ktm metrics correlate
Run ktm metrics correlate to detect correlation, distribution shift, and lag across InfluxQL queries with configurable analysis modes.
Detect cross-metric relationships: correlation, distribution shift, lag
Usage
ktm metrics correlate [flags]Examples
# Quick correlation (default)
ktm metrics correlate --operator acme \
--query "SELECT mean(duration_seconds) FROM test WHERE time > now() - 24h GROUP BY time(1h)" \
--query "SELECT mean(response_time) FROM action WHERE time > now() - 24h GROUP BY time(1h)"
# Full analysis with lag detection
ktm metrics correlate --operator acme --mode full \
--query "SELECT mean(duration_seconds) FROM test WHERE time > now() - 7d GROUP BY time(1h)" \
--query "SELECT count(error) FROM event WHERE time > now() - 7d GROUP BY time(1h)"
# Distribution shift only (which metrics changed?)
ktm metrics correlate --operator acme --mode shift \
--queries '["SELECT mean(duration_seconds) FROM test WHERE time > now() - 7d GROUP BY time(1h)", "SELECT mean(response_time) FROM action WHERE time > now() - 7d GROUP BY time(1h)"]'Flags
| Flag | Type | Default | Description |
|---|---|---|---|
--mode | string | quick | Analysis mode: quick (correlation only), full (correlation + shift + lag), shift (distribution shift only) |
--operator | string | Operator slug (= InfluxDB database name) | |
--queries | string | JSON array of InfluxQL query strings | |
--query | string[] | InfluxQL query (repeatable — use multiple --query flags instead of --queries JSON) |
Details
Run 2-10 InfluxQL queries and detect cross-metric relationships.
Provide queries via --query (repeatable) or --queries (JSON array). Queries are run in parallel and aligned by timestamp.
3 analysis modes (--mode): quick (default): pairwise Spearman/Pearson correlation + p-values. Answers: "are these metrics correlated?" full: adds per-query KS distribution shift detection and lag scan (±50 steps). Answers: "did they change together?" and "does one cause the other?" shift: per-query KS distribution shift only (no pairwise correlation). Answers: "which metrics changed recently?"
Use matching GROUP BY time() intervals across queries for best alignment.
IMPORTANT: Action-type measurements (ethernet_, cellular_, smartphone_*) only contain data from tests that use those action types — they are NOT infrastructure metrics. Verify which monitors populate a measurement before correlating: SHOW TAG VALUES FROM "<measurement>" WITH KEY = "monitoring_name" To understand what a monitor does: ktm monitors list → ktm monitors get --id <id>. Measurements may also contain orphaned data from deleted monitors. Check freshness: SELECT last(duration_seconds), time FROM "<measurement>"
Related
Global flags (--output, --debug, --host, …) apply to every command. See the command reference overview.
What's next?
Last updated on