Success of running Dataiku in HA on EKS
• Mark Eschbach
Perhaps I spoke a little to early. First of the large problems I faced was properly mounting the persistent volumes.
At the root of the mount was the standard *nix directory lost+found
. This caused Dataiku DSS to refuse to perform any
further work. Verification took the place of using a job
to issue the specific commands to explore the mount.
Resolving this utilizes a subPath
mount option. Next up was resolving the group issues, which was done with the
securityContext
element set to the appropriate user.
spec:
template:
spec:
securityContext:
runAsUser: 1000
fsGroup: 1000
containers:
- name: dataiku
image: dataiku/dss:5.0.5
ports:
- containerPort: 10000
name: http
volumeMounts:
- mountPath: /home/dataiku/dss
name: data-store
readOnly: false
subPath: dss-data-dir
In order to get the Deployment
descriptor to operate correctly strategy
needs ot be set to Recreate
. When the Deployment
is updated this will result in the old Pod
being destroyed and
the volume being detached. Otherwise we’ll find ourselves in a livelock waiting for the volume to be detached from the
old Pod.