> Lately, I've witnessed several cases of files turning out to be
> corrupted after system crashes. I know it CAN happen, but what I
> think is strange is that:
>
> - the vast majority of cases occur in Solaris-2.5 and Solaris-2.5.1
> machines (This is a way to say that I can't remember it occurring in
> Solaris1 machines, but I can't tell for sure. For the records, we
> have about the same number (~200) of machines running each system.),
> and that
>
> - the vast majority of cases (again) affect system configuration files
> like /etc/hosts and /var/yp/binding/`domainname`/ypservers that are
> updated nightly.
>
> I begun to suspect the fsflush process is malfunctioning after a test.
> First, I created a file with vi and waited more than 60 seconds before
> switching the power of my machine off. After the reboot, the file was
> corrupted. Then, I did it again, but, this time, I waited a whole
> hour before switching the power off. This second time the file was OK
> after the reboot. (I'm sure nobody sync'ed the file system
> meanwhile.)
>
> I couldn't find much information about fsflush but, according to the
> System Administration Guide, Volume II of the Solaris 2.5 System
> Administrator Answerbook, page 1454, the default rate at which fsflush
> is run is 30 seconds. The problem is that I waited more the 60
> seconds (in the first test) and, apparently, the file hadn't been
> flushed to disk yet.
>
> Am I missing something? Can somebody help me with a clue?
I've received several responses, almost all of them telling me to
apply patch #103640-08. Apparently this is a know problem with
vanilla Solaris-2.5 AND Solaris-2.5.1. I'm getting all the
Recommended patches for Solaris-2.5.1 (103640 is one of them) and
intend to apply all of them as soon as possible.
Thanks to the following people:
Ric Anderson
Russ Poffenberger
Glenn Satchell
Casper Dik
Per Boussard
Rachel Polanskis
Michael Kohne
Marc S. Gibian
Mark Bergman
Bruce Rossiter
Vinay Somashekar
Bismark Espinoza
Gustavo.