Skip to main content
The agent is the eyes of the panel on the node. It regularly sends three types of reports: heartbeat (host sign and metadata), a snapshot of system metrics and service statuses from a fixed directory. From this data, the panel builds tabs Metrics and Services, calculates the status of the node and applies speed limits.
For whom: Panel Administrator, SRE. Required rights: Panel Administrator (reading metrics and services).

Frequency

The frequency of reports to the agent is reported by the server during registration; default values are: If the heartbeat doesn’t come in about 90 seconds, the panel tags the offline agent.

Metrics.

Each photo metric contains:
  • CPU - load in percentage, load average in 1/5/15 minutes, CPU pressure (PSI);
  • Memory - all / used, swap all / used;
  • Disk - total / used, read / write speed;
  • Network Accumulated rx/tx bytes and current speeds, errors and packet losses per second, TCP retransmits per second
  • Connections — TCP established/listen/time-wait, UDP sockets, counter and conntrack limit;
  • System system Open file descriptors and limit, number of processes, uptime, availability and inclusion of BBR, active queue discipline (qdisc).
The agent also transmits telemetry to the anti-abuse panel system, which is used by abuse detectors; it is not separately displayed in the node metrics. Storage: Metrics are stored by 7 days (by default), old records are cleaned automatically. Time series reading: The since setting defines the sample window; a maximum of 24 hours.

Channel quality and track verification

When measured, the panel shows RTT, jitter, and packet loss, separately from the speed of incoming and outgoing traffic. Channel estimation can take into account the bandwidth declared by the provider. The missing measurement is not replaced by zero and does not confirm a good channel. In the node card, you can run the path diagnostics: trace and measure MTU. The TTL tip relies on the measured path; it’s not a universal value for all directions. Check the selected target and the time of the result. Measuring latency or speed through Flow refers to a particular tunnel path and checkpoint.It does not guarantee the same speed to each site, nor does it equal the node interface.

Catalogue of services

The agent monitors the supported service directory, and the following are the main options; availability and operations depend on the agent’s version and installation method: Service statuses: If a service is not installed, it appears as absent (“Uninstalled”) rather than as “dead.” Rebooting the service from the panel takes into account its shape: containers are restarted through Docker, systemd units are restarted through systemd. Status picture:

Installation and operation of services

If the service is not installed and is supported by the directory, the interface may suggest an installation. Before the action, check the required capabilities of the agent and the chosen method of launch. An unavailable operation should not be considered successful just because the task is sent. Targeting one service doesn’t mean that the other services have disappeared from the node; match the last query area with the full image.

Heartbeats

Heartbeat is a sign of the life of the agent plus host and OS metadata and a health brief.Heartbeat is also involved in confirming updates: an agent update task is considered successful only after heartbeat with the target version (see Updates).

QoS: speed limit

When the anti-abuse panel system assigns a specific subscription speed limit, the agent on each node regularly (about every 30 seconds) picks up the list of active restrictions and applies them to at the level of the node core - the participation of proxy software is not required. Behavior:
  • Idempotent Synchronization - Restrictions removed on the server are automatically removed on the node on the next pass.
  • Inaccessible use If the node does not have the necessary tools, the agent reports that he can not use QoS, and the panel shows the restriction as “saved, but not applied”, rather than pretending that it works. Installation of an agent).

Verification

Tab Metrics. Shows fresh dots (not older than a minute) and tab Services current statuses, including absent for unspecified services.

Rollback

The agent’s reporting is telemetry, it has no separate switch; the only way to stop it is to remove the agent from the node (see below). Remove the nodeQoS restrictions roll back on the panel side: the removed restriction agent automatically removes from the node during the next synchronization cycle.

Limitations and risks

  • Metrics are stored for 7 days - for longer storage, unload them with external means.
  • The metric sample window is limited to 24 hours per request.
  • The capabilities of the agent (metric collection, application of QoS, etc.) are self-declared and depend on the tools installed on the node; the missing basic packages can be additionally installed by the provision operation from the Agent tab.

Typical errors

Check if the agent is online (Agent, systemctl status axottle-agent on the node). The procedure is in Diagnostics.
absent means “not installed on the node” - the agent simply did not find it. Check on the node if the service is installed and in the same form (systemd vs. Docker container).
The node does not have the necessary tools for QoS. Perform provision of missing packets from the tab Agent or reinstall the agent with a script - it resets the dependencies.
Related pages: Overview of nod, Updates, Diagnostics.