Re: Re: General question

On Tue, 7 Jun 2005 19:14:22 +0200 (MET DST)
"Peter T. Breuer" <[email protected]> wrote:

> "Also sprach Stephan von Krawczynski:"
> > How can I mark a certain mirror partition as new (increase event counter
> > manually)?
> 
> I'm afraid I don't REALLy know, but I suspect if I looked into the
> latest version of mdadm, the option would be there to manipulate things
> like that.

I hoped that, but was not able to find a corresponding option so far.

> 
> > I would like to be able to write to a secondary mirror partition mounted
> > locally, while primary is dead. Then, after primary comes up again, md
> > should know somehow that the primary is outdated ...
> 
> Err - I think you are trying to dictate the direction in which resync is
> done?

Yes, thats right.

> I'm pretty sure that everything should work right automatically.

It does not, I checked that. Basic reason is:

Jun  7 18:50:15 mail-a01 kernel: md: ndb's event counter: 00000000
Jun  7 18:50:15 mail-a01 kernel: md: nda's event counter: 00000000
Jun  7 18:50:15 mail-a01 kernel: md: md0: raid array is not clean -- starting
background reconstruction
Jun  7 18:50:15 mail-a01 kernel: md: RAID level 1 does not need chunksize!
Continuing anyway.
Jun  7 18:50:15 mail-a01 kernel: md0: max total readahead window set to 124k
Jun  7 18:50:15 mail-a01 kernel: md0: 1 data-disks, max readahead per
data-disk: 124k
Jun  7 18:50:15 mail-a01 kernel: raid1: device ndb operational as mirror 1
Jun  7 18:50:15 mail-a01 kernel: raid1: device nda operational as mirror 0
Jun  7 18:50:15 mail-a01 kernel: raid1: raid set md0 not clean; reconstructing
mirrors
Jun  7 18:50:15 mail-a01 kernel: raid1: raid set md0 active with 2 out of 2
mirrors
Jun  7 18:50:15 mail-a01 kernel: md: syncing RAID array md0

And after that the data I wrote raid-offline to ndb is gone....

> The event cunt on the primary will be old, so it will be resynced from
> the secndary when the primary is brought into the mirror.

The problem is that the event counter is obviously not updated if the fs is not
mounted via enbd but locally. As seen above both counters are zero, although I
wrote 30 M data to the partition that ndb uses.
 
> But you can always force such things with mkraid --dangerous-no-resync
> followed by a bit of setfaulty, remove, and add.

Well, manually: yes. Only I wanted to be able to flag/mark simply the newer
mirror partition so that the normal resync knows what to do. This does not
actually sound complicated, but I can't see the correct way ...

Regards,
Stephan