The SRE Manifesto
Site Reliability Engineering Practice
Time-series Data Analysis through Percentiles
| Practice code | Practice area(s) | Practice name | Practice description | Practice applicability | Practice technology(ies) | Implementation steps |
|---|---|---|---|---|---|---|
| DSC100 | [x] Data Science; [x] Observability | Time-series Data Analysis through Percentiles | Analyze time-series data, mainly monitoring metrics, by applying percentiles or quantiles to the dataset. | Applicable to all industries and systems | Time-series DB-based monitoring systems like Prometheus | 1. Collect monitoring metric data points as a time-series dataset; 2. Analyze the series by applying percentile or quantile methods; 3. Calculate the relevant percentiles or quantiles (preferably based on an SLO) to uncover outliers; 4. Isolate outliers to see if they are errors or problems. |
Source: Google Book, Prometheus project