Lustre discuss - Mar 2009 - Beginners Question: Mapping Lustre Node <-> DNS Hostname?

Hello out there!

For quite a while I''m exploring the Lustre file system and become more
and more fan of it. I''m familiar with other distributed multiple writer
file systems as well and have worked with IBM GPFS, Sun QFS and StorNext
file system (now Quantum, former ADIC).

In Lustre, I frequently experience the following "problem", where I
couldn''t find an answer in the manual nor in the Lustre FAQ. Maybe
I''m
just blind: 

My setup ist TCP based only. Is there an easy way to match the Lustre
node name (testfs-OST0001) with the corresponding IP hostname/domain
(something like lussrv1.some.domain)? I also tried to find something in
the /proc structure on MGS/MDT, OST and Client without success so far.

Thanks for any hints.
wolfgang stief

On Tue, 2009-03-31 at 13:38 +0200, Wolfgang Stief wrote:
> Hello out there!
Hi.
> My setup ist TCP based only. Is there an easy way to match the Lustre
> node name (testfs-OST0001)
testfs-OST0001 is not a node.  It''s a lustre target.  An OST to be
specific.  Think disk in a node, rather than a node.
> with the corresponding IP hostname/domain
> (something like lussrv1.some.domain)?
I guess what you really want to know is "what is the IP address (and by
extension, DNS name) of the machine that is currently providing the
testfs-OST0001 target?".

But again, the answer is not so simply an IP address as IP is only one
of several networking protocols that Lustre works on and as such it
(LNET) does not use IP addresses as node identifiers but uses NIDs
instead.  For some protocols, the NID does use the IP address, but this
is just a convenience of that given protocol.  So really, the only thing
you can really ask is "what NID is serving the target
testfs-OST0001?".

I''m afraid I don''t think there is any easy way to ask this
currently,
other than brute-force searching through the OSSes.

Maybe somebody else can remember something I''m forgetting.

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090331/00c91c03/attachment.bin

On Tue, 2009-03-31 at 10:20 -0400, Brian J. Murrell
wrote:
> On Tue, 2009-03-31 at 13:38 +0200, Wolfgang Stief wrote:
> > Hello out there!
> 
> Hi.
> 
> > My setup ist TCP based only. Is there an easy way to match the Lustre
> > node name (testfs-OST0001)
> 
> testfs-OST0001 is not a node.  It''s a lustre target.  An OST to be
> specific.  Think disk in a node, rather than a node.
> 
> > with the corresponding IP hostname/domain
> > (something like lussrv1.some.domain)?
> 
> I guess what you really want to know is "what is the IP address (and
by
> extension, DNS name) of the machine that is currently providing the
> testfs-OST0001 target?".
> 
> But again, the answer is not so simply an IP address as IP is only one
> of several networking protocols that Lustre works on and as such it
> (LNET) does not use IP addresses as node identifiers but uses NIDs
> instead.  For some protocols, the NID does use the IP address, but this
> is just a convenience of that given protocol.  So really, the only thing
> you can really ask is "what NID is serving the target
testfs-OST0001?".
> 
> I''m afraid I don''t think there is any easy way to ask
this currently,
> other than brute-force searching through the OSSes.
> 
> Maybe somebody else can remember something I''m forgetting.
lustre_createcsv maybe?

Cheers,
 Arne

On Tue, 2009-03-31 at 16:26 +0200, Arne Wiebalck wrote:
> 
> lustre_createcsv maybe?
Yeah, the information that OP is looking for could surely be gleaned out
of a generated CSV file, but that createcsv script just boils down to
the brute-force search I suggested.

That''s not to say that everyone should not run that script and tuck the
generated CSV away for safekeeping, and even for reference, to answer
such questions.

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090331/47130214/attachment.bin

On Mar 31, 2009  10:20 -0400, Brian J. Murrell wrote:
> On Tue, 2009-03-31 at 13:38 +0200, Wolfgang Stief wrote:
> > with the corresponding IP hostname/domain
> > (something like lussrv1.some.domain)?
> 
> I guess what you really want to know is "what is the IP address (and
by
> extension, DNS name) of the machine that is currently providing the
> testfs-OST0001 target?".
> 
> But again, the answer is not so simply an IP address as IP is only one
> of several networking protocols that Lustre works on and as such it
> (LNET) does not use IP addresses as node identifiers but uses NIDs
> instead.  For some protocols, the NID does use the IP address, but this
> is just a convenience of that given protocol.  So really, the only thing
> you can really ask is "what NID is serving the target
testfs-OST0001?".
> 
> I''m afraid I don''t think there is any easy way to ask
this currently,
> other than brute-force searching through the OSSes.
You can use:

$ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid

e.g.

