spark standalone cluster
- statically defined cluster in conf...boo
- master sshes to slaves to do things...could overcome via running sshd on dyno private ip port 22 and having keys set up, but see above
spark on mesos: boils down to running mesos on dogwood
- tough to get working binaries without full docker images, probably surmountable
- dealbreaker: mesos master has a replicated log file. so restarting all master 'dynos' at once kills the cluster.
- workaround: feature-flag in bind-mounting a directory in runtimed that survives dyno restarts, store replicated log there
still less than ideal