Murder question : restarting the mupdate master daemon, mailbox list incoherent on frontend

[Rob Siemborski]

On Fri, 26 Sep 2003, Etienne Goyer wrote:

> I was playing around in IMP and deleted a few folders.  Then, I tried
> recreating a folder I just deleted with the same name.  It complained
> that the folder already exist.  I check with cyradm on the backend and
> can see that they are really gone.  I then checked on the frontend and,
> to my surprise, all the folder I thought I had deleted where still
> shown in cyradm!  I restarted cyrus on the frontend, checked again with
> cyradm and now they are gone.  I can then recreate my folder correctly.

How long did you wait between restarting all of these services and then
performing mailbox operations?  If the frontend loses its connection to
the mupdate master, it'll be a minute or two before it reattempts the
connection.  After that, the resynch process takes some time before
updates will again be processed.

> I guess what happened is that the information the frontend held about
> the mailbox list and their location was stale.  I thought that the
> mupdate master server would "push" change in folder list back to
> frontend but apparently, it's the frontend that "pull" (or at least,
> initiate the update) that information instead.  Is that right ?  If yes,
> does that imply that when cyrus need restarting across a murder, we
> should do backend first, mupdate master second and frontend last ?

Changes are pushed from backend to mupdate master to frontend.  The
frontend initiates the connection to the mupdate master, but the mupdate
master independently pushes the updates down (sort of like the IMAP IDLE
extention).

Its probably not a terribly good idea to restart a backend if the mupdate
master is down, but I don't think that will explicitly hurt anything.
They should all be independently restartable.  Certainly, when I do
upgrades locally, I generally do the mupdate master first, then I restart
backends and frontends at the same time.

-Rob

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Rob Siemborski * Andrew Systems Group * Cyert Hall 207 * 412-268-7456
Research Systems Programmer * /usr/contributed Gatekeeper