On 7 October 2015 at 11:40, Graham Leggett <[email protected]> wrote: > On Fri, 2015-10-02 at 10:51 -0400, Ted Ross wrote: > >>> https://netprototalk.wordpress.com/2015/10/01/amqp-as-a-network-proto >>> col >>> >>> This is the first of a planned series of articles about AMQP, message >>> routing, and distributed-system use cases. > > We tried to use AMQP v1.0 as a network protocol, and ran into a host of > problems. > > Firewalls hate long lived TCP connections, and AMQP doesn’t seem to have a > mechanism to deal with this. After 15 minutes the TCP connection would be > silently dropped, the server assumed the client had died, and the client > assumed the server was just being quiet. Our system would only work for 15 > minutes at a time. > > We eventually changed our model to create a brand new connection to the AMQP > server, attempt to receive exactly one message, then gracefully shut down the > connection on each request, then rinse repeat. > > What we were doing that was special was dealing with small numbers of very > expensive messages (each message triggered an expensive processing job that > took many minutes to complete). AMQP didn’t handle that scenario well at all. > > Regards, > Graham > — >
You didn't elaborate much on the setup being used or how it was configured. One question would be whether this was or without AMQP heartbeating/idle-timeout handling enabled on the connections? Clients and servers can for example each use that mechanism to detect connection failures, and/or help keep them 'active' to begin with. Robbie --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
