Test a probe
A key with write:metrics or write:config may call it. Validates source_config with the same rules as a metric write, then runs the probe once: on the given agent (on its network, with its env-var secrets), or, with no agent_id, from Observer’s cloud for a public HTTP check. Waits up to wait_seconds (default 20) for the agent; if it has not answered by then the response is 202 with a job to poll via getProbeTest. Nothing is stored: no metric, no sample, no status change. The result carries only the value, its unit, the status the given thresholds would produce, a reason code and the duration; never response bodies, rows or headers.
Value units by source_type:
- prometheus: Value: scalar result of the PromQL query, as returned by the query; Observer does not rescale it.
- http: Value: ms: response time in milliseconds. With json_path set: the number extracted from the JSON response at json_path (booleans read as 1/0); unit as in the response.
- tcp: Value: ms: TCP connect time in milliseconds.
- dns: Value: ms: DNS resolve time in milliseconds.
- icmp: Value by interpretation: interpretation=latency (default) → ms: average round-trip time in milliseconds; interpretation=packet_loss → %: percentage of pings lost, 0 to 100; interpretation=reachability → 1 when any ping returned, 0 when none did.
- tls_cert: Value: days: days until the certificate expires.
- grpc: Value by interpretation: interpretation=health_state (default) → 1 when the health check reports SERVING, 0 when NOT_SERVING; interpretation=latency → ms: health check round-trip time in milliseconds.
- websocket: Value by interpretation: interpretation=handshake_latency (default) → ms: handshake time in milliseconds; interpretation=round_trip_latency → ms: message round-trip time in milliseconds; interpretation=connection_success → 1 when the socket opened, 0 when it did not.
- mtls_http: Value: ms: response time in milliseconds (deprecated alias of http). With json_path set: the number extracted from the JSON response at json_path (booleans read as 1/0); unit as in the response.
- database: Value: scalar returned by the read-only query (one row, one column), as returned by the query; Observer does not rescale it.
- otlp: Value: aggregated value of the pushed OTLP data points, in the instrument’s own unit.
- cloudwatch: Value: latest CloudWatch datapoint for the chosen statistic, in the metric’s CloudWatch unit.
- custom: Value: number returned by the registered probe function, in whatever unit it uses.
- loki: Value: scalar result of the LogQL aggregation, as returned by the query; Observer does not rescale it.
- elasticsearch: Value: value of the named aggregation, as returned by the query; Observer does not rescale it.
- host: Value by metric: metric=cpu (default) → %: CPU utilisation percentage, 0 to 100; metric=memory → %: used memory percentage, 0 to 100; metric=filesystem → %: used disk space percentage of the mountpoint, 0 to 100; metric=network → B/s: network throughput in bytes per second (received + sent); metric=load → 1-minute load average divided by the CPU count.
- manual: Value: no measured value; the status is set by hand or by an open incident.
- heartbeat: Value: s: seconds since the last successful ping.
- alert: Value: worst firing alert as a score: 0 none, 1 degraded, 2 unhealthy.
Authorizations
An organization API key (obs_pub_…) sent as Authorization: Bearer <key>. Each key carries scopes; an endpoint lists the scope it needs. read:change_events is reserved: keys can carry it, but no endpoint needs it yet.
Body
The probe type. custom, otlp (push), manual, heartbeat and alert cannot be tested.
prometheus, http, tcp, dns, icmp, tls_cert, grpc, websocket, mtls_http, database, cloudwatch, loki, elasticsearch, host The same object a metric's source_config takes for this source_type (see the SourceConfig schemas used by applyConfig); validated the same way before anything runs.
The Observer Agent to run the test on (listAgents). It must be online and run agent 1.7.0 or later. Omit it to test a public HTTP check from Observer's cloud instead (header_refs secrets are not sent then); every other source type needs an agent.
Optional. With all four set, status says what the metric's status would be. Strict comparisons: over is >, under is <, equal is =.
How long to wait for the agent's answer before returning 202 with a job to poll. Cloud tests always answer at once.
0 <= x <= 20Response
the probe ran
What the probe measured, in unit. null when it got no value (see reason).
Unit of value (ms, %, days, s, B/s), or null for a bare number. Same rules as a metric's unit.
The status these thresholds give the value; no_data when the probe got no value. null when a value was measured but thresholds were not all given.
healthy, degraded, unhealthy, no_data, null How long the probe took, in milliseconds.
agent: ran on the agent's network with its secrets. cloud: a public HTTP check run from Observer's cloud.
agent, cloud What the value is, in words.
Why there is no value: the agent's reason code (e.g. ETIMEDOUT, unexpected_status:503). null on success.
Short human description of reason.
What usually causes reason and how to fix it.
The agent job (agent mode only).
Cloud mode only: header_refs names that were not sent, because their values live on an agent host.

