Daniel J McDonald wrote:
Ok, I get it. The master is listed in the wgDBServer
variable, and the
slaves in the wgDBServers array of arrays. Upon failure of the master,
it's a manual process to get writes working again,
Yes; we're not losing sales if the site's partially broken for a few minutes, so
it hasn't yet been worth it to invest time and money in setting up automatic
transparent failover for something that's very rare.
but reads should function fine.
Likely not; because reads need to function properly even when slaves are
slightly lagged, some items will check the master. For instance, on a page view
the tiny 'page' record is read from the master, then revision metadata, text,
and link-coloring information are pulled from a slave. Otherwise you can edit a
page and have it serve you back the previous version -- which we don't want then
stuck in a cache!
-- brion vibber (brion @
pobox.com)