一个轻巧高效的多线程c++stream风格异步日志(一)

一个轻巧高效的多线程c++stream风格异步日志

---

前言

本文主要实现muduo的异步日志.省去boost库中特性的引用,轻巧简洁。

对于编程而言,日志是必不可少的,对于关键进程,日志通常需要记录：
1.收到的每条内部消息id.
2.收到的每条外部消息全文.
3.发出的每条消息全文,每条消息都有全局唯一的id.
4.关键内部状态变更，等等.
5.每条日志都有时间戳,以方便的追踪分布式系统中一个时间的来龙去脉.

c++日志大体上有两种API风格:
C/JAVA的printf(fmt, ...)风格：
log_info(“received %d bytes from %s”, len, getName.c_str());
c++的stream<<风格：
LOG_INFO << “received << len << bytes from << getName.c_str();

本文实现的是c++的stream风格, 优点是使用起来更自然，不用保持格式的一致性.

功能需求

常规的通用日志库如log4j/logback通常会提供丰富的功能,但不一定是必须的功能.

1.日志有多种级别：TRACE、DEBUG、INFO、WARN、ERROR、FATAL等。
2.日志可能会有多个目的地：文件、socket、SMTP等。
3.日志消息格式可配置。
4.可设置运行时过滤器(filter)、控制不同的组件的日志消息级别和目的地.

这四项中,除了第一项之外,其余并非是必须的功能.本文内容只包含第一项功能.
对于分布式系统中的服务进程而言,日志的目的地只有一个：本地文件,往网络写日志消息是不可靠的，另一个坏处就是网络带宽的消耗.

muduo日志格式如下，本日志对格式有所改动，但基本信息不变。

日期          时间   微妙   线程  级别  函数名   正文       源文件名: 行号
20180802 12:42:50.621354 44321 WARN  main just test - test.cpp:445
20180802 12:42:50.621359 44321 INFO  main just test - test.cpp:446
20180802 12:42:50.621364 44321 ERROR main just test - test.cpp:447

• 时间戳精确到微妙，每条消息通过gettimeofday(2)获取当前时间.这样做不会有什么性能损失,在x86-64Linux上,gettimeofday()不是系统调用,不会陷入内核
• 打印线程ID,便于分析多线程程序时序,也可以检测死锁(未加入)
• 打印日志级别,通过查看有无ERROR日志快速定位
• 打印文件名和行号,修复问题不至于搞错对象
• 每条日志尽量占一行,便于分析日志

性能需求

只有日志库足够高效，程序员才敢在代码中输出足够多的诊断信息，减小运维难度，提升效率.它体现在：

• 每秒几千万条日志没有明显的性能损失
• 能应对一个进程产生大量日志数据的场景，如1GB/min。
• 不阻塞正常的执行流程
• 在多线程程序中，不造成争用

为了实现性能指标，本日志设计中几点优化：

• 时间戳字符串中的日期和时间两部分是缓存的,一秒之内的多条日志只需格式化微妙即可
• 日志消息 的前四段是定长的,可避免运行时字串长度的计算
• 线程id是预先格式化的字符串
• 原文件名部分通过编译期计算来获得,避免运行期strrchr()的开销

Logger实现

LogStream类

LogStream类，主要作用是重载<<操作符，将基本类型的数据格式成字符串通过append接口存入LogBuffer中。

#ifndef _LOG_STREAM_HH
#define _LOG_STREAM_HH
#include <stdio.h>
#include <string.h>

const int kSmallBuffer = 4096;
const int kLargeBuffer = 4096;

template<int SIZE>
class LogBuffer
{
public:
    LogBuffer(): m_cur(m_data){
    }

    ~LogBuffer(){
        //printf("%s", m_data);
    }

    void append(const char* /*restrict*/ buf, size_t len){
    // append partially
        if (/*implicit_cast<size_t>*/(avail()) > len)
        {
            memcpy(m_cur, buf, len);
            m_cur += len;
        }
    }

    // write in m_data directly
    char* current() {  return m_cur; };
    int avail() const { return static_cast<int> (end() - m_cur); }
    void add(size_t len) { m_cur += len; }
    int length() const {return m_cur - m_data;}

    const char* data() const { return m_data; }

private:
    const char* end() const { return m_data + sizeof(m_data); }

    char m_data[SIZE];
    char* m_cur;
};

class LogStream{
public:
    LogStream();
    ~LogStream();

    typedef LogBuffer<kSmallBuffer> Buffer;
    typedef LogStream self;
    self& operator<<(bool v);

    self& operator<<(short);
    self& operator<<(unsigned short);
    self& operator<<(int);
    self& operator<<(unsigned int);
    self& operator<<(long);
    self& operator<<(unsigned long);
    self& operator<<(long long);
    self& operator<<(unsigned long long);

    self& operator<<(const void*);

    self& operator<<(float v);
    self& operator<<(double);

    self& operator<<(char v);
    self& operator<<(const char *);

    void append(const char* data, int len) { return m_buffer.append(data, len); }
    const Buffer& buffer() const { return m_buffer; }

private:
    LogStream(const LogStream& ls);         //no copyable
    LogStream& operator=(const LogStream& ls);

    template<typename T>
    void formatInteger(T v);

