I spoke a bit too soon about the upgrade being flawless. It turns out
that one subsystem started failing after the upgrade, although it is
not an out-of-the-box one.
I've included the description here in case someone runs into something
similarly weird (especially with scripts calling MySQL?) and because
although I have a work-around, I'm not sure (but can guess) as to the
actual cause of the change.
This subsystem is on a pacemaker/corosync cluster, which as one
resource group is running a resource agent (RA) for a bacula-fd
instance. (The bacula-fd RA takes care of backing up those
DRBD-mounted filesystems that are part of the resource group.)
The bacula-fd was failing because as a run-before command on
the backup job, it was executing the mysqlblasy.pl script to
back up some MySQL databases. After the upgrade, mysqlblasy.pl
started failing when running from bacula-fd, even though it
could be run directly as root.
However, *none* of these components were upgraded during the 5.5->5.6
mysql server & client
After suitable diagnosis, I was able to determine that with CentOS 5.5,
the environment provided to the RA had $HOME set, but that with
CentOS 5.6 $HOME was *not* set. In particular, since this RA runs
as root, in 5.5 HOME was set to "/" (as opposed to "/root"). Since
there was no change to pacemaker, I suspect that there was either a
change in the way daemons are started up, or a change to the fork
or set[re]uid model (or something similar).
Ensuring that HOME was (re)set properly before invoking mysqlblasy.pl
caused the backups to work again as expected.
CentOS mailing list