The data used to generate the results of a metric query are driven by the following factors:
- Metric and statistic
- Time range
- Server instances included in the result
- Included dimension values
- Histogram range
Every query returns results for a single statistic and of a single metric. A query must include the time range used to generate the results. Time ranges can either be absolute dates (in ISO-8601 format) or relative dates (such as -30m). A relative start time offset is relative to the end time. A relative end time offset is relative to the current time. When no end time is specified, the server includes results up to the current time.
The time range and the desired number of points (for pivot by time) dictates the resolution of data used to process the query. For example, the finest granularity of data, one second resolution, is only kept for a few hours. It will not be used to satisfy a query spanning multiple days.
By default, all server instances that produce the metric are used to calculate the query results. However, the metric query can be restricted to one of the following:
- A specific list of servers
- Servers of a given type, such as PingDirectory server
- Servers within a specific location
For metrics that include one or more dimensions, a query can be evaluated across a subset
of dimension values. For example, the results returned for the
response-time metric can be restricted to just the search and
modify values of the op-type dimension.
For discrete-valued metrics that break their values down into histogram
ranges, a query can count statistics applied to a subset of histogram buckets by
specifying a minimum and/or maximum histogram value. For example, a query on the
response-time metric could return a count of operations that took
longer than 100 milliseconds.