Chris Adams <[email protected]> wrote: > There are 4 VMs for outbound postfix, and 4 separate VMs for policyd > (each are load-balanced). There's a separate dedicated MySQL server > that everything talks to. > > I've been looking at my logs, and I see other instances of the quota > being reset to 1, not just under this spam flood, although under > moderate rates. For example, one user had the quota go from 1 to 38 > between 03:10:03 and 03:11:02, and then at 03:11:03 it went back to 1. > That user's limit is 50 messages per 30 minutes; the "reset to 1" > behavior is not consistent; a little later that user hit their 50 and > got rejects. > > I do see the entries in the session_tracking table. I don't know why > the Quotas module didn't see them.
I'm just wondering if there's a race condition where multiple servers are updating the quota, and one update overrides another ? I have a smaller 'cluster' of 3 machines which are all Xen PV guests, sharing a backend. I run one instance of PolicyD on the backend and originally had my Postfix instances accessing the single MySQL instance on the backend as well. it all worked perfectly under the loads I was able to impose as tests - then it all fell apart when I tried making it live. Once under real tests, I found that MySQL lookups were failing and so there were random mail rejections - so I had to revert back to the old server. I've put slave MySQL instances on each host, but as yet haven't found the out-of-hours time (or TBH, inclination to do OOH work) to try making it live again. _______________________________________________ Users mailing list [email protected] http://lists.policyd.org/mailman/listinfo/users_lists.policyd.org
