xfs
[Top] [All Lists]

random file write failure...anyone else seen this?

To: Linux-Xfs <linux-xfs@xxxxxxxxxxx>
Subject: random file write failure...anyone else seen this?
From: lawalsh <xfs@xxxxxxxxx>
Date: Sat, 29 May 2004 23:00:11 -0700
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 0.6 (Windows/20040502)
This has happened on _rare_ occasions (rare likely because I am UPS backed-up and
usually only run even series kernels on this particular machine).

But the problem is thus. I'll write to a file -- pre-existing file or, I think has happened on new files as well. I write them to disk and think nothing more
of them....

Then...a _few_ days pass... (file if often recoverable from backups!)...and system crashes (like UPS not charged...minor details...may be getting near EOL)...whatever.
Maybe just a mysterious crash no-apparent details -- but what is screwey
is that some files will be filled with binary zeros, or sections of files
will be filled with binary zeros. Huh? I know in this last case, the file that croked was my fstab (slightly inconvenient) I'd edited a few _days_ prior. Filename was still there but filled with all zeros. Sometimes I'll find other files...log
files and such that also are zeroed out.

Is there any circumstance where xfs wouldn't actually write something to disk for
a few days?

Not just for xfs, but I wish there was a user command to sync all buffers to disk on the system. Sync used to do that but with delayed writes and multiple buffering schemes -- even mounting some fs's "async" so a specific program won't slow down, but something other than "umount" since that requires, sometimes, shutting down the machine. But I know it had been 2-3 days since I wrote to that file before the system crashed. It might have been open for read in the middle of a send crash (was testing circuits to do some wiring). Might that be a reason? Though I'm sure I've had it happen on files like ".profile", or something like that that wouldn't
have been getting read but that had been changed hours or days earlier.

VERY rare, this happens, and it happens randomly -- other files I wrote to after
that file were fine.  so I dunno what gives...

weird -- but just wanted to know if any one else has seen any such weirdness.

I saw this under 2.4 patched kernels as well as the more recent 2.6.5 kernel....

-l


<Prev in Thread] Current Thread [Next in Thread>
  • random file write failure...anyone else seen this?, lawalsh <=