Hi, On Mon, Nov 24, 2008 at 05:33:22PM +1100, Anibal Monsalve Salazar wrote: > On Mon, Oct 27, 2008 at 11:18:39AM +0100, Lars Marowsky-Bree wrote: > >On 2008-10-27T13:31:41, Anibal Monsalve Salazar <[EMAIL PROTECTED]> wrote: > > > >>Please add the sgi-ipmi plugin > >> > >>The attached file is the sgi-ipmi plugin. > >> > >>It supports SGI x86_64 systems based on both Intel and Supermicro BMCs. > > > >Hi Anibal, > > > >thanks for this submission! Before merging it, one or two questions: > > > >First, we have an external/ipmi and a ipmilan fencing RA already; could > >this not be folded into these? It seems extremly similar to the > >external/ipmi one. > > We plan to use ipmilan. > > >># External STONITH module for ipmi. > >># This STONITH module takes two arguments: > >># $1 is the operation > >># $2 is the node to operate on > >># It takes a single environmental variable: nodelist > >># which is a space separated list of node definitions > >># Each Node definition consists of: > >># nodename;userid;passwd;bmctype;ipaddr1[;ipaddrn] > >># Where: > >># nodename is the hostname of a node > >># userid is the userid to use on the IPMI device > >># passwd is the password to use on the IPMI device > >># bmctype is the bmctype of the IPMI device (intel or supermicro) > >># ipaddr is the IP address of the IPMI device > > > >I wonder if this is a good interface for configuration. > > > >Would it not be cleaner to have one such resource per "nodelist" entry, > >so that these would indeed be separate parameters, which the user then > >could configure as such using the GUI/CLI? > > I'll start using ipmilan soon with the goal to raplace the sgi-ipmi > plugin with it. I have two questions. > > We have a minimum of two IPMI IP addresses per node. If one of them is > not responding, we use the other one for the reset operation. But > before the IMPI reset command, we ping the IPMI IP address. We do the > ping because we want to know in less than 3 seconds if the IPMI IP > address is responding to pings instead of waiting 50 seconds for an > IPMI command timeout. I'm planning to add a ping function to > ipmilan_command.c, is it a good idea?
Fine with me. But please have it as an option, sth like monitor_with_ping which would default to false. > How can I tell ipmilan that I have two two IPMI IP addresses per node? No way. Unless somebody implements it :) Isn't it a bit of an overkill to have two lights-out devices per node? If the device doesn't work the next monitor operation would fail and the admin should notice that. Thanks, Dejan > _______________________________________________________ > Linux-HA-Dev: [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev > Home Page: http://linux-ha.org/ _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
