How can you specify an exec command with arguments in k8s deployment yaml?

Question

I have a cli that does a few different operations on pod startup and can be run from inside the pod like kubectl -n default -exec my-deployed-pod-xyz -- /path/my_cli -command health and runs fine like that, it logs success messages and actually does what I want it to.

In that commmand it does os.Exit(0) if it succeeds and os.Exit(1) if an error occurs, as per usual (it's written in go but that's just a detail).

Unfortunately when I specify the command (the part to the right of -- in the example above) as an exec in my kubernetes deployment yaml, it appears to always fail. To remedy this I made a cmd_wrapper.sh that just has:

#! /bin/bash

if /path/my_cli -command health ; then
    exit 0
else
    exit 1
fi

(maybe I don't need the if else but I thought I'd make sure).

Anyway when I use that as the exec command in k8s it works fine, which made me think the problem is the multiple words in the original command. I tried:

        readinessProbe:
          exec:
            command:
            - /path/my_cli -command health

Also:

command:
 - /path/my_cli 
 - -command 
 - health

And:

command: ["/path/my_cli -command health"]

Also:

command: ["/path/my_cli", "-command", "health"]

Unfortunately none of these seemed to work. I can still use the bash wrapper script but would be nice to know what's going on here...

Edit: thanks to larsks I found the error log in the journal of the worker node, it looks like this:

Dec 18 14:32:33 mcrumzy-worker-1.nasa.gov kubelet[1068]:         rpc error: code = Unknown desc = command error: time="2023-12-18T14:32:33-08:00" level=error msg="exec failed: container_linux.go:380: starting container process caused: exec: \"/app/my_cli -command health\": stat /app/my_cli -command health_check: no such file or directory"

So that doesn't help too much, just makes it look like it can't parse the command somehow. Escaping maybe?

Answer 1

This...

        readinessProbe:
          exec:
            command:
            - /path/my_cli -command health

...is not a valid configuration. The argument to command (just like the argument to command in a container spec) must be a list of arguments. The error you're getting in your logs...

exec failed: container_linux.go:380: starting container process caused: exec:
\"/app/my_cli -command health\": stat /app/my_cli -command health_check: no
such file or directory"

...occurs because the system is looking for a file in $PATH named literally /app/my_cli -command health. There is no file with that name!

One of your attempts looks correct:

        readinessProbe:
          exec:
            command:
              - /path/my_cli
              - -command
              - health

(You can see that matches the syntax shown in the documentation.)

Update your Deployment manifest to use the above syntax, and then look and see if there's a new error message.