fix(net): Stop() 幂等化, 解决请求后 server 崩溃

[NeverENG]

Summary

• Connection.Stop() 用 sync.Once 改成幂等, 替代 isClose 标志
• 移除 Stop 末尾的裸 recover() 与 close(msgChan)/close(msgBuffChan): worker 仍可能在 SendBuffMsg, close 会触发 panic: send on closed channel (= "数据静态" 现象, server 处理首个请求后崩溃)
• ExitBuffChan 在 Stop 内改为非阻塞 send, 防止 buffer 已满时永阻塞

Background

原 Stop 末尾用 defer recover() 吞掉 close 已关闭 channel 的 panic, 同时 close 了 msgChan/msgBuffChan — 但 worker 还在通过 SendBuffMsg 往 msgBuffChan 写数据。删除 recover 后这个 panic 暴露, 修复方案是不 close 这两个 channel, 让它们随 Connection GC 即可。

Test plan

• go build ./... 通过
• go vet ./network/... 通过
• 本地试运行: 起服务 + 客户端 PUT/GET/DEL 多轮, server 不再崩溃 (4 次连接周期都 clean exit, 无 panic)

试运行发现一个独立 bug ("Error getting value: incomplete response"), 与 msgID 解析相关, 属于 PR-7 协议改造范畴, 本 PR 不处理。

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Enhanced connection shutdown mechanism for improved stability and reliability during graceful cleanup operations.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b778b5e1-7082-4691-831a-4c7556591f34

📥 Commits

Reviewing files that changed from the base of the PR and between 33e4b8e and 8a66edd.

📒 Files selected for processing (1)

• network/banNet/connection.go

---

📝 Walkthrough

Walkthrough

The PR refactors Connection's shutdown mechanism from a simple boolean flag to a sync.Once guard, making Stop() idempotent and eliminating "send on closed channel" panics during concurrent sender activity. The method now signals shutdown via non-blocking send rather than closing channels.

Changes

Idempotent Connection Shutdown

Layer / File(s)	Summary
Shutdown guard and initialization network/banNet/connection.go	Connection replaces the isClose bool field with stopOnce sync.Once to guard one-time stop execution. NewConnection reorganizes initialization to place ctx, cancel, and msgChan in the same control fields block.
Idempotent Stop() method network/banNet/connection.go	Stop() wraps all shutdown steps in stopOnce.Do(...) to ensure single execution. Channel closing logic is removed entirely; instead, a non-blocking send notifies ExitBuffChan to trigger worker shutdown without risking panics from concurrent senders.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

A rabbit hops through channels with grace,
No more closed doors in the shutdown race,
With stopOnce guarding the final call,
One gentle signal ends it all. 🐰

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/conn-stop-once

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}