Duplicate lease, different ip.

[Simon Hobson]

Bruce Hudson wrote:

>  > I have some strange things happening with my dhcp server.
>>  I am running SLES 10 with isc-dhcpd-V3.0.3

>     The difference between the two leases is that the first request does
>not include a client identifier and the second does. In theory, the first
>(without an identifier) should always get one address and the request with
>an identifier should alway get a second since it is a different client.
>
>     However, the first time the server see a request with an identifier it
>will "upgrade" an existing lease for that client address that does not have
>one by adding it. The process is not reversible so any requests there-after
>without an idetifier are a different client. There was an ancient message
>to the list from Ted Lemon that called this "tragically unavoidable".
>
>     This doesn't really change anything in your case, except for which IP
>is kept in use. Setting up a host stanza with "deny duplicates" should take
>care of your problem. It tells the server to discard any existing leases
>for the specified hardware address, even if the client identifiers do not
>match.

Alternatively, you have two other options :

1) Configure your clients so that they all use the same client 
identifier - although that does somewhat negate some of the benefits 
of DHCP.

2) There is a patch that 'synthesises' a client-id if none is 
supplied (I think that's how it works). From the archives :

>Date: Tue, 12 Dec 2006 07:43:57 +0200
>From: Yedidyah Bar-David <didi at cs.tau.ac.il>
>To: dhcp-users at isc.org
>Subject: Re: status update
>
>On Mon, Dec 11, 2006 at 04:35:15PM -0500, SCOTT BURGIN wrote:
>  > Hello.  Running V3.0.4 server on RHEL.
>  >
>  > We have a remote site using PXE to image workstations.
>  > What we're seeing is the initial PXE boot (with a certain uid) is
>  > causing a lease to be written to the leases file as active.
>  > After the image is laid down and the machine boots the second time, the
>  > machine sends another Discover using a different uid, getting a
>  > different lease.
>  >
>  > I understand this is correct behavior per RFC, but have read that this
>  > is widely acknowledged to be an implementation PIA.
>  > What's fuzzy to me after searching the archives is what everyone agrees
>  > to be options currently to this dilemna for me.
>  > I've heard of patches, but can't seem to find them...other options ?
>
><http://www.cs.tau.ac.il/~didi/dhcp/>

I know nothing more about the currency or otherwise of this patch.


For the future, I understand that the database key will be user 
definable. It will default to "pick first (client-id, hardware)" as 
it is now, but can be changed (primarily to allow the use of relay 
agent circuit ids etc) to just use the hardware address if you so 
wish.

I can't remember if this is in 3.1, or just on the to-do list.