Design: decompose ModelCache into a per-cluster ModelCacheHydration - #362

Closed

dennis-upbound wants to merge 11 commits into

modelplaneai:mainfrom

dennis-upbound:design/modelcache-hydration

Commits on Jul 21, 2026

Add design for decomposing ModelCache into ModelCacheHydration
dennis-upbound
committed
4feb19b
Address hydration failure gating and 63-char naming in design

dennis-upbound
and
claude
committed
908b6de

Commits on Jul 22, 2026

Rework footprint to pre-warm-authoritative per review

dennis-upbound
and
claude
committed
ab36095

Commits on Jul 28, 2026

State plainly what approving the ModelCache design covers

dennis-upbound
and
claude
committed
c12281a

Commits on Aug 4, 2026

Constrain placement to the cache footprint; inject env, not engine flags

dennis-upbound
and
claude
committed
b01b991
Resolve open questions and tighten the alternatives

dennis-upbound
and
claude
committed
bbfd46e
Format the two-contract and rejected-alternative lists for scanning

dennis-upbound
and
claude
committed
74470db
Ground the two engine contracts in SGLang and TensorRT-LLM, not just vLLM

dennis-upbound
and
claude
committed
f65af87
Match the voice and sentence rhythm of the existing design docs

dennis-upbound
and
claude
committed
390a674

Commits on Aug 5, 2026

Show the env-injection mechanism with a ModelDeployment example

dennis-upbound
and
claude
committed
e25ff61

Commits on Aug 6, 2026

Lock the injected var name and show the $(MODELPLANE_MODEL) UX

dennis-upbound
and
claude
committed
d21ce3d