[gfarm-discuss:14630] Re: ldap server failing

[Matt <mattmoran76@xxxxxxxxx>]

Hi Osamu,

Thanks for the reply.  I've moved it over to Postgresql, and sure enough it is going through inserting the files, however, this is taking much time :-(

Bottleneck appears to be Postgresql which is on the mds server with the process postmaster constantly at 99 cpu.

I'm only 20% through the number of files and it's been running for 5 days now.

I'm thinking if there are any improvements for to alter this, or is it just that gfarm v1 is not designed for millions of files?

Thanks,

Matt

On 11/10/2007, Osamu Tatebe <tatebe@xxxxxxxxxxxxxxxx> wrote:

Hi Matt,

OpenLDAP server is not so robust regarding frequent updates.
I suggest to use PostgreSQL for a backend server since it is
much robust.

BTW, creating many small files is not fast enough in Gfarm v1,
while it is much much improved by Gfarm v2.  Still Gfarm v2
lacks several functionalities for operation and administration,
but it has basic functionality and robustness including
gfarm2fs (GfarmFS-FUSE for Gfarm v2).

I will announce a Gfarm v2 preview release pretty soon.

Thanks,
Osamu

On Wed, 10 Oct 2007 10:48:51 +0100
Matt <mattmoran76@xxxxxxxxx> wrote:

> Hi,
>
> Further on to this I realised it was because the slapd was falling over.  I
> enabled logging and tried to start gfarm-slapd and got the following in the
> log:
>
> PANIC: fatal region error detected; run recovery
> Oct 10 10:05:46 sl08 slapd[11611]: bdb_db_open: dbenv_open failed:
> DB_RUNRECOVERY: Fatal error, run database recovery (-30978)
> Oct 10 10:05:46 sl08 slapd[11611]: backend_startup: bi_db_open(0) failed!
> (-30978)
> Oct 10 10:05:46 sl08 slapd[11611]: slapd shutdown: initiated
>
> So looks like the db got corrupted.
>
> Copying large files doesn't appear to be a problem, but when this happened I
> was copying 100,000s of small files, which is obviously creating a large
> ldap db, the ldap db directory was at 8.3GB when this happened.
>
> I've wiped, resetted, and ran again and hit the same problem.
>
> I don't really want to alter the gfarm settings for the bdb or ldap, anyone
> got any suggestions? would using postgresql be better?
>
> Thanks,
>
> Matt
>
> ---------- Forwarded message ----------
> From: Matt <mattmoran76@xxxxxxxxx>
> Date: 8 Oct 2007 10:35
> Subject: /: invalid argument
> To: gfarm-discuss@xxxxxxxxxx
>
> Hi folks,
>
> Left a cp going from a gfarm client using fuse over the weekend.  When not
> trying to ls my /mnt/gfarm I receive invalid argument.
>
> So tried fusermount -u /mnt/gfarm
> then mount again with gfarm -sl -N2 /mnt/gfarm and receive /: invalid
> argument
>
> Restarted gfarm services on mds, the three datanodes, and rebooted a
> client.  Still getting the same error.
>
> Anyone have any idea what's going on?
>
> Thanks,
>
> Matt
>