ARTEMIS-239 disable ttl for STOMP connections

[jscheid]

Leave dead connection detection up to the heart-beating logic.

[clebertsuconic]

there is some tests changes I have to do for this...

I will keep this open for now

[jscheid]

Hi @clebertsuconic, is there anything else I can do to move this forward?

[themerius]

Would be nice if this fix goes into the next release of Artemis! Or into the branch, so I can test it.

[jbertram]

@themerius, I'm not sure if @clebertsuconic is done with the test changes, but in any case you can add @jscheid as a git remote and cherry-pick his commit into your own branch for testing purposes.

[jscheid]

We've discussed this more on IRC and it turns out this is a bad idea.

According to the STOMP protocol, the client can ask for no heartbeats (and STOMP 1.0 doesn't support heartbeats at all) and so the server runs in danger of leaking connections when dealing with badly behaved clients when there is no hard upper limit on connection TTL.

We've agreed the correct fix is instead to improve the documentation to point out that TTL needs to be configured correctly on the server to suit the use case, and that clients can't expect long heart beat intervals (or no heart beating at all) to work. Also, it would be good if Artemis could add a mechanism to configure per-acceptor TTL but that is a separate issue.