Fix: Kubes Cluster in Container Crashing Container (IS#923)

[Michaelgathara]

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update

Motivation and Context

This fixed issue #923

I fixed a race condition in ConnectHandler.swift where an asynchronous network connection could complete after the handler had already been removed from the pipeline.

This:

• Added a removed flag to track if the handler is still active.
• Updated the connection callback to check this flag; if the handler was removed while connecting, we now immediately close the new connection instead of trying to use it.

This prevents the EXC_BREAKPOINT crash in container-runtime-linux that occurred when kinc (Kubernetes in Container) created rapid connections. As we saw in the issue attached

Testing

• Tested locally
• Added/updated tests
• Added/updated docs

[jglogan]

@Michaelgathara Thanks! Could you set up commit signing and update your PR with signed commits?

See: https://github.com/apple/containerization/blob/main/CONTRIBUTING.md#pull-requests

[Michaelgathara]

@Michaelgathara Thanks! Could you set up commit signing and update your PR with signed commits?

See: https://github.com/apple/containerization/blob/main/CONTRIBUTING.md#pull-requests

Hi, sorry for missing that. Should be fixed now!

[jglogan]

Can you explain the race a little more clearly? The sequence of events I think you're trying to guard against is:

1. channelActive event, ConnectHandler will be in the handler chain, this method gets called, and this block runs when the backend connection completes
2. this handler gets removed from the chain via removeHandler()
3. the backend connection setup completes successfully, and we call glue() to replace ConnectHandler with GlueHandler.

How does step 2 come about? The only way that I see removeHandler() getting called and self.removed being set to true is from glue(), which only happens if execution has proceeded past your check.