IssueTracker

Backend_connection_closed_before_data_sent_to_client errors show on logs, This correlates with a VM starting up as errors are showing after a new instance starts responding back with a 200, after a while the app engine load balancer* intermittently sends 502 erros with the message mentioned before.

Health checks are returning 200 messages while the error is ongoing. it was possible to get the same error message once using a "hello world" with a 1 per sec request using the command

gcloud app deploy --version=20240217t004624

To emulate automatic rolling updates.

Interestingly, while these errors are happening, there are some random calls that receive a 200 response from the affected instance, A screenshot[2] shows an example of how calls in between are sent to the client's code and processed accordingly

The errors are registered as coming from the load balancer

This issue seems to happen when / is as the path for the load balancers to check, when using a default configuration instead the issue stops happening.