[luckyq]

Series map (Phase 1 harness, one PR per step; each is based on the previous one)

PR	Step	Test cases
#137	Harness core: stack, client, telemetry, reference engine, CI unit tests; phase1 job in vllm-e2e.yml on the two-GPU runner (TC-01; #138/#139 extend --cases)	TC-01 integration generation correctness
#138	Local DRAM flow and hinted peer transfer, hybrid geometry flags	TC-02, TC-03
#139	Guard protection, takeover, replacement	TC-04
#140	Streaming client, performance arms, Guard recovery timing	TC-05, TC-06
#141	Sidecar kvcr_service, 2P+1D, two nodes, k8s / Slurm / Docker runners; phase1-2p1d CI job (2P+1D on the two-GPU runner, prefill workers share GPU 0)	all cases on those topologies
#142	docs/ci.md: what the two CI lanes verify, how to run them elsewhere, how to read evidence	—
#143	Test plan, run history, bare-vLLM probes	—
#180	--engine sglang: dynamo.sglang + HiCache KVCR storage backend (quick-start config), SGLang telemetry mapping, deterministic mode; no Guard/disagg	TC-01..03, TC-05 on SGLang

Replaces #103. All seven are drafts until the per-PR GPU runs on the current main image are done.

Problem

• Part 2 of the stacked Phase 1 series (on top of the core + TC-01 PR). The harness so far proves only that a worker reuses its own KVCR tier; the local DRAM flow with partial prefixes and the router-hint peer transfer are untested.

What this adds

• TC-02 local DRAM full flow: full prompt, partial-prefix prompt, absent prompt; the hit length is checked against what the KVCR-disabled reference engine itself could reuse, with the hybrid-model arms (--tc02-seed-prefix, --tc02-extra-shared-sections, --tc02-evict-target) that exposed the SSM-checkpoint over-transfer (dynamo#16794).
• TC-03 request hints and peer transfer: a hinted and a fresh request reach worker 1 together; the hinted one hits with remote_deliver == source_write == hit blocks and equal bytes, the fresh one is cold; no failed/partial KVCR metrics.
• Hybrid geometry flags --hit-block-tokens 576 --ssm-groups 3 --page-bytes 18874368 (Qwen3.5-4B) make the delivered-pages and bytes checks exact; without them they are lower bounds.
• Unit tests: prompt prefix layout, expected hit blocks, missing-SSM-pages verdict.

Validation

• ruff, pytest tests/ci: see CI (unit tests run on a cluster login node, not locally).
• GPU: TC-02/03 passed on 2×H100 with KVCR v0.1.1, dense Qwen3-8B and hybrid Qwen3.5-4B, as part of the full test: KVCR Phase 1 end-to-end harness (CI image build, TC-01..06, dlcluster and K8s runners) #103 run.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.