MySQL 几种调式分析利器

目录

---

---

pstack

获取堆栈信息
问题线程的定位
负载较低

mysql_pid=4522
pstack $mysql_pid>pstack.info

pt-pmp对堆栈信息排序
pt-pmp pstack.info | less

也可以直接执行pt-pmp
pt-pmp --pid 4522
如

     10 __io_getevents_0_4(libaio.so.1),linuxaiohandler::collect(os0file.cc:2502),linuxaiohandler::poll(os0file.cc:2648),os_aio_linux_handler(os0file.cc:2704),os_aio_handler(os0file.cc:2704),fil_aio_wait(fil0fil.cc:5835),io_handler_thread(srv0start.cc:311),start_thread(libpthread.so.0),clone(libc.so.6)
      3 pthread_cond_wait,wait(os0event.cc:165),os_event::wait_low(os0event.cc:165),srv_worker_thread(srv0srv.cc:2520),start_thread(libpthread.so.0),clone(libc.so.6)
      2 pthread_cond_wait,native_cond_wait(thr_cond.h:140),my_cond_wait(thr_cond.h:140),inline_mysql_cond_wait(thr_cond.h:140),per_thread_connection_handler::block_until_new_connection(thr_cond.h:140),handle_connection(connection_handler_per_thread.cc:329),pfs_spawn_thread(pfs.cc:2190),start_thread(libpthread.so.0),clone(libc.so.6)

gdb

堆栈跟踪
等待分析
侵入性大

mysqld_pid 4522
gdb -p $mysqld_pid
(gdb) info thread 显示运行的所有线程
(gdb) thread 4 切换到某个线程
bt 显示调用栈
如

select b.thread_os_id as mysqld_os_thread_id,a.id as processlist_id,a.user,a.host,a.db,a.command,a.time,a.state,left(a.info,150) as statement from information_schema.processlist a inner join performance_schema.threads b on a.id=b.processlist_id where a.id !=connection_id();

+---------------------+----------------+-------+-----------+------+---------+------+---------------------------------+----------------------------------------+
| mysqld_os_thread_id | processlist_id | user | host | db | command | time | state | statement |
+---------------------+----------------+-------+-----------+------+---------+------+---------------------------------+----------------------------------------+
| 27015 | 4 | admin | localhost | null | sleep | 2597 | | null |
| 28252 | 6 | admin | localhost | test | query | 1166 | waiting for table metadata lock | alter table test.test_1 drop index idx |
+---------------------+----------------+-------+-----------+------+---------+------+---------------------------------+----------------------------------------+
(gdb) thread 4
[switching to thread 4 (thread 0x7f8d107f8700 (lwp 28252))]
#0 0x00007f8d3b147d12 in pthread_cond_timedwait@@glibc_2.3.2 () from /lib64/libpthread.so.0

bt
........
#4 mdl_wait::timed_wait (this=0x7f8ce4000958, owner=0x7f8ce40008c0, abs_timeout=0x7f8d107f37e0, set_status_on_timeout=false,
wait_state_name=) at /export/home/pb2/build/sb_0-27500212-1520171728.22/mysql-5.7.22/sql/mdl.cc:1861

• processlist_id 为6（ os中 mysqld的线程id为28252）正在等待mdl 元数据锁，通过gdb堆栈跟踪发现，该连接正在执行的内部函数为 pthread_cond_timedwait

注意

• pstack 和gdb都可以获取堆栈信息，区别是gdb是交互式的（需要手动退出），pstack是快照形式（执行3s自动退出）
• 对mysqld使用gdb后，已经建立的连接无法再执行语句，新连接无法建立

strace

查看进程的系统调用信息
负载较高

查看系统调用
strace -cp $mysqld_pid

查看执行语句
mysqld_pid=4522
strace -f -f -ff -o mysqld-strace -s 1024 -p $mysqld_pid
find ./ -name "mysqld-strace" -type f -print |xargs grep -n "select.from"

查看读写的文件
strace -o /tmp/strace_output.txt -t -tt -f -e trace=read,open -p “mysqld_pid”

perf

进程内部函数调用情况
负载较低

mysqld内部函数整体消耗情况
perf top -p 4522
如

samples: 66  of event 'cpu-clock', event count (approx.): 6520074                                                                 
overhead  shared object       symbol                                                                                              
  18.42%  mysqld              [.] sync_array_print_long_waits_low
  10.83%  [kernel]            [k] _raw_spin_unlock_irqrestore
   7.88%  libaio.so.1.0.1     [.] 0x0000000000000645
   7.30%  [kernel]            [k] finish_task_switch
   6.84%  [kernel]            [k] system_call_after_swapgs
   6.22%  [kernel]            [k] aio_read_events
   6.02%  libc-2.17.so        [.] __memset_sse2

记录所有内部函数调用
该命令会输出记录到perf.data中
perf record -p 4522

解析并查看perf.data内容
perf script -i perf.data>perf.log && less perf.log

参考

mysql所有操作hang住问题的故障排查


debugging a mysql server
what to do if mysql keeps crashing