Reconsider BridgeDB's pool assignment file implementation and deployment
While deploying the new BridgeDB feature that dumps bridge pool assignments to disk, I had two ideas for improving that feature. Neither of these ideas are critical, and I wanted to see how well the current implementation works before making it even more perfect. Maybe we should revisit these ideas in a month or two from now.
We only write to assignments.log after parsing a new network status file, but not after dumping unallocated bridges to file buckets. That means we might miss the fact that, say, Twitter distributes new bridges for up to 30 minutes between dumping to file buckets and reading the next network status. Does that matter? Should we also write to assignments.log after dumping to file buckets?
Appending to a single assignments.log file and rsyncing that to the host that sanitizes it won't scale forever. We could run a monthly or weekly cronjob that runs
"mv assignments.log assignments.log.old". metrics-db can handle multiple input files and will read both files, as long as they are rsync'ed correctly.