Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Master doesn't put to SchedulingDisabled mode #639

Closed
shamil opened this issue Oct 12, 2016 · 9 comments
Closed

Master doesn't put to SchedulingDisabled mode #639

shamil opened this issue Oct 12, 2016 · 9 comments
Labels
Milestone

Comments

@shamil
Copy link
Contributor

shamil commented Oct 12, 2016

Hi,
Since upgrading to kops 1.4.1, the master is no longer put into Ready,SchedulingDisabled.
Initially starts as Ready,SchedulingDisabled, but after minute switches to back to Ready.

@krisnova
Copy link
Contributor

Just curious - but what problems is this causing?

@chrislovecnm
Copy link
Contributor

Yah I noticed this. Without it being in SchedulingDisabled mode pods are scheduled to the master. Typically you don't want your master acting as a minion.

@krisnova
Copy link
Contributor

krisnova commented Oct 13, 2016

Understood. Are replication steps fairly straight forward? Is it safe to say this can be easily replicated with a fresh copy of kops 1.4.1 and a simple k8s deployment?

Attempting to replicate now, but just wondering what you have noticed - and want to make sure I am replicating the same issue.
Cheers

@justinsb
Copy link
Member

The master is tainted, so normal pods should not schedule on it. Daemonsets don't honor taints though, but we also label it, so that you can target a daemonset only to run on the master. We don't have a label on the nodes yet, so you could target a daemonset to run only on the nodes, but we probably should.

This is the "new way" of doing things, but we're ahead of - well - everyone here. And perhaps we shouldn't have done that. Would be interested to hear what problems this causes - we should open issues in the kubernetes repo where we are finding fault with this model.

We do have two problems already:

NodePort on the master: kubernetes/kubernetes#33884

kubectl get nodes is awkward: kubernetes/kubernetes#33533

Neither is a deal-breaker AFAIK though

@shamil
Copy link
Contributor Author

shamil commented Oct 13, 2016

2 questions:

  1. Are taints supported in k8s 1.3.x? or we need to upgrade to 1.4
  2. The NodePort issue is deal-breaker no? Usually masters use smaller instances, which may affect networking as well.

@dkapanidis
Copy link

For those interested to activate again the SchedulingDisabled flag until this is fixed you can do on each master:

kubectl patch node MASTER_NAME -p "{\"spec\":{\"unschedulable\":true}}"

The issue we're facing is that the ELBs are pointing also to masters. This will disconnect the masters from the ELBs and redirect traffic only to nodes.

@chrislovecnm
Copy link
Contributor

What is the status on this issue? Can we close?

@chrislovecnm
Copy link
Contributor

Closing, please reopen if needs be.

@mausch
Copy link
Contributor

mausch commented Mar 5, 2018

I lost track of how k8s changed "taints"/"schedulable" flags, but I just provisioned a new cluster with kops 1.8.1 10 days ago and got an unwanted deployment on a master. Had to drain it manually.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

6 participants