You configure a monitor when you create it, and can change some parts of its configuration when you edit it. Options are on the three panes of a monitor configuration: Define, Train, and Preview.
You can change settings on the Define pane before you save a monitor. All settings are listed below, but you'll only see settings that work with the monitor type you create (determined by Symptom to Detect).
|Metrics||Choose the metric/metric slice to monitor||Pick from a list of existing metrics/slices.|
|Symptom to Detect||Specify how monitors determine that metric values are out of bounds.||- Manual threshold: Select this if you want to set thresholds.|
- Values out of expectation: This is what you should select if you want a generic anomaly monitor. The monitor sets thresholds automatically.
- Sharp change: An anomaly monitor that allows for gradual changes, generating incidents only when it detects a sudden change.
- Slow burn trend change: An anomaly monitor that allows for short-term variability, generating incidents only when it detects changing trends.
|Account for trend||Include trends in anomaly calculations||Enable/disable. Mutually exclusive with Advanced forecasting.|
|Advanced forecasting||Forecasting of metric values||Enable/disable. Mutually exclusive with Account for trend.|
|Backfill Incidents Starting||Identify how far in the past incidents will be detected, once the monitor goes live||Select a date|
|Manage alerts||Identify a notification channel for incident alerts||Select a notification channel|
|Notifications muting schedule||Identify a schedule during which notifications will be muted||Select a schedule|
|Mute notifications||Turn off incident notifications for the monitor||Enable/disable|
The configuration options on the Train pane depend on the monitor type: some are common to any monitor, some are just for anomaly detection monitors, and some just for manual threshold monitors.
|Drift Duration||Set the drift duration: the amount of time that a detected anomaly must last before a monitor will log an incident.||Specify an amount and a time unit, e.g., 10 seconds.|
|Recovery Duration||Set the recovery duration. Once a metric is within threshold, the recovery duration is the amount of time the metric must remain within threshold for the incident to end. Longer recovery durations can cause several short incidents to merge into one.||Specify an amount and a time unit, e.g., 10 seconds.|
|Aggressiveness||Set how sensitive the monitor should be. Low aggressiveness causes the monitor to pad the actual thresholds, and high aggressiveness causes the monitor to detect incidents when metric values are somewhat inside thresholds.||Select a value from 1 to 10. A value of 7 causes the monitor to strictly obey thresholds.|
|Drift Direction||Set whether to detect upward/downward drift.||Select an arrow to enable/disable detection in that direction.|
|Incident Learning||Set whether to use incident resolution to adjust anomaly detection.||Enable/disable (recommended setting is enabled)|
|Training periods||Include training periods with only known-good metric values.||- Select + to add a new period|
- Select an existing training period to edit or delete it (to ensure it has known-good data)
|Type||Set whether to use values or percentage change for thresholds.||- Value|
- Percent Change
|Lower/Upper Threshold Value||Values/percentage changes of the thresholds, e.g., from 2 to 12 or from -2% to 10%.||Set at least one threshold.|
|Period||Only applies to Percent Change thresholds. Set the periods of time to compare to determine the percent change.||- Day over Day|
- Week over Week
- Sample over Sample
|Start time/End time||Set the time range to use to preview the monitor.|
Updated 13 days ago