Skip to content

Latest commit

 

History

History
97 lines (86 loc) · 4.37 KB

CHANGELOG.md

File metadata and controls

97 lines (86 loc) · 4.37 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

v0.8.0 - 2021-08-15

Changed

  • Updated to Custom Pod Autoscaler v2.2.0.
  • Updated to Horizontal Pod Autoscaler CPA v0.7.0.

Fixed

  • Linear regression no longer fails on first run.

v0.7.0 - 2021-04-05

Added

  • Holt Winters values can now be fetched at runtime, rather than simply being hardcoded.

Fixed

  • Fixed slow shutdown of PHPA due to ignoring SIGTERM from K8s.

Changed

  • Switched from Golang to Python for calculating statistical predictions for Linear Regression and Holt-Winters.
  • Holt-Winters now calculated using statsmodels, opening up statsmodels configuration options for tuning.
    • trend - Either add/additive or mul/multiplicative, defines the method for the trend element.
    • seasonal - Either add/additive or mul/multiplicative, defines the method for the seasonal element.
    • dampedTrend - Boolean value to determine if the trend should be damped.
    • initializationMethod - Which initialization method to use, see statsmodels for details, either estimated, heuristic, known, or legacy-heuristic
    • initialLevel - The initial level value, required if initializationMethod is known.
    • initialTrend - The initial trend value, required if initializationMethod is known.
    • initialSeasonal - The initial seasonal value, required if initializationMethod is known.
  • Holt-Winters seasonLength variable renamed to seasonalPeriods.
  • Holt-Winters method split into trend and seasonal variables.
  • Switched docs theme to material theme.

v0.6.0 - 2020-08-31

Changed

  • Update Custom Pod Autoscaler version to v1.0.0.
  • Update Horizontal Pod Autoscaler version to v0.6.0.

v0.5.0 - 2020-03-27

Changed

  • Evaluation from HPA now included in list of predicted replica counts, rather than being treated separately at the end. Now included in mean, median, minimum calculations rather than just the maximum calculation.

v0.4.0 - 2020-03-10

Added

  • Documentation as code; configuration reference.
  • New decision type median, returns the median average of the predictions.
  • JSON support for configuration options.
  • Can now configure tolerance, initialReadinessDelay and initializationPeriod that are available to be configured in the K8s HPA.
  • Default downscaleStabilization set to 300 (5 minutes) to match K8s HPA.

Changed

  • Metric specs now defined in predictiveConfig rather than in their own section.
  • Update Custom Pod Autoscaler version to v0.11.0.
  • Update Horizontal Pod Autoscaler version to v0.5.0.
  • Default interval set to 15000 (15 seconds) to match K8s HPA.
  • Default minReplicas set to 1 to match K8s HPA.
  • Default maxReplicas set to 10 to match K8s HPA.

v0.3.0 - 2020-02-17

Added

  • Multiplicative method for Holt-Winters time series prediction.

Changed

  • Update Custom Pod Autoscaler version to v0.10.0.
  • Update Horizontal Pod Autoscaler version to v0.4.0.
  • Holt-Winters no longer additive by default, must specify a method, either additive or multiplicative in the Holt-Winters configuration.

v0.2.0 - 2019-12-19

Added

  • Holt-Winters time series based prediction model.

v0.1.0 - 2019-12-9

Added

  • Added the ability to use Linear Regression models to predict future scaling.