vllm: support mooncake_kv_store + expose mooncake_master metrics

[esmeetu]

Summary

Extends mooncake_master orchestration (previously SGLang-only) to the vLLM backend so vLLM workers can use the in-process MooncakeStoreConnector for cross-process KV sharing. Also wires the master's admin HTTP server (/metrics, /health, etc.) on a configurable port so ops can scrape Prometheus metrics.

What's new

• VLLMMooncakeKVStoreConfig (backends/vllm.py): mirrors the SGLang mooncake_kv_store: block and adds a vLLM-specific store_config: section. srtslurm renders that section into the MOONCAKE_CONFIG_PATH JSON file vLLM's MooncakeStoreConnector reads on startup (MooncakeStoreConfig.load_from_env()); the env: map is injected on every vLLM worker for in-process MC_* knobs (e.g. MC_ENABLE_DEST_DEVICE_AFFINITY, MC_STORE_CLIENT_METRIC, MC_TE_METRIC).
• Shared launch constants moved to a new backends/mooncake.py; the SGLang module re-exports them so existing imports keep working.
• start_mooncake_master now fires for vLLM as well as SGLang, writes the per-job store JSON next to the other log artifacts, and stamps the master RPC + HTTP-metadata endpoints on each worker.
• Admin HTTP server: master srun now passes --enable_metric_reporting=true --metrics_port=9003 (upstream default in master.cpp), and start_mooncake_master waits on the metrics port like it already waits on RPC + HTTP-metadata. The flag toggles a periodic stdout log thread only — MasterAdminServer listens unconditionally — the new MOONCAKE_METRICS_PORT constant's comment documents that to prevent future reordering. Exposes /metrics, /metrics/summary, /health, /role, /ha_status, /leader, /query_key.

Test plan

• make check (ruff + pytest, 732 passed / 2 skipped)
• New cases in tests/test_e2e.py cover vLLM mooncake env injection and JSON rendering
• New cases in tests/test_dry_run.py cover the new config surface in srtctl dry-run output

[esmeetu]

Thanks for the review @qiching — addressed all three in 8306191:

1. endpoints_to_processes signature — added port_allocator: NodePortAllocator | None = None to match BackendProtocol, and threaded it through to topology.endpoints_to_processes(...) (both the standard-TP and DP-mode paths) so per-job port jitter from NodePortAllocator.from_job_id applies uniformly to vLLM jobs as well.
2. Unrelated NIXL env var — removed VLLM_NIXL_SIDE_CHANNEL_HOST and the get_hostname_ip import; that change isn't mooncake-related and shouldn't have been in this PR.
3. store_config typing — widened to dict[str, Any] (both the field on VLLMMooncakeKVStoreConfig and the return / local in build_mooncake_store_config). Added a short comment explaining that vLLM's MooncakeStoreConfig is a mix of str, int, and human-readable size strings, so dict[str, str] would force users to quote numeric values unnecessarily.

make check clean (732 passed, 2 skipped).

[esmeetu]

@qiching It should be clean now.

[qiching]

LGTM. Good job. Thank you! @esmeetu

[codecov-commenter]

Codecov Report

❌ Patch coverage is 83.58209% with 11 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@d042570). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/srtctl/cli/do_sweep.py	38.46%	8 Missing ⚠️
src/srtctl/core/schema.py	89.65%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #157   +/-   ##
=======================================
  Coverage        ?   65.63%           
=======================================
  Files           ?       68           
  Lines           ?     8589           
  Branches        ?        0           
=======================================
  Hits            ?     5637           
  Misses          ?     2952           
  Partials        ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.