Stack logging in Docker prevents container recreation

[thbkrkr]

When deploying Elasticsearch in Docker with ES_LOG_STYLE=file to use stack logging (#65778), the container may end with the CrashLoopBackOff status if the container is restarted recreated.

Step to reproduction with ECK on k8s:

• Deploy Elasticsearch using this manifest:

apiVersion: elasticsearch.k8s.elastic.co/v1
kind: Elasticsearch
metadata:
  name: ananas
spec:
  version: 7.12.0
  nodeSets:
  - name: default
    count: 1
    config:
      node.store.allow_mmap: false
    podTemplate:
      spec:
        containers:
        - name: elasticsearch
          env:
          - name: ES_LOG_STYLE
            value: file

• Wait for ES to be deployed:

> k get es,pods
NAME                                                HEALTH   NODES   VERSION   PHASE   AGE
elasticsearch.elasticsearch.k8s.elastic.co/ananas   green    1       7.12.0    Ready   39s

NAME                      READY   STATUS    RESTARTS   AGE
pod/ananas-es-default-0   1/1     Running   0          42s

• Force a container restart:

> k exec ananas-es-default-0 -- kill -15 1

The container status then oscillates between Error and CrashLoopBackOff

> > k get es,pods
NAME                                                HEALTH    NODES   VERSION   PHASE             AGE
elasticsearch.elasticsearch.k8s.elastic.co/ananas   unknown           7.12.0    ApplyingChanges   4m49s

NAME                      READY   STATUS             RESTARTS   AGE
pod/ananas-es-default-0   0/1     CrashLoopBackOff   4          4m46s

What's going on?

The ES log reveals that docker-entrypoint.sh fails because log4j2.file.properties has already been moved.

> k logs ananas-es-default-0               
mv: cannot stat '/usr/share/elasticsearch/config/log4j2.file.properties': No such file or directory

Here is the culprit:

elasticsearch/distribution/docker/src/docker/bin/docker-entrypoint.sh

Line 67 in cbb2ca2

mv /usr/share/elasticsearch/config/log4j2.file.properties /usr/share/elasticsearch/config/log4j2.properties

Could we just replace the mv by a cp -f?

[elasticmachine]

Pinging @elastic/es-delivery (Team:Delivery)