一、 插件作用
众所周知，pg是没有像oracle那样的ash视图的，因此要回溯历史问题不太方便。pgsentinel插件会将pg_stat_activity与pg_stat_statements视图内容定期快照，并存入pg_active_session_history和pg_stat_statements_history视图中。
1. pg_active_session_history视图字段
| Column | Type | 备注 |
|---|---|---|
| ash_time | timestamp with time zone | 采样时间 |
| datid | oid | |
| datname | text | |
| pid | integer | |
| leader_pid | integer | 若有并行，其leader进程的pid |
| usesysid | oid | user id |
| usename | text | |
| application_name | text | |
| client_addr | text | |
| client_hostname | text | |
| client_port | integer | |
| backend_start | timestamp with time zone | |
| xact_start | timestamp with time zone | |
| query_start | timestamp with time zone | |
| state_change | timestamp with time zone | |
| wait_event_type | text | |
| wait_event | text | |
| state | text | |
| backend_xid | xid | |
| backend_xmin | xid | |
| top_level_query | text | 执行函数、存储过程时的外层SQL（开pg_stat_statements.track = all才会有区别） |
| query | text | |
| cmdtype | text | |
| queryid | bigint | |
| backend_type | text | |
| blockers | integer | blockers数量 |
| blockerpid | integer | |
| blocker_state | text |
2. pg_stat_statements_history视图字段
与对应版本的pg_stat_statements视图字段含义相同
| Column | Type | 备注 |
|---|---|---|
| ash_time | timestamp with time zone | |
| userid | oid | |
| dbid | oid | |
| queryid | bigint | |
| calls | bigint | |
| total_exec_time | double precision | |
| rows | bigint | |
| shared_blks_hit | bigint | |
| shared_blks_read | bigint | |
| shared_blks_dirtied | bigint | |
| shared_blks_written | bigint | |
| local_blks_hit | bigint | |
| local_blks_read | bigint | |
| local_blks_dirtied | bigint | |
| local_blks_written | bigint | |
| temp_blks_read | bigint | |
| temp_blks_written | bigint | |
| blk_read_time | double precision | |
| blk_write_time | double precision | |
| plans | bigint | |
| total_plan_time | double precision | |
| wal_records | bigint | |
| wal_fpi | bigint | |
| wal_bytes | numeric |
二、 插件安装配置
1. 下载
GitHub - pgsentinel/pgsentinel: postgresql extension providing Active session history
2. 安装
bash
# poatgres用户执行
unzip pgsentinel-master.zip
cd pgsentinel-master/src
make
# root用户执行（要配环境变量，参考下面）
make install具体安装过程
-bash-4.2$ unzip pgsentinel-master.zip
-bash-4.2$ cd pgsentinel-master/src
-bash-4.2$ make
gcc -std=gnu99 -Wall -Wmissing-prototypes -Wpointer-arith -Wdeclaration-after-statement -Werror=vla -Wendif-labels -Wmissing-format-attribute -Wformat-security -fno-strict-aliasing -fwrapv -fexcess-precision=standard -g -O2 -fPIC -I. -I./ -I/data/postgres/base/14.0/include/server -I/data/postgres/base/14.0/include/internal -D_GNU_SOURCE -c -o pgsentinel.o pgsentinel.c
gcc -std=gnu99 -Wall -Wmissing-prototypes -Wpointer-arith -Wdeclaration-after-statement -Werror=vla -Wendif-labels -Wmissing-format-attribute -Wformat-security -fno-strict-aliasing -fwrapv -fexcess-precision=standard -g -O2 -fPIC -I. -I./ -I/data/postgres/base/14.0/include/server -I/data/postgres/base/14.0/include/internal -D_GNU_SOURCE -c -o get_parsedinfo.o get_parsedinfo.c
