Cannot access env vars in pod cmd #1309
Comments
|
This is a good issue. I don't see an easier way to make the join you;re describing without bouncing through the shell or doing text subst ourselves (which we really don't want to do). Can anyone see any better way to do this? |
|
I figured out a legitimate workaround. { "command": [ "sh", "-c", "redis-server /etc/redis/redis.conf --slaveof $SERVICE_HOST $REDIS_MASTER_SERVICE_PORT"] } |
|
This would be good in a FAQ somewhere. |
Closed
Gurpartap
added a commit
to Gurpartap/kubernetes
that referenced
this issue
Sep 18, 2014
Port of current guestbook's README.md Fix guestbook pod and service names Add go boilerplate Use role label for redis pods Give service.containerPort a name based value This gives better env variable keys like REDIS_MASTER_SERVICE_REDIS_SERVER_ADDR, etc. Avoid unnecessarily long names for service (esp in env vars) Adding guestbook deploy/destroy scripts for k8s These are probably only useful for quick testing. Maybe remove them before merging the pull request. Part of avoiding long names for services Update Dockerfiles to git clone from Google's repo Use correct service names while deleting them Fix the script usage path. K8s is not go gettable. Use git clone instead. Using my fork in the Dockerfile to release and update to the docker hub image. Doesn't delete all pods if you remove controller too soon Run the command in a shell to substitute env vars. Workaround for kubernetes#1309 GoogleCloudPlatform in lieu of my fork in Dckrfile Some directory structure changes for guestbook src README that explains the build process for source Strip down the html and css to absolute essentials Reformat JS according to Google's guidelines Also added code to set random colors for elements. Handle repetitive error checks using a common func Also uses @roberthbailey’s stripped down code for reading env vars infoString isn't really a string. Use info instead Remove deploy.sh/destroy.sh scripts Bind submit instead of keypress to capture submit Add links to /env and /info in the footer Reformat the JS Incorporating suggestions by @filbranden License boilerplate and some fixes to release.sh Update README.md Update README.md Add building on boot2docker info to README Accept docker bin path as a param for building Use kubernetes user to host the image on registry Don't get included in k8s's recursive build deps kubernetes#1299 (comment) 17638061
pdecat
added a commit
to pdecat/cloudsql-proxy
that referenced
this issue
Feb 28, 2017
…les cannot be evaluated directly in command The region part of the fully qualified instance parameter was missing: `project:region:name`. Also, I removed the $ sign to avoid letting us think we can evaluate environment variable specified in command directly. A workaround would be to wrap the command into `sh -c` but this docker image does not include any shell (`FROM scratch`). cf. kubernetes/kubernetes#1309 (comment)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I found this while working on using the same standard redis image for both the master and the salve. While the master pod is content with the image's defaults, the slave needs the master's address and port.
One way of doing this is by overriding the image's CMD to add
--slaveof $ADDR $PORTin the pod template. Assuming$ADDRand$PORTare the env variables defined by kubernetes.This, however, will only work if the command is passed to docker as a string to run it under
sh -c(which is the only easy way — right? — for the executable to have the substituted values for the env vars supplied). Using the array based command executes it without shell, where these env vars are not substituted and the slave is passed "$ADDR" and "$PORT" directly as strings.From the docker docs:
Pull request #1299 depends on this.
The text was updated successfully, but these errors were encountered: