Good article explaining my argument: http://openlife.cc/blogs/2012/september/failover-evil
TL;DR Cluster failover sucks. On Mar 12, 2013 9:21 PM, "Jason Davis" <scr...@gmail.com> wrote: > Galera MySQL might be an option, depending on how fast the round trip is > between the two sites. > > I will confess that I am NOT a fan of Linux HA clustering. I would > recommend doing replication at the MySQL layer. Linux HA clustering is > complex and fragile... Our big LMS at the university runs on such a cluster > (RHEL cluster + DRBD) and its more trouble than its worth. I am working to > hopefully move this to Galera. In our testing it has been way easier and > more resistant to fault. > On Mar 12, 2013 9:00 PM, "Mathias Mullins" <mathias.mull...@citrix.com> > wrote: > >> Hi John, >> >> So in a lot of the installations that we have worked with, we are seeing a >> lot of work with DRBD (Active/Standby, Active/Active w/Pacemaker). So far >> it's been one of the better solutions that we have seen installations >> stable with. I think a lot of people would like to see this functionality >> written into ACS to support something in addition to MySQL Master/Slave >> replication. >> >> Personally I like the DRBD solution because it deals with the replication >> at a block level and gives us the flexibility of active/passive and >> active/active ability versus having to do a lot of manual work to revert >> the master/slave relationship. >> >> Thanks, >> Matt >> >> >> On 3/12/13 9:32 PM, "John Kinsella" <j...@stratosec.co> wrote: >> >> >Thought I'd throw this question out on the list - we're working on >> >geographically (1500 miles) replicated databases for customers who really >> >don't want their stuff to go down. >> > >> >In this particular case, how they've architected their DB schemas means >> >they're really not friendly to the standard transactional replication >> >(this is MSSQL server) so we're looking at replicating the whole block >> >device the db is stored on. >> > >> >We've tried usingŠ >> > * Ceph - know it's not meant for geo-repŠloving it locally, though! >> > * Gluster - When clustering across the WAN, performance was weak. Don't >> >have enough disks/nodes to create a gluster cluster to replicate with >> >gluster-geo-replicate >> > * drbd in active/standby - Fairly decent performance. I hear it'd be >> >better with the drbd-proxy, but don't feel like spending the $$ yet. I've >> >been previously shot in the foot enough times with drbd active/active >> >that I won't try that again. >> > >> >Currently using drbd (I'm pondering writing management of the >> >primary/secondary stuff into ACS) but curious if others have found ways >> >of doing this with ACS that they like? >> > >> >John >> > >> >Stratosec - Secure Infrastructure as a Service >> >o: 415.315.9385 >> >@johnlkinsella >> > >> >>
