IO Cache 对于 Mysql Binary log 写入的影响

 

由于工作需要，需要确认IO_CACHE对于mysql binary log的安全性的影响，在系统崩溃的情况下对于存在IO_CACHE是否会影响数据的安全性。本文主要根据mysql 源代码分析mysql binary log 中IO_CACHE的相关原理的应用。

 

在mysql的源码上MYSQL_BIN_LOG类有2份实例，分别对应mysql binary log和 mysql relay log.本文主要对 mysql binary log 的write方法进行分析，观察在写入是由于缓存的存在是否对系统造成数据不能持久化行为。

 

相关代码如下：

sql/log.c (暂时去除row based 模式的binary log代码)

/** Write an event to the binary log. */ bool MYSQL_BIN_LOG::write(Log_event *event_info) { THD *thd= event_info->thd; bool error= 1; DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)"); if (thd->binlog_evt_union.do_union) { /* In Stored function; Remember that function call caused an update. We will log the function call to the binary log on function exit */ thd->binlog_evt_union.unioned_events= TRUE; thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt; DBUG_RETURN(0); } ...... /* Write the SQL command */ if (event_info->write(file) || DBUG_EVALUATE_IF("injecting_fault_writing", 1, 0)) goto err; if (file == &log_file) // we are writing to the real log (disk) { if (flush_and_sync()) goto err; signal_update(); rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED); } error=0; err: if (error) set_write_error(thd); } pthread_mutex_unlock(&LOCK_log); DBUG_RETURN(error); }  

由此可见在涉及IO_CACHE时有2个动作值得关注，及在log event 类中的write动作和本类中的flush_and_sync动作。我从客户端出发了一个insert动作，服务器端跟踪 log event中的write动作可以观察到:

 

 

 

 

// Query_log_event 为log event 的一个子类 bool Query_log_event::write(IO_CACHE* file) { .... return (write_header(file, event_length) || my_b_safe_write(file, (uchar*) buf, QUERY_HEADER_LEN) || write_post_header_for_derived(file) || my_b_safe_write(file, (uchar*) start_of_status, (uint) (start-start_of_status)) || my_b_safe_write(file, (db) ? (uchar*) db : (uchar*)"", db_len + 1) || my_b_safe_write(file, (uchar*) query, q_len)) ? 1 : 0; } } 

 

 

在Query_log_event::write 前部为向buffer中放入需要写入的header 和body,最后调用my_b_safe_write 写入。下面为my_b_safe_write函数。

 

Mf_iocache.c

int my_b_safe_write(IO_CACHE *info, const uchar *Buffer, size_t Count) { /* Sasha: We are not writing this with the ? operator to avoid hitting a possible compiler bug. At least gcc 2.95 cannot deal with several layers of ternary operators that evaluated comma(,) operator expressions inside - I do have a test case if somebody wants it */ if (info->type == SEQ_READ_APPEND) return my_b_append(info, Buffer, Count); return my_b_write(info, Buffer, Count); }  

 

最后调用宏 my_b_write， 此宏中只是做了IO_CACHE的memory copy动作，数据到此为止并不安全。

#define my_b_write(info,Buffer,Count) / ((info)->write_pos + (Count) <=(info)->write_end ?/ (memcpy((info)->write_pos, (Buffer), (size_t)(Count)),/ ((info)->write_pos+=(Count)),0) : / (*(info)->write_function)((info),(uchar *)(Buffer),(Count))) 

 

而真正的写入是在前面所讲的flush_and_sync动作中

// log.c 先调用flush_io_cache 最后sync数据到磁盘 bool MYSQL_BIN_LOG::flush_and_sync() { int err=0, fd=log_file.file; safe_mutex_assert_owner(&LOCK_log); if (flush_io_cache(&log_file)) return 1; if (++sync_binlog_counter >= sync_binlog_period && sync_binlog_period) { sync_binlog_counter= 0; err=my_sync(fd, MYF(MY_WME)); } return err; } // My_sys.h #define flush_io_cache(info) my_b_flush_io_cache((info),1) //Mf_iocache.c 中最后写入disk int my_b_flush_io_cache(IO_CACHE *info, int need_append_buffer_lock) { ... if (my_write(info->file,info->write_buffer,length, info->myflags | MY_NABP)) info->error= -1; ... }  

 

下面是对应的堆栈：

 

 

(gdb) bt #0 my_b_safe_write (info=0xe10270, Buffer=0x46c890c0 "/f6蒐/002/001", Count=19) at mf_iocache.c:1623 #1 0x000000000066dfab in Log_event::write_header (this=0x46c894f0, file=0xe10270, event_data_length=<value optimized out>) at log_event.cc:959 #2 0x000000000066ec61 in Query_log_event::write (this=0x46c894f0, file=0xe10270) at log_event.cc:2341 #3 0x000000000064eb75 in MYSQL_BIN_LOG::write (this=0xe0ff20, event_info=0x46c894f0) at log.cc:4369 #4 0x000000000058a8f9 in THD::binlog_query (this=0x1b70fbb0, qtype=THD::ROW_QUERY_TYPE, query_arg=0x1b764dd0 "insert into example value(1,2)", query_len=30, is_trans=false, suppress_use=false, errcode=0) at sql_class.cc:3961 #5 0x000000000062e0de in mysql_insert (thd=0x1b70fbb0, table_list=<value optimized out>, fields=@0x1b711f90, values_list=@0x1b711fd8, update_fields=@0x1b711fc0, update_values=@0x1b711fa8, duplic=DUP_ERROR, ignore=false) at sql_insert.cc:921 #6 0x00000000005b2845 in mysql_execute_command (thd=0x1b70fbb0) at sql_parse.cc:3206 #7 0x00000000005b30ae in mysql_parse (thd=0x1b70fbb0, inBuf=0x1b764dd0 "insert into example value(1,2)", length=30, found_semicolon=0x46c8add0) at sql_parse.cc:5994 #8 0x00000000005b432a in dispatch_command (command=COM_QUERY, thd=0x1b70fbb0, packet=0x1b760da1 "insert into example value(1,2)", packet_length=<value optimized out>) at sql_parse.cc:1241 #9 0x00000000005b5316 in do_command (thd=0x1b70fbb0) at sql_parse.cc:882 #10 0x00000000005a4f58 in handle_one_connection (arg=<value optimized out>) at sql_connect.cc:1134 #11 0x00000033e12064a7 in start_thread () from /lib64/libpthread.so.0 #12 0x00000033e0ad3c2d in clone () from /lib64/libc.so.6 

 

由代码可见，IO_CACHE的存在对于binary log 的数据安全存在一定的隐患，但是相比与innodb中的buffer cache来说，binary log 在写入数据后立即flush和sync动作保证了数据不会大量丢失。