clouderra
Service status
For a standby node the interesting metric is not uptime but freshness: how far behind the primary this copy is allowed to drift, and whether it actually stays inside that bound.
Components
| Component | State | 30d availability | Lag behind primary |
|---|---|---|---|
| answers udp | operational | 99.997% | — |
| answers tcp | operational | 99.996% | — |
| transfer from primary | current | 99.94% | 0 serials |
| probe agent (eu) | operational | 99.91% | — |
| promotion readiness | held | — | — |
Promotion is held on purpose, not blocked by a fault. The node will stay out of the
delegation until the data comparison in verify has been clean for
thirty consecutive days.
Change and incident log
| Date | Duration | Component | Summary |
|---|---|---|---|
| 2026-06-28 | — | tls | Certificate for the web front-end reissued. Unrelated to DNS service, which does not use it, but worth noting because the two are often confused in reports. |
| 2026-06-09 | — | rate limiting | Response rate limiting enabled after this address appeared as a spoofed source in a small amplification attempt. No impact on legitimate resolvers, which retry over TCP. |
| 2026-05-31 | 3 h 12 m | transfer | Transfers refused by the primary after a key rotation on our side that was not mirrored there. Served copies stayed correct throughout; the lag reached 1 serial before the key was fixed. Rotation is now a two-sided checklist. |
| 2026-05-04 | 22 m | answers udp | Host reboot for a kernel update. TCP recovered first; the gap between the two is why the availability figures above are tracked per protocol. |
| 2026-03-16 | — | probe agent | European probe endpoint added on this node. Before that, all RTT figures in our capacity planning were measured from Moscow only, which flattered the European paths. |
How lag is measured
Every five minutes a checker asks the primary and this node for the SOA of each zone and compares serials. Lag is reported in serials, not in seconds, because a zone that has not changed in a week is not stale — it is simply unchanged, and a time-based figure would misrepresent that.
A lag above one serial for longer than one refresh interval raises an operator alert. That threshold is what the held state above is guarding: a node that cannot reliably stay within it has no business being in a delegation.