distributed system pattern to perform repairs in the write path, post-failure data (as hints - temporary records that hold write operations) comes from backup node (coordinator) to target node when it is healthy again
some DBMSs store hints for a certain time frames, and flush to disk-based storage, some DBMSs store in local directory
hints can be rejected if target node remains unavailable, or gets decommissioned?
sloppy quorum - hinted handoff
traditional quorum is not fault-tolerant when network partition happens or some nodes become unavailable (constraint of r/w nodes cannot be reached)
sloppy quorum leverages hinted handoff (temporary backup nodes accept writes/hints) to reach quorum -> keep availability for writes
drawback of sloppy quorum is clients may read stale data (delayed consistency - reads come before hints are delivered to designed nodes), DMBSs dont allow read directly from coordinators