gcc -std=gnu99 -Wall -Wmissing-prototypes -Wpointer-arith -Wdeclaration-after-statement -Werror=vla -Wendif-labels -Wmissing-format-attribute -Wformat-security -fno-strict-aliasing -fwrapv -fexcess-precision=standard -g -O2 -fPIC -shared -o pgsentinel.so pgsentinel.o get_parsedinfo.o -L/data/postgres/base/14.0/lib -Wl,--as-needed -Wl,-rpath,'/data/postgres/base/14.0/lib',--enable-new-dtags -lm
[root@linux01 ~]# vi .bash_profile
.bash_profile
Get the aliases and functions
if [ -f ~/.bashrc ]; then
. ~/.bashrc
fi
User specific environment and startup programs
PATH=PATH:HOME/bin
export PATH
export PGHOME=/data/postgres/base/14.0
export PGDATA=/data/postgres/pg5432/data
export PATH=PGHOME/bin:PATH:$HOME/bin
export LD_LIBRARY_PATH=LD_LIBRARY_PATH:PGHOME/lib
export LANG=en_US.UTF-8
~
[root@linux01 ~]# source .bash_profile
[root@linux01 ~]#
[root@linux01 ~]# cd .../pgsentinel-master/src
[root@linux01 src]# make install
/usr/bin/mkdir -p '/data/postgres/base/14.0/lib'
/usr/bin/mkdir -p '/data/postgres/base/14.0/share/extension'
/usr/bin/mkdir -p '/data/postgres/base/14.0/share/extension'
/usr/bin/install -c -m 755 pgsentinel.so '/data/postgres/base/14.0/lib/pgsentinel.so'
/usr/bin/install -c -m 644 .//pgsentinel.control '/data/postgres/base/14.0/share/extension/'
/usr/bin/install -c -m 644 .//pgsentinel--1.0.sql '/data/postgres/base/14.0/share/extension/'
创建插件
sql
CREATE EXTENSION pgsentinel;3. 插件配置
- 必须配置
bash
vi postgresql.conf
shared_preload_libraries = 'pg_stat_statements,auto_explain,pgsentinel'若未配置，查询会报错
sql
postgres=# select * from pg_active_session_history ;
ERROR: pg_active_session_history must be loaded via shared_preload_libraries重启db生效
bash
pg_ctl stop -m fast
pg_ctl start -D $PGDATA
postgres=# select * from pg_active_session_history ;
(0 rows)- 可选配置
也可以直接在postgresql.conf中修改
| 参数名 | Description | 默认值 | 建议值 |
|---|---|---|---|
| pgsentinel_ash.sampling_period | 采样时间（秒） | 1 | 10（高负载） |
| pgsentinel_ash.max_entries | pg_active_session_history视图在内存中的缓冲区大小（字节） | 1000 | 10000 |
| pgsentinel.db_name | 数据存在哪个db中 | postgres | pgawr |
| pgsentinel_ash.track_idle_trans | 是否记录 idle in transaction 状态会话 | false | true |
| pgsentinel_pgssh.max_entries | pg_stat_statements_history 视图在内存中的缓冲区大小（字节） | 10000 | 10000 |
| pgsentinel_pgssh.enable | 启用 pg_stat_statements_history | false | true |
这部分对应源码
cpp
/* GUC variables */
static int ash_sampling_period = 1;
static int ash_max_entries = 1000;
static int pgssh_max_entries = 10000;
static bool pgssh_enable = false;
static bool ash_track_idle_trans = false;
static int ash_restart_wait_time = 2;
char *pgsentinelDbName = "postgres";- 其他相关参数
查询语句保留长度
bash
# 为每个活动会话的pg_stat_activity.query字段所保留的内存量（字节，默认1024）
track_activity_query_size = 2048跟踪层级
pgsentinel依赖于pg_stat_statements插件的数据，如果想要更详细，可以调整相应参数（但必须注意对系统的负载）
bash
# 记录函数和存储过程中的子语句
pg_stat_statements.track = all四、 实现原理
插件最核心的就是pg_active_session_history,pg_stat_statements_history两个视图，所以源码中最重要的，也就是这两个视图的创建。
1. 视图创建
源码中的 pgsentinel--1.0.sql，可以看到这两个视图内容来自两个函数，并进行授权
sql
CREATE VIEW pg_active_session_history AS
