On Tue, Oct 20, 2009 at 8:55 AM, Tundra Slosek <ivoryring at gmail.com> wrote: > I have not yet. Am I misunderstanding my reading of the documentation, that a > quorum server is a single point of failure for the cluster (i.e. the > cluster's availability won't be any higher than the quorum server's). As it > stands, I'm a little surprised because my '4 node HA cluster' seems to be > less tolerant of failure than a pair of 2-node clusters would seem to be - > with one of the four nodes out of the cluster, a reboot of one of the > remaining 3 causes the other 2 to panic and reboot. I think I understand that > this is intentional to avoid a partition, but it really feels like '4 node > cluster' is no more available than '3 node cluster'. If I'm going to add a > 5th machine for no purpose other than to be the quorum server, would I be > better off making the 5th machine a 5th node?
The QS is a very light weight process (about on par with NTP) that can run on pretty much any other server outside of the cluster nodes with no impact. It can also be used as the QS for multiple clusters. Also, if you are truly worried about it being a SPoF, you could have more than one of them on different machines - or even an HA clustered QS. Though that would probably be taking things a bit far. > My current experiment, which I'm working on setting up, is to have the > Private Interconnect occur over physical NICs that are not shared for any > other purpose. If that doesn't work, I'll try a quorum server. Either way, > I'll keep this thread updated as I go. Unless you are using VLAN tagging, you are required to use dedicated NICs for the Interconnect. fpsm
