[DRBD-user] Initializing shortcut

Lars Ellenberg Lars.Ellenberg at linbit.com
Wed May 10 18:32:19 CEST 2006

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


/ 2006-05-10 09:10:05 -0700
\ Martin Fick:
> --- Lars Ellenberg <Lars.Ellenberg at linbit.com> wrote:
> 
> > / 2006-05-09 11:06:04 -0700
> > \ Martin Fick:
> ...
> > > What I mean is that if a device has never had any
> > > data on it, let's say the partitions were just 
> > > created, allow a startup feature that would not 
> > > attempt to sync the two partitions.  In this case,
> 
> > > only newly written data to the disks need be
> > > synced, so as soon as mkfs or what not is run on
> > > the primary it would sync over the fs layout and
> > > continue to operate as normal.
> 
> > yes it is possible.
> > no we probably won't make it easier.
> > 
> > as experience shows, people are very tempted to
> > shoot themselves, and would even use that option
> > because it is available and mentioned in some
> > "howto" thing, even if it is not applicable.  it
> > would initially appear to work ok, then they would
> > do a failover, complain about fs-corruption and
> > declare that drbd had eaten their data...
> 
> I understand the desire to have people avoid shooting
> themselves in the foot, a goal that I am very glad
> drbd takes seriously, but I have to wonder if there
> wouldn't still be a safe option for drbd to do this
> somehow.  
> 
> If  a partition is starting off fresh, can't drbd be
> sure of that?

in drbd8, we have the command drbdmeta, which is much
more flexible and powerful than write_gc.pl .
it has commandline parameters to initialize the meta
data, or dump/restore it to some ascii representation.

> Afterall there is a period in time when
> there is no metadata on the partitions and drbd has
> decided that it is OK to initialize the partition
> (--do-what-i-say).

this option is seriously misunderstood, too.
it has been renamed to "--overwrite-data-of-peer", to
better express what it actually does.

> At this point in time it surely
> would be no worse to use an algorythm that simply
> shortcutted the syncing process would it?

what, if you had an existing single-server setup with
existing data, and you extend it with drbd to form a
cluster, so you have existing filesystems with
important data on it?
is it ok to skip the initial sync there, too?

so, in drbd8, you have to initialize the meta data with
drbdmeta, (well, drbdadm will call that for you), and
you have to confirm some annoying questions and stuff.

you would be able to use drbdmeta with the appropriate
command line parameters to initialize the meta data
into a state of your choice.

we can probably put some more annoying questions in
there, so you could confirm that you want to skip the
initial sync because you mkfs anyways.

but then again, once we implement the online
verification, we would _need_ to have this initial
sync, because the drives _have_ to be bitwise identical
(thats one of our "guarantees", after all).

-- 
: Lars Ellenberg                                  Tel +43-1-8178292-0  :
: LINBIT Information Technologies GmbH            Fax +43-1-8178292-82 :
: Schoenbrunner Str. 244, A-1120 Vienna/Europe   http://www.linbit.com :
__
please use the "List-Reply" function of your email client.



More information about the drbd-user mailing list