Memory leak on connection loss about lettuce HOT 12 CLOSED

I pushed 37c9423 to branch issue-26-memory-leak-fix. Can you give it a try? Deallocation happens using netty the same way resources are allocated.

from lettuce.

You can set the log level to DEBUG for the logger com.lambdaworks.redis.protocol (or even com.lambdaworks) to inspect the raw data. ASCII 111 is o which looks like some data was either in the buffer or the reply was not expected from redis. Guess, this will need further investigation.

from lettuce.

I will give it a shot. Sadly I don't have a good way to reproduce yet.
The fix looks good and makes sense. The allocation/release is now symmetrical.
Other invalid bytes I had were 'l', 'o', 'u', 'C'.
This might be related to the way we wrap the connection in SSL.

from lettuce.

hm. Looks like a feature request :)

from lettuce.

SSL wrapping? That might have actually helped avoid all of these issues. Lettuce would have known the connection is not really up and it's only the SSL tunnel that's up.

from lettuce.

Yep. Created issue #28

from lettuce.

Another thing that could have helped avoid this is a more thorough handshake that verified Lettuce is talking to Redis on the other side before beginning to dispatch commands. Maybe a ping/pong sequence first.

from lettuce.

No luck reproducing yet.

from lettuce.

Still not able to reproduce the exact issue, but I was able to confirm this fix is good. Instead of bad bytes, I got plain old connection reset by peer. When Redis goes down and Lettuce reconnects close() is never called. I've also added assert buffer == null at the top of of channelRegistered() and without this fix it fires multiple times on reconnection.

from lettuce.

Setting this one to resolved. The SSL issue is addressed with issue #28

from lettuce.

3.0.3.Final released

from lettuce.

Thanks! :)

from lettuce.