php 开启coredump,GDB分析PHP连接Memcached 导致coredump问题

最近一项目压力比较大，502超多，PHP进程coredump比较频繁

于是学习了下coredump的调试

讲下PHP  coredump的调试，coredump先准备代码比较好，因为PHP是开源的，这点好办。我的是存在/usr/local/src//usr/local/src/php-5.3.10/ 下

第一种当然是通用的调试

gdb /usr/local/php/sbin/php-fpm -c /data/core_files/core-php-fpm-28436-1370008094

第二种是PHP特有的调试

获得一个PHP正在运行的进程

#ps -ef|grep php-fpm|head

#gdb --pid=$PID

>source /usr/local/src//usr/local/src/php-5.3.10/.gdbinit

这种方式能调试一些PHP特有的东西，比如包括参数的函数调用栈的信息.

先讲通用方式调试

(gdb) bt

#0 0x00007ffed0603a75 in raise () from /lib/libc.so.6

#1 0x00007ffed06075c0 in abort () from /lib/libc.so.6

#2 0x00007ffed342ef7c in memcached_io_slurp (ptr=0x2ef9f50) at libmemcached/io.cc:510

#3 0x00007ffed3431da6 in memcached_quit_server (ptr=0x2ef9f50, io_death=false) at libmemcached/quit.cc:95

#4 0x00007ffed3431eed in send_quit (ptr=0x2d34e90) at libmemcached/quit.cc:137

#5 0x00007ffed342ff39 in _free (ptr=0x2d34e90, release_st=true) at libmemcached/memcached.cc:130

#6 0x00007ffed3430397 in memcached_free (ptr=0x2d34e90) at libmemcached/memcached.cc:284

#7 0x00000000005d00e0 in php_memc_destroy (m_obj=0x2e3bf60, persistent=0 '\000') at /usr/local/src/php-5.3.10/ext/memcached/php_memcached.c:2519

#8 0x00000000005d145f in php_memc_free_storage (i_obj=0x2dea580) at /usr/local/src/php-5.3.10/ext/memcached/php_memcached.c:2530

#9 0x000000000081817a in zend_objects_store_del_ref_by_handle_ex (handle=, handlers=) at /usr/local/src/php-5.3.10/Zend/zend_objects_API.c:220

#10 0x0000000000818193 in zend_objects_store_del_ref (zobject=0x2dea620) at /usr/local/src/php-5.3.10/Zend/zend_objects_API.c:172

#11 0x00000000007e9fb5 in _zval_dtor (zval_ptr=) at /usr/local/src/php-5.3.10/Zend/zend_variables.h:35

#12 _zval_ptr_dtor (zval_ptr=) at /usr/local/src/php-5.3.10/Zend/zend_execute_API.c:447

#13 0x00000000008020eb in zend_hash_destroy (ht=0x2d21e20) at /usr/local/src/php-5.3.10/Zend/zend_hash.c:529

#14 0x00000000007f5416 in _zval_dtor_func (zvalue=0x2d21df0) at /usr/local/src/php-5.3.10/Zend/zend_variables.c:43

#15 0x00000000007e9fb5 in _zval_dtor (zval_ptr=) at /usr/local/src/php-5.3.10/Zend/zend_variables.h:35

#16 _zval_ptr_dtor (zval_ptr=) at /usr/local/src/php-5.3.10/Zend/zend_execute_API.c:447

#17 0x00000000008023f3 in zend_hash_clean (ht=0x2d21cb0) at /usr/local/src/php-5.3.10/Zend/zend_hash.c:561

#18 0x00000000007ee2cb in zend_cleanup_op_array_data (function=) at /usr/local/src/php-5.3.10/Zend/zend_opcode.c:133

#19 zend_cleanup_function_data (function=) at /usr/local/src/php-5.3.10/Zend/zend_opcode.c:140

#20 0x0000000000801ea4 in zend_hash_reverse_apply (ht=0x29e5c30, apply_func=0x7ee2a0 ) at /usr/local/src/php-5.3.10/Zend/zend_hash.c:757

#21 0x00000000007ea2f3 in shutdown_executor () at /usr/local/src/php-5.3.10/Zend/zend_execute_API.c:296

#22 0x00000000007f6563 in zend_deactivate () at /usr/local/src/php-5.3.10/Zend/zend.c:891

#23 0x00000000007a47a0 in php_request_shutdown (dummy=) at /usr/local/src/php-5.3.10/main/main.c:1661

#24 0x0000000000884442 in main (argc=, argv=) at /usr/local/src/php-5.3.10/sapi/fpm/fpm/fpm_main.c:1886

(gdb) f 2

#2 0x00007ffed342ef7c in memcached_io_slurp (ptr=0x2ef9f50) at libmemcached/io.cc:510

510 libmemcached/io.cc: No such file or directory.

in libmemcached/io.cc

为了调试memcached，会碰到了提示说 libmemcached/io.cc不存在，于是也把libmemcached的源码也下载回来。

注意版本得配对，否则容易造成找代码行数找不准，就白瞎了。

warning: Source file is more recent than executable.

