Skip to content

Fleet monitoring

Once an agent is enrolled, it reports basic host telemetry back to Sencai. This page covers where to see it and how to read it. For patch and software data specifically, see Patch management and Software inventory.

Two related but different things show up in the app:

  • Connection status - shown on Fleet → Agents (/gravity/agents): whether an agent is currently connected, and when it last sent a heartbeat. This comes from the agent’s own heartbeat, sent every 30 seconds, and is what tells you a host is reachable at all.
  • Host telemetry - shown on Fleet → Agent Telemetry (/gravity/fleet/telemetry): CPU, memory, and disk usage per agent, refreshed every 60 seconds.

A host can show as connected on the Agents page while its telemetry table is empty or stale - the two are collected on separate cycles, and telemetry collection is a platform-side setting for your environment rather than something you turn on per agent.

ColumnMeaning
Agent ID / Agent NameWhich host the row is for
CPU %CPU utilization at the time of the snapshot
MemMemory used, shown in MB or GB
Disk %Disk utilization at the time of the snapshot
Last SeenWhen this snapshot was recorded

CPU above 80% and disk above 90% are highlighted so a host under real pressure stands out without you having to scan every row.

The table refreshes automatically every 60 seconds while you have it open, and you can force an immediate refresh with the Refresh button.

On Fleet → Agents, a host marked active is treated as stale - and flagged visually - if its last heartbeat is more than about 90 seconds old. In practice this usually means one of:

  • A brief network blip - the agent’s own reconnect logic (with backoff) generally recovers within a minute or two without you doing anything
  • The host is under enough load that the agent process itself is struggling to run its own heartbeat loop on schedule
  • The agent process has stopped or the host has gone down entirely

If a host stays stale for more than a few minutes, it’s worth checking directly (SSH, or the browser terminal if it’s a registered cloud instance) rather than assuming it will recover on its own.

Telemetry here is read-only - it tells you a host is under pressure, but doesn’t do anything about it by itself. If a host is consistently showing high disk usage, for example, you can dispatch a cleanup action against it directly from Remote execution (the built-in clear_disk_space runbook action covers exactly this case) rather than SSHing in by hand.

Telemetry here is a snapshot, not a live stream - treat a single reading as a point in time, not a trend. If you need a history of a metric over time, or threshold-based alerting so you don’t have to watch this table yourself, see Monitoring under Operations, which covers platform-wide alerting and incidents.