The key difference is that the kubelet must remember all plugin instances
because it could always happen that the new instance dies and leaves only the
old one running.
The endpoints of each instance must be different. Registering a plugin with the
same endpoint as some other instance is not supported and triggers an error,
which should get reported as "not registered" to the plugin. This should only
happen when the kubelet missed some unregistration event and re-registers the
same instance again. The recovery in this case is for the plugin to shut down,
remove its socket, which should get observed by kubelet, and then try again
after a restart.
Currently, when a Kubelet Plugin is being added in the DesiredStateOfWorld,
a timestamp is saved in the PluginInfo. This timestamp is then updated on
subsequent plugin reregistrations.
The Reconciler, when it detects different timestamps for a Plugin in its
DesiredStateOfWorld and ActualStateOfWorld, it will then trigger a Plugin unregister
and then a new Plugin registration.
Basically, the timestamp is being used to detect whether or not a Plugin needs to
be reregistered or not. However, this can be an issue on Windows, where the time
measurements are not as fine-grained. time.Now() calls within the same ~1-15ms
window will have the same timestamp. This can mean that Plugin Reregistration events
can be missed on Windows [1]. Because of this, some of the Plugin registration unit
tests fail on Windows.
This commit updates the behaviour, instead of relying on different timestamps,
the Reconciler will check the set PluginInfo UUID to detect a Plugin Reregistration.
With this change, the unit tests mentioned above will also pass on Windows.
[1] https://github.com/golang/go/issues/8687
v1.43.0 marked grpc.WithInsecure() deprecated so this commit moves to use
what is the recommended replacement:
grpc.WithTransportCredentials(insecure.NewCredentials())
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
