"""
分析缓存命中情况的脚本
"""

import json
import sys
from pathlib import Path


def analyze_cache(trace_dir):
    """分析指定 trace 的缓存情况"""
    events_file = Path(trace_dir) / "events.jsonl"

    if not events_file.exists():
        print(f"错误: 找不到 {events_file}")
        return

    print("=" * 80)
    print("缓存命中情况详细分析")
    print("=" * 80)
    print()

    # 统计数据
    total_messages = 0
    assistant_messages = 0
    tool_messages = 0
    total_prompt_tokens = 0
    total_cache_creation = 0
    total_cache_read = 0

    cache_events = []

    with open(events_file) as f:
        for line in f:
            event = json.loads(line)
            if event.get("event") == "message_added":
                msg = event["message"]
                role = msg.get("role")
                total_messages += 1

                if role == "assistant":
                    assistant_messages += 1
                    seq = msg["sequence"]
                    prompt = msg.get("prompt_tokens", 0)
                    creation = msg.get("cache_creation_tokens", 0)
                    read = msg.get("cache_read_tokens", 0)

                    total_prompt_tokens += prompt
                    total_cache_creation += creation
                    total_cache_read += read

                    cache_events.append({
                        "seq": seq,
                        "prompt": prompt,
                        "creation": creation,
                        "read": read,
                    })
                elif role == "tool":
                    tool_messages += 1

    # 打印统计
    print(f"总消息数: {total_messages}")
    print(f"  - Assistant: {assistant_messages}")
    print(f"  - Tool: {tool_messages}")
    print(f"  - 其他: {total_messages - assistant_messages - tool_messages}")
    print()

    print(f"Token 统计:")
    print(f"  - Total prompt tokens: {total_prompt_tokens:,}")
    print(f"  - Cache creation: {total_cache_creation:,} ({total_cache_creation/total_prompt_tokens*100:.1f}%)")
    print(f"  - Cache read: {total_cache_read:,} ({total_cache_read/total_prompt_tokens*100:.1f}%)")
    print()

    # 计算缓存命中率
    if total_prompt_tokens > 0:
        hit_rate = total_cache_read / total_prompt_tokens * 100
        print(f"缓存命中率: {hit_rate:.1f}%")
    print()

    # 详细的缓存事件
    print("=" * 80)
    print("详细缓存事件")
    print("=" * 80)
    print()
    print(f"{'Seq':>4} | {'Prompt Tokens':>14} | {'Cache Creation':>15} | {'Cache Read':>11} | {'Hit Rate':>9}")
    print("-" * 80)

    for event in cache_events:
        seq = event["seq"]
        prompt = event["prompt"]
        creation = event["creation"]
        read = event["read"]
        rate = f"{read/prompt*100:.1f}%" if prompt > 0 else "0%"

        # 标记特殊事件
        marker = ""
        if creation > 0:
            marker = " 🔵 创建"
        elif read == 0:
            marker = " ❌ 失效"
        elif read > 0:
            marker = " ✅ 命中"

        print(f"{seq:4d} | {prompt:14,d} | {creation:15,d} | {read:11,d} | {rate:>9s}{marker}")

    print()

    # 分析缓存失效的原因
    print("=" * 80)
    print("缓存失效分析")
    print("=" * 80)
    print()

    miss_count = sum(1 for e in cache_events if e["read"] == 0 and e["creation"] == 0)
    if miss_count > 0:
        print(f"发现 {miss_count} 次缓存失效")
        print("可能的原因:")
        print("  1. Level 2 压缩导致消息序列改变")
        print("  2. 缓存过期（5 分钟）")
        print("  3. 消息内容发生变化")
    else:
        print("没有发现缓存失效")
    print()


if __name__ == "__main__":
    if len(sys.argv) < 2:
        print("用法: python3 analyze.py <trace-directory>")
        print("示例: python3 analyze.py .trace/abc123...")
        sys.exit(1)

    trace_dir = sys.argv[1]
    analyze_cache(trace_dir)