On Thu, Mar 27, 2014 at 06:23:39PM +0000, MV wrote: > >Viktor Dukhovni wrote: > > Furthermore, because "*" is cached, you really don't want to use > > "*" at all for dynamic transport resolution. > Thanks for your input RE the caching of the special pattern "*" results. > > > I answered your question upthread, use: > > sender_dependent_default_transport_maps > > For some reason you're still looking elsewhere... > > I can't use sender_dependent_default_transport_maps because I don't > want to use sender-based static "routes". I'm looking for a "random" > or round-robin-ish split of smtp that provides consistent "helo .. > hostname .. ip .. reverse-dns-lookup"
Of course you can. You're just not listening carefully. Your sender dependent maps would actually largely ignore the sender, and just provide a round-robin response. The important part is that this mechanism returns a "default_transport" which never overrides local or other more specific transport information. This mechanism (being a lookup on the sender, which is a message, not recipient property) can help avoid unnecessarily splitting of the envelope for multi-recipient mail. You should try to send all the recipients of a message to the same default transport, therefore you should have an modestly sized LRU cache within the round-robin process that returns a cached answer for a cached sender, but returns and caches a random answer for a new sender. If many messages come predominantly from a single sender and (since your traffic is likely single-recipient bulk mail, solicited or otherwise) multi-recipient messages are rare, you may not need the cache, since envelope splitting will not be useful, but the cache might defeat the load-balancing you want. > Anyways, I've got the round-robin working now, using transport_maps > and this script https://gist.github.com/mvsantos/9813415 > But I'm finding the solution very ugly. There must be a more elegant > way to do dynamic smtp deliveries... Use socketmap (or the obsolete tcp table) with: sender_dependent_default_transport_maps Don't bother logging the queries except briefly if you're unsure it is working. -- Viktor.
