Evaluation reportGenerated by the project's CI evaluation on held-out data.About the projectCode on GitHub

Kubernetes management platform: evaluation

Admission policies on manifests with known violations, and rightsizing judged on the weeks after the recommendation.

1.00policy F1 on forms written after freezing (searching the text: 0.34)
59.6%less CPU requested on held-out workloads
0 vs 92OOM-killed leaking and batch workloads (average x 1.2)
0violations introduced by automatic fixes

Admission policies (F1)

ManifestsPolicy engineSearching the YAML textAuto-fixed
Forms the engine was written against1.0000.725339 of 339
Same forms, new manifests1.0000.707348 of 348
Forms written after freezing1.0000.335142 of 142

Rightsizing

WorkloadsMethodCPU vs todayMemory vs todayTime above requestOOM-killed
Held-out workloadsRecommended (this platform)40.4%52.9%0.1%0 of 300
Held-out workloadsAverage x 1.226.9%49.9%21.1%0 of 300
Held-out workloadsToday's requests100.0%100.0%0.0%0 of 300
Leaks and batch (held back)Recommended (this platform)19.3%54.5%1.3%0 of 200
Leaks and batch (held back)Average x 1.29.4%39.3%18.2%92 of 200
Leaks and batch (held back)Today's requests100.0%100.0%0.0%0 of 200

Gate (regression limits)

CheckValueLimitResult
heldout_policy_f1_min1.00.99pass
fresh_policy_f1_min1.00.98pass
autofix_introduced_max00pass
heldout_rightsizing_time_above_request_max0.0010.01pass
held_back_oom_killed_max00pass
heldout_cpu_vs_current_max0.4040.6pass