Our strategy has always been to use as little higher abstractions from cloud providers as possible. Glad we went this way, saved us quite a bunch of SLA breaches today! I am confident to say that it's "best of both worlds". We get great availability zone redundancy by AWS without having to rely on and pay for all those PaaS stuff the cloud giants offer. Also, we can "fairly easy" migrate to any other cloud provider because we only need Debian instances running.