David J N Begley wrote:
On Mon, 31 Jan 2005, Wilkins, Vern wrote:
Note: Biased response below.
Of course, real failure may only be once a year or even more rare, but if it
leads to the loss of important data then that's one loss too many - hence
people freaking out about XFS (even if the real incidents of loss are few).
I agree with you that the risk of losing data should be higher then I
have encountered so far.
It's even worse. The database server we had running under Red Hat 7.3
using XFS for the 190GB filesystem containing our Progress 9 databases
suffered both kernel crashes and UPS failures but it never led to
problems. This was all during daytime with people actively working in it.
We currently run RHEL 2.1 on the same host with ext3 for the database
filesystem (data=ordered). We had one power failure, at night, with
nobody logged in, with nobody active in the database and the backup
running (tape and rsync), only doing reads.
When I came into the entirely dark server room the next morning (not
much I could do at 3:45 anyways) the databases residing on this
supposedly "safe" "journaled" "robust" "supported" filesystem were
damaged beyond repair and we had to restore them.
People can tell me all they want about what's better and what not but I
don't trust it. Not now and not never. Backups are a good thing.
My luck was that at 3:34 the rsync of the production database to the
remote (well 500m actually) backupserver was done.
This meant that we could restore about 20GB of databases in about 30-45
minutes and let people login again without dataloss.
All it takes is a bit of static and some coffee on the keyboard, to
which I have lost far more then XFS in the past 3 years.