Mailing List Archive: LinuxFailSafe information

LinuxFailSafe information

Mar 17, 2000, 9:02 PM

Post #1 of 17 (4602 views)

Hi,

My *long* overdue set of notes from my trip to SGI last week are
attached. Hopefully, things at my end are on a little more even keel
for a while, and these kinds of delays won't recur soon. My apologies
for the delay.

I spent two days last week at SGI in a review of their cooperative
effort with SuSE to release their FailSafe project into the Linux world
as an open source development project.

I got an overview of their architecture, and spent a good bit of time
talking to various team members about the Linux community. Since their
product isn't *yet* open sourced, the discussion was made under a
non-disclosure agreement.

A quick summary of my visit:

+ They were very enthusiastic about starting up the community-based
LinuxFailSafe project.

+ They are working through their internal processes to release the
code and docs as quickly as they can.

+ They have started the port to Linux and have made good progress
on it so far.

+ Although I haven't seen the source yet, I did like what I saw.
The architecture makes sense, and the demos were nice.
It seems to have a lot of promise.

+ A very high-level document or two will be available to everyone
(not just SuSE) soon, and a few lower-level docs are on their
way to Lars and me.

+ Once they've completed their process, ALL code and docs will be
available to EVERYONE. After all, this is a community-based
open source project.

+ I set up a mailing list for the project. You can subscribe to it
here: http://lists.tummy.com/mailman/listinfo/linuxfailsafe

I hope this helps!

Thanks!

-- Alan Robertson
alanr@suse.com

LinuxFailSafe information [ In reply to ]

Mar 18, 2000, 8:23 AM

Post #2 of 17 (4593 views)

Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
Seeing 4-5 journaling filesystems comming and LVM integrated in the
kernel...

/Michael

LinuxFailSafe information [ In reply to ]

Mar 18, 2000, 8:56 AM

Post #3 of 17 (4578 views)

Michael Liljeblad wrote:
>
> Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> Seeing 4-5 journaling filesystems comming and LVM integrated in the
> kernel...
>
> /Michael

I would guess that LinuxFailSafe will support ReiserFS and XFS in its
early incarnations. After that, it is an open source development
project, and patches will be accepted. Including for, and especially
for, other filesystems.

-- Alan Robertson
alanr@suse.com

LinuxFailSafe information [ In reply to ]

Mar 18, 2000, 8:58 AM

Post #4 of 17 (4580 views)

On 2000-03-18T16:23:55,
Michael Liljeblad <mld@kyla.kiruna.se> said:

> Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> Seeing 4-5 journaling filesystems comming and LVM integrated in the
> kernel...

Yes. GFS is a good candidate for one, ext3/reiserfs on top of drbd another.

XLV hasn't even been ported yet, though I wouldn't object to seeing this too
;) I hear work is being done on making the Linux LVM cluster aware too, but I
have no current status on this.

Sincerely,
Lars Marowsky-Brée <lmb@suse.de>
Development HA

--
Perfection is our goal, excellence will be tolerated. -- J. Yahl

LinuxFailSafe information [ In reply to ]

Mar 18, 2000, 9:11 AM

Post #5 of 17 (4583 views)

Alan Robertson wrote:
>
> Michael Liljeblad wrote:
> >
> > Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> > Seeing 4-5 journaling filesystems comming and LVM integrated in the
> > kernel...
> >
> > /Michael
>
> I would guess that LinuxFailSafe will support ReiserFS and XFS in its
> early incarnations. After that, it is an open source development
> project, and patches will be accepted. Including for, and especially
> for, other filesystems.

Although I can't imagine that the differences between filesystems will
matter much.

-- Alan Robertson
alanr@suse.com

LinuxFailSafe information [ In reply to ]

wiegand at suse

Mar 19, 2000, 8:09 PM

Post #6 of 17 (4580 views)

On Sat, 18 Mar 2000, Alan Robertson wrote:

> Alan Robertson wrote:
> >
> > Michael Liljeblad wrote:
> > >
> > > Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> > > Seeing 4-5 journaling filesystems comming and LVM integrated in the
> > > kernel...
> > >
> > > /Michael
> >
> > I would guess that LinuxFailSafe will support ReiserFS and XFS in its
> > early incarnations. After that, it is an open source development
> > project, and patches will be accepted. Including for, and especially
> > for, other filesystems.
>
> Although I can't imagine that the differences between filesystems will
> matter much.
>
Ummm, methinks it makes a significant difference. The filesystem will
have to be mounted/unmounted or be prepared for concurrent access. Both
are a challenge right now. For the first we need journaling filesystems,
and JournalingReiserFS might be the nearest to production use, although
XFS and ext3 are also close, of course. The latter would mean GFS+LVM or
cXFS. But both are not around yet, for different reasons (LVM is not yet
cluster aware, and cXFS is not running on Linux yet).

