Sudden (huge) spike in Oplog GB/hour on primary member rendered secondaries in "RECOVERY" state. There was no spike in Opcounters, so no idea what went wrong there. Since it's a production db, I wasn't able to wait to collect more information about the issue.

I have added the relevant MMS graphs. Notice the "Oplog GB/hour". First one is primary member, others are secondaries - all running v3.0.9 WiredTiger.