Skip to main content

If the cloud or the internet is unreachable

The Collector keeps monitoring your devices while it can’t reach the cloud:
  • Telemetry and events are buffered on disk and sent, in order, as soon as the connection returns. The buffer holds up to 72 hours or 256 MB, whichever comes first, and survives a restart of the Collector or its host. Once the buffer is full, the oldest data is discarded first.
  • Alerts raised during the outage reach the portal when the connection returns. Notifications go out only for alerts less than 15 minutes old, so nobody is paged hours late about an alert that may already have cleared.
  • The device list is cached on disk, encrypted with the Collector’s key. If the Collector restarts during an outage, it carries on polling the devices the cloud last gave it.
  • Commands sent from the portal wait up to 10 minutes for the Collector. If it doesn’t pick a command up in that time, the command is cancelled and the portal says so. A command can’t run hours later, when the Collector comes back.
Buffer locations: /var/lib/av-bridge/spool on Linux, and a spool folder next to the state file on Windows. To change the limits, set cloud.spool_dir, cloud.spool_max_bytes and cloud.spool_max_age in config.yaml.

Back up and restore

The Collector keeps only a small amount of local state: the last-known status of each device, the cached device list, and any buffered telemetry. When it restarts from scratch, it downloads its device list from the cloud and rebuilds current device status within one poll cycle.
You don’t need to back up the Collector. Redeploying is safe. The only file worth keeping is the configuration, because it holds the Collector’s identity and HMAC key. If it’s lost, re-enrol the Collector with a new token (see Secure the configuration).

Replace a failed Collector

If a Collector’s host fails, you don’t need to recreate its devices. Either:
  • Reuse the same Collector. Install the Collector on the new host and enrol it with a fresh token from the Collectors page. Enrolling issues a new key, so the old host can’t report as this Collector any more, even if it comes back.
  • Enrol the new host as a new Collector, then move the devices. On the Collectors page, select Replace on the failed Collector and choose the new one. Every device moves across with its history, alerts and settings. Both Collectors pick up the change within seconds. Then delete the old Collector.
To move a single device, edit it and choose a different Collector. A move is refused if the target Collector already has a device with the same reported ID.

Automatic failover with a standby Collector

For sites where monitoring must continue through the loss of a Collector host, add a standby: a second Collector, on a separate machine that can reach the same devices, which takes over automatically.
  1. On the Collectors page, select Add standby on the Collector you want to protect.
  2. Install the standby on the second machine with the enrolment command shown, exactly as for any Collector.
The two machines then work as a pair:
  • One machine polls at a time. The active machine polls the devices; the standby stays connected to M.A.R.C.U.S. Cloud but polls nothing. Each shows Active or Standby on the Collectors page. Devices only ever have one poller, which matters for devices that accept a single control connection.
  • Failover is automatic. If the active machine stops reporting, the standby takes over its devices within about 90 seconds. Device history, alerts, settings and pending commands carry on as normal, and the change is recorded in the audit log.
  • No automatic failback. When the original machine comes back, it becomes the standby. To move the devices back, select Make active on it; the active machine stops first, and the other starts about 10 seconds later.
  • Safe when a machine is cut off. An active machine that loses contact with M.A.R.C.U.S. Cloud stops polling its devices after 40 seconds, before the standby can take over. If the whole site loses its internet connection, a pair therefore pauses monitoring until the connection returns; a single Collector keeps monitoring and buffering instead.
  • Alerts. Losing one machine of a pair raises a warning, not a critical alert, because the other is still covering: “Standby collector offline” means there’s no failover available until it’s back.
Devices always belong to the primary Collector, so add and move devices there. To retire a pair, delete the standby first.

Updated 23 Oct 2026 15:20 · page 4f43977