I suppose the problem should be quite obvious...
any technique deal with this kind of delay?
On 05/01/07, howard chen howachen@gmail.com wrote:
I suppose the problem should be quite obvious...
any technique deal with this kind of delay?
Yes, the software itself. MediaWiki's load balancer is quite clever, in that it will pick a slave server which isn't too lagged, and make that available for read requests...our maintenance scripts will also wait for production servers to catch up when running large operations.
You might want to look at includes/LoadBalancer.php and includes/Database.php, although most of the intelligence exists in the former.
Rob Church
Howard,
I suppose the problem should be quite obvious...
any technique deal with this kind of delay?
Though we do have some delay detection and avoiding code, the primary principle is not to have replication delay on servers that are used for mediawiki. Let it cost a bit of consistency in case of huge crashes, but we have pretty async configuration of our db servers.
Domas
wikitech-l@lists.wikimedia.org