github/client-go
1
0
Fork 0
mirror of https://github.com/kubernetes/client-go.git synced 2025-09-04 16:40:21 +00:00
Code Issues Projects Releases Wiki Activity

Fix Panic Condition

Browse Source 
Currenlty an event recorder can send an event to a
broadcaster that is already stopped, resulting
in a panic.  This ensures the broadcaster holds
a lock while it is shutting down and then forces
any senders to drop queued events following
broadcaster shutdown.

It also updates the Action, ActionOrDrop,  Watch,
and WatchWithPrefix functions to return an error
in the case where data is sent on the closed bradcaster
channel rather than panicing.

Lastly it updates unit tests to ensure the fix works correctly

fixes: https://github.com/kubernetes/kubernetes/issues/108518

Signed-off-by: Andrew Stoycos <astoycos@redhat.com>

Kubernetes-commit: 6aa779f4ed3d3acdad2f2bf17fb27e11e23aabe4
This commit is contained in:
Andrew Stoycos
2022-02-11 14:50:19 -05:00
committed by Kubernetes Publisher
parent b4b8e5e1e5
commit f19a514ff5
3 changed files with 21 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tools/cache/testing/fake_controller_source.go vendored
View File

@@ -275,11 +275,11 @@ func (f *FakeControllerSource) Watch(options metav1.ListOptions) (watch.Interfac
// clients).
changes = append(changes, watch.Event{Type: c.Type, Object: c.Object.DeepCopyObject()})
}
return f.Broadcaster.WatchWithPrefix(changes), nil
return f.Broadcaster.WatchWithPrefix(changes)
} else if rc > f.lastRV {
return nil, errors.New("resource version in the future not supported by this fake")
}
return f.Broadcaster.Watch(), nil
return f.Broadcaster.Watch()
}
// Shutdown closes the underlying broadcaster, waiting for events to be