-
Notifications
You must be signed in to change notification settings - Fork 38.7k
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
Support Restart policy in the kubelet (pre-design) #544
Comments
Seems like #137 is a dep of this, agree @smarterclayton ? |
@pmorie makes sense to me, yeah |
Part of this will be to put in place the framework for #137 - killed in the triple represents some of that data. Agree this should set the foundation for that. |
I'll probably also introduce PodInstanceID at the same time as ContainerAttemptID (which is the DockerID) from https://github.com/GoogleCloudPlatform/kubernetes/blob/master/docs/identifiers.md |
Not sure I understand that statement; in #127 there's an open discussion re: per-pod/per-container policy. Which direction are you headed? |
I'll update to say per container. |
I know Dawn was starting to look at this too - might be worth making sure I do think restart should be a per-container thing. I also want to think a bit about HA. Maybe it is not important here but We solved this by moving restarts out of the agent proper and into a third In the first containervm we solved it by wrapping each container run in a Like I said - not sure it matters here, but worth thinking about. What is
|
As I was thinking about this problem, I think I'd argue for a PodControl abstraction that the sync loop invokes to manage the details of the actual pod - that allows a separate subsystem to manage restart policy as well down the road. Dawn, let me know what you'll cover and whether there are pieces you want to split up. |
Clayton, I started a PR simply adding a restart policy for the container to container manifest with a default that is equivalent to the current API behavior (restart always). Initially I planed to put more effort on restart policy, but I am fine that you take over the whole issue or handle over the entire one to me. |
Up to you - if you've got other items you'd like to take I'm happy to cover this one. |
If you don't mind, I would like to take over this working item. I am still new to kubernete and golang, want to take this over to familiar with the entire workflow and codebase. Thanks! |
Two more important pieces:
This branch contains a very quick n' dirty run-once policy implementation and also handles container cleanup. It doesn't implement the more efficient/comprehensive design outlined above, but it at least demonstrates a lot of points of modification for the overall change. The hacky policy implementation is just incidental to the other POC work going on, so mostly just check out the modifications to |
@dchen1107 any updates on where this is? |
Dawn is out all this week. We need to try to reconcile with On Fri, Aug 8, 2014 at 8:03 AM, Clayton Coleman notifications@github.com
|
Any updates on this? |
There maybe two dawn chen in github,sorry I'm not the one in this thread
|
I have a pending one in branch, but not finish it. Will concentrate on this the rest of week. |
In the previous PR you added RestartPolicy to PodState and you added it as
or else a struct with optional fields (proto style) for possible extensions
We also might choose to rephrase "never" to "run once". We also should think about enums and strings: "ON_ERROR" vs "OnError". On Thu, Aug 21, 2014 at 12:24 PM, Dawn Chen notifications@github.com
|
@thockin 1) You are right. My current working PR did move RestartPolicy to ContainerManifest. 2) enum-in-a-struct model is for extension. RestartOnFailure could take extra parameters, for example, max failures. Also all restart should include an configurable parameter -- minimal interval between each restarts, etc. |
So then I think it should look something like LivenessProbe: type RestartPolicy { On Mon, Aug 25, 2014 at 11:34 AM, Dawn Chen notifications@github.com
|
I believe this is now fixed. Re-open if its not. |
It was fixed
|
* Fix interface problem & apiserver bind address
* Fix interface problem & apiserver bind address
…approvers Revert OWNERS changes
…ecker Separate linux/windows health checker files. Build health checker plugin for Windows
In order to begin to enable #127 (as soon as #356 is merged) I'm going to start down the following path:
Work not included here:
The text was updated successfully, but these errors were encountered: