共计 7186 个字符，预计需要花费 18 分钟才能阅读完成。
这篇文章主要讲解了“PostgreSQL 同步复制主库挂起分析”，文中的讲解内容简单清晰，易于学习与理解，下面请大家跟着丸趣 TV 小编的思路慢慢深入，一起来研究和学习“PostgreSQL 同步复制主库挂起分析”吧！
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)感谢各位的阅读，以上就是“PostgreSQL 同步复制主库挂起分析”的内容了，经过本文的学习后，相信大家对 PostgreSQL 同步复制主库挂起分析这一问题有了更深刻的体会，具体使用情况还需要大家实践验证。这里是丸趣 TV，丸趣 TV 小编将为大家推送更多相关知识点的文章，欢迎关注！
正文完