$ lctl get_param osc.*-OST0000*.ost_conn_uuid
osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp

or
$ lctl get_param osc.*.ost_conn_uuid
osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
osc.myth-OST0001-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
osc.myth-OST0002-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
osc.myth-OST0003-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
osc.myth-OST0004-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

On Tue, 2009-03-31 at 14:41 -0600, Andreas Dilger wrote:
> 
> You can use:
> 
> $ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid
> 
> e.g.
> 
> $ lctl get_param osc.*-OST0000*.ost_conn_uuid
> osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> 
> or
> $ lctl get_param osc.*.ost_conn_uuid
> osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> osc.myth-OST0001-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> osc.myth-OST0002-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> osc.myth-OST0003-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> osc.myth-OST0004-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
Cool.

Sheila, this is probably good material for the manual.

Thanx!

b.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090331/8c58f247/attachment.bin

Should this work with 1.6.4.3?  My lctl doesn''t even have a get_param,
the closest thing I have is getattr, and it doesn''t like the syntax
below...
> -----Original Message-----
> From: lustre-discuss-bounces at lists.lustre.org [mailto:lustre-discuss-
> bounces at lists.lustre.org] On Behalf Of Brian J. Murrell
> Sent: Tuesday, March 31, 2009 2:45 PM
> To: lustre-discuss
> Subject: Re: [Lustre-discuss] Beginners Question: Mapping Lustre Node
> <-> DNS Hostname?
> 
> On Tue, 2009-03-31 at 14:41 -0600, Andreas Dilger wrote:
> >
> > You can use:
> >
> > $ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid
> >
> > e.g.
> >
> > $ lctl get_param osc.*-OST0000*.ost_conn_uuid
> > osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> >
> > or
> > $ lctl get_param osc.*.ost_conn_uuid
> > osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > osc.myth-OST0001-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > osc.myth-OST0002-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > osc.myth-OST0003-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > osc.myth-OST0004-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> 
> Cool.
> 
> Sheila, this is probably good material for the manual.
> 
> Thanx!
> 
> b.

Hi there!

On Tue, 31 Mar 2009 14:41:59 -0600
Andreas Dilger <adilger at sun.com> wrote:
> On Mar 31, 2009  10:20 -0400, Brian J. Murrell wrote:
> > On Tue, 2009-03-31 at 13:38 +0200, Wolfgang Stief wrote:
> > > with the corresponding IP hostname/domain
> > > (something like lussrv1.some.domain)?
> > 
> > I guess what you really want to know is "what is the IP address
> > (and by extension, DNS name) of the machine that is currently
> > providing the testfs-OST0001 target?".
Yes, that was my intention.
> You can use:
> 
> $ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid
Works great, thank you!
What would the syntax be for a MDT?

Is there a list of valid values for get_param? Seems to me, that
this command is useful for gathering some other informations as
well. Couldn''t find something in the wiki nor in the manual. Even some
piece of source code would be fine...

wolfgang

On Mar 31, 2009  14:55 -0600, Lundgren, Andrew wrote:
> Should this work with 1.6.4.3?  My lctl doesn''t even have a
get_param, the closest thing I have is getattr, and it doesn''t like the
syntax below...
The get_param/set_param helpers were added in (maybe) 1.6.5 and should be
preferred in the future for scripts, instead of digging through /proc
directly.  For servers/clients running on Solaris or Windows there is no
/proc interface.  In either case, using "lctl set_param" can be MUCH
easier
than doing things with /proc directly

	lctl set_param osc.*.max_dirty_mb=512
vs
	for O in /proc/fs/lustre/osc/*/max_dirty_mb; do
		echo 512 > $O
	done

There is a direct mapping between entries in /proc and their get_param
equivalents, for ease of converting scripts:

/proc/{fs,sys}/{lustre,lnet}/foo/bar/baz := foo.bar.bax
> > On Tue, 2009-03-31 at 14:41 -0600, Andreas Dilger wrote:
> > >
> > > You can use:
> > >
> > > $ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid
> > >
> > > e.g.
> > >
> > > $ lctl get_param osc.*-OST0000*.ost_conn_uuid
> > > osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > >
> > > or
> > > $ lctl get_param osc.*.ost_conn_uuid
> > > osc.myth-OST0000-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > > osc.myth-OST0001-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > > osc.myth-OST0002-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > > osc.myth-OST0003-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
> > > osc.myth-OST0004-osc-f1579000.ost_conn_uuid=192.168.20.1 at tcp
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

On Apr 01, 2009  15:31 +0200, Wolfgang Stief wrote:
> On Tue, 31 Mar 2009 14:41:59 -0600
> Andreas Dilger <adilger at sun.com> wrote:
> > You can use:
> > 
> > $ lctl get_param osc.${fsname}-${OSTname}*.ost_conn_uuid
> 
> Works great, thank you!
> What would the syntax be for a MDT?
Pretty much the same: mdc.${fsname}-${MDTname}*.mds_conn_uuid
> Is there a list of valid values for get_param? Seems to me, that
> this command is useful for gathering some other informations as
> well. Couldn''t find something in the wiki nor in the manual. Even
some
> piece of source code would be fine...
Currently "ls -lR /proc/fs/lustre", or "lctl get_param -N
{wildcard pattern}",
but there will be a "lctl list_param" in an upcoming release.  This is
just
being phased in as a transition for direct /proc access.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.