Comment Twice a Week - Transparently via Replicated sites (Score -1) 182
We release on average twice a week at a new startup funded by AAA - heavy DevOps shop with fast developing online travel product suite (www.tstllc.net).
Infrastructure runs on a mix of local private cloud and public cloud. MySQL replication between the two geographically different locations. Only 1 site takes traffic at a time. Replication is stopped before a push, new code deployed to cold site, automated QA testing verifies code site updates are working correctly, traffic gets flipped across datacenters via DNS Failover (see dnshat.com for example), backup site goes hot, release is pushed to other site, verified, traffic shifted back to primary site again via DNS Failover, and lastly replication is resumed. Replication is stopped only if there are schemas changes that prevent the new release from going live with replication running (rare now we have the Devs up to speed on the failover architecture). Pushes are done via automated scripts - Java/Scala/Play apps. Total release time 1 hour - smooth as butter - transparent to site visitors and AAA club clients pulling webservices to their automated backend systems (we make sure their backend systems obey's DNS TTLs like end user browsers so that the release failovers are transparent).
Within a year the whole thing will automated for multiple releases per day if needed.