On Sat, Mar 01, 2003 at 05:10:42AM +0000, Louis C. Candell wrote: > >after several sysadmins had used it with no problems - even power outages. > >In fact before they implemented it they did extensive tests by pulling the > >plug. XFS came back up and ran without a hitch and that has been born out > >through actual experience. One of it's strong points is the robustness and > >ability to take hits ext2 can't handle. Another nice feature is it recovers > >from outages quickly where on big file systems ext2 can take 30+ minutes to > >fsck. > > Heh, say that to my mp3 pentium2 gentoo box which now registers mod_php > as pornview.... > > Pull the plug out of the box right after compilation and while it is > installing the program, libraries and files in their respective > locations and tell me what happens. > > Also, do some work on a file for 3 hours without saving and then go > ahead and save while pulling the plug out of the box 1/4-1/2 second > after doing C-x C-s or pressing save and you tell me how much of the > file is there... I would be more than willing to bet the last 3-5 > seconds worth of saving, work and general changes you made before an > abrupt shutdown have been L O S T.
Personally, I've had great luck with reiserfs, but I don't think the cases you cite are problems in XFS's behavior. In fact the one about losing your changes is by design. XFS can only try to ensure that either _all_ your changes to a file get saved or none of them do. That's why you'd lose all the work, because it didn't get far enough to save all of it. With the problem where libraries got half installed, the filesystem can only ensure that changes to a single file happen or don't happen. It has no way of viewing an entire emerge operation as a transaction. If you pull the plug when half the libraries are installed and half are not, it's going to break no matter what filesystem you're using. > This box is 90% XFS on LVM on RAID-0... trust me, I *enjoy* using XFS, > but have found abrupt losses of power have you lose the last 3-5 > seconds of work you were doing on the system. I guess the 3 to 5 seconds might be somewhat large, so if you're talking about a server whose logs have to be as up-to-date as possible before a crash happens, then XFS might be a bad choice. But that must be configurable somewhere. It's got to be just a buffer size or buffer flushing frequency setting of some sort. > I understand the extensive tests, but I'm also speaking out of my own > personal experience. Yes, a box on XFS comes RIGHT UP after a hard > shutdown, but things are and will be missing. > > If you dont take my word for it, maybe you will believe: > > http://www.gentoo.org/doc/en/gentoo-x86-install.xml > > XFS is a filesystem with metadata journaling that is fully supported > under Gentoo Linux's xfs-sources kernel, but is generally not > recommended due to its tendency to lose recently-modified data if > your system locks up or unexpectedly reboots (as a result of power > failure, for instance) Wow, that's weird. For some special applications, like the one I mentioned above, I could see that losing changes made 3-5 seconds before a system crash might be bad, but I can't see how it would be such a big deal for most setups. Hopefully your system doesn't crash very often, and if it does, what are the chances that it happened right after you finally saved that 3 hours worth of work? - richard -- Richard Kilgore [EMAIL PROTECTED] -- [EMAIL PROTECTED] mailing list
