Tim Bannister
Tim Bannister
Changelog suggestion ```diff -Automatically recover the hung service by restarting the kubelet, if the kubelet health check fails and systemd watchdogs are configured. +Added kubelet support for systemd watchdog integration....
BTW, can you disable this feature?
Even if we think you can disable it by editing a unit, we should make sure our tests cover this well enough (it would be bad to ship a kubelet...
> One option will be to enable this on all our e2e tests. Like here: Even better IMO: have at least one unit test that launches kubelet as a `Type=simple`...
I think if we haven't parsed config yet, we probably shouldn't start the watchdog timer anyway. But then, readers may be wondering, how do we handle failure during kubelet startup?...
Not a change to the live site, but I'd prefer not to approve this one myself. @katcosgrove @tengqm @salaxander PTAL
I agree with all this, but the `notin` operator is perhaps even more important in one respect: you can only emulate it client side by doing a full list (or...
This issue also reminds me of https://github.com/kubernetes/kubernetes/issues/1348 (Reverse lookup by labels). Some of the solutions we could develop feasibly solve both problems. For example, if we added an aggregated API...
Also see https://github.com/kubernetes/kubernetes/issues/1362
Looking forward to a KEP for this!