Hi Michael, In response to your response below: I do not have omapi enabled.
I tried removing every parameter from the bond interfaces, so the only thing left was the interface field (named "bond0" and "bond1" for the two interfaces). This made no difference, sync still timed out. So it doesn't look like a problem with settings. The only thing that works is removing all of the bond interfaces entirely. So I have a workaround: I wrote a script that removes all the bond interfaces from every server, then I run cobbler sync, then I add all the bond interfaces back. This works but is obviously not a great solution. Also, the sync (without the bond interfaces) takes over 2 minutes. It was taking around 24 seconds before, but I'm not sure what I changed in the meantime, if anything. >> It definitely has something to do with the bond interfaces. If I add >> a system with 4 interfaces, sync takes about 6 seconds. If I add two >> bond interfaces, the sync time goes to around 20 seconds. If I add a >> second system with two bond interfaces, it times out (3 minutes). If >> I remove those bond interfaces, it syncs in around 6 seconds again. >> Note that I am not touching the slave interfaces, they are still >> pointing to bond masters that no longer exist. >> >> If I add 7 systems with a total of 34 interfaces plus 14 bond >> interfaces, the sync times out. If I remove all the bond interfaces >> and try again, the sync takes about 24 seconds. >> >> I went into dhcp.template and removed all the cheetah stuff. It made >> no difference at all, with one system or 7, bond interfaces or no. So >> the long delays occur before cobbler starts working with the dhcp >> template. >> >> Does anyone know what cobbler is doing at this point? What is >> different in how it treats bond interfaces? >> >> FYI I'm on RHEL 4.6 x86_64. I brought in a newer syslinux (version >> 3.72-2.el4.rf, the latest I could find). I went searching for a newer >> syslinux because the cobbler docs recommend doing that. As I test I >> went back to the syslinux version that came with RHEL 4.6 (version >> 2.11-1) and it made no difference in the timing. >> >> For a while I had another version of syslinux but it caused a core >> dump when cobbler tried to run gethostip. So I'm wondering if >> syslinux is causing these long delays. What does cobbler sync use it for? >> >> -- >> ... Chris Weaver > >My first thought it to turn off omapi_enabled in /etc/cobbler/settings >to see if this improves things -- OMAPI is rather buggy anyway and it >could be sending more commands than needed with the new version. (I'm >thinking complete removal of the feature, rather than deprecation, would >be a good idea WRT OMAPI support). > >syslinux is used to supply our network bootloader mainly, of course. >gethostip is also used to generate the filename segments in tftpboot, >particularly for encoded IP's when a MAC is used as the boot criteria >instead of the IP. > >I'm not sure why it would be slower when dealing with the bonded >interfaces, though a copy of "cat >/var/lib/cobbler/config/systems.d/$systemname" for one of the systems in >question may help me better understand what the inputs might be. > >First though, let me know whether omapi is enabled or not. That may >be choking on bad input, as everything else should just be restarting >the service once. > >It could also be that another component cobbler uses was broken by a yum >update. > >--Michael -- ... Chris Weaver _______________________________________________ cobbler mailing list [email protected] https://fedorahosted.org/mailman/listinfo/cobbler
