Agent not reporting: host down, agent dead, or the network path broke

Agent not reporting: the host vanished from Infrastructure

What you see: the host was there, now it is gone or stale. datadog.agent.up is missing or not OK.

Triage order

1. Is the host alive? SSH in. If the box is down, this is not a Datadog problem.
2. Is the agent running? systemctl status datadog-agent (or the platform equivalent). If dead, check why: OOM killer, disk full (the agent buffers to disk), a bad config push that fails validation on restart.
3. Did the config change? A recent datadog.yaml edit with bad indentation prevents startup. The agent logs the parse error; read them before reinstalling.
4. Network path to Datadog. From the host, reach the intake endpoints for your site. Proxy changes and egress firewall rules are the usual silent killers.
5. Clock skew. TLS to the intake fails with a wildly wrong system clock. NTP issues are in the troubleshooting index for a reason.

The flapping variant

Host appears and disappears every few minutes: usually duplicate hostnames (two machines reporting as one, each overwriting the other) or the agent restarting in a loop (config error plus a supervisor restarting it). Check the agent uptime in status: if it is always a few minutes old, it is crash-looping.

Fix and confirm

Fix the underlying cause, restart, then datadog-agent status should show the forwarder accepting payloads and the host reappears in Infrastructure within a couple minutes. Set a monitor on datadog.agent.up per host so the next one pages you instead of being discovered in a dashboard.

The agent getting-started docs define the datadog.agent.up service check as OK when the agent connects to Datadog, and datadog.agent.check_status as CRITICAL when a check cannot send metrics; these back Infrastructure presence. Verified against live docs 2026-09-26.