Martin Sustrik wrote: > On 06/01/2011 07:33 PM, Pieter Hintjens wrote: > > >> Now to avoid memory exhaustion, the server sets a HWM. Each client >> output queue is independent, each is limited to the same HWM. >> >> We now get the situation where one slow client reaches its queue >> limit. At this stage, the server either blocks on send, or receives >> EAGAIN if it's doing non-blocking sends. >> >> Consequence: one slow client blocks all clients. >> >> What's the best way around this? >> > > If it behaves that way, it's a bug. The correct behaviour should be to > remove the blocked queue from load balancing. Ie. any further requests > are going to be sent to the other peers until the blocked peer reads the > some pending requests and thus creates a space in its queue. >
isnt this essentially related to the fact that xrep_t::has_out always returns true? > Martin > _______________________________________________ > 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
