On Thu, 2006-07-13 at 14:45 -0500, Stephen Kirkpatrick wrote: > Lon Hohberger wrote: > > On Thu, 2006-07-13 at 12:14 -0500, Stephen Kirkpatrick wrote: > > > > > >> Thanks for the reply. I wondered if it was possible to have HA for > >> stateful applications. At least > >> I know my options better now. I can tolerate brief downtime, like the > >> time it would take a failover > >> node to take over, although no downtime would have been preferred. > >> > >> Any suggestions on HA solutions for a non-SAN deployment? > >> Stephen, Coming in on this thread late (so I may be missing some details), stateful HA application failover can be achieved by using something like rgmanager + checkpointing, or availability management framework + checkpointing. The rgmanager code is in the current cluster tree. AMF and Checkpointing APIs are available via the openais project (http://developer.osdl.org/dev/openais). At the moment rgmanager is in production state and AMF is more alpha-ish. checkpointing is production. No downtime is not possible - this would mean availability=100%. There is always some time associated with detection and recovery from a fault. The above openais project can detect application or node faults in the 15-30 msec range on 2.6 kernels. Node faults can be tuned to 40-60msec depending on number of nodes. This still provides high levels of availability and if an application is properly designed and tested, with standard models should provide 6 to 7 nines of availability. Regards -steve > > > > You don't need a SAN if you don't intend to share data. > > > > If you intend to share data but do not want a SAN, you'll need to set up > > DRBD or CLVM+Cluster Mirroring+GNBD (Note: cluster mirroring is still in > > a bit of development). The main advantage to the latter approach is > > simultaneous read/write access from both machines in the cluster. The > > advantage to DRBD is that it is easier to set up. > > > > Alternatively, you could use NFS mounts from a good NAS (a NetApp filer, > > as an example) as your source for shared dat. > > > > You can use linux-cluster or Linux-HA to do failover with or without a > > SAN; the choice is up to you. Linux-HA supports DRBD out of the box, > > but CLVM will still need parts of linux-cluster running in order to > > operate. > > > > If you want support, you can use Red Hat Cluster Suite (which is > > effectively linux-cluster w/ support). > > > > Good luck! > > > > -- Lon > > > > -- > > > > Linux-cluster@xxxxxxxxxx > > https://www.redhat.com/mailman/listinfo/linux-cluster > > > > > > > Thanks for the info. If I understood correctly what I was reading > about DRDB, it sounds like it runs entirely over an IP network. > This sounds like what I had originally envisioned for this project, > as it is a small deployment. > > I was envisioning SCSI storage in two nodes with individual > internal RAID arrays as depicted in the diagram below, > > > +----------------+ +------------------+ > | Primary server | | Secondary server | > | | | | > | ----------- |/ \| ----------- | > | RAID-1 |------------| RAID-1 | > | ----------- |\ | /| ----------- | > | | | | | > +----------------+ | +------------------+ > | > +-- Data sync over IP network? > > > I was thinking fibre channel storage when I said no SAN. The > cost of deploying fibre storage is a bit much for this small > project. However, if there is a good way to utilize shared > SCSI storage and a good HA solution to go along, I would be open > to researching it. > > In your response, you indicated that Linux-HA supports DRDB out > of the box. Does the Red Hat Cluster Suite support this too? > The Cluster Project FAQ just mentions the CLVM/GNBD solution. > > Thanks, > Stephen Kirkpatrick > > -- > > Linux-cluster@xxxxxxxxxx > https://www.redhat.com/mailman/listinfo/linux-cluster -- Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