We NEED to discuss the relationship between all those projects ASAP with
their impact on IRIS FailSafe for Linux.

> -- Alan Robertson
> alanr@suse.com
>
Volker

--
Volker Wiegand Voice: +1-510-628-3380 ext 5029
SuSE Inc. Fax: +1-510-628-3381
580 Second Street, Suite 210 Mobile: +1-510-376-0302
Oakland, CA 94607 USA E-Mail: wiegand@suse.com

Volker Wiegand Phone: +49 (0) 6196 / 50951-24
SuSE Linux AG Fax: +49 (0) 6196 / 40 96 07
Mergenthalerallee 45-47 Mobile: +49 (0) 179 / 292 66 76
D-65760 Eschborn E-Mail: Volker.Wiegand@suse.de
++ Only users lose drugs. Or was it the other way round? ++

LinuxFailSafe information [ In reply to ]

Mar 19, 2000, 8:22 PM

Post #7 of 17 (4584 views)

Volker Wiegand wrote:
>
> On Sat, 18 Mar 2000, Alan Robertson wrote:
>
> > Alan Robertson wrote:
> > >
> > > Michael Liljeblad wrote:
> > > >
> > > > Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> > > > Seeing 4-5 journaling filesystems comming and LVM integrated in the
> > > > kernel...
> > > >
> > > > /Michael
> > >
> > > I would guess that LinuxFailSafe will support ReiserFS and XFS in its
> > > early incarnations. After that, it is an open source development
> > > project, and patches will be accepted. Including for, and especially
> > > for, other filesystems.
> >
> > Although I can't imagine that the differences between filesystems will
> > matter much.
> >
> Ummm, methinks it makes a significant difference. The filesystem will
> have to be mounted/unmounted or be prepared for concurrent access. Both
> are a challenge right now. For the first we need journaling filesystems,
> and JournalingReiserFS might be the nearest to production use, although
> XFS and ext3 are also close, of course. The latter would mean GFS+LVM or
> cXFS. But both are not around yet, for different reasons (LVM is not yet
> cluster aware, and cXFS is not running on Linux yet).

But the APIs for each are the same. Testing is different. Preparing a
filesystem for access is done with "mount". On every filesystem. So,
the system won't care which it's using - as long as it works. It will
just care that the API for doing them. This isn't a cluster aware
issue. This is just a failover primary/secondary configuration being
discussed.

> We NEED to discuss the relationship between all those projects ASAP with
> their impact on IRIS FailSafe for Linux.

We've been doing just that. That's part of what I want to discuss at
the seminar.

-- Alan Robertson
alanr@suse.com

LinuxFailSafe information [ In reply to ]

Mar 20, 2000, 7:54 AM

Post #8 of 17 (4579 views)

Alan Robertson wrote:
> Michael Liljeblad wrote:
> >
> > Will LinuxFailSafe include support for other FS/LV than XFS/XLV?
> > Seeing 4-5 journaling filesystems comming and LVM integrated in the
> > kernel...
> >
> > /Michael
>
> I would guess that LinuxFailSafe will support ReiserFS and XFS in its
> early incarnations. After that, it is an open source development
> project, and patches will be accepted. Including for, and especially
> for, other filesystems.

From a technical perspective LinuxFailSafe has no dependency on any
underlying filesystem or volume manager. You could use ext2 or XFS. The
difference is that on failover, you will have to wait for fsck if you
used ext2. This may or not be acceptible depending on what you were trying
to do and how big your file system is.

Simon.

LinuxFailSafe information [ In reply to ]

jetienne at arobas

Mar 20, 2000, 8:16 AM

Post #9 of 17 (4594 views)

On Mon, Mar 20, 2000 at 09:54:31AM -0500, Simon Patience wrote:
> From a technical perspective LinuxFailSafe has no dependency on any
> underlying filesystem or volume manager. You could use ext2 or XFS. The
> difference is that on failover, you will have to wait for fsck if you
> used ext2.

does this mean that XFS doesnt need fsck in this case ?

LinuxFailSafe information [ In reply to ]

Mar 20, 2000, 8:25 AM

Post #10 of 17 (4591 views)

On 2000-03-20T10:16:06,
Jerome Etienne <jetienne@arobas.net> said:

> > From a technical perspective LinuxFailSafe has no dependency on any
> > underlying filesystem or volume manager. You could use ext2 or XFS. The
> > difference is that on failover, you will have to wait for fsck if you
> > used ext2.
> does this mean that XFS doesnt need fsck in this case ?

JFS, ext3, reiserfs, XFS and GFS (if memory serves) are journaled and don't
require a full fsck - sync'ing the log is enough.

