[Top] [All Lists]

Re: Stale XFS mount for Kernel

To: "Ngo, Andrew" <andrew.ngo@xxxxxxxx>
Subject: Re: Stale XFS mount for Kernel
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Tue, 14 Oct 2008 16:53:48 +1100
Cc: v9fs-developer@xxxxxxxxxxxxxxxxxxxxx, xfs@xxxxxxxxxxx, "Johnson, Je" <je.johnson@xxxxxxxx>
In-reply-to: <8604545CB7815D419F5FF108D3E434BA3BD631@xxxxxxxxxxxxxxxxxxxxx>
Mail-followup-to: "Ngo, Andrew" <andrew.ngo@xxxxxxxx>, v9fs-developer@xxxxxxxxxxxxxxxxxxxxx, xfs@xxxxxxxxxxx, "Johnson, Je" <je.johnson@xxxxxxxx>
References: <8604545CB7815D419F5FF108D3E434BA3BD626@xxxxxxxxxxxxxxxxxxxxx> <20081013035939.GB10716@disturbed> <8604545CB7815D419F5FF108D3E434BA3BD628@xxxxxxxxxxxxxxxxxxxxx> <8604545CB7815D419F5FF108D3E434BA3BD62F@xxxxxxxxxxxxxxxxxxxxx> <20081014021534.GI10716@disturbed> <8604545CB7815D419F5FF108D3E434BA3BD631@xxxxxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.18 (2008-05-17)
On Mon, Oct 13, 2008 at 11:40:17PM -0400, Ngo, Andrew wrote:
> What's the storage structure (DM, MD, iSCSI, etc)?
> Output of 'xfs_info <mtpt>'?
> [Andrew Ngo]xfs_info /mtpt
> meta-data=/dev/sda3 isize=256 agcount=16 agsize=672219 blks
>          =          sectsz=512 attr=0
> data     =          bsize=4096 blocks=10755504, imaxpct=25
>          =          sunit=0; swidth=0 blks, unwritten=1
> naming   = version2 bsize=4096
> log      = internal bsize=4096 blocks=525, version=1
>                     sectsz=512  sunit=0 blocks
> realtime = none     extsz=65536 blocks=0 rtextents=0

I think your cut-n-paste has dropped characters there - the log
cannot be that small - it must be at least 10MB in size
(2560 blocks, IIRC). Can you check this, please?

> What are you really trying to acheive with the remount command?
> [Andrew Ngo]This server hosts the patch directory for the developers.
> When the developers are not patching software, the file system is ro.
> When the developers needs to patch the software, the file system is
> changed to rw.

Ok. So when a software update needs to be done, the filesystem
is made writable, correct? But once the update is done, it
then gets made RO again? how much data gets written when an
update is run?

> Is the filesystem busy at the time the ro,remount is run?
> (e.g. any large background writes occurring?)
> [Andrew Ngo]Yes.  

So you try to make a busy, dirty filesystem read only while an
application is still writing to it?

> Note that I make a script that performs the "mount -o rw,remount <mtpt>"
> and "mount -o ro,remount <mnpt>".  When the system is not being used, I
> can finish the loop of 100 mount operations; however, when the system is
> being used, a couple of the above mount commands, even manually, will
> cause the system to hang.
> Does the problem go away if you do:
> # xfs_freeze -f <mtpt>
> # xfs_freeze -u <mtpt>
> # mount -o ro,remount <mtpt>
> [Andrew Ngo] When the mount command hangs, the above commands continue
> the hang the system.

I meant using that sequence of commands instead of just a single
"mount -o ro,remount <mtpt>" to make the fs RO. They won't magically
fix anything once something is broken.

> Or does the first freeze command trigger the same problem?
> [Andrew Ngo] Yes, the xfs_freeze -f command hang the system, just like
> the stale mount command does.

Is that before or after a remount has already hung?

Next time you hang the system, please attach the output that
appears in your syslog from:

# echo t > /proc/sysrq-trigger


Dave Chinner

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