SELECT * FROM pg_active_session_history();
GRANT SELECT ON pg_active_session_history TO PUBLIC;
CREATE VIEW pg_stat_statements_history AS
SELECT * FROM pg_stat_statements_history();
GRANT SELECT ON pg_stat_statements_history TO PUBLIC;而这两个函数实际是用c语言编写的
2. 函数创建
sql
CREATE FUNCTION pg_active_session_history(
OUT ash_time timestamptz,
OUT datid Oid,
OUT datname text,
OUT pid integer,
OUT leader_pid integer,
OUT usesysid Oid,
OUT usename text,
OUT application_name text,
OUT client_addr text,
OUT client_hostname text,
OUT client_port integer,
OUT backend_start timestamptz,
OUT xact_start timestamptz,
OUT query_start timestamptz,
OUT state_change timestamptz,
OUT wait_event_type text,
OUT wait_event text,
OUT state text,
OUT backend_xid xid,
OUT backend_xmin xid,
OUT top_level_query text,
OUT query text,
OUT cmdtype text,
OUT queryid bigint,
OUT backend_type text,
OUT blockers integer,
OUT blockerpid integer,
OUT blocker_state text
)
RETURNS SETOF record
AS 'MODULE_PATHNAME', 'pg_active_session_history'
LANGUAGE C STRICT VOLATILE PARALLEL SAFE;
-- Register a view on the function for ease of use.
CREATE VIEW pg_active_session_history AS
SELECT * FROM pg_active_session_history();
GRANT SELECT ON pg_active_session_history TO PUBLIC;
CREATE FUNCTION pg_stat_statements_history(
OUT ash_time timestamptz,
OUT userid Oid,
OUT dbid Oid,
OUT queryid bigint,
OUT calls bigint,
OUT total_exec_time double precision,
OUT rows bigint,
OUT shared_blks_hit bigint,
OUT shared_blks_read bigint,
OUT shared_blks_dirtied bigint,
OUT shared_blks_written bigint,
OUT local_blks_hit bigint,
OUT local_blks_read bigint,
OUT local_blks_dirtied bigint,
OUT local_blks_written bigint,
OUT temp_blks_read bigint,
OUT temp_blks_written bigint,
OUT blk_read_time double precision,
OUT blk_write_time double precision,
OUT plans bigint,
OUT total_plan_time double precision,
OUT wal_records bigint,
OUT wal_fpi bigint,
OUT wal_bytes numeric
)
RETURNS SETOF record
AS 'MODULE_PATHNAME', 'pg_stat_statements_history'
LANGUAGE C STRICT VOLATILE PARALLEL SAFE;既然如此，我们看看源码中究竟是怎么实现的这些函数
五、 源码学习
1. pg_active_session_history函数内容
它有两个分支，另外根据不同pg版本有不同语句（这里只挑了一个版本）：
- 启用pgsa_query_no_track_idle，即只记录active会话
cpp
select act.datid, act.datname, act.pid, act.usesysid, act.usename, \
act.application_name, text(act.client_addr), act.client_hostname, \
act.client_port, act.backend_start, act.xact_start, act.query_start, \
act.state_change, case when act.wait_event_type is null then 'CPU' \
else act.wait_event_type end as wait_event_type,case when act.wait_event is null \
then 'CPU' else act.wait_event end as wait_event, act.state, act.backend_xid, \
act.backend_xmin, act.query, act.backend_type,(pg_blocking_pids(act.pid))[1], \
cardinality(pg_blocking_pids(act.pid)),blk.state,gpi.*, act.leader_pid \
from pg_stat_activity act left join pg_stat_activity blk \
on (pg_blocking_pids(act.pid))[1] = blk.pid,get_parsedinfo(act.pid) gpi \
where act.state ='active' and act.pid != pg_backend_pid()";- 启用 pgsa_query_track_idle，即记录active和idle in transaction会话
sql