    Buffer m_buffer;
    static const int kMaxNumericSize = 32;

};

class Fmt{
public:
    template<typename T>
    Fmt(const char* fmt, T val);

    const char* data() const { return m_buf; }
    int length() const { return m_length; }

private:
    char m_buf[32];
    int m_length;
};

inline LogStream& operator<<(LogStream &s, const Fmt& fmt){
    s.append(fmt.data(), fmt.length());
    return s;
}

#endif

Logger类

Logger类，使用LogStream的<<操作符将文件名，日志级别，时间等信息格式好提前写入LogBuffer中.
它实际上是一个临时对象,通过宏定义临时构造一个logger对象，构造的时候即将上面信息写入LogBuffer,之后通过stream()接口引用LogStream的<<操作符写入我们的日志信息.
优点是 ：对于日志级别低的log那么它将是一个空的操作.

// CAUTION: do not write:
//
// if (good)
//   LOG_INFO << "Good news";
// else
//   LOG_WARN << "Bad news";
//
// this expends to
//
// if (good)
//   if (logging_INFO)
//     logInfoStream << "Good news";
//   else
//     logWarnStream << "Bad news";
//
#define LOG_TRACE if (Logger::logLevel() <= Logger::TRACE) \
    Logger(__FILE__, __LINE__, Logger::TRACE, __func__).stream()
#define LOG_DEBUG if (Logger::logLevel() <= Logger::DEBUG) \
    Logger(__FILE__, __LINE__, Logger::DEBUG, __func__).stream()
#define LOG_INFO if (Logger::logLevel() <= Logger::INFO) \
    Logger(__FILE__, __LINE__).stream()
#define LOG_WARN Logger(__FILE__, __LINE__, Logger::WARN).stream()
#define LOG_ERROR Logger(__FILE__, __LINE__, Logger::ERROR).stream()
#define LOG_FATAL Logger(__FILE__, __LINE__, Logger::FATAL).stream()
#define LOG_SYSERR Logger(__FILE__, __LINE__, false).stream()
#define LOG_SYSFATAL Logger(__FILE__, __LINE__, true).stream()

它的成员结构如下图所示:
Impl对象是个私有类,主要是为了闭源，构造Impl的时候后写入时间戳，日志级别，文件名等等.
logLevel()、setlogLevel()、为两个静态成员函数，用于设置日志级别.
setOutput()也是个静态成员函数，主要作用是重定向日志的输出.默认是输出至stdout.

#ifndef _LOGGER_HH
#define _LOGGER_HH

#include <string.h>
#include "LogStream.hh"
#include "TimeStamp.hh"

// CAUTION: do not write:
//
// if (good)
//   LOG_INFO << "Good news";
// else
//   LOG_WARN << "Bad news";
//
// this expends to
//
// if (good)
//   if (logging_INFO)
//     logInfoStream << "Good news";
//   else
//     logWarnStream << "Bad news";
//
#define LOG_TRACE if (Logger::logLevel() <= Logger::TRACE) \
    Logger(__FILE__, __LINE__, Logger::TRACE, __func__).stream()
#define LOG_DEBUG if (Logger::logLevel() <= Logger::DEBUG) \
    Logger(__FILE__, __LINE__, Logger::DEBUG, __func__).stream()
#define LOG_INFO if (Logger::logLevel() <= Logger::INFO) \
    Logger(__FILE__, __LINE__).stream()
#define LOG_WARN Logger(__FILE__, __LINE__, Logger::WARN).stream()
#define LOG_ERROR Logger(__FILE__, __LINE__, Logger::ERROR).stream()
#define LOG_FATAL Logger(__FILE__, __LINE__, Logger::FATAL).stream()
#define LOG_SYSERR Logger(__FILE__, __LINE__, false).stream()
#define LOG_SYSFATAL Logger(__FILE__, __LINE__, true).stream()

class Logger
{
public:
    enum LogLevel
    {
        TRACE,
        DEBUG,
        INFO,
        WARN,
        ERROR,
        FATAL,
        NUM_LOG_LEVELS,
    };

//compile time calculation of basename of source file
    class SourceFile
    {
    public:
        template<int N>
        inline SourceFile(const char (&arr)[N])
            :m_data(arr),
             m_size(N-1){
            const char* slash = strrchr(m_data, '/'); // builtin function
            if (slash){
                m_data = slash + 1;
                m_size -= static_cast<int>(m_data - arr);
            }
        }

        explicit SourceFile(const char* filename)
            : m_data(filename){
            const char* slash = strrchr(filename, '/');
            if (slash){
                m_data = slash + 1;
            }
            m_size = static_cast<int>(strlen(m_data));
        }

        const char* m_data;
        int m_size;
    };

    Logger(SourceFile file, int line);
    Logger(SourceFile file, int line, LogLevel level);
    Logger(SourceFile file, int line, LogLevel level, const char* func);
    Logger(SourceFile file, int line, bool toAbort);
    ~Logger();

    static void setLogLevel(LogLevel level);
    static LogLevel logLevel();

    LogStream& stream() { return m_impl.m_stream; }

    typedef void (*outputFunc)(const char *msg, int len);
    typedef void (*flushFunc)();

    static void setOutput(outputFunc);
    static void setFlush(flushFunc);

private:
    Logger(const Logger &lg);           //no copyable
    Logger& operator=(const Logger &lg);

    class Impl
    {
    public:
        typedef Logger::LogLevel LogLevel;
        Impl(LogLevel level, int old_errno, const SourceFile& file, int line);
        void formatTime();
        void finish();

        TimeStamp m_time;
        LogStream m_stream;
        LogLevel m_level;
        int m_line;
        SourceFile m_fileBaseName;
    };

    Impl m_impl;

};

#endif

完成Logger即可直接使用了。
直接通过宏LOG_DEBUG << 即可打印格式化好的日志信息. 只是默认输出到stdout.
后面文章会给出日志文件类及异步线程类的实现.

LogStream及Logger cpp源码

//LogStream
#include <stdio.h>
#include <assert.h>
#include <algorithm>
#include "LogStream.hh"

LogStream::LogStream(){
}

LogStream::~LogStream(){
}

LogStream& LogStream::operator<<(bool v){
    m_buffer.append(v ? "1" : "0", 1);
    return *this;
}


LogStream& LogStream::operator<<(short v){
    *this << static_cast<int>(v);
    return *this;
}

LogStream& LogStream::operator<<(unsigned short v)
{
  *this << static_cast<unsigned int>(v);
  return *this;
}

LogStream& LogStream::operator<<(int v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(unsigned int v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(long v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(unsigned long v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(long long v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(unsigned long long v)
{
  formatInteger(v);
  return *this;
}

LogStream& LogStream::operator<<(const void*){
    printf("undefine\n");
}

LogStream& LogStream::operator<<(float v)
  {
    *this << static_cast<double>(v);
    return *this;
  }

LogStream& LogStream::operator<<(double v){
    if(m_buffer.avail() >= kMaxNumericSize){
        int len = snprintf(m_buffer.current(), kMaxNumericSize, "%.12g", v);
        m_buffer.add(len);
    }
    return *this;
}

LogStream& LogStream::operator<<(char v){
    m_buffer.append(&v, 1);
    return *this;
}

LogStream& LogStream::operator<<(const char *str){
    if(str){
        m_buffer.append(str, strlen(str));
    }else{
        m_buffer.append("(NULL)", 6);
    }

    return *this;
}

const char digits[] = "9876543210123456789";
const char* zero = digits + 9;

//convert to str
template<typename T>
size_t convert(char buf[], T value){
    T i = value;
    char *p = buf;

    do{
        int lsd = static_cast<int>(i % 10);
        i /= 10;
        *p++ = zero[lsd];
    } while(i != 0);

    if(value < 0){
        *p++ = '-';
    }

    *p = '\0';
    std::reverse(buf, p);

    return p - buf;
}

template<typename T>
void LogStream::formatInteger(T v)
{
    if(m_buffer.avail() >= kMaxNumericSize){
        size_t len = convert(m_buffer.current(), v);
        m_buffer.add(len);
    }
}

template<typename T>
Fmt::Fmt(const char* fmt, T val)
{
    m_length = snprintf(m_buf, sizeof(m_buf), fmt, val);
    assert(static_cast<size_t>(m_length) < sizeof(m_buf));
}

