Hi Volkan, On Mon, 16 Sept 2024 at 20:19, Volkan Yazıcı <vol...@yazi.ci.invalid> wrote: > Gary has a failure on L361, that is, it retries every 100ms to succeed with > `logger.info()` for at most 2mins. I doubt if more waiting will solve the > problem. I tried to improve that test several times (see its history), but > Windows just behaves weird with sockets. I'd appreciate it if Windows users > can share some tips on what else we can do.
IIRC Windows has a very rude firewall: if the connection to a port is allowed, but the port is closed, the firewall drops all packages instead of reporting an appropriate error. Maybe we should add more logging to the reconnector? For example it would be probably appropriate to log all connection errors at the `WARN` level and all connection attempts at the `INFO` level. > Piotr, you say "a proven history of flakiness", but the link you shared > states, out of 1.26k times, in the last 90 days, it has failed only 2 > times, and was flaky only 16 times. What is our definition of flakiness > here? For me a test is either flaky or not. The SocketAppenderReconnectTest is a very good test (especially when compared with the `RollingFile` appender tests), but it is still missing some synchronisation. BTW I have also tried to improve it last year, but some flakiness remains. Piotr
