Skip to content
This repository was archived by the owner on Sep 18, 2020. It is now read-only.

pkg/agent: avoid exiting on watch termination#89

Merged
euank merged 2 commits into
coreos:masterfrom
euank:watch-retry
Jun 22, 2017
Merged

pkg/agent: avoid exiting on watch termination#89
euank merged 2 commits into
coreos:masterfrom
euank:watch-retry

Conversation

@euank

@euank euank commented Jun 21, 2017

Copy link
Copy Markdown
Contributor

Works around #75. Properly, it would also distinguish between
"retryable" and "unretryable" errors, but client-go doesn't give us
great granularity.

In the future, adding further granularity would be a good idea.

This also tweaks whether delete failures are fatal based on a discussion @aaronlevy and I had a while ago.

cc @dghubble

Testing done: none yet 😦

euank added 2 commits June 21, 2017 16:11
Works around coreos#75. Properly, it would also distinguish between
"retryable" and "unretryable" errors, but client-go doesn't give us
great granularity.
In practice I haven't observed this, but a pod that can't be deleted is
no reason to not update.
@euank euank changed the title Watch retry pkg/agent: avoid exiting on watch termination Jun 21, 2017
@aaronlevy

Copy link
Copy Markdown

LGTM assuming testing :)

@dghubble

Copy link
Copy Markdown
Contributor

@euank did you push this anywhere already? We can manually test on a cluster, until #90 is addressed.

@dghubble

dghubble commented Jun 22, 2017

Copy link
Copy Markdown
Contributor

https://quay.io/repository/dghubble/container-linux-update-operator?tab=tags. Dirty to fix the problems in the build scripts #92.

@dghubble

Copy link
Copy Markdown
Contributor

I've tested this on a Kubernetes 1.6.4 cluster today (bare-metal, masked locksmithd) for both update-operator and update-agent. Tested fake D-bus signals and real update_engine updates from an older version of stable, no major regressions. I'll try to post something about this testing process soon, to formalize the process.

@euank euank merged commit 0b5e619 into coreos:master Jun 22, 2017
@euank euank deleted the watch-retry branch June 22, 2017 21:47
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants