The semantics of a metric tells what its value is and how it adds up over time.
The detailed metric pages give it in their Semantics column, the MCP tools with each metric (hpa_search_metrics); the DPS
pages call it the kind of the metric.
| Semantics | DPS kind | val of an interval |
count of an interval |
Over time | Examples |
|---|---|---|---|---|---|
| GAUGE | Average | average | calls, or samples for a clocked percentage | average weighted by count | Average Response Time (ms), Total Transaction Time |
| DELTA | Interval counter | total | events (= val) |
sum | Responses Per Interval, processedPerInterval, SystemEdge In Octets 64-bit |
| STATE_COUNTER | Fluctuating counter | highest value | samples or changes | highest | Concurrent Invocations, runningJobs |
| CUMULATIVE | Other | value | values recorded | depends on the Introscope type | none today |
To know:
- metric type VALUE is an average only for a GAUGE
- STATE_COUNTER, CUMULATIVE: a restart resets the value; for CUMULATIVE, use end minus begin
HpaInfrastructure agent metrics
Each metric page gives the semantics of the SystemEdge|... metrics, e.g. Host network metrics:
- counters (bytes, packets, segments, errors…) are DELTA: the value of the last 15 s, although the name or the SNMP standard may suggest a total since boot
- readings (link speed, status, open connections, rates per second) are GAUGE, but recorded as a fluctuating counter: over time, the DPS keeps their highest value
- after a restart of the agent, the first sample of a counter holds the whole count since the machine boot: one abnormally high point, to filter out before any aggregation