watch: prevent done callback from being called twice on connection loss #2389
Conversation
Avoids multiple invocations of the done callback when a watch connection is closed unexpectedly (e.g., a premature socket close). The fix sets doneCalled = true before calling controller.abort(), to ensure that any AbortError triggered by aborting does not result in another call to the done callback. A regression test is included to simulate an abrupt connection termination. It uses a real HTTP server that flushes headers and then destroys the connection immediately. This reliably triggers the combination of AbortError and Premature close needed to reproduce the issue. Refs: kubernetes-client#2387
k8s-ci-robot
added
the
cncf-cla: yes
|
Welcome @bverhoeven! |
k8s-ci-robot
added
the
size/M
|
The changes LGTM. I'll give the other maintainers a chance to weigh in on the test since it's different from what we normally do (binding a port instead of using nock). Personally, I'm good with it since nock has been giving us issues lately particularly around these types of error edge cases. |
|
Thanks! Yeah, I spent quite a bit of time trying to get this working with nock, but couldn’t get it to reliably simulate an abrupt disconnect. As mentioned in this comment, this setup seemed like the most dependable way to reproduce the issue. Happy to adjust if maintainers prefer a different approach, but I figured it was worth using something that reliably triggers the edge case. |
|
This looks good to me thanks for adding the test. /lgtm |
k8s-ci-robot
added
the
lgtm
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: brendandburns, bverhoeven The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
k8s-ci-robot
added
the
approved
Avoids multiple invocations of the done callback when a watch connection is closed unexpectedly (e.g., a premature socket close). The fix sets
doneCalled = truebefore callingcontroller.abort(), to ensure that anyAbortErrortriggered by aborting does not result in another call to the done callback.A regression test is included to simulate an abrupt connection termination. It uses a real HTTP server that flushes headers and then destroys the connection immediately. This reliably triggers the combination of
AbortErrorand "Premature close" needed to reproduce the issue.Refs: #2387