[EMAIL PROTECTED] [EMAIL PROTECTED] wrote: > = As usual, the system operating normally isn't half as interesting > = as the system failing :-) Could you describe acceptable and > = unacceptable consequences of failures? > = > = For instance, if one database goes down, is it acceptable to have: > = - a few seconds dataloss? > > Yes, a few seconds is fine, considering that the databases are presumed to be > solid performers, and DB server failures would be rare.
That makes it a lot easier. And considering you found MS SQL Server clustering expensive it might be good to know that this can also make it a lot cheaper. You should consider a setup where your website connects to a master DB and performs any queries it needs to do there. That DB is replicated to another DB on another machine using a replication daemon. Replication takes place every (few) second(s) by the replication daemon. From the webserver, monitor the database server. If you notice database connection errors or the number of queued queries suddenly jumping you know something is up with the database. At that point, you know you want to switch to the secondary database. Take an exclusive lock, wait 10 seconds just to be sure and change the application / server variables that contain the datasource information to point to the second database server. Downside: you probably need changes to the application code. > If there is a complete, brief outage, then that's fine, but I'd need to avoid > unsynchronized writes to both servers. That depends a bit on the database and the replication daemon. I am not sure if the replication build into MySQL is fully transactional. The Slony-1 replication engine for PostgreSQL is transactional so that shouldn't be a problem there. > = - a restore from last backup situation? > > That one sounds sketchy. Depends on the dataloss and the time to restore, I > guess. Again this is a point against MySQL Cluster. It is a memory-only solution and they have not yet implemented transaction logging so if too many nodes go down, you loose data. ISTM you basically have the choice to go the DIY route or ask for a quote from MS / Oracle / IBM. The wildcard might be the clustering in MaxDB, anybody have any experience with that? Jochem ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Logware (www.logware.us): a new and convenient web-based time tracking application. Start tracking and documenting hours spent on a project or with a client with Logware today. Try it for free with a 15 day trial account. http://www.houseoffusion.com/banners/view.cfm?bannerid=67 Message: http://www.houseoffusion.com/lists.cfm/link=i:4:203046 Archives: http://www.houseoffusion.com/cf_lists/threads.cfm/4 Subscription: http://www.houseoffusion.com/lists.cfm/link=s:4 Unsubscribe: http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=89.70.4 Donations & Support: http://www.houseoffusion.com/tiny.cfm/54
