Failure and Latency at a Glance¶
| Event | single |
multi-node |
leader |
|---|---|---|---|
| Node loses primary role | idle, waits for promotion | n/a | n/a |
| Peer node dies (clean) | n/a | its rows adopted after wal_receiver_timeout |
next-lowest node becomes leader after wal_receiver_timeout |
| Sustained partition | one primary only (your HA layer fences the rest) | isolated node's own rows may duplicate | entire backlog may duplicate |
| Adoption/takeover latency | promotion time (set by your HA layer) | ~wal_receiver_timeout (default 60 s) |
~wal_receiver_timeout (default 60 s) |
This table pulls together everything covered in How Each Mode Works, Deciding If a Node Is Dead, and The Split-Brain Risk — use it as a quick reference once you already understand the detail behind each row.
Continue to A Note on the Spock-Facing SQL.