On Mon, 30 May 2011 11:06:45 -0400, Digimer <linux@xxxxxxxxxxx> wrote: > On 05/30/2011 10:49 AM, Hiroyuki Sato wrote: >> Hello Digimer. >> >> Thank you for your advice. >> It is very very useful information for me. >> >>> a) Forcing a node to power off, or does it just start an ACPI shutdown? >> >> Maybe ok. I'll test it. > > To test, hang the host (echo c > /proc/sysrq-trigger), then try to force > it to power off with wol. If this succeeds, you are in business. I have > my doubts though. > >>> b) Can you check that the node is successfully off using wol? >> >> I'm not sure, I'll test it. > > Please do. If you can though, it will make IPMI far less needed. :) > >> Could you tell me one more thing. >> >> Where fenced will call fence agent?? >> It is mean that the following >> >> * Can I check where fenced daemon will call fence_agent when I >> execute fence_node?? >> (that message send to master fenced, or localhost??) >> * And Can I check ``where are master'' with command?? (If fenced is >> master-slave type) >> * Can I control master priority. >> (for example I want to specify gfs1, gfs2, gfs3 as fenced master) >> >> Thanks again >> >> Regards. > > I'm not sure about the internals of cman, so I am not sure which machine > actually sends the fence command. I do know that it has to come from a > machine with quorum, and I do believe it is handled by the cluster > manager. It's not like pacemaker where a DC is clearly defined. > > I'll try to sort out how the internals work and will let you know. Not sure where i got this information from (i think it was on this list), but for sure: the node with the lowest ID, which is quorate, will take the responsibility to call the fencing script -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
