* Viktor Dukhovni <postfix-us...@dukhovni.org>: > On Thu, Nov 07, 2013 at 08:58:47PM -0600, Stan Hoeppner wrote: > > This would require too much complex code for what is a simple Postfix > > operation. Your example is "poor man's round robin". That's the best > > Postfix can do without serious code changes. But why add such code when > > others have already solved this problem? I.e. do it in your DNS server, > > which supports true round robin. > > DNS round-robin won't help. Postfix re-shuffles the RR's since > not all DNS servers do. The result *will* be random. If you don't > like random you need a load-balancer in the network path with a > round-robin policy for handing off the TCP connections to a real > server.
Is a load balancer really necessary? If I understand the documentation correctly, up to smtp_mx_address_limit addresses will be tried during a message delivery. If one or more of these addresses do not respond, exhibit timeouts or respond with temporary failure (codes e.g. due to overload), Postfix will simply try another address. We should advise the OP to carefully evaluate if introducing the additional complexity of a load balancer is really necessary. The Performance Tuning[1] and Bottleneck Analysis[2] documents both contain valuable guidelines that I'd consider relevant to this topic before I'd go ahead and add a load balancer to my setup. Stefan [1] http://www.postfix.org/TUNING_README.html [2] http://www.postfix.org/QSHAPE_README.html
signature.asc
Description: Digital signature
