Victor Duchovni <[EMAIL PROTECTED]> wrote: > Your queue manager is crashing. [...] > Queue manager re-starts.
Is that a bug? Should it handle this situation more gracefully? Now, a more interesting question: What is the incremental memory cost of each smtp process? According to pmap -d, the shared/writeable portion of each smtp is 712k. According to top & ps, resident size is 2276-2384k for each smtp, and top reports a shared size is 1788-1864k for each smtp, with the most common ones clustering around 2300k resident, 1800k shared. DATA is 448k for each smtpd, which is close to resident - shared. CODE for each smtpd is 316. I think this is included in pmap's 712k, but is shareable, so that would suggest pmap is saying each smtp costs 396k... So is the incremental cost of more smtp processes around 400k-500k, or 712k, or the full 2300k resident size? The latter doesn't make sense to me logically (because resident size looks like it includes a lot of shareable segments that are shared by all smtp processes), ... but trial and error suggests that might be right: On servers that currently have 750 smtp processes and about 1.6-1.8 gigabytes of free memory, all of them ran into the qmgr crashing problem when they had master.cf set to allow 1500 smtp processes. If the incremental cost of each smtp were a little over 2300k then 750 additional processes would use up about that much free memory. Ideas? -- Cos
