This would have solved our problems in a recent demo. We would publish data and it would work fine until we paused for a little while. I thought about putting heartbeats in, but we don't really need to guarantee data (using PUB) or the complexity (adding the heartbeat handling to all components).
Joshua On Jan 23, 2012, at 2:02 PM, Ivan Pechorin wrote: > 2012/1/23 Martin Sustrik <[email protected]>: >> On 22/01/12 13:02, Sergey Matveychuk wrote: >> >>> I think the patch will resolve my problem. >> >> Are you sure about that? >> >> First, keepalives are meant for the situation where the network >> connectivity between peers is broken (cable pulled out from the socket, >> switch turned off etc.) Is that your use case? > > Some VPN boxes are configured to close idle sessions after some > timeout. Some firewalls are configured to drop idle connections > ("idle" connection here means no traffic passes through it). For > instance, if I remember correctly, in popular Cisco ASA firewalls the > idle timeout is set to 30 minutes by default. > > Enabling TCP keepalives with reasonable TCP_KEEPIDLE (like 5 or 15 > min) and TCP_KEEPINTVL usually solves this issue. > > >> Moreover, TCP keep-alive interval is set to 2 hours or more (see RFC >> 1122). Is that sufficient for you? > > TCP_KEEPIDLE and TCP_KEEPINTVL should be set to something smaller than > the default idle timeout in firewalls. > _______________________________________________ > zeromq-dev mailing list > [email protected] > http://lists.zeromq.org/mailman/listinfo/zeromq-dev _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
