Synchronize gateway liveness (#15096)

[officialasishkumar]

Problem

RemoteClusterServiceWatcher updates gateway liveness from a background goroutine while mirror endpoint reconciliation reads the same state without synchronization. This leaves the multicluster service mirror vulnerable to race detector failures and inconsistent readiness updates under gateway flapping.

Solution

Protect gateway liveness with a dedicated RWMutex-backed accessor, route the liveness loop through a helper that uses those accessors, and update existing tests to use the synchronized setter. Add a race-focused regression test that exercises the liveness watcher concurrently with endpoint readiness updates.

Validation

• go test ./multicluster/service-mirror/...
• go test -race ./multicluster/service-mirror/...
• go test -race ./multicluster/service-mirror -run TestGatewayAliveSynchronization -count=1

Fixes #15096

Signed-off-by: Asish Kumar officialasishkumar@gmail.com

[raykroeker]

The lack of gatewayAlive synchronization was discussed as part of the initial code's introduction.

I agree with the original conversation, this is not practically a data-race even though static analysis suggests it is.

[adleong]

Since this field is just a single bool, did you consider using an atomic bool instead of adding a mutex?

[raykroeker]

Since this field is just a single bool, did you consider using an atomic bool instead of adding a mutex?

I false-started an implementation that used an atomic bool before I found the attached PR
@officialasishkumar do you mind adjusting your approach?

[officialasishkumar]

Replaced the mutex-protected bool with sync/atomic.Bool in de0bb5f and kept access through the existing helpers.