We are in the process of designing a High Availability PACS solution using PacsOne on a 2-node Windows 2003 server cluster attached to a shared SCSI storage.
The goal is to provide not only hardware failover, but also automated application failover for PacsOne, MySQL and possibly Apache.
Under Windows, PacsOne is very well supported by the cluster service, and is being passed from node to node flawlessly.
By simply registering the PacsOne service as a resource in the Cluster Administrator console, if PacsOneSrv.exe were to fail, the cluster service would start it on the failover node, and tranfer ownership of the SCSI shared storage.
For MySQL, we currently configured multi-master replication: both node continuously replicate each other.
This option works well for Network Load Balancing clusters and essentially provides continuous back up of the database.
Another design option would be to have the database tables on the shared storage, and register MySQL as a cluster resource along with PacsOne; this would provide automated failover of MySQL in the event of a crash.
We are still debating which of the two options to go for, and we'd like to have input from the community:
What seems the most common failure scenario with MySQL?
Database corruption, or plain crash?
Thanks.