Now that coronelli's back online, curly (the server Bomis loaned to cover coronelli's squid job briefly) will serve as a MySQL slave until the New Server to be ordered gets here.
We'll have to take the database offline for about 2 hours to make a clean transfer, but once up this will protect us against potentially much worse outages in the case of hardware or software failure on suda.
The transfer is currently scheduled for midnight UTC Sunday night/Monday morning (5pm PDT, 8pm EDT, 2 AM CEST). Make sure this is known...
The database will be _offline completely_. Squid should still push out cached pages to random visitors during this time.
-- brion vibber (brion @ pobox.com)
When the New Server gets here, will it be necessary to have this kind of outage again? It seems like a big flaw in the replication system if we have to go completely down to put a new machine online.
It's not even possible to have read-only access to the database while the new server is getting synced?
Brion Vibber wrote:
Now that coronelli's back online, curly (the server Bomis loaned to cover coronelli's squid job briefly) will serve as a MySQL slave until the New Server to be ordered gets here.
We'll have to take the database offline for about 2 hours to make a clean transfer, but once up this will protect us against potentially much worse outages in the case of hardware or software failure on suda.
The transfer is currently scheduled for midnight UTC Sunday night/Monday morning (5pm PDT, 8pm EDT, 2 AM CEST). Make sure this is known...
The database will be _offline completely_. Squid should still push out cached pages to random visitors during this time.
-- brion vibber (brion @ pobox.com)
Wikitech-l mailing list Wikitech-l@Wikipedia.org http://mail.wikipedia.org/mailman/listinfo/wikitech-l
On Apr 18, 2004, at 18:09, Jimmy Wales wrote:
When the New Server gets here, will it be necessary to have this kind of outage again? It seems like a big flaw in the replication system if we have to go completely down to put a new machine online.
This won't be necessary since at that point we'll be able to take the _existing_ slave offline to make a full copy of it, and let both slaves re-sync from the master's logs without interruption of service.
The problem is that our old slave backup fell out of sync while we were having all kinds of troubles, so it was necessary to make a clean, complete copy to start the new slave from. If the database is active while we're copying the files, the copy will be corrupt and completely useless.
(MySQL sells a commercial tool for doing complete hot backups of InnoDB databases, but we don't have it.)
-- brion vibber (brion @ pobox.com)
Suda and all wikis back online. I'll set up syncing on the slave after dinner.
-- brion vibber (brion @ pobox.com)
Curly is synced up. Hopefully we won't need it until it's time to transfer everything peacably to Mme. Geoffrin II...
-- brion vibber (brion @ pobox.com)
wikitech-l@lists.wikimedia.org