[Rocks-Discuss] no free leases

[Rob Hutten]

Hi all,

A Rocks newbie here; hopefully this isn't an inappropriate forum for
this sort of question.

I'm working on my first full Rocks install and have successfully
configured 5 of the 20 compute nodes in our cluster. In attempting to
add the sixth node (using PXE, like the others), I'm suddenly seeing "no
free leases" messages on the head node. The PXE error on the compute
node console is:

PXE-E51: No DHCP or proxyDHCP offers were received.

The server's /var/log/daemon message is:

Feb 12 13:50:39 xxhostnamexx dhcpd: DHCPDISCOVER from
00:1d:09:f2:89:1c via eth0: network 10/8: no free leases

I've tried restarting dhcpd, blowing away /var/lib/dhcp/dhcpd.leases,
and half-a-dozen blade reboots and power cycles.

Can anyone see anything I might be missing? I've dug through the usual
resources to no avail.

Thanks,

-Rob

Rob Hutten

Senior HPC Engineer

Flagstone RE

Suite 700, Cogswell Tower

2000 Barrington Street

Halifax, Nova Scotia B3J 3K1

www.flagstonere.bm

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080212/a3fe349d/attachment.html

[Tim Carlson]

On Tue, 12 Feb 2008, Rob Hutten wrote:

You are probably not running

insert-ethers

on the frontend.

Tim

Tim Carlson
Voice: (509) 371 6435 (new 1/12/08)
Email: Tim.C...@pnl.gov
Pacific Northwest National Laboratory
HPCaNS: High Performance Computing and Networking Services

[Mason J. Katz]

Insert-ethers detectsthe PXE/DHCP request. At this point it sounds
like networking. Direct connect the compute node to the frontend and
try again. If this works it's your switch.

On Feb 12, 2008 11:47 AM, Rob Hutten <rhu...@flagstonere.bm> wrote:

>
> Tim Carlson wrote:
> > You are probably not running
> >
> > insert-ethers
> >
> > on the frontend.
>

> Nope, that's not it - insert-ethers running. I'm using the exact same
> technique that successfully brought the first five compute nodes online.
>
> Where does Rocks have the dhcpd configuration? /etc/dhcpd.conf is
> dynamically generated, with new sections added for each compute node as
> it comes online. What I can't find is a config file that defines the
> address space used for the static IP assignments.

>
>
> -Rob
>
> Rob Hutten
> Senior HPC Engineer
> Flagstone RE
> Suite 700, Cogswell Tower
> 2000 Barrington Street
> Halifax, Nova Scotia B3J 3K1
> www.flagstonere.bm
>
>

--
-mjk

[Rob Hutten]

Tim Carlson wrote:
> You are probably not running
>
> insert-ethers
>
> on the frontend.

Nope, that's not it - insert-ethers running. I'm using the exact same

technique that successfully brought the first five compute nodes online.

Where does Rocks have the dhcpd configuration? /etc/dhcpd.conf is
dynamically generated, with new sections added for each compute node as
it comes online. What I can't find is a config file that defines the
address space used for the static IP assignments.

-Rob

[Rob Hutten]

Klaus Steden suggested:

> If for some reason you've swapped or changed IPs on your front end,
you'll
> see this issue as well.

Nope, that's not it.

Is it significant that /var/lib/dhcp/dhcpd.leases is empty, other than
some commented-out text?

-Rob

[Klaus Steden]

If for some reason you've swapped or changed IPs on your front end, you'll
see this issue as well.

Klaus

On 2/12/08 11:47 AM, "Rob Hutten" <rhu...@flagstonere.bm>did etch on stone
tablets:

[Rob Hutten]

Folks,

I reseated the blade in question and the rocks install is happily
churning along as I type this.

Very weird - the "no free leases" error was the red herring that sent me
off on a wild goose chase. Sorry for taking your time with what was in
essence a hardware issue.

Thanks to Greg Bruno, Mason Katz, Klaus Steden and Tim Carlson for their
input.
Cheers,

[Greg Bruno]

the 'no free leases' message is normal for a rocks cluster. in rocks,
we don't specify a dhcp 'pool', so when a new node enters the system,
this message will be generated.

you may want to try:

- stop insert-ethers

- restart syslogd:

# service syslogd restart

- start insert-ethers:

# insert-ethers

then try to PXE boot your compute node.

- gb

[Mason J. Katz]

For future reference what is the hardware that gave you the issue?

--
-mjk

[Rob Hutten]

Mason Katz asked:

> For future reference what is the hardware that gave you the issue?

Dell 1955 blades in a 1855/1955 enclosure with a 1950 headnode.

-Rob