kubectx
kubectx copied to clipboard
ahmetb
Proposal: Support copying or aliasing contexts
Example:
An engineering team using multiple GKE clusters may reasonably have these desires:
- Common tooling used across the team which needs to rely on predictable cluster naming conventions, eg:
gke_project_zone_cluster-01 - Ability to reference these clusters by friendlier shorternames such as
cluster-01,cluster-02, and so on.
Currently kubectx supports renaming clusters which would address #2. However, renaming clusters would break use case #1.
Providing an easy mechanism for copying contexts would allow for a straightforward way for the team to address both desires. By not renaming and keeping the full GKE name the team can distribute common tooling that relies on the full context name. And copying to a friendlier name makes it easier for the team to operate against multiple clusters, eg: kubectl get pods --context cluster-01 or kubectx cluster-01
Implementing "copy" functionality would satisfy both desires.
As I said in #101 I don't think it's a common case to both rely on GKE context names on dev machines AND provide user-friendly names for them.
For a GKE advocate, I would highly discourage anyone from relying on gke context name pattern (i.e. gke_PROJECT_ZONE_CLUSTERNAME as you said) in any programming environment or scripting. Please use multiple kubeconfig files like the following for scripting/tooling:
KUBECONFIG=cluster-01.yaml gcloud container clusters get-credentials cluster-01
However, this issue may have some merit. So I'll keep it open for a while to see if there's enough demand for it and what are their use cases.
We would probably support this with a syntax similar to current renaming syntax (A=B). So maybe A==B or something like that may work.
As you noted in your PR #101, this complicates the context deletion since they would be sharing user/cluster entries.
I can confirm @joemiller's example, we are running into similar issues where people want to use peronalized aliases for clusters. People from different teams work with different sets of cluster, some just with "their" clusters ("prod", "dev"), SRE do support multiple teams so they need "team1.prod", "team1.dev", "team2.prod" ... etc.
We have implemented some developer workstation configuration management that does rollout team specific sets of kubeconfigs and the script itself is idempotent. If developers are suppossed to edit kubeconfigs for having cluster nicknames it would break any kind of kubeconfig configuration management. Having an alias for cluster names would still allow tools to do regenerate kubeconfigs while keeping the cluster nickname.
I also am in need of this feature.
This was actually implemented in the past, judging from https://github.com/ahmetb/kubectx/commit/9ed6690978df71de8961f2b35bd86a6c01db14ec#diff-10bafbbcaaa2bf26d2f237f58c279417
What was the rationale behind the decision to replace aliasing with renaming, instead of keeping both functions?
Every time you alias, you end up with more contexts. And context name is meant to be used for human friendlines. This is also the reason why kubectl has a rename-context method.
Can you explain your use case? Why do you need to clone your contexts?
Every time you alias, you end up with more contexts.
I am aware of that :)
context name is meant to be used for human friendlines
Yet prominent cloud providers generate kubeconfigs via their CLI sdk with names that are not exactly user friendly. GKE has been cited above; DOKS is another one, for example.
Why do you need to clone your contexts?
because original ones are generated programmatically by third party tools, and I want to keep those untouched because I may need to re-run these tools, e.g. for credentials rotation or other configuration updates.
I think as you said, I'd recommend doing what was deleted in https://github.com/ahmetb/kubectx/commit/9ed6690978df71de8961f2b35bd86a6c01db14ec#diff-10bafbbcaaa2bf26d2f237f58c279417 as a workaround. I'm hesitant about increasing the interaction surface for a feature that only a few people would use.
The algorithm was:
- for $CONTEXT, read its $USER and $CLUSTER
- $USER:
kubectl config view -o=jsonpath="{.contexts[?(@.name==\"${CONTEXT}\")].context.user}" - $CLUSTER=
kubectl config view -o=jsonpath="{.contexts[?(@.name==\"${CONTEXT}\")].context.cluster}"
- $USER:
- create a $NEW context with
kubectl config set-context $NEW --cluster="$CLUSTER" ---user="$USER"
Please do this as a bash function in the meanwhile as a workaround.
