observe_metrics_query

Tool

observe_metrics_query

Read one metric of the current product as a time series.

Read one metric of the current product as a time series. Cumulative counters are differenced at read time, so a process restart reads as its own increase rather than as a negative spike; histogram percentiles are interpolated across the emitter's buckets. Up to 100 series and 500 points each, over at most 31 days; truncated says when a ceiling was reached.

Behavior

HTTP: GET /observe/metrics/query

Arguments

Name	Type	Required	Description
to	string	—	End of the window, ISO 8601; default now
from	string	—	Start of the window, ISO 8601; default one hour ago
name	string	•	The metric to read, exactly as observe_metrics_list names it
step	string	—	Bucket width; default the finest that renders at most 500 points
groupBy	array	—	Attribute keys to keep as the series labels, aggregating every other attribute into them; default one series per distinct attribute set
service	string	—	Only points from this service
attributes	object	—	Only points whose attributes match every one of these key/value pairs
aggregation	string	•	How to aggregate each step. A counter takes "rate" (per second), "increase" (over the step) or "value" (its last reading); a histogram takes "p50", "p90", "p99", "avg" or "count". A pair that does not match the metric is rejected.
environment	string	—	Only points from this environment

Response

Field	Type	Always	Description
to	string	•	End of the window that was read, ISO 8601
from	string	•	Start of the window that was read, ISO 8601
name	string	•	The metric that was read
step	string	•	The bucket width the series is rendered at
type	string	•	What the metric holds: a counter ("sum") or an explicit-bucket histogram
unit	string	•	Unit as the emitter declared it; empty when none
series	array	•	One series per label set, ordered stably
truncated	boolean	•	True when a ceiling cut the answer — more series exist than were returned
aggregation	string	•	The aggregation that was applied
stepSeconds	number	•	That width in seconds — what a rate was divided by

Try it

Read one metric of the current product as a time series.

When to use

Example

Ask your agent something like this:

Read one metric of the current product as a time series.

The agent will invoke observe_metrics_query with these arguments:

{
  "name": "<name>",
  "aggregation": "<aggregation>"
}