k8s(llm-serving): stop managing the coder InferenceService #106

Merged
rock merged 1 commits from retire-coder-isvc into main 2026-11-03 22:07:58 +00:00
Owner

coder.yaml requests 2 V100s and never reaches Ready, permanently occupying half the GPU pool. candle-vllm serves the coder model on the other two cards. ArgoCD has prune:false, so this removes management only; the live object is deleted manually after sync.

coder.yaml requests 2 V100s and never reaches Ready, permanently occupying half the GPU pool. candle-vllm serves the coder model on the other two cards. ArgoCD has prune:false, so this removes management only; the live object is deleted manually after sync.
rock added 1 commit 2026-11-03 22:07:33 +00:00
coder.yaml requests 2 V100s and never reaches Ready, so it permanently
occupies half the GPU pool. candle-vllm serves the coder model on the
other two cards; the unmanaged predictor starves clef-serve and blocks
its own rollout on every sync.

ArgoCD has prune:false, so this removes management only. The live object
is deleted manually after the sync, otherwise selfHeal recreates it.
rock merged commit c5099e3a8f into main 2026-11-03 22:07:58 +00:00
Sign in to join this conversation.