Stop updating boundPods objects. #5320
Conversation
|
I saw a integration test failure once. Then a bunch of passes. Trying to repro. |
erictune
force-pushed
the
no_bound_pods_2
branch
from
af98a71 to
26715c6
Compare
|
Doesn't kubelet still watch the boundpods? How will it handle this? |
|
No. Kubelet watches Pods. |
|
Okay, this will pass integration test reliably once #5337 merges. Wait for that. |
Does not clean up your existing boundPods records. Does not clean up all the dead code. Future PRs from me will do that.
erictune
force-pushed
the
no_bound_pods_2
branch
from
26715c6 to
e9b6c75
Compare
|
Okay, IMO this is ready. |
|
LGTM. This does break cloud providers not using salt, or have they all switched over? Do we need to wait until "breakage day" to merge? |
|
All cloud providers have had one of the following done to them:
Additionally, I've been asking recently-added cloud providers to pin their install instructions to a specific release, so that people don't accidentally walk into this. Also, service env vars have been broken for like a month or more for people who used etcd-based distros. Also, a PR which disabled the use of the flags that use etcd already went in. So people will get the idea from that that something is wrong. |
|
So, we don't need to wait for breakage day. |
|
Awesome! LGTM. A number of other changes are waiting on this, so merging. |
Stop updating boundPods objects.
|
Ref #2483 |
Does not clean up your existing boundPods records.
Does not clean up all the dead code. Future PRs from me
will do that.