Cook Versions Save

Fixed

• Fixed bug in parallelized Kubernetes watch processing, from @scrosby

Changed

• Make prometheus JVM metrics use compute cluster name, from @samincheva

Changed

• Parallelize Kubernetes watch processing, from @scrosby

Fixed

• Do not set scalar-requests to pool specific resources for Kenzo pods, from @ahaysx

Changed

• Reverted parallel Kubernetes watch processing for future release

Added

• Prometheus metrics
  • JVM metrics, from @samincheva
  • Ring metrics, from @samincheva
  • Parity for remaining codahale metrics, from @samincheva
• Direct-to-Kubernetes scheduler (Kenzo)
  • Use backpressure of scheduling pods to moderate launching new pods for real jobs, from @ahaysx
  • Prometheus metrics parity, from @ahaysx

Changed

• Parallelize Kubernetes watch processing, from @scrosby
• Optimize getting the nodename from a node, from @scrosby

Added

• Initial implementation for submitting jobs directly to Kubernetes Scheduler, from @ahaysx
• Better error handling in scheduler/write functions, from @ahaysx
• Launch tasks similarly for both Fenzo and K8s Scheduler pools, from @ahaysx
• Adding prometheus metrics to remaining modules, from @samincheva

Fixed

• Optimizing total pod count metric and fixing mismatched metric labels, from @samincheva

Added

• Make Fenzo config pool-specific and set up for other schedulers, from @ahaysx
• Prometheus metrics for the kubernetes, API, and tools module, from @samincheva
• Update of synthetic pods counter metric even if the current match cycle doesn't autoscale, from @samincheva

Changed

• Performance optimization for add-starting-pods, from @scrosby
• Remove metatransaction filter from match, from @scrosby
• Make job resource lookup more efficient in miss path, from @scrosby
• Relazy some list generation in rank cycle, from @scrosby
• Parallelize autoscale to run at the same time as main job launches, from @scrosby

Added

• Add new JobSubmissionModifier and refactor JobRouter, from @laurameng

Changed

• Prometheus metrics
  • Updated match cycle metric logic for 0 considerable case, from @samincheva
  • Added prometheus metric for synthetic pods count, from @samincheva
• Use a factory fn for creating (future) different types of pool handlers, from @ahaysx

Changed

• Configured the /metrics endpoint to have a separate rate limit, from @samincheva

Added

• Prometheus, from @samincheva
• Adding match cycle metrics to prometheus, from @samincheva
• Adding prometheus metric for jobs launch count, from @samincheva

Fixed

• Use pools & submit pools in /jobs list endpoint, from @laurameng