Hardcoded call to /usr/local/bin/kubectl fails

[loeschzwerg]

I have kubectl installed at /usr/bin/kubectl via dnf

$ punq install -i punq.minikube.local
Do you really want to install punq to 'minikube' context? [Y/n]: y
Creating punq namespace ...
Created punq namespace. ✅
Creating punq RBAC ...
Created punq RBAC. ✅
Creating punq deployment ...
Created punq deployment. ✅
Determining cluster provider ...
Determined cluster provider: 'MINIKUBE'. ✅
Creating new punq-context secret ...
Created new punq-context secret. ✅
Creating punq service ...
Created punq service. ✅
Creating TRAEFIK punq ingress (punq.minikube.local) ...
Created TRAEFIK punq ingress (punq.minikube.local). ✅
Creating TRAEFIK middleware (punq.minikube.local) ...
Error: failed to execute command (/bin/bash -c /usr/local/bin/kubectl apply -f /tmp/traefik-middleware.yaml3688081572): exit status 127
/bin/bash: line 1: /usr/local/bin/kubectl: No such file or directory

A code review revealed there are several places where kubectl is called on ¬Windows using the full path /usr/local/bin/kubectl. This is fine and desirable, but only when the fully qualified command is resolved, stored and checked.

Using a hardcoded call to /usr/local/bin/kubectl will fail, when the caller doesn't have kubectl installed at /usr/local/bin. This can be the case if it is installed via a package manager, like dnf in my case, or when a custom location is chosen, e.g. when developing on kubectl.

There are four solutions and a local quick fix:

1. Every call to /usr/local/bin/kubectl should be resolved via path, instead of being hard-coded.
2. On startup, punq checks whether kubectl is available using PATH but only on CLI calls which use kubectl. kubectl is defined once, and referred afterwards.
3. (Like 2. but with a singleton pattern which raises 4.)
4. Instead of relying on an external CLI Tool like kubectl, punq should talk directly with the API server, e.g. by using https://github.com/kubernetes/client-go/. This is already a dependency.

NOTE: 1, 2, 3 will create a mess, when the behaviour ofkubectl is altered during execution, unlikely, but possible: e.g. kubectx or kubens changes.

Quick fix:

ln -s /usr/bin/kubectl /usr/local/bin/kubectl

At last, when one of the install commands fails, I suggest punq should do a clean up itself or reminds to do a clean install.

[beneiltis]

Very good points. Thx a lot for the detailed explaination :-) Fix is iscoming in a few hours. Stay tuned.

[beneiltis]

Ok from a few hours to over a month. what a shame. sorry for that 😞

For simplicity reasons we will go with solution no1.

solution number 4 would be my first choice but we still need to use kubectl because the describe functionallity is not easily available by using the kubernetes go-lang-package as a third party import.

[beneiltis]

I am gone close this now because it should be solved.