Would this happen to be running under Xen HVM? When starting a new guest under Xen HVM, the time is apparently initialized from the Xen hypervisor itself, not from the good time kept in Dom0. Xen's time is not corrected by NTP. When the physical host has been running for a very long time, drift would be significant. I have observed this issue on Xen hosts I manage myself. I've never taken the time to discuss this upstream, or confirm the issue still exists with current Xen versions. The thing is, this has caused me problems as it is causing you, and you are a paying Amazon customer. Why don't you try to get them to contribute a fix?
Meanwhile, you can set up ntpdate to run and hard-step the clock right before NTP starts and certainly before any mail-related, database or any important record-keeping things start up. Thanks for these suggestions. These are certainly running under Xen, and this is likely the nature of the problem. I believe I have a workaround ready for the next time I get an outage on these servers. I'll have to consider whether or not it's worth punching Amazon over. -- ## List details at https://lists.exim.org/mailman/listinfo/exim-users ## Exim details at http://www.exim.org/ ## Please use the Wiki with this list - http://wiki.exim.org/