Closing a Finagle service leaves live, non-daemon threads

[stuhood]

Starting a Finagle client using ClientBuilder (and the ServiceToClient Thrift generation in this case), using it, and then closing it down will cause the process to stay alive, probably because of non-daemon threads opened by Finagle. Attaching a stack from after calling close on the Service underlying the ServiceToClient object.

Thanks!

[stuhood]

https://gist.github.com/806609

[mariusae]

this should be fixed in 1.1.10 — could you verify? thanks, marius.

[stuhood]

I think the 1.1.10 release is causing other problems. With or without our custom Protocol shim, we're getting: https://gist.github.com/808398 ... on the first request.

[mariusae]

thanks for the stack trace. i was able to reproduce & fix.

80a0794

this is now included in the 1.1.11 release.

[stuhood]

The problem from 1.1.10 is fixed, but I'm still seeing the original issue with 1.1.11: I've updated this gist https://gist.github.com/806609 with the new stack trace I see after exit. In particular, there is one thread in some Netty code for "HashedWheelTimer" that looks suspicious.

[stuhood]

Sorry, don't know how this got closed: as the last comment indicates, it is still a problem.

[mariusae]

ok, cool. 1.1.11 squashed the NIO worker threads, but i still need to refcount the timer thread.

i'll fix this on monday.

[mariusae]

this should be resolved in 1.1.15.

let me know.

[stuhood]

Indeed it is! Thank you very much Marius.