Yes, in the moment I hadn't thought about mis/poorly-configured spanning tree holding a port down. I agree it would be more robust to background a poll of ethtool for each interface and then bring dhclient up (and perhaps down?) as appropriate.
Perhaps a cleaner approach would be something like ifplugd or netplug. Although there are drawbacks - ifplugd requires libdaemon and netplug depends on the per-driver implementation in the Linux kernel. On 03/15/2013 03:49 PM, Jarrod B Johnson wrote: > I agree with the ethtool check. Was more thinking about the DHCPv6 > stuff. Was hoping to make IPv6 'just work' without configuration if > available without a lot of exotic configuration. > > I'd have ethtool probably sit in a loop on the various nics periodically > checking for link. Not only are there cases where a cable is actually > unplugged. With some NICs and link scenarios link negotiation actually > delays link up long enough to falsely detect as unplugged even that late > in the cycle. > Inactive hide details for "Daniel M. Weeks" ---03/15/2013 02:00:19 > PM---I can agree to some extent - you likely will have the n"Daniel M. > Weeks" ---03/15/2013 02:00:19 PM---I can agree to some extent - you > likely will have the necessary information on the DHCP server. Howe > > From: "Daniel M. Weeks" <[email protected]> > To: xCAT Users Mailing list <[email protected]> > Cc: Jarrod B Johnson/Raleigh/IBM@IBMUS > Date: 03/15/2013 02:00 PM > Subject: Re: [xcat-user] genesis: limit DHCP startup to connected interfaces > > ------------------------------------------------------------------------ > > > > I can agree to some extent - you likely will have the necessary > information on the DHCP server. However, if you put yourself in the > place of an installer or sysadmin: would you want the additional > complexity of looking in another separate log when diagnosing a problem > or "following along" during a boot, having to match timestamps or > interleave 2+ logs? Furthermore, consider a situation where that DHCP > server is handled by a different person or group within the organization > and access to those logs is restricted - there would be another big > hurdle to solving what could otherwise be very straight-forward. > > Besides that, isn't it just silly to bring up DHCP on an interface that > doesn't have the cable? (Yes, always having it running covers the > situation where a machine is booted and then a cable is plugged in > afterwards. However, I think connecting to the node on the boot NIC > (which will always have DHCP running) and starting DHCP on the second > interface is an order of magnitude less work than connecting the cable > already was. This is even more true if it's multiple systems or multiple > cables thanks to psh.) > > On 03/15/2013 01:39 PM, Jarrod B Johnson wrote: >> I would personally be leaning toward suppressing dhclient induced >> logging messages or at most logging them to ramdisk. Given the fact >> that the server also logs and the dhcp server is almost always a server >> within convenient reach, I don't think the syslog having dhclient >> messages from genesis is a big enough value given the cost. >> >> Inactive hide details for "Daniel M. Weeks" ---03/15/2013 01:24:48 >> PM---This is a follow-up to my last patch for syslog, motiva"Daniel M. >> Weeks" ---03/15/2013 01:24:48 PM---This is a follow-up to my last patch >> for syslog, motivated mainly by the amount of log cruft present >> >> From: "Daniel M. Weeks" <[email protected]> >> To: xCAT Users Mailing list <[email protected]> >> Date: 03/15/2013 01:24 PM >> Subject: [xcat-user] genesis: limit DHCP startup to connected interfaces >> >> ------------------------------------------------------------------------ >> >> >> >> This is a follow-up to my last patch for syslog, motivated mainly by the >> amount of log cruft present after resolving that issue. >> >> dhclient is started on all interfaces during /bin/doxcat whether those >> interfaces are connected or not. Since ethtool is present in the genesis >> image for x86 the attached patch uses it to check for a link and only >> starts dhclient on connected interfaces. >> >> If there is a good way to configure/determine if IPv6 is NOT used in the >> environment it would also be good to skip starting the v6 copies of >> dhclient when it is unnecessary. Before I go and add something >> unnecessary, is there currently a clean way to make this determination >> (perhaps guided by the site table)? Thanks. >> >> -- >> Daniel M. Weeks >> Systems Administrator >> Computational Center for Nanotechnology Innovations >> Rensselaer Polytechnic Institute >> Troy, NY 12180 >> 518-276-4458 >> [attachment "doxcat-only_connected.patch" deleted by Jarrod B >> Johnson/Raleigh/IBM] >> > ------------------------------------------------------------------------------ >> Everyone hates slow websites. So do we. >> Make your web apps faster with AppDynamics >> Download AppDynamics Lite for free today: >> > http://p.sf.net/sfu/appdyn_d2d_mar_______________________________________________ >> xCAT-user mailing list >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/xcat-user >> >> >> >> > ------------------------------------------------------------------------------ >> Everyone hates slow websites. So do we. >> Make your web apps faster with AppDynamics >> Download AppDynamics Lite for free today: >> http://p.sf.net/sfu/appdyn_d2d_mar >> >> >> >> _______________________________________________ >> xCAT-user mailing list >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/xcat-user >> > > > -- > Daniel M. Weeks > Systems Administrator > Computational Center for Nanotechnology Innovations > Rensselaer Polytechnic Institute > Troy, NY 12180 > 518-276-4458 > > -- Daniel M. Weeks Systems Administrator Computational Center for Nanotechnology Innovations Rensselaer Polytechnic Institute Troy, NY 12180 518-276-4458 ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_mar _______________________________________________ xCAT-user mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/xcat-user
