ram, > On Fri, 2007-09-07 at 08:36 -0400, Matt Kettler wrote: > > No, because all the DNS tests are run together as a batch. > > (this way the lookups run in parallel) > > Besides, if completewhois is the only "slow" RBL, the timeout for it is > > going to effectively be 4 seconds anyway. (ie: if 90% of the DNS lookups > > finish in the first second, SA will only wait 3 more for the remaining > > lookups before dropping them)
True. But if anyone cares to pursue the matter, the current code in trunk already drags along individual timeouts for each request. They all default to rbl_timeout, but in principle each one could have its own individual timeout value, all that is missing is a configuration front-end to be somehow able to specify these timeouts, perhaps by a domain name of a zone, or maybe by rule. > But I am seeing the DNS tests taking more than 10s on my servers > from the spamassassin -D logs > Even though I have in my prefs file > rbl_timeout 5 The 10 s from the time at which the _first_ dns query was launched, or 10 s from the time this particular query was launched? If the first, it is normal, not all queries are launched at the same time, and some of them (specially the URI tests) are launched quite late. See also the: http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5589 and the patch there: http://issues.apache.org/SpamAssassin/attachment.cgi?id=4081 It is applicable to 3.2.3, and individualizes query timeouts, benefiting from HiRes::Time perl module (recommended to be installed). Try it and see if you get a more predictable behaviour. Mark
