[he-yufeng]

Description

The TCP connection pool has been enabled by default since #3569, but the docs never caught up:

• design/transfer-engine/index.md still presents MC_TCP_ENABLE_CONNECTION_POOL as an opt-in flag.
• troubleshooting/troubleshooting.md tells operators to export MC_TCP_ENABLE_CONNECTION_POOL=1 (now a no-op), and its "two known limitations" describe the removed dynamic pool (cleanupIdleConnections(), pool_mutex_). The flag has routed to the lane-based implementation since the lanes landed, so those limitations no longer describe any shipping code.

This PR rewrites both spots against the current implementation: default-on with 0/false/no as opt-out, bounded per-peer lanes (MC_TCP_LANES_PER_PEER, default 4), reuse only after a cleanly completed exchange, dirty sockets closed and re-established. The troubleshooting entry keeps the old-builds note (=1 for pre-flip releases) and replaces the stale limitations with the lane caveats that actually apply: no idle reaping (up to lanes-per-peer sockets held per peer for process lifetime), no keepalive probe before reuse (a middlebox-dropped idle connection stalls the next transfer instead of failing fast), and dirty-socket discard with lane cooldown retry.

Surfaced while reviewing #4464; the behavioral half of that PR is the already-merged #3569.

Module

• Transfer Engine (mooncake-transfer-engine)
• Mooncake Store (mooncake-store)
• Mooncake Conductor (mooncake-conductor)
• Reshard (mooncake-reshard)
• Mooncake EP (mooncake-ep)
• Mooncake PG (mooncake-pg)
• Integration (mooncake-integration)
• P2P Store (mooncake-p2p-store)
• Python Wheel (mooncake-wheel)
• Common (mooncake-common)
• Mooncake RL (mooncake-rl)
• CI/CD
• Docs
• Other

Type of Change

• Bug fix
• New feature
• Refactor
• Breaking change
• Documentation update
• Performance improvement
• Other

How Has This Been Tested?

Docs-only change. Claims were verified against upstream/main (0d1a8040): the env parsing in TcpTransport::TcpTransport() (0/false/no opt-out, default on), lane defaults in the same constructor, dirty-socket handling in handleLaneTerminal, absence of an idle reaper or SO_KEEPALIVE in the lane/session code, and session keep-open behavior in tcp_transport_session_impl.h.

Test commands:

cd docs && uv venv && uv pip install -r requirements-docs.txt
python -m sphinx -b html -q source /tmp/mc_docs_build
pre-commit run --files docs/source/design/transfer-engine/index.md docs/source/troubleshooting/troubleshooting.md

Test results:

• Unit tests pass
• Integration tests pass (if applicable)
• Manual testing done (describe below)

Sphinx HTML build completed with no warnings on either edited page; all applicable pre-commit hooks pass (codespell, end-of-files, etc.).

Checklist

• I have performed a self-review of my own code
• I have formatted my code using ./scripts/code_format.sh
• I have run pre-commit on the files changed in this PR and all hooks pass
• I have updated the documentation (if applicable)
• I have added tests to prove my changes are effective
• For changes >500 LOC: I have filed an RFC issue

AI Assistance Disclosure

• No AI tools were used
• AI tools were used (specify below)

Drafted with AI assistance (Kimi Code): the tool read the current lane implementation and traced which doc statements no longer match shipping code. I reviewed every changed line against the sources cited above and can defend the change end-to-end.

[alogfans]

LGTM