http://bugzilla.spamassassin.org/show_bug.cgi?id=4246
------- Additional Comments From [EMAIL PROTECTED] 2005-04-14 09:35 ------- could you paste the interesting bits from the cpanel forum discussions? ;) anyway, my take on this is that, in a way, the exim SA support code is doing the right thing. I presume (Tony?) the design is that exim passes msgs to SA to scan *during the SMTP transaction*, so that rejects will be sent back correctly to the sending client, and not to some random third party (if the MAIL FROM is forged). what's happening is that under high load, SA is the bottleneck in this pipeline, and it's impossible for some of the Exim processes to get their msg scanned in a timely fashion. this is entirely plausible -- if there are 100 incoming connections, and Exim can handle that many conns, it should try to do so -- but there's no way it can possibly get SA to scan 100 msgs in parallel if the hardware can't keep up, because SA is a lot more heavyweight. Exim then has these options: 1. drop the messages silently. obviously a terrible idea 2. deliver the messages as nonspam, unscanned. this is the approach that "spamc" uses, and at least ensures the mail gets through. 3. return a 421 so the SMTP client can retry in a short while. 4. return a 550 and the SMTP client will bounce them as a delivery failure. also obviously a terrible idea. Given the choice between #2 and #3, #3 sounds like a reasonable choice. (personally, I would tend to go for #2 myself though.) if you're seeing this a lot, the correct fix -- if you have the horsepower -- is to increase the number of spamd processes available to scan messages. a lot of 421s is indicative of your mail load being higher than the number of SA scanners available to deal with it -- in other words, a sign that tuning is required. (As you noted, when the hosting co increased the maxprocs setting, it helped dramatically.) BTW, having said that, it does sound like *some* of the people reporting this bug -- but not all! -- are seeing something like bug 3828, the cPanel dual-CPU spamd hang bug. that would, of course, cause spamd to lock up, and exim would therefore time out attempting to contact it. Some of the threads online are referring to this (specifically the ones picking out dual-CPU), and some are not -- the problem is that the main symptom (Exim's BSMTP 421 errors) are identical without further digging. ------- You are receiving this mail because: ------- You are the assignee for the bug, or are watching the assignee.
