hinted handoff pattern
When a node is temporarily down, other nodes accept its writes on its behalf and store hints. When the node comes back, the hints are forwarded to it. Trades a bit of complexity for higher write availability during failures.
When a node is temporarily down, other nodes accept its writes on its behalf and store hints. When the node comes back, the hints are forwarded to it. Trades a bit of complexity for higher write availability during failures.
symptoms
- write rejections during transient node failures
- convergence delays after node recovery
causes
- leader-or-replica unavailable during write
- dynamo-style availability requirement
fixes
- adjacent nodes accept writes + store hints
- deliver hints back when target recovers
- TTL on hints for safety
you might say
- take the hint
- buffered write for the dead node
- hinted handoff queue