Appearance
Monitoring
How the KriptoNyx chain and the services around it are watched, and how a failure is told apart from noise. The live result is on the status page.
Overview
A health sweep runs every 60 seconds and makes 39 checks across four areas: the chain and node, the RPC endpoint, each public service, and the infrastructure underneath them. Every sweep is recorded, whether or not anything failed.
The monitor only reads. It holds no keys and signs nothing, so it can be restarted at any time without affecting the network.
What is checked
Chain and node
| Check | Fails when |
|---|---|
| Node service | The node process is not running |
| Synced | The node reports that it is still catching up |
| Blocks being produced | The newest block is more than 60 seconds old. The chain produces one about every two seconds |
| Height advancing | The chain height has not moved since the previous check |
| Peers | The node has no peers |
Why "height advancing" is a separate check
A stalled node still answers every request, so checking that it responds proves nothing. Only the height standing still reveals a wedged node — which is why liveness is measured by the chain moving, not by a port being open.
RPC
Checked on both the public endpoint, https://rpc.testnet.kriptonyx.com, and the node's own RPC.
| Check | Fails when |
|---|---|
| Answers | A block-height request does not return a height |
| Response time | A request takes longer than two seconds |
| Serving chain 3009 | The endpoint reports any chain other than 3009 |
| Up to date | The public endpoint is more than 10 blocks behind the node |
Why the chain ID and lag are checked
The dangerous RPC failure is not being down — it is answering with the wrong chain, or with stale state. A wallet cannot tell either from the truth, so an endpoint doing that is worse than one that is plainly unreachable.
Public services
Each public site is fetched and checked against the response it is meant to give, not merely for a successful reply: the explorer, staking, DEX, ICO, faucet, bridge, API and RPC hosts.
Some hosts are correct to refuse a plain request — the API answers 404 at its root and the RPC endpoint answers 405 to a GET. Those are their expected responses, so they pass. Treating them as outages would produce alerts nobody could act on, and teach everyone to ignore alerts.
Infrastructure
| Check | Covers |
|---|---|
| Host resources | Disk, memory and load, each against a set limit |
| Application processes | Every process that serves a public site is running. Restart counts are recorded, so a process crash-looping while still reported as running is visible in the log |
| TLS certificates | Every public host's certificate is read directly and flagged at 14 days from expiry |
Certificates renew automatically. They are checked independently of that renewal, so a renewal that silently fails is caught well before the certificate lapses rather than discovered when it does.
Failure detection
A check is reported down only after failing three consecutive times. One failed request is noise; three minutes of failures is an outage.
| Rule | Effect |
|---|---|
| Three consecutive failures | The check goes down and an incident opens |
| While it stays down | Nothing further is reported |
| First healthy check afterwards | The incident closes, reported once |
| An inconclusive check | Recorded as unknown. It never opens an incident, and never closes one |
So a flapping endpoint produces two events — down and recovered — rather than one on every check.
This is the output of the monitor's own failure-detection logic, run over a check that fails three times, is once inconclusive, then recovers:
| Sweep | Check result | Event | Shown as |
|---|---|---|---|
| 1 | Failing | — | degraded |
| 2 | Failing | — | degraded |
| 3 | Failing | down — incident opens | down |
| 4 | Inconclusive | — the incident stays open | unknown |
| 5 | Healthy | recovered — incident closes | operational |
| 6 | Healthy | — | operational |
The degraded state exists so a failure is visible from the first check, even though it is not yet counted as an outage.
Logs
- Every check is recorded, not only failures. The record shows what was healthy as well as what was not, which is what makes an incident's start and end times trustworthy.
- Kept for 30 days, rotated daily and compressed.
- The incident history is rebuilt from this log at start-up, so restarting the monitor loses nothing.
- Node, application and web-server logs are kept separately by each service.
What is and is not public
The status page shows the state of each check, never its measurement. It says the host is healthy; it does not say how full the disk is. Incidents are published under the same coarse labels, so the public history never names an internal process, disk or port.
Full measurements, thresholds and the detailed operational runbook are kept by the operators and are available on request.
Monitoring your own node
This page describes how the network's operators monitor the platform. If you run a validator, monitor your own node too — see Operating a node.