[2xdevv]

Description

The Store KV benchmark rejects configurations with zero memory and NoF replicas, even when a DFS replica is requested. Allow DFS-only placement while continuing to reject configurations with all replica counts set to zero.

• Add regression coverage for DFS-only preparation and reads across read_perf, verify_write, and mixed_rw, using plain/zcopy APIs and single/batch requests.
• Expand the usage guide with scenario behavior, concurrency, read selection, CLI options, replay limitations, and output interpretation.
• Document DFS setup for both memory+DFS and DFS-only placement.

This follows the Store's DFS-only placement support merged in #4637. Related PR #4645 changes lane-startup cleanup in the same benchmark files; this PR changes replica validation and documentation, without modifying that cleanup path or the DFS adapter implementation.

Suggested Merge Priority

• P2 — Normal

Module

• Mooncake Store (mooncake-store)
• Docs

Type of Change

• Bug fix
• Documentation update

How Has This Been Tested?

Test commands:

pre-commit run --files \
  mooncake-store/benchmarks/store_kv_bench.py \
  mooncake-store/benchmarks/test_store_kv_bench.py \
  mooncake-store/benchmarks/store_kv_bench.md

/usr/bin/python3 -m unittest discover \
  -s mooncake-store/benchmarks \
  -p test_store_kv_bench.py

Test results:

• File-scoped pre-commit passed.
• All 17 unit tests passed.
• All 16 live smoke cases passed using a fresh worktree build with USE_3FS=ON.

The live matrix covered POSIX and native HF3FS USRBIO, memory+DFS and DFS-only placement, plain/zcopy APIs, and verify_write/read_perf. Each case used two worker lanes, wrote nine 64 KiB objects, and verified nine reads, with zero misses or verification failures. verify_write used single-object requests; read_perf used batches of up to four objects, including a final partial batch.

Client logs confirmed DFS reads for DFS-only placement and memory selection for memory+DFS placement. These were functional smoke tests, not performance measurements. The local smoke harness, build artifacts, and logs are excluded from the PR.

Checklist

• I have performed a self-review of the changes.
• Pre-commit passes on all changed files.
• Documentation has been updated.
• Regression tests have been added.

C/C++ formatting and an architectural RFC are not applicable.

AI Assistance Disclosure

• No AI tools were used.
• AI tools were used.

OpenAI Codex assisted with implementation, documentation, regression tests, and local validation. The human submitter reviewed the changes.

[fcczzz]

LGTM