看版本用 -i来看

#/usr/local/php/sbin/php-fpm -i|grep libmemcached

libmemcached version => 1.0.4

#wget https://launchpad.net/libmemcached/1.0/1.0.4/+download/libmemcached-1.0.4.tar.gz

把代码解压到php源码目录下  ，如果出错可能是路径不对

继续回来gdb

(gdb) f 2

#2 0x00007ffed342ef7c in memcached_io_slurp (ptr=0x2ef9f50) at libmemcached/io.cc:510

510 in libmemcached/io.cc

看到是libmemcached/io.cc  的510行报错

#vim  libmemcached/io.cc +510

看到代码如下：

memcached_return_t memcached_io_slurp(memcached_server_write_instance_st ptr)

{

assert_msg(ptr, "Programmer error, invalid memcached_server_write_instance_st");

assert(memcached_is_udp(ptr->root) == false);

if (ptr->fd == INVALID_SOCKET)

{

assert_msg(int(ptr->state) <= int(MEMCACHED_SERVER_STATE_ADDRINFO), "Invalid socket state"); #line 510

return MEMCACHED_CONNECTION_FAILURE;

}

ssize_t data_read;

这是一个无效的连接？ 除非说是网络非常繁忙，或压力很大。

往下看看吧

(gdb) f 3

#3 0x00007ffed3431da6 in memcached_quit_server (ptr=0x2ef9f50, io_death=false) at libmemcached/quit.cc:95

warning: Source file is more recent than executable.

95 while (memcached_continue(rc_slurp= memcached_io_slurp(ptr))) {} ;

是在quit函数中，进代码看看

WATCHPOINT_ASSERT(rc == MEMCACHED_SUCCESS or rc == MEMCACHED_FETCH_NOTFINISHED);

/* read until socket is closed, or there is an error

* closing the socket before all data is read

* results in server throwing away all data which is

* not read

*

* In .40 we began to only do this if we had been doing buffered

* requests of had replication enabled.

*/

if (ptr->root->flags.buffer_requests or ptr->root->number_of_replicas)

{

memcached_return_t rc_slurp;

while (memcached_continue(rc_slurp= memcached_io_slurp(ptr))) {} ; #line 95

WATCHPOINT_ASSERT(rc_slurp == MEMCACHED_CONNECTION_FAILURE);

}

/*

* memcached_io_read may call memcached_quit_server with io_death if

* it encounters problems, but we don't care about those occurences.

* The intention of that loop is to drain the data sent from the

* server to ensure that the server processed all of the data we

* sent to the server.

*/

ptr->server_failure_counter= 0;

}

memcached_io_close(ptr);

}

slurp的意思是：大量读取

看看注释的意思：等待socket 正常关闭或没有读完数据但是出现了错误引起的异常关闭，结果是服务器扔掉了所有没有读完的数据。

95行的代码是一个条件循环，条件是memcached_io_slurp 的结果

继续分析函数  memcached_io_slurp

发现是等待读完buffer的过程

总体的意思是说服务端开启了buffer，PHP发送了退出请求，所以memcached扩展也被要求退出，然而buffer可能没有读完，所以memcached等待buffer读完，而php等待memcached退出后才能退出

(gdb) f 7

#7 0x00000000005d00e0 in php_memc_destroy (m_obj=0x2e3bf60, persistent=0 '\000') at /usr/local/src/php-5.3.10/ext/memcached/php_memcached.c:2519

2519 in /usr/local/src/php-5.3.10/ext/memcached/php_memcached.c

一路往后f

(gdb) f 24

#24 0x0000000000884442 in main (argc=, argv=) at /usr/local/src/php-5.3.10/sapi/fpm/fpm/fpm_main.c:1886

1886 php_request_shutdown((void *) 0);

看到是 fpm请求退出进程 php_request_shutdown

fastcgi_request_done:

if (primary_script) {

efree(primary_script);

}

if (request_body_fd != -1) {

close(request_body_fd);

}

request_body_fd = -2;

if (EG(exit_status) == 255) {

if (CGIG(error_header) && *CGIG(error_header)) {

sapi_header_line ctr = {0};

ctr.line = CGIG(error_header);

ctr.line_len = strlen(CGIG(error_header));

sapi_header_op(SAPI_HEADER_REPLACE, &ctr TSRMLS_CC);

}

}

fpm_request_end(TSRMLS_C);

fpm_log_write(NULL TSRMLS_CC);

STR_FREE(SG(request_info).path_translated);

SG(request_info).path_translated = NULL;

php_request_shutdown((void *) 0); #line 1886

if (exit_status == 0) {

exit_status = EG(exit_status);

}

requests++;

if (max_requests && (requests == max_requests)) {

fcgi_finish_request(&request, 1);

if (max_requests != 1) {

/* no need to return exit_status of the last request */

exit_status = 0;

好像还没有结果。虽然还没查到具体的原因，大概问题归结在memcached扩展和memcached服务器通讯的时候，数据还未传完，

PHP进程就要结束，导致coredump。