Mateusz Gozdek

Interesting observation: - Rewriting tests to use events-based approach makes tests more fragile, as they are more implementation specific. Sleep based tests are more implementation agnostic. - Optimizing sleep based...

See also https://github.com/kinvolk/Flatcar/issues/273.

Note, that https://github.com/flatcar-linux/flatcar-linux-update-operator/pull/97 adds full test suite and does many improvements to the code, so it is likely that this code should be taken to the separate repo.

I can confirm that this is still happening: ``` I0927 12:48:56.022773 1 agent.go:237] Deleting 0 pods I0927 12:48:56.023058 1 agent.go:267] Node drained, rebooting panic: runtime error: invalid memory address or...

>update_engine - If we want to be able to cancel the reboot process at every stage, each stage must take update_engine status into account. Simply put: If we run before-reboot...

The fix for this could be: ```diff diff --git a/pkg/updateengine/client.go b/pkg/updateengine/client.go index 6306077e..bda237c6 100644 --- a/pkg/updateengine/client.go +++ b/pkg/updateengine/client.go @@ -15,6 +15,7 @@ package updateengine import ( + "context" "fmt" "os"...

>I'm not sure if it is desired to have 65534 (nobody) be able to reboot. Agreed. Perhaps there should be a dedicated user created on host OS, which is only...

Not only this, but once this happens it also won't stop gracefully after closing the control channel.

We should also consider compatibility with k8s versions before merging.

Just hit some issue with this code: 1. Draining failed because one workload couldn't satisfy the PDB. 2. Waiting for volume detachment never finished. Perhaps we should also have some...