┌─────────────┐     Crash Signal      ┌──────────────────┐
│  主进程      │ ───────────────────▶ │  守护进程/独立上报进程 │
│ (业务逻辑)   │   (SIGSEGV/SIGABRT)  │ (最小化依赖、无锁队列) │
└─────────────┘                        └────────┬─────────┘
                                                 │
                                                 ▼
                                    ┌────────────────────────┐
                                    │  本地落盘 (mmap + CRC) │
                                    │  断点续传、指数退避重试 │
                                    └───────────┬────────────┘
                                                │ HTTPS / QUIC
                                                ▼
                                    ┌────────────────────────┐
                                    │  网关层：限流、鉴权、脱敏 │
                                    └───────────┬────────────┘
                                                ▼
                                    ┌────────────────────────┐
                                    │  存算分离：ClickHouse + │
                                    │  对象存储 (原始 dump)   │
                                    └────────────────────────┘

# CMake 统一基线（Android/iOS/Windows 通用）
add_compile_options(
    -g                       # 生成调试信息
    -fno-omit-frame-pointer  # 保留帧指针，便于 unwind
    -fvisibility=hidden      # 减少导出符号干扰
    -fstack-protector-strong # 安全加固，不影响符号化
)

# Release 仍保留行表（体积仅增 3%-5%）
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -gline-tables-only")

// 使用 memmap2 + crossbeam-channel 实现生产者-消费者零拷贝
let mut mmap = MmapMut::map_anon(4 * 1024 * 1024)?; // 4MB 环形缓冲
let (tx, rx) = crossbeam_channel::bounded(1024);

// 信号处理器（仅 async-signal-safe 操作）
signal_hook::flag::register(signal::SIGSEGV, move || {
    let ptr = mmap.as_mut_ptr().wrapping_add(offset.load(Ordering::Relaxed));
    // 直接写入寄存器上下文 + 栈回溯指针，无 malloc、无锁
    unsafe { ptr.write_unaligned(capture_context()) };
    offset.fetch_add(RECORD_SIZE, Ordering::Relaxed);
    tx.try_send(()).ok(); // 通知写盘线程
})?;

// 写盘线程：批量 fsync，配合 O_DIRECT 绕过页缓存
std::thread::spawn(move || {
    let file = OpenOptions::new().write(true).create(true).open("crash_ring.bin")?;
    while rx.recv().is_ok() {
        file.write_all(&mmap[..offset.load(Ordering::Acquire)])?;
        file.sync_data()?; // 保证掉电不丢
    }
    Ok(())
});

// unified_symbol.proto —— 所有平台符号表统一转换目标
message SymbolFile {
  string build_id = 1;           // UUID / Build ID / PDB GUID
  string arch = 2;               // arm64, x86_64, wasm32
  repeated Function functions = 3;
  repeated InlineSite inlines = 4; // 统一内联展开信息
}

message Function {
  uint64 start_addr = 1;
  uint64 end_addr = 2;
  string name = 3;               // 已 demangle
  string file = 4;
  uint32 line = 5;
  repeated InlineRef inlines = 6; // 调用内联点
}

// 统一查询接口，屏蔽下游存储差异
service SymbolResolver {
  rpc ResolveBatch(ResolveRequest) returns (ResolveResponse);
  rpc HealthCheck(Empty) returns (HealthStatus);
}

message ResolveRequest {
  string build_id = 1;
  string arch = 2;
  repeated uint64 addresses = 3; // 批量查询，单次 ≤ 1024 地址
  bool need_inline = 4;          // 是否展开内联帧
}

# 伪代码：灰度决策引擎核心逻辑
def evaluate_canary(metrics: CrashMetrics, baseline: Baseline) -> Decision:
    # 1. 统计显著性检验（Sequential Probability Ratio Test, SPRT）
    p_value = sprt_test(
        treatment_crashes=metrics.new_version_crashes,
        treatment_sessions=metrics.new_version_sessions,
        control_crashes=baseline.crashes,
        control_sessions=baseline.sessions,
        min_lift=0.2  # 仅关注崩溃率上升 ≥ 20%
    )
    
    # 2. 新指纹突变检测
    new_fingerprints = set(metrics.fingerprints) - set(baseline.fingerprints)
    critical_new = [fp for fp in new_fingerprints if fp.severity == 'P0']
    
    # 3. 核心链路守护（入会、推流、屏幕共享）
    core_path_regression = any(
        fp.match_path(['JoinMeeting', 'StartPublish', 'ScreenShare']) 
        for fp in critical_new
    )
    
    if p_value < 0.01 or critical_new or core_path_regression:
        return Decision.ROLLBACK_IMMEDIATELY
    elif metrics.new_version_sessions > 10000 and p_value < 0.05:
        return Decision.PAUSE_EXPAND  # 暂停扩量，人工复核
    return Decision.CONTINUE

# .gitlab-ci.yml 片段
stages:
  - build
  - canary_1pct
  - canary_10pct
  - full_rollout

canary_analysis:
  stage: canary_1pct
  image: python:3.11
  script:
    - pip install crash-analytics-sdk
    - |
      python - <<'EOF'
      from crash_analytics import CanaryEvaluator
      evaluator = CanaryEvaluator(
          app_id="meeting-client",
          new_version=os.environ["CI_COMMIT_TAG"],
          baseline_version=os.environ["LAST_STABLE_TAG"],
          window_minutes=60
      )
      decision = evaluator.run()
      if decision != "CONTINUE":
          print(f"::error ::Canary failed: {decision}")
          sys.exit(1)
      EOF
  rules:
    - if: $CI_COMMIT_TAG =~ /^vd+.d+.d+$/
  timeout: 90m  # 观察窗口 60min + 缓冲

# 实时崩溃率对比（新版 vs 基线）
sum(rate(crash_total{version=~"$new_version"}[5m])) by (fingerprint)
/
sum(rate(session_start_total{version=~"$new_version"}[5m]))

# 新增指纹趋势
increase(crash_fingerprint_new_total{version=~"$new_version"}[1h])

-- ClickHouse TTL + Kafka 触发合规清理
ALTER TABLE crash_raw_dumps
MODIFY SETTING TTL event_time + INTERVAL 7 DAY TO DISK 'hdd',
                 event_time + INTERVAL 90 DAY TO VOLUME 'archive',
                 event_time + INTERVAL 365 DAY DELETE
WHERE privacy_level = 'PII_FREE'; -- 含 PII 的表走独立合规流程