Reference
How it works
For IT and for the curious: what crosses the network, how often, what happens when the network goes away, and what protects the gateway.
What it sends, and when
Every call goes from the gateway to nexa.com.tn over HTTPS, signed in with the gateway token. Nexa never connects to the plant.
| What | When | Contents |
|---|---|---|
| Tag links | At start, then every minute | Fetched from Nexa: which tag is which signal of which machine. |
| Minute data | About 5 s after each minute | The minute’s values for every linked tag. See below. |
| Heartbeat | Every 60 s | “Still here”, with the gateway’s status: each OPC UA server up or down, the buffer, the last error, free disk, the sending rules and the update state. |
| Live values | Every 2 s, only while watched | Current values of the machines whose Live tab someone has open. Shown, never stored. |
| Tag list | When you click Send to Nexa | Tag paths and their current values, so you can link them. |
| Diagnostics | On request | Version, settings and status, and the end of the log, with tokens and passwords removed. Sent from the Run tab, or when Nexa support asks for it. |
Minute data
The gateway reads every change of a tag, then sums each minute up on the PC before sending. That keeps the traffic small however fast the machine runs:
- Measurements: the minimum, maximum, average and last value of the minute.
- Counters: how much the counter went up in the minute, so a PLC counter that wraps or is reset does not show as a negative production.
- Faults: when each fault started and cleared.
A minute goes in batches of at most 2,500 readings, so even a large plant sends a few small calls a minute.
Network cuts
Every batch is written to a buffer on the PC (%ProgramData%\nexa-edge\outbox.db) before it is sent, and removed only once Nexa has confirmed it. When the network goes, the gateway keeps reading and buffering, retries with a growing pause (up to two minutes), and sends the backlog, oldest first, when the network is back.
- The buffer holds up to 20,000 batches, about two weeks for a gateway sending one batch a minute. Past that, the oldest are dropped first.
- While the PC is off or asleep nothing is read, and Nexa shows a gap, not zeros.
- A service restart, for an update for example, carries the open minute and every counter over, so no production count is lost.
- If Nexa refuses the token, the gateway stops and says so rather than retrying forever. If the plan lacks telemetry, it keeps buffering.
How long Nexa keeps it
| Data | Kept |
|---|---|
| Minute data | 90 days |
| Monthly archive of the minute data | 2 years |
| Hourly summaries | 5 years |
| Alarms, downtime and setting changes | As long as the machine exists in Nexa |
Security
- Read-only. Nexa Edge has no way to write to a machine: there is no write call anywhere in it.
- Outbound only. One destination, HTTPS on port 443. No inbound port, no VPN, no remote access to the plant.
- Encrypted OPC UA. Signed and encrypted sessions (Basic256Sha256) with a user name, and a certificate each server must trust.
- Least privilege. The service runs as a virtual account with no password and no rights beyond its own folder.
- Secrets stay on the PC. The token and passwords are encrypted with Windows DPAPI for this PC, in a folder only administrators can read. Nexa itself keeps only a fingerprint of the token.
- Signed updates. An update installs only if it carries Nexa’s signature. See How an update is checked.
- The plant decides what leaves. Sending rules are set on the PC, and Nexa cannot change them.