search

radanalyticsio / openshift-spark

72 stars 83 forks source link

Move nss-wrapper logic into dockerfile entry-point #17

Closed erikerlandson closed 7 years ago

erikerlandson commented 7 years ago

I'm wondering if it would be cleaner to publish openshift-spark containers that did all the "nss-wrapper" goo in a custom entrypoint. Then people using our containers could get it without having to worry about doing it themselves if they customized startup logic.

mattf commented 7 years ago

see https://github.com/radanalyticsio/openshift-spark/pull/18

mattf commented 7 years ago

'oc rsh' and 'oc exec' do not invoke the container's entrypoint.

recommendation from openshift team is to provide a command to manually run or try using the shell profile.

erikerlandson commented 7 years ago

That is disappointing. I wonder if it would be possible to create oshinko rsh and oshinko exec, that would wrap the nss logic around the rsh and exec arguments.

oshinko exec <args> ==> oc exec entrypoint.sh <args>

mattf commented 7 years ago

i'm still of the belief that the way to address this is closer to the source. maybe not in hadoop itself but at least in spark.

mattf commented 7 years ago

21 addresses this for the 'oc run' case. 'oc rsh' and 'oc exec' are still an issue.

we should simply always recommend 'oc run' instead of shelling into an existing container w/ rsh or exec.