environment. I am passing these env vars to elastic container, may be they are wrong..

        environment:
          - cluster.name=es-cluster
          - transport.host=0.0.0.0
          - discovery.zen.minimum_master_nodes=1

I tried new supported images docker.elastic.co/elasticsearch/elasticsearch:5.3.0 and clustering is working with a couple of environment variable in docker-compose

        environment:
          - cluster.name=es-cluster
          - "discovery.zen.ping.unicast.hosts=dh-elk-02"

Trying to understand the problem in images which we are using in jhipster. Should I focus to solve the problem in elasticsearch:5.1.2-alpine image (deprecated) or we should upgrade images to supported ones.

I can try to upgrade and get issue #49 also which is related..

[PierreBesson]

Wow very nice. If you can PR a working clustering setup, it would be awesome. Yes you can upgrade the version.

[ygokirmak]

Ok.. I am working on that

[ygokirmak]

@PierreBesson instead of installing curl on Kibana image we can use another short-lived image to install dashboards etc?

they did smilar here; https://github.com/elastic/stack-docker/blob/master/docker-compose.yml#L75

Putting installing dashboard and index logic into dockerfile has some drawback, warn me if I am mistaken. When we put load configuration logic in kibana dockerfile than we need to rebuild kibana image;

• when we remove/clean elasticsearch image .
• or when we restart elasticsearch image and it doesn't have any volume for persisting indexes

[PierreBesson]

@ygokirmak This is a really good idea ! The reason I didn't do it was that I was a docker noob at the time. This is definitely the way to go. This totally fits the docker philosophy of "one container for one thing" and sidekick containers.

[PierreBesson]

Love your link by the way, we should move to something like this.
[EDIT] However we need to make sure the setup can also work in Kubernetes, Rancher, etc...

[rifatdover]

+1