Am 12.07.2013 21:25, schrieb Marc A. Pelletier: > On 07/12/2013 03:26 PM, Platonides wrote: >> Except that we could have several IPs per cluster (as TS does). > > I'm pretty sure we have no plans to do that ever; if we ended up running > out of resources, it'd be much simpler to spin off some databases off to > a new server entirely than set up multiple replication with roundrobin.
No redundancy for failover? You just trust the one instance to give us five nines? > In fact, that's a very good example of a situation where trusting that > there is a stable map between shard and database would bite your behind. > :-) There is no assumption that this map is stable. There's just a central place for a lookup. I understand that you are reluctant to add maintenance overhead for the db-to-shard mapping, but my experience tells me that there will be a lot more maintenance pain if you don't. The difference is just that it'll be the users who suffer it. -- daniel _______________________________________________ Labs-l mailing list Labs-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/labs-l
