Every raw metric point actually carries four numbers on a given timestamp — an aggregate computed by the Introscope agent runtime from however many raw samples landed in that collection interval (e.g. a 15-second window), not something HPA itself computes:
-
val— the aggregate value for this interval; which aggregate it actually is depends on the metric itself, not something the portal controls or can tell just by looking at a number, - the max of all values recorded on this interval,
- the min of all values recorded on this interval,
- the count of underlying samples for that interval.
Illustration (assuming average-type instrumentation) — several raw samples can land in the same interval; they get folded into the one point stored for that interval:
| Interval 1 (10:00:00) | Interval 2 (10:00:15) | Interval 3 (10:00:30) | |
|---|---|---|---|
| Values seen | 1, 3, 4, 5, 1 | 2, 2, 6 | 10 |
| Point stored | aggregate: 2.8, max: 5, min: 1, count: 5 | aggregate: 3.33, max: 6, min: 2, count: 3 | aggregate: 10, max: 10, min: 10, count: 1 |
(The aggregate depends on the metric’s own nature — e.g. a response time metric aggregates as
a weighted average, so its val genuinely is an average; other kinds of metric aggregate
differently:
aggregate means as… |
count means |
Example metric |
|---|---|---|
| Average on an interval, weighted average on multiple interval | Number of completions (the average’s denominator) | Average Response Time (ms) |
| Sum | Number of completions during the interval (equal to the value) | Responses Per Interval, Errors Per Interval |
| Highest value seen | Number of increments/decrements during the interval | Stall Count, Concurrent Invocation, Bytes In Use |
So aggregate is only a true average for average-style metrics like response times — for a
per-interval counter it’s a sum, and for a stall/gauge-style metric it’s the highest value seen
during the interval, not an average at all.)
The Metric type field picks which one of those four becomes “the value” of each point for the rest of the pipeline — everything downstream (filtering, reducing, resampling) only ever sees this single chosen number.
Effective metric only (checked by default) additionally drops interval whose count is zero before
they ever reach this selection.
The optional Time period selector picks which time window is evaluated: leave it on
Use current to always evaluate against whatever period is currently selected in the portal, or
pin it to a specific saved period (e.g. “Last hour”, or a fixed incident window).