ITPub博客

首页 > 数据库 > PostgreSQL > PostgreSQL 源码解读(157)- 后台进程#9(同步复制主库挂起分析)

PostgreSQL 源码解读(157)- 后台进程#9(同步复制主库挂起分析)

原创 PostgreSQL 作者:husthxd 时间:2019-03-26 14:04:56 0 删除 编辑

在Streaming Replication环境中PostgreSQL主节点设置为同步复制,如standby节点没有启动或者网络出现问题没法连接到主节点时,主节点如执行DML则进程会挂起,下面分析这个挂起的问题.

一、数据结构

Latch
Latch结构体应被视为opaque”不透明的”,并且只能通过公共的函数访问.在这里定义是运行把Latchs作为更大的结构体的一部分.


//通常情况下,int类型的变量通常是原子访问的,也可以认为 sig_atomic_t就是int类型的数据,
//因为对这些变量要求一条指令完成,所以sig_atomic_t不可能是结构体,只会是数字类型。
typedef int __sig_atomic_t;
/*
 * Latch structure should be treated as opaque and only accessed through
 * the public functions. It is defined here to allow embedding Latches as
 * part of bigger structs.
 * Latch结构体应被视为"不透明的"opaque,并且只能通过公共的函数访问.
 * 在这里定义是运行把Latchs作为更大的结构体的一部分.
 */
typedef struct Latch
{
    sig_atomic_t is_set;
    bool        is_shared;
    int         owner_pid;
#ifdef WIN32
    HANDLE      event;
#endif
} Latch;

二、源码解读

N/A

二、跟踪分析

启动master节点,不启动standby节点,使用psql连接数据库,执行SQL,Session挂起:


testdb=# drop table t1;

使用gdb跟踪挂起的进程


[xdb@localhost ~]$ ps -ef|grep postgres
xdb       1318     1  0 12:14 pts/0    00:00:00 /appdb/xdb/pg11.2/bin/postgres
xdb       1319  1318  0 12:14 ?        00:00:00 postgres: logger   
xdb       1321  1318  0 12:14 ?        00:00:00 postgres: checkpointer   
xdb       1322  1318  0 12:14 ?        00:00:00 postgres: background writer   
xdb       1323  1318  0 12:14 ?        00:00:00 postgres: walwriter   
xdb       1324  1318  0 12:14 ?        00:00:00 postgres: autovacuum launcher  
xdb       1325  1318  0 12:14 ?        00:00:00 postgres: archiver   
xdb       1326  1318  0 12:14 ?        00:00:00 postgres: stats collector   
xdb       1327  1318  0 12:14 ?        00:00:00 postgres: logical replication launcher  
xdb       1331  1318  0 12:15 ?        00:00:00 postgres: xdb testdb [local] DROP TABLE waiting for 0/5D07B668
[xdb@localhost ~]$ gdb -p 1331
GNU gdb (GDB) Red Hat Enterprise Linux 7.6.1-100.el7
...

查看调用栈


(gdb) bt
#0  0x00007f4636d48903 in __epoll_wait_nocancel () from /lib64/libc.so.6
#1  0x000000000088e668 in WaitEventSetWaitBlock (set=0x21640e8, cur_timeout=-1, occurred_events=0x7ffc96572f40, nevents=1)
    at latch.c:1048
#2  0x000000000088e543 in WaitEventSetWait (set=0x21640e8, timeout=-1, occurred_events=0x7ffc96572f40, nevents=1, 
    wait_event_info=134217761) at latch.c:1000
#3  0x000000000088dcec in WaitLatchOrSocket (latch=0x7f462d5b44d4, wakeEvents=17, sock=-1, timeout=-1, 
    wait_event_info=134217761) at latch.c:385
#4  0x000000000088dbcd in WaitLatch (latch=0x7f462d5b44d4, wakeEvents=17, timeout=-1, wait_event_info=134217761)
    at latch.c:339
#5  0x0000000000863e2d in SyncRepWaitForLSN (lsn=1560786536, commit=true) at syncrep.c:286
#6  0x0000000000546279 in RecordTransactionCommit () at xact.c:1359
#7  0x0000000000546da3 in CommitTransaction () at xact.c:2074
#8  0x0000000000547a3f in CommitTransactionCommand () at xact.c:2817
#9  0x00000000008be250 in finish_xact_command () at postgres.c:2523
#10 0x00000000008bbf45 in exec_simple_query (query_string=0x20a1d78 "drop table t1;") at postgres.c:1170
#11 0x00000000008c0191 in PostgresMain (argc=1, argv=0x20cdcd8, dbname=0x20cdb40 "testdb", username=0x209ea98 "xdb")
    at postgres.c:4182
#12 0x000000000081e06c in BackendRun (port=0x20c3b10) at postmaster.c:4361
#13 0x000000000081d7df in BackendStartup (port=0x20c3b10) at postmaster.c:4033
#14 0x0000000000819bd9 in ServerLoop () at postmaster.c:1706
#15 0x000000000081948f in PostmasterMain (argc=1, argv=0x209ca50) at postmaster.c:1379
#16 0x0000000000742931 in main (argc=1, argv=0x209ca50) at main.c:228
(gdb)

kill进程,重新进入在WaitLatch上设置断点进行跟踪


#########
[xdb@localhost ~]$ kill -9 1331
#########
testdb=# select pg_backend_pid();
 pg_backend_pid 
----------------
           1377
(1 row)
#########
[xdb@localhost ~]$ gdb -p 1377
...
(gdb) b WaitLatch
Breakpoint 1 at 0x88dbac: file latch.c, line 339.
(gdb) 
#########
testdb=# drop table t1;
ERROR:  table "t1" does not exist
testdb=# create table t1(id int);

