Robert Mustacchi wrote:
For what it's worth, at Joyent we do not try and do HA failover with shared storage at the OS-layer. Instead it's done at the granularity of the application layer and whatever makes the most sense. While it's true that there are many legacy applications that don't support this, we've found that going down the path of doing this in the application not only simplifies things, but also can cause less downtime for consumers of those services. It's also partially from the school of hard knocks from a bunch of us who worked on the Fishworks boxes at Sun and have gone down the centralized storage in other lives.
I quite agree with you. The place where I'm having trouble with application-layer redundancy is mail handling - specifically SMTP and our list manager - too many queues and items in flight, without checkpoints (looking for models, but haven't found any good ones yet). So, for those, running smtp and the list manager in a virtual machine, with DRBD synchronous replication to a failover machine works nicely, and reliably - but with a bit of annoying complexity (raid+LVM+DRBD on the storage side) - ZFS + DRBD over a pair of ZVOLs would sure simplify things.
Thanks, Miles -- In theory, there is no difference between theory and practice. In practice, there is. .... Yogi Berra ------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be Modify Your Subscription: https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4 Powered by Listbox: http://www.listbox.com