// Explicit instantiations

template Fmt::Fmt(const char* fmt, char);

template Fmt::Fmt(const char* fmt, short);
template Fmt::Fmt(const char* fmt, unsigned short);
template Fmt::Fmt(const char* fmt, int);
template Fmt::Fmt(const char* fmt, unsigned int);
template Fmt::Fmt(const char* fmt, long);
template Fmt::Fmt(const char* fmt, unsigned long);
template Fmt::Fmt(const char* fmt, long long);
template Fmt::Fmt(const char* fmt, unsigned long long);

template Fmt::Fmt(const char* fmt, float);
template Fmt::Fmt(const char* fmt, double);



//Logger

#include <time.h>
#include <stdint.h>
#include <stdlib.h>
#include <assert.h>
#include <stdio.h>
#include <errno.h>
#include <string.h>
#include "TimeStamp.hh"
#include "Logger.hh"

__thread char t_time[64];
__thread time_t t_lastSecond;
__thread char t_errnobuf[512];

const char* strerror_tl(int savedErrno)
{
  return strerror_r(savedErrno, t_errnobuf, sizeof(t_errnobuf));
}

Logger::LogLevel g_logLevel = Logger::INFO;

void Logger::setLogLevel(LogLevel level){
    g_logLevel = level;
}

Logger::LogLevel Logger::logLevel(){
    return g_logLevel;
}

const char* LogLevelName[Logger::NUM_LOG_LEVELS] =
{
    "[TRACE]",
    "[DEBUG]",
    "[INFO ]",
    "[WARN ]",
    "[ERROR]",
    "[FATAL]",
};

// helper class for known string length at compile time
class T
{
 public:
  T(const char* str, unsigned len)
    :m_str(str),
     m_len(len)
  {
    assert(strlen(str) == m_len);
  }

  const char* m_str;
  const unsigned m_len;
};

void defaultOutput(const char *msg, int len){
    size_t n = fwrite(msg, 1, len, stdout);
    (void)n;
}

void defaultFlush(){
    fflush(stdout);
}

Logger::outputFunc g_output = defaultOutput;
Logger::flushFunc g_flush = defaultFlush;

void Logger::setOutput(outputFunc out){
    g_output = out;
}

void Logger::setFlush(flushFunc flush){
    g_flush = flush;
}

Logger::Logger(SourceFile file, int line)
    : m_impl(INFO, 0, file, line){
}

Logger::Logger(SourceFile file, int line, LogLevel level)
    : m_impl(level, 0, file, line){
}

Logger::Logger(SourceFile file, int line, bool toAbort)
    : m_impl(toAbort? FATAL:ERROR, errno, file, line){
}

Logger::Logger(SourceFile file, int line, LogLevel level, const char* func)
    : m_impl(level, 0, file, line){
    m_impl.m_stream << '[' << func << "] ";
}

Logger::~Logger(){
    m_impl.finish();
    const LogStream::Buffer& buf(stream().buffer());
    g_output(buf.data(), buf.length());
    if (m_impl.m_level == FATAL)
    {
        g_flush();
        abort();
    }
}

Logger::Impl::Impl(LogLevel level, int savedErrno, const SourceFile& file, int line)
    : m_time(TimeStamp::now()),
      m_stream(),
      m_level(level),
      m_fileBaseName(file),
      m_line(line)
{
    formatTime();
    m_stream << LogLevelName[level] << ' ';
    m_stream << '[' << m_fileBaseName.m_data << ':' << m_line << "] ";
    if (savedErrno != 0)
    {
        m_stream << strerror_tl(savedErrno) << " (errno=" << savedErrno << ") ";
    }
}

void Logger::Impl::finish()
{
    m_stream<< '\n';
}

void Logger::Impl::formatTime()
{   
    int64_t microSecondsSinceEpoch = m_time.microSecondsSinceEpoch();
    time_t seconds = static_cast<time_t>(microSecondsSinceEpoch / TimeStamp::kMicroSecondsPerSecond);
    int microseconds = static_cast<int>(microSecondsSinceEpoch % TimeStamp::kMicroSecondsPerSecond);
    if (seconds != t_lastSecond){
        t_lastSecond = seconds;
        struct tm tm_time;

        ::gmtime_r(&seconds, &tm_time); // FIXME TimeZone::fromUtcTime

        int len = snprintf(t_time, sizeof(t_time), "%4d-%02d-%02d %02d:%02d:%02d",
        tm_time.tm_year + 1900, tm_time.tm_mon + 1, tm_time.tm_mday,
        tm_time.tm_hour + 8, tm_time.tm_min, tm_time.tm_sec);
        assert(len == 19); (void)len;
    }

    Fmt us(".%06d ", microseconds);
    assert(us.length() == 8);
    m_stream << t_time << us.data();
}