进入断点


(gdb) b WaitLatch
Breakpoint 1 at 0x88dbac: file latch.c, line 339.
(gdb) c
Continuing.
Breakpoint 1, WaitLatch (latch=0x7f462d5b44d4, wakeEvents=17, timeout=-1, wait_event_info=134217761) at latch.c:339
339     return WaitLatchOrSocket(latch, wakeEvents, PGINVALID_SOCKET, timeout,
(gdb)

进入WaitLatchOrSocket


(gdb) step
WaitLatchOrSocket (latch=0x7f462d5b44d4, wakeEvents=17, sock=-1, timeout=-1, wait_event_info=134217761) at latch.c:359
359     int         ret = 0;
(gdb) 
(gdb) p *latch
$1 = {is_set = 0, is_shared = true, owner_pid = 1377}

构建等待事件集


(gdb) n
362     WaitEventSet *set = CreateWaitEventSet(CurrentMemoryContext, 3);
(gdb) n
364     if (wakeEvents & WL_TIMEOUT)
(gdb) 
367         timeout = -1;
(gdb) 
369     if (wakeEvents & WL_LATCH_SET)
(gdb) p *set
$2 = {nevents = 0, nevents_space = 3, events = 0x2181eb8, latch = 0x0, latch_pos = 0, epoll_fd = 37, 
  epoll_ret_events = 0x2181f00}
(gdb) p *set->events
$3 = {pos = 0, events = 0, fd = 0, user_data = 0x0}
(gdb) p *set->epoll_ret_events
$4 = {events = 0, data = {ptr = 0x0, fd = 0, u32 = 0, u64 = 0}}
(gdb) 
$5 = {events = 0, data = {ptr = 0x0, fd = 0, u32 = 0, u64 = 0}}
(gdb) n
370         AddWaitEventToSet(set, WL_LATCH_SET, PGINVALID_SOCKET,
(gdb) 
373     if (wakeEvents & WL_POSTMASTER_DEATH && IsUnderPostmaster)
(gdb) 
374         AddWaitEventToSet(set, WL_POSTMASTER_DEATH, PGINVALID_SOCKET,
(gdb) 
377     if (wakeEvents & WL_SOCKET_MASK)
(gdb) 
385     rc = WaitEventSetWait(set, timeout, &event, 1, wait_event_info);
(gdb) p *set
$6 = {nevents = 2, nevents_space = 3, events = 0x2181eb8, latch = 0x7f462d5b44d4, latch_pos = 0, epoll_fd = 37, 
  epoll_ret_events = 0x2181f00}
(gdb) p *set->events
$7 = {pos = 0, events = 1, fd = 11, user_data = 0x0}
(gdb) p *set->epoll_ret_events
$8 = {events = 0, data = {ptr = 0x0, fd = 0, u32 = 0, u64 = 0}}
(gdb)

进入WaitEventSetWait


(gdb) step
WaitEventSetWait (set=0x2181e90, timeout=-1, occurred_events=0x7ffc96572f40, nevents=1, wait_event_info=134217761)
    at latch.c:925
925     int         returned_events = 0;
(gdb)

输入参数


(gdb) n
928     long        cur_timeout = -1;
(gdb) p *set
$9 = {nevents = 2, nevents_space = 3, events = 0x2181eb8, latch = 0x7f462d5b44d4, latch_pos = 0, epoll_fd = 37, 
  epoll_ret_events = 0x2181f00}
(gdb) p *occurred_events
$10 = {pos = 35135068, events = 0, fd = -1772664741, user_data = 0x7ffc96572fa0}
(gdb)

执行相关判断和设置参数


(gdb) n
930     Assert(nevents > 0);
(gdb) 
936     if (timeout >= 0)
(gdb) 
943     pgstat_report_wait_start(wait_event_info);
(gdb) 
946     waiting = true;
(gdb)

未有事件出现,则循环


951     while (returned_events == 0)
(gdb)

不符合set->latch->is_set为T的条件,继续循环


982         if (set->latch && set->latch->is_set)
(gdb) p *set->latch
$11 = {is_set = 0, is_shared = true, owner_pid = 1377}
(gdb)

进入WaitEventSetWaitBlock


(gdb) n
1000            rc = WaitEventSetWaitBlock(set, cur_timeout,
(gdb) step
WaitEventSetWaitBlock (set=0x2181e90, cur_timeout=-1, occurred_events=0x7ffc96572f40, nevents=1) at latch.c:1042
1042        int         returned_events = 0;
(gdb)

调用epoll_wait,挂起


(gdb) n
1048        rc = epoll_wait(set->epoll_fd, set->epoll_ret_events,
(gdb) p *set
$12 = {nevents = 2, nevents_space = 3, events = 0x2181eb8, latch = 0x7f462d5b44d4, latch_pos = 0, epoll_fd = 37, 
  epoll_ret_events = 0x2181f00}
(gdb) 
(gdb) n

启动standby节点


####
[xdb@localhost ~]$ pg_ctl start
pg_ctl: another server might be running; trying to start server anyway
...

接收到信号


Program received signal SIGUSR1, User defined signal 1.
0x00007f4636d48903 in __epoll_wait_nocancel () from /lib64/libc.so.6
(gdb) 
(gdb) n
Single stepping until exit from function __epoll_wait_nocancel,
which has no line number information.
procsignal_sigusr1_handler (postgres_signal_arg=-1) at procsignal.c:262
262 {
(gdb)

DONE!

四、参考资料

PG Source Code

来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/6906/viewspace-2639372/,如需转载,请注明出处,否则将追究法律责任。

全部评论
ITPUB数据库版块资深版主,对Oracle、PostgreSQL有深入研究。

注册时间:2007-12-28

  • 博文量
    1504
  • 访问量
    3974686