Highlights
Anomaly detector
Anomaly
is a new analytic unit type, which can detect anomalies like this:
Algorithm smoothes data and makes a confidence interval using:
Alpha
is a magnitude of smoothness. Bigger alpha value means less previous values to take for exponential smoothing algorithm.
![](./img/alpha.gif)
Confidence
defines the range where metric value considered as "normal". Each value that falls outside this interval is an anomaly.
![](./img/confidence.gif)
Seasonality
Let's say we have a metric:
We can see 1-day seasonality here.
If we apply anomaly detector, we'll get "false" detections every 24 hours.
![](./img/seasonality-false.png)
We can set Seasonality
period and label a segment that is not an anomaly:
![](./img/seasonality-labeled.png)
Values from the labeled segment will be added to the confidence interval every "seasonality" times.
Learn more about anomaly detector on Wiki page.
Inspect mode and HSR
HSR -- is a new concept which stands for Hastic Signal Representation
.
HSR purpose is to inspect how hastic-server actually processes the metric. It can be different from actual metric because of aggregation interval. For example:
![](./img/hsr.png)
HSR has a different look for anomaly detector. It includes upper and lower bounds:
![](./img/hsr-anomaly.png)
P.S. It is our anniversary release :)
Changelog
Hastic Server
Added
Fixed
- Thresholds error: from is NaN #621
- Anomaly detector: Need at least 1 labeled segment #667
- Queue or drop learning task on new learning task #664
- Anomaly detector: wrong seasonality offset #671
- Not ending learning for anomaly detector #665
Changed
- Detect thresholds on the whole dataset instead of 1 value #505
- Start learning after analytic unit update #647
- Set ready status for span only after inserting segments #648
- Remove old detected segments and detection spans when learning starts #650
- Merge threshold segments #624
- Remove old detected segments and detection spans when learning starts #650
- Anomaly detector: send confidence bounds instead of smoothed data #656
Hastic Grafana App
Added
- HSR: Hastic signal representation #258
- HSR in Inspect mode #279
- UI for anomaly detector configuration #270
- Labeling for anomalies #281
- HSR for anomaly analytic unit #302
- Save and cancel buttons for analytic units #310
Fixed
- Thresholds error: from is NaN #277
- Broken Detect button in Threshold / Anomaly analytic units #283
- Color picker is broken #275
- Confusing Save / Label / Apply / Detect buttons #280
- HSR is not updated when anomaly analytic unit options are changed #292
- Seasonality checkbox is disabled on clearing seasonality period input #304
- Missing detections on a bigger zoom #269