Talk:Flow/Architecture/Memcache

• "We must write to memcached before we write to the database." -- Keep in mind that this can result in temporary consistency issues. DB writes may fail do to lock timeouts or other reasons. Since reads always hit memcached first, this could result in keys being read and utilized within some contexts, without ever having been persisted. --Afeldman (talk) 21:15, 2 July 2013 (UTC)

Yes, this is one of my many concerns with the caching ideas. The idea that things should be written to memcached before the db came from terry, apparently he has used it in prior organizations. I am just now starting to flesh out what the caching system should look like from a programming standpoint, hopefully will be able to pick his brain regarding things like this because i'm not sure how you could possibly "undo" an update to memcache after a db failure without serious race conditions. One possibility that came to mind is Write-Behind, but I really think thats massively over-complicated and likely to create more issues than it solves. Ebernhardson (talk) 21:55, 16 July 2013 (UTC)