[zeromq-dev] Requirements for reliability
john.apps at hp.com
Wed Apr 7 16:17:00 CEST 2010
Interesting that 29West do this. HP has had a product called the Reliable Transaction Router (RTR) for many, many years now and this has been a feature for at least 15 of them.
In addition, RTR can use the data gathered to perform a reply of transactions to a hot-standby server, should that crash or simply be taken down for maintenance.
29West borrowed the idea, no doubt about it! Nothing wrong with that, just a shame that so few people know about RTR:(
-- John.Apps at hp.com | +491718691813 | http://twitter.com/johnapps --
From: zeromq-dev-bounces at lists.zeromq.org [mailto:zeromq-dev-bounces at lists.zeromq.org] On Behalf Of Steven McCoy
Sent: Wednesday, April 07, 2010 11:28
To: Martin Sustrik
Cc: zeromq-dev at lists.zeromq.org
Subject: Re: [zeromq-dev] Requirements for reliability
On 7 April 2010 17:09, Martin Sustrik <sustrik at imatix.com<mailto:sustrik at imatix.com>> wrote:
Pieter Hintjens wrote:
> Reliable fanout needs subscriber info in the reliability layer, which
> we can't do now but I'm thinking that some kinds of fanout can be done
> in the application layer anyhow (since they involve more complex
> matching algorithms and message formats).
My understanding is that reliable fanout directly implies that a single
slow consumer can bring whole data distribution network to a standstill.
No sane person would want that kind of thing so we'll probably have to
do with unreliable fanout.
I like the idea 29West took, have a node on the network listening to all the communication and keeping a log. Slow consumers are told to shutdown NAKs and can slowly recover from the network log, when a consumer has recovered and can keep up with the live source it can start sending NAKs again.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the zeromq-dev