If more people report this I might be open to adding, so keeping the issue open. 👍
+1 for aliases. Oracle's OKE also programatically generates kubeconfigs with randomised contexts, and I'd rather not touch the default configuration for reasons similar to those mentioned by @stefanotorresi. Renaming and aliasing should co-exist, since they cater to different use cases.
+1 for aliases and the use case @cruschke mentioned where different teams have different aliases for the same clusters, and may have helper scripts that depend on those aliases.
Yes, they should use a config file to store the cluster names, but they don't and my group doesn't have the ability to force them to.
+1 we ran into the same issue with internal tooling expecting real cluster names instead of aliases. I'll do the work around and share with others, but I can't imagine a downside for making this the default behavior for aliasing contexts..
strike that. I was able to get this working by recreating the original context. I now have the aliased context and the original context listed, so I can reference the original name or the alias.
Came here to file a similar request and saw this so just adding my vote in here.
Use case: we have standardized, long-form context names that our (internal) tooling generates and expects (e.g. cluster.prod.us.domain.com).
I just want to be able to reference it when doing ad hoc CLI work by going: kubectx us
Usually, when I create a new k8s cluster, it provides a new k8s config for this cluster, I have to manually merge the new context to ~/.kube/config, then kubectx switch context work normally.
If improve merge or copy new context into the current k8s config, will be very helpful.
# Assume default k8s config contains 5 contexts
kubectx | wc -l
5
# There is a new k8s config file contains 1 context
grep "\- context" ./config | wc -l
1
# Merge the new context into default config file
kubectx -m ./config
# The default k8s config will contain 6 contexts
kubectx | wc -l
6
@zhangsean kubectl already lets you merge kubeconfigs. You could easily search and find it. https://ahmet.im/blog/mastering-kubeconfig/
Furthermore this issue is clearly about duplicating/aliasing contexts which is NOT related to what you want. Please do not conflate these two things. Your request is a separate issue, and won’t be implement since it’s more suitable for kubectl itself or kubectl plugins (which I believe already exist).
In case anyone is still interested in this functionality, the tool kubeswitch might solve the alias problem for you by transparently defining the alias without changing the underlying kubeconfig file. Full disclosure: I am the author of it.
Come across here when looking for a similar feature to manage my contexts. Thanks @danielfoehrKn for your sharing and I like the idea of creating alias separately other than modifying kubeconfig directly. It also allows the one-to-many mapping between alias and context when needed. Actually I was using a similar approach when I was creating a homegrown tool for this. The only thing makes it a bit tricky is that to maintain the mapping, we need to be careful when the context is deleted.
OTOH, I'm still thinking to have this feature inside kubectx would be a good idea as it doesn't have to be thousands of contexts which can have this issue. For example, when I was working w/ OpenShift, any time when I use oc login to log into a cluster, it will auto-create a context for me where the name is not user friendly. If I rename it to something else, next time when I re-login for some reason e.g. token expired, oc login will generate a new context that's duplicated. This is annoying.
@ahmetb If that makes sense, I'd be glad to spend some time to dig into it and submit a PR. Or, if that's not appropriate to have such stuff in kubectx, would it make sense to have a separate kubectl plugin such as ctxalias on top of ctx?
Somewhere in your $PATH kubectx:
#!/bin/bash
#
# my wrapper script for kubectl ctx
# No argument, just do the normal thing
case $1 in
'') kubectl ctx
exit $?;;
esac
CONTEXTS="$(kubectl ctx)"
SELECT="$(echo "$CONTEXTS" | grep -i "$1")"
COUNT=$(echo "$SELECT" | wc -l)
case $COUNT in
0) echo "No context found for '$1'"
exit 1;;
1) kubectl ctx "$SELECT"
exit 0;;
*) echo "Multiple contexts found for '$1':"
echo "$SELECT"
exit 1;;
esac
