xfs
[Top] [All Lists]

Re: Errors using amanda/xfsdump

To: Justin Tripp <justin@xxxxxxxxx>
Subject: Re: Errors using amanda/xfsdump
From: Steve Lord <lord@xxxxxxx>
Date: Thu, 26 Apr 2001 13:19:11 -0500
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: Message from Justin Tripp <justin@xxxxxxxxx> of "Wed, 25 Apr 2001 14:22:47 MDT." <Pine.BSF.4.33.0104251253290.43901-100000@xxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
Having thought about this some more, I think the problem is being caused by
the combination of local and remote access to the same files - lookups 
coming in from both directions in parallel almost instantaneously. xfsdump
will spin through inodes very quickly during some parts of its operation,
the news feed is doing the same thing from the other end.

I would still like to see some stack traces if you have time, but I think
I know what is going on.

Steve

> 
> I have been trying to backup xfs partitions using amanda, and there seems
> to be a problem with xfsdump.  Amanda suprisingly recognized xfsdump and
> seems to tried to do the backup correctly.  On the other hand, whenever we
> attempted to backup the machine would belly-up hard.
> 
> The problem is vexing, because sometimes it will fail consistently and
> quickly, yet other times it seems to take all day to fail.  I am doing to
> following to replicate the amanda backup:
> 
> ssh xfs_machine -l root "/usr/sbin/xfsdump -F -l 0 - /dev/sda1" | gzip -6
> - > file.gz
> 
> At the same time the xfs_machine is serving up a partition nfs and the
> partition is being read and written, by two independant news spools.
> Doing a usenet news spool over nfs onto xfs, may not be the best
> performance-wise, but I think most may agree that a news server can hit
> disks pretty hard when it comes to file ops.
> 
> After xfsdump caused the machine to fail, got the following error message
> on the console:
> 
> xfs_iget_core: ambiguous vns: vp/0xc3a543c8 invp/0xcf89d948
> 
> Unable to handle kernel NULL pointer dereference at virtual address
> 00000008
> 
> printing eip:
>   d08f90bf
> *pdc=00000000
> 
> Entering kdb (current=0xc96dc00, pid 803) on processor 1
> Oops: Oops
>   due to oops @ 0xd08490bf ...
> 
> The process listing showed that xfs was pid 803.
> 
> I have been unable to reliable recreate the failure.  Sometimes it fails,
> and some times it does not.  (It does seem to fail more reliably in the
> morning :) )  I am backing up about 2G of files produced by the news
> servers.  Any ideas?
> 
> The machine is a Dual 500 MHz PIII, and the filesystems run on top of the
> 3ware IDE raid card with 4 46G disks running in RAID level 5.  (138G
> filesystem available...)  The XFS is the 2.4.3 version from April 5th.
> 
> 
> 
>                               .justin.
> 
> ------------------------------------------------------------------------
> Justin Leonard Tripp                                   justin@xxxxxxxxxx
> Configurable Computing Laboratory Research Assistant      CB 461 x8-7206
> Electrical and Computer Engineering Department  Brigham Young University
> 
> 



<Prev in Thread] Current Thread [Next in Thread>