When running upgrades from Serverless 1.29 to 1.30 with full Service Mesh configured (SM is not being upgraded during the test), the "Prober" upgrade test from Knative Serving fails.

This can be seen in this CI run that is run for this pull request

The PreUpgrade and PostUpgrade tests work correctly. It's just the test that sends requests continually that fails.

The queue-proxy logs show that during upgrade the new version of Deployment for the service is created and it immediately returns status code 200. But the previous version that is being shutdown will temporarily return status code 502. This happens for about 2-3 seconds. When the old version is removed all requests start to pass again. So this is only temporary.

Attaching a piece of logs that shows this sequence.