On 01/12/2007, Noam Meltzer <[EMAIL PROTECTED]> wrote: > > On Dec 1, 2007 2:01 AM, Amos Shapira <[EMAIL PROTECTED]> wrote: > > > > On 30/11/2007, Noam Meltzer <[EMAIL PROTECTED]> wrote: > > > Hi, > > > > > > I had some experience with RHCS. Personally, it will be one of my last > > > choices for a good opensource HA software. (linux-ha first, actually) > > > > That's what I tried - but at least the packages on my test systems > > just crash and burn even running a simple "stonith -h" or > > "BasicSanityCheck". > Never felt the need for using stonith, so can't really know. Linux-HA just > worked for me great without it.
Me neither. I don't even have the hardware to use it. But it appears that "service heartbeat start" depends on it being installed and functioning. Or maybe I'm mixing up the failure of crmd to start on that system with something else... > > > > > > > > > It is very lacking in the area of Ethernet heartbeats (allows only one!) > and > > > > What do you mean by "allows only one"? > Exactly what it sounds like. That it allows only one heartbeat on the ether? Allows only one heartbeat medium or what? Because from what I've seen so far it's possible to have multiple cluster's heartbeats on the same network without interruption - just use different ports or different multicast groups or unicast, and the documentation recommends using multiple parallel heartbeat mediums (serial + ether, for instance) so a single medium failure won't cause a split-brain situation (where both nodes are alive and think that they are the master). > > > > > > > > > Anyhow, RHCS does have one good point which will make it the first OSS > > > candidate on certain scenarios which is GFS. > > > > Well - can't you configure GFS without it? > > No. GFS requires RHCS, it's one bundle. I see. Thanks for the clarification. I intend to configure it as master/slave anyway (at least at first stage) so I don't need a cluster-aware filesystem, but I'll keep your comments in mind if and when I need it. > > > > > > Also - I got the impression the OCFS is more recommanended by users on > > the net, but it doesn't come as a CentOS package - is it worth the > > trouble to install? > You need OCFS2. You can download RPMs from Oracle's website. > Though, I *think* that you will require oracle's cssd to use it. OK, thanks. > > > > > > > > > There is much more into it, but in order to give more information I > would > > > need more information about your application and its architecture. > > > > The main application just updates a SQLite file a lot. All we want is > > to have a stand-by node to take over and keep providing network > > service (reading and updating the file) if/when the primary node goes > > down. For this we share the file on DRBD but of course we need > > something to tell the secondary node to take control of the service. > > Uhm. Don't have actual experience with DRBD. Though, I find this technology > a bit scary. Personally I would prefer a real storage device (SAN or NAS) > for this kind of work (and make sure that the storage is highly available). > Then I can go to sleep at night ;-) > At this point I would prefer Linux-HA (or if you really have the money: > Veritas Cluster) That's the point - I don't have the money. Someone pointed me to a new 3000$ SAN not long ago (Geoffry?) but it's not in my budget for now. Also my previous employer (a very large company with hundreds of RedHat servers) used it extensively with no problem. Right now I finished "make" of the 2.1.2 sources but got a breakage in "make rpm". When I get that RPM built I'll try using it instead of the ones coming with CentOS and see what happens. Cheers, --Amos ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
