TCP Keepalive or Stomp level ping about stomp

[gmallard]

KEEPALIVE might help. I would want to think about how easy / difficult that would be to implement. Thiago probably has some thoughts here.

STOMP 1.1 defines Heartbeats, which would help. But it requires both client and server be 1.1 capable.

The 1.1 spec will make it out of draft status 'real soon now' (I think).

Regards, G.

[morellon]

Try adding
used_socket.setsockopt(Socket::SOL_SOCKET, Socket::SO_KEEPALIVE, true)
to the code in the ope_socket method and see if it helps. If it does, I can't see a reason to make this the default behavior.
Otherwise, you could make your client subscribe one "ping" queue to which itself sends a message every now and then (actually the subscribe part would not be so necessary). I also recommend that you monitor your consumers and broker, so you can get a faster feedback on your connections and messages.
Let me know if the SO_KEEPALIVE helps.

Cheers, Thiago

[gmallard]

One other thought: please make sure that your 'firewall' and 'operating system' engineers agree on idle timeout settings and that those settings are logical. I see this difficulty at 'real' work a lot.

Regards, G.

[ripienaar]

We have all the timeouts and stuff in line yes, but Stomp connections can be idle for days nothing will really help a whole lot there :(

I'll try the socket option and let you know

[gmallard]

My thinking was along these lines.

I will guess that your Rehat/CentOS TCP stacks are configured for defaults:

net.ipv4.tcp_keepalive_time = 7200
net.ipv4.tcp_keepalive_probes = 9
net.ipv4.tcp_keepalive_intvl = 75

And further: if your firewalls are configured to time out in less than 75 minutes, then I believe you will still have problems, even with the added call to setsockopt.

[gmallard]

OK, that is 75 seconds. I am fretting too much.

[gmallard]

so .... any updates on experiments with the proposed help?

[ripienaar]

Been very busy last 2 weeks, its on my list of things to test

[ripienaar]

Sorry for the delay, I've had a chance to test this and it solves the problem well.

A configurable option that's on by default would be awesome.

thanks

[morellon]

I think we could make this as the default behavior. I don't know if we need to make it a configurable option. What do you think Guy?

[gmallard]

Let's try it. If it breaks something for someone else, we will hear about it.

My only concern is side effects in a Windows environment. I am not concerned about Unix flavors in general.

[gmallard]

OK, this is now on the master branch.

[morellon]

Nice, did you try it in the windows environment Guy?
Even if you didn't, let's publish it and wait for feedbacks from windows users.

[gmallard]

Thiago - It was tested on:

a) 2 different Linux systems, with several Ruby versions on each system
b) Vista64 Home Premium, Ruby 1.9.1p378 (only WIndows I have)
c) OSX Darwin 10.7.1. Ruby 1.8.7p174 (only Mac I have)

I am going to wrap this up and publish 1.1.8 later today or early tomorrow unless you have objections.

[morellon]

No objections at all. :)

[gmallard]

KEEPALIVE is now the default for 1.1.8.