> On Thu, 10 Jan 2002, Victor wrote: > > > Hi Rob. Thanks a lot for replying. CMU user base is probably huge. I wanted > > to ask you what is your procedure for rolling out updates? Do you have a > > procedure? I wanted to see what would be involved in upgrading cyrus in a > > production environment without significant downtime and was curious what > > procedures established sites have come up with. > > In addition to the systems we use for development of the source, we use > one machine which is basically set up as a beta server, which a few people > in our group use. That is, before anything moves to the production > machine, it gets tested by a small number of real users. We also > occasionally run load tests against this machine trying to simulate what > real activity looks like. > > Then, we schedule downtime, upgrade the binaries on the production server, > and reboot (if needed). This is, of course, provided that the data files > haven't changed format between the two versions ;)
This is exactly the issue I was concerned about :) I suppose there would be tools then to convert. However if the set of mail is huge (or distributed over servers), it can be a real hassle to convert everything. > Some of the tools we use for deploying arbitrary software collections are > desribed at: http://andrew2.andrew.cmu.edu (Namely, Depot and The > Environment Maintenance Tool) > > I'm not entirely sure this answers your question, but I hope it helps. > > -Rob Yes, it heped. If you don't mind, I also wanted to ask you about your setup. It is not recommended that Cyrus run over NFS. So I assume you have it running on one server (or many server using the special function to break message storage onto many servers). Do you just hold backups and if a server fails, restore? Or do you hot like a hot standby box? I am curious what steps can be taken to make cyrus be High Availability.
