Node health
Figures below come from the lease table on page load. Incidents older than 90 days move to the archive; nothing is quietly removed.
| Opened | Scope | Summary | Duration |
|---|---|---|---|
| 2026-07-14 03:22 | node | NTP source drifted; skew reached 61 ms and the node refused lease decisions until resynchronised. No cluster lost quorum. | 22 m |
| 2026-05-30 18:07 | parcel-nl | Both voters partitioned from each other and from us. Neither side could reach quorum, which is the correct outcome; service resumed on healing. | 1 h 14 m |
| 2026-04-02 11:48 | node | Kernel upgrade and reboot inside the maintenance window. Clusters ran without a witness for the duration, as designed. | 6 m |
| 2026-01-19 09:31 | ledger-fra | Repeated elections from a flapping voter; 14 votes granted in 90 seconds. Rate limit added per cluster. | 3 h 02 m |
Availability counts the vote endpoint only. A cluster losing quorum for its own reasons does not open an incident here unless the failure was ours.
Median decision time is measured at the edge and excludes the network leg to the caller, which for most clusters dominates. Treat 11 ms as the floor, not the experience.
During maintenance the node returns 503 immediately so callers fail fast rather than waiting out a timeout.