From Bugzilla Helper: User-Agent: Mozilla/4.75 [en] (X11; U; OSF1 V4.0 alpha) Description of problem: We're running a cluster of ~150 machines here. They get both IP addresses and host names from two servers, one set up for odd IPs, the other for even ones. All machines get their IP address fine, but when ifup is run for eth0 and dhcpcd is started (with options -n -H eth0), the hosts get random host names. A way to avoid this is to have a delay of 3--5 seconds between the starts. Removing -H results in all hosts being named "localhost". Version-Release number of selected component (if applicable): How reproducible: Always Steps to Reproduce: 1. Shut down all DHCP clients (slave nodes of the cluster). 2. wake-on-lan them without delays. 3. run hostname on each node Actual Results: Nodes get the right IP address, but a random name. Names may occur several times. Expected Results: Since there is an 'option host-name "xxx"' for each host entry in dhcpd.conf, I'd expect the client to get this information. dhcpcd runs with -n -H eth0 Additional info: The relevant snippet from dhcpd.conf: host n002 { hardware ethernet 00:02:B3:B6:48:01; fixed-address 192.168.100.002; option host-name "n002"; } host n004 { hardware ethernet 00:02:B3:B8:29:80; fixed-address 192.168.100.004; option host-name "n004"; }
Using tcpdump and dhcpdump, we found out that the bug is not caused by dhcpd, but dhcpcd on different nodes sends out requests with the same XID so replies from the server(s) get misinterpreted. This has been fixed in a later version of dhcpcd (1.3.22pl2). The site www.phystech.com/download/ already has 1.3.22pl3 ready for d/l. Please upgrade the package from 1.3.18pl8 to at least 1.3.22pl2. I attach a snippet from http://www.phystech.com/download/dhcpcd_changelog.html: 09/21/02 - v.1.3.22-pl2 ... Michal Dobes <dobes> pointed out there could be situation with many dhcpcd clients starting at the same time on different machines on the same network in which case they will generate the DHCP messages with the same XID and potentially configure the same IP addresses. Fixed by including CLientHwAddr (MAC address) into the seed for srandom(). -S.V. ...
I have just taken over support of dhcp in RedHat, but my investigation shows that we have switched to using isc version of dhcp in 8.0. dhcp-3.0pl1-9 from www.isc.org. Have you tried this version? If so does it fix your problem? Dan
Hi Dan, looking at the tcpdumps, it seems not to be dhcp's fault (but there's no way to select more than one package in the bug report). I already changed the package to dhcpcd. Please look at www.phystech.com Steffen Bug reassigned. (to whoever is responsible for dhcpcd)
I am responsible for both.dhcp-3.0pl1-9 has both the client and server code in it. Red Hat moved away from the other distribution because they had a lot of problems with it, so I am told. Please try dhcp-3.0pl1 as the server and tell me if this solves the problem. Dan
So I'll set the component as dhcp again. Note that this bug report is against 7.2, which did use dhcpcd still.