It's usually possible in a murder configuration to do a zero downtime upgrade by xfer-ing users to a backend running the new version. The admin is then in control of how much index upgrade load to inflict on the new machine.
:wes On 10 May 2011, at 11:41, Henrique de Moraes Holschuh wrote: > On Tue, 10 May 2011, Per Steinar Iversen wrote: >> Running a reconstruct on user.% before allowing in new mail, after the >> upgrade, seems to be a better and more orderly option than allowing >> sendmail to do the job? > > That's certainly how we always do it here. We pause the new email feed at > the MTAs (so that it will just queue), firewall off any user access, run a > full backup, do the upgrade, trigger a full reconstruct and quota rebuild, > and test everything before we restore service. > > Better the planned downtime, than surprise breakages. ---- Cyrus Home Page: http://www.cyrusimap.org/ List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