Sincerely,
Lars Marowsky-Brée <lmb@suse.de>
Development HA

--
Perfection is our goal, excellence will be tolerated. -- J. Yahl

LinuxFailSafe information [ In reply to ]

Mar 20, 2000, 8:30 AM

Post #11 of 17 (4583 views)

In message <20000320162518.H18304@pointer.teuto.de>you wrote:
> On 2000-03-20T10:16:06,
> Jerome Etienne <jetienne@arobas.net> said:
>
> > > From a technical perspective LinuxFailSafe has no dependency on any
> > > underlying filesystem or volume manager. You could use ext2 or XFS. The
> > > difference is that on failover, you will have to wait for fsck if you
> > > used ext2.
> > does this mean that XFS doesnt need fsck in this case ?
>
> JFS, ext3, reiserfs, XFS and GFS (if memory serves) are journaled and don't
> require a full fsck - sync'ing the log is enough.

Sorry I didn't mean to confuse, I was using those ext2 or XFS as examples.
LinuxFailSafe doesn't care what the filesystem is, but the application
might, is what I was trying to convey.

Simon.

LinuxFailSafe information [ In reply to ]

jetienne at arobas

Mar 20, 2000, 8:39 AM

Post #12 of 17 (4582 views)

On Mon, Mar 20, 2000 at 04:25:18PM +0100, Lars Marowsky-Bree wrote:
> JFS, ext3, reiserfs, XFS and GFS (if memory serves) are journaled and don't
> require a full fsck - sync'ing the log is enough.

so they all need fsck.

btw if the log is corrupted, even journaled fs still need to scan the
entire fs, what you call a 'full fsck'

LinuxFailSafe information [ In reply to ]

Mar 20, 2000, 9:38 AM

Post #13 of 17 (4592 views)

Jerome Etienne wrote:
> On Mon, Mar 20, 2000 at 04:25:18PM +0100, Lars Marowsky-Bree wrote:
> > JFS, ext3, reiserfs, XFS and GFS (if memory serves) are journaled and don't
> > require a full fsck - sync'ing the log is enough.
>
> so they all need fsck.

No. XFS does not need fsck. The log is re-run on mount.

> btw if the log is corrupted, even journaled fs still need to scan the
> entire fs, what you call a 'full fsck'

If the log is corrupted then you have a whole set of problems more than
fsck!

Simon.

LinuxFailSafe information [ In reply to ]

Mar 20, 2000, 9:43 AM

Post #14 of 17 (4588 views)

> On Mon, Mar 20, 2000 at 09:54:31AM -0500, Simon Patience wrote:
> > From a technical perspective LinuxFailSafe has no dependency on any
> > underlying filesystem or volume manager. You could use ext2 or XFS. The
> > difference is that on failover, you will have to wait for fsck if you
> > used ext2.
>
> does this mean that XFS doesnt need fsck in this case ?

Correct. Log based file systems do not use fsck. When you mount the
filesystem they re-run any log entries to bring the file system up to
date. fsck is needed because transactions can be half completed when
the system crashed leaving the file system corrupted. Because the log
is updated atomically with respect to file system transactions, simply
rerunning the log is all you need. This time will be proportional to
the size of the log and not the size of the filesystem, and will take
less than a second even for terrabyte file systems.

Simon.

LinuxFailSafe information [ In reply to ]

jetienne at arobas

Mar 20, 2000, 9:54 AM

Post #15 of 17 (4584 views)

On Mon, Mar 20, 2000 at 11:38:52AM -0500, Simon Patience wrote:
> > so they all need fsck.
>
> No. XFS does not need fsck. The log is re-run on mount.

ok

> > btw if the log is corrupted, even journaled fs still need to scan the
> > entire fs, what you call a 'full fsck'
>
> If the log is corrupted then you have a whole set of problems more than
> fsck!

does this mean that most journaled fs assume that the log is never
corrupted ? so if one sector of my log is broken, how can i fix
the corrupted fs ?

ps: this discussion may appear off topic, but to know what happen
in case of faillure is the heart of high availability.

LinuxFailSafe information [ In reply to ]

jetienne at arobas

Mar 20, 2000, 10:07 AM

Post #16 of 17 (4595 views)

On Mon, Mar 20, 2000 at 11:43:53AM -0500, Simon Patience wrote:
> Correct. Log based file systems do not use fsck. When you mount the
> filesystem they re-run any log entries to bring the file system up to
> date.

ok i see what i didnt understood.
for you 'fsck' meant the unix tool, for me 'fsck' meant the feature
i.e. check if the fs is clean and if not, try to fix it (what is done
on mount by xfs).

LinuxFailSafe information [ In reply to ]

Mar 21, 2000, 9:28 AM

Post #17 of 17 (4585 views)

Hi,

On Mon, Mar 20, 2000 at 11:54:37AM -0500, Jerome Etienne wrote:
>
> does this mean that most journaled fs assume that the log is never
> corrupted ? so if one sector of my log is broken, how can i fix
> the corrupted fs ?

You run fsck on it, and the fsck tells you if it was able to repair
the damage or not. Just like for any other sort of abnormal filesystem
failure.

Cheers,
Stephen