[Bug] Fix external metric fetching for GPU optimizer autoscaling and replace dead scale-down annotation in samples - #2616

Merged

varungup90 merged 5 commits into

vllm-project:mainfrom

yaojiejia:yaojiejia/fix-optimizer-kpa-autoscaling

Aug 26, 2026

Commits on Aug 26, 2026

Replace the unparsed kpa.autoscaling.aibrix.ai/scale-down-delay annotation with autoscaling.aibrix.ai/scale-down-cooldown-window in all sample manifests
yaojiejia
committed
5c91d92
Fetch external autoscaling metrics directly from the configured endpoint and path instead of the central metric registry, and switch samples from the deprecated domain source type to external
yaojiejia
committed
a9ef25c
Merge branch 'main' into yaojiejia/fix-optimizer-kpa-autoscaling
varungup90
authored
93718d5
Stop the retry backoff timer on context cancellation instead of leaving a pending time.After timer behind
yaojiejia
committed
45a47ac
Stop counting external metric fetch failures in the llm_engine_metrics_query_fail engine counter
yaojiejia
committed
0230459