select act.datid, act.datname, act.pid, act.usesysid, act.usename, \
act.application_name, text(act.client_addr), act.client_hostname, \
act.client_port, act.backend_start, act.xact_start, act.query_start, \
act.state_change, case when act.wait_event_type is null then 'CPU' \
else act.wait_event_type end as wait_event_type,case when act.wait_event is null \
then 'CPU' else act.wait_event end as wait_event, act.state, act.backend_xid, \
act.backend_xmin, act.query, act.backend_type,(pg_blocking_pids(act.pid))[1], \
cardinality(pg_blocking_pids(act.pid)),blk.state,gpi.*, act.leader_pid \
from pg_stat_activity act left join pg_stat_activity blk \
on (pg_blocking_pids(act.pid))[1] = blk.pid,get_parsedinfo(act.pid) gpi \
where act.state in ('active', 'idle in transaction') and act.pid != pg_backend_pid()";2. pg_stat_statements_query函数内容
也有版本区分，这里只取其中一版
sql
select userid, dbid, queryid, calls, total_exec_time, rows, shared_blks_hit, \
shared_blks_read, shared_blks_dirtied, shared_blks_written, local_blks_hit, \
local_blks_read, local_blks_dirtied, local_blks_written, temp_blks_read, \
temp_blks_written, blk_read_time, blk_write_time, \
plans, total_plan_time, wal_records, wal_fpi, wal_bytes \
from pg_stat_statements \
where queryid in (select queryid from pg_active_session_history \
where ash_time in (select ash_time from pg_active_session_history \
order by ash_time desc limit 1))";3. 记录内容
每一行记录叫做一个entry
- pg_active_session_history对应叫ashEntry
- pg_stat_statements_query对应叫pgsshEntry
cpp
/* ash entry */
typedef struct ashEntry
{
int pid;
#if PG_VERSION_NUM >= 130000
int leader_pid;
#endif
int client_port;
uint64 queryid;
TimestampTz ash_time;
Oid datid;
Oid usesysid;
char *usename;
char *datname;
char *application_name;
char *wait_event_type;
char *wait_event;
char *state;
char *blocker_state;
char *client_hostname;
int blockers;
int blockerpid;
char *top_level_query;
char *query;
char *cmdtype;
char *backend_type;
char *client_addr;
TransactionId backend_xmin;
TransactionId backend_xid;
TimestampTz backend_start;
TimestampTz xact_start;
TimestampTz query_start;
TimestampTz state_change;
} ashEntry;
/* pg_stat_statement_history entry */
typedef struct pgsshEntry
{
TimestampTz ash_time;
Oid userid;
Oid dbid;
uint64 queryid;
int64 calls;
double total_time;
int64 rows;
int64 shared_blks_hit;
int64 shared_blks_read;
int64 shared_blks_dirtied;
int64 shared_blks_written;
int64 local_blks_hit;
int64 local_blks_read;
int64 local_blks_dirtied;
int64 local_blks_written;
int64 temp_blks_read;
int64 temp_blks_written;
double blk_read_time;
double blk_write_time;
#if PG_VERSION_NUM >= 130000
int64 plans;
double total_plan_time;
int64 wal_records;
int64 wal_fpi;
uint64 wal_bytes;
#endif
} pgsshEntry; 每个字段有一个buffer变量，记录共享内存用量，例如
cpp
static char *AshEntryUsenameBuffer = NULL;
static char *AshEntryDatnameBuffer = NULL;
static char *AshEntryAppnameBuffer = NULL;ash_entry_memsize和pgssh_entry_memsize估算entry所需内存
参考：
GitHub - pgsentinel/pgsentinel: postgresql extension providing Active session history