[FQuest Alert] BENDER Server

[Bob]

The BENDER server has experienced an overload condition and technicians are looking into the issue at this time.

As soon as additional information is available we will post again.

Our apologies,
Bob

[Bob]

BENDER was power cycled and has been returned to full service. It took longer then normal as the server had been up for 188 days and forced a file system check.

It appeared to be an I/O hang.

Event Start: 9:32 AM ET
Event End: 9:59 AM ET
Duration: 27 minutes

Again our apologies for any inconvenience,
Bob

[hobbes]

Are there any techniques that can be used to minimize fsck times?

[Kevin]

Are there any techniques that can be used to minimize fsck times?
Sure, we could limit the number of files that site owners are allowed to have instead of just megabytes :EG:

It is likely that in the future newer filesystems will allow fsck to run in the background while the server finishes booting. Right now xfs in Linux can already do that however it has other issues that make it inappropriate for our environment.

[Matt]

Sure, we could limit the number of files that site owners are allowed to have instead of just megabytesLame.

Why is XFS inappropriate?

[Terra]

Because XFS is designed for handling large files, not small ones... The vast majority of files on the server, that Apache has to read, are small files... Another problem with XFS is that it shows incredibly bad behavior on power failures where it will eat data and/or transpose the contents of files that were active...

Small files is where ReiserFS really shined, but unfortunately due to a myriad of other consistency problems with that FS, we had to dump it long ago...

We ended up taking the middle road with ext3, which is a compromise between the strengths of ReiserFS and XFS...

We are excited to see ext4 coming of age, though I will admit I'm salivating over the prospect of btrfs after using ZFS for awhile now on the backup servers...

[Kevin]

There is another reason that XFS is inappropriate...

One of the design "features" of XFS is to ensure file security even if a system crashes while the file is being written. One of SGI's fears was that a file may be written insecurely and then secured at the end. While that is of course the wrong thing to do they planned for it. Unfortunately their solution is to simply empty every file that was open when the system went down. We tried XFS on one of the email servers a while back and learned very quickly that when a system with XFS goes down there is a bloody trail of carnage left behind.