PostgreSQL ash —— pgsentinel插件 学习与踩坑记录

零、 注意事项

       测试发现，pgsentinel插件在pg_active_session_history视图记录条数较多时，存在严重的内存占用问题，群里的其他朋友反馈还可能存在严重的内存泄漏问题。本文仅用于学习和测试，未用于生产环境。

       设置 pgsentinel_ash.max_entries=10000000，启动DB报错需请求28G内存。对于负载稍高的数据库，例如每秒40个活跃会话，按照每秒收集一次的频率，1000万行也仅够保存不到3天的数据，该插件就需要占掉28G内存，实用性太低。

-bash-4.2$ pg_ctl start -D $PGDATA
waiting for server to start....2023-10-07 19:43:09.865 CST [2210] FATAL:  could not map anonymous shared memory: Cannot allocate memory
2023-10-07 19:43:09.865 CST [2210] HINT:  This error usually means that PostgreSQL's request for a shared memory segment exceeded available memory, swap space, or huge pages. To reduce the request size (currently 29601538048 bytes), reduce PostgreSQL's shared memory usage, perhaps by reducing shared_buffers or max_connections.
2023-10-07 19:43:09.865 CST [2210] LOG:  database system is shut down
 stopped waiting
pg_ctl: could not start server
Examine the log output

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（重启数据库其中数据会被清空）。

1. pg_active_session_history视图字段

重启数据库其中数据会被清空，大部分字段与对应版本的pg_stat_activity视图字段含义相同

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

重启数据库其中数据会被清空，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

# 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

CREATE EXTENSION pgsentinel;

3. 插件配置

• 必须配置

vi postgresql.conf
shared_preload_libraries = 'pg_stat_statements,auto_explain,pgsentinel'

若未配置，查询会报错

postgres=# select * from pg_active_session_history ; 
ERROR:  pg_active_session_history must be loaded via shared_preload_libraries

重启db生效

pg_ctl stop -m fast
pg_ctl start -D $PGDATA

postgres=# select * from pg_active_session_history ;
(0 rows)

• 可选配置

可以直接在postgresql.conf中修改，也可以alter system设置

 alter system set pgsentinel_pgssh.enable=on;

参数名	参数含义	默认值	建议值	备注
pgsentinel_ash.sampling_period	采样时间（秒）	1	10，视业务负载及需求而定	reload生效
pgsentinel_ash.max_entries	pg_active_session_history 最大记录条数（占用ring buffer大小，单位为字节）	1000	视业务负载及需求而定。注意非常耗内存，设置1000万约占内存28G	重启生效。设置过大可能内存不足，DB启动失败
pgsentinel.db_name	数据存在哪个db中	postgres	pgawr	重启生效
pgsentinel_ash.track_idle_trans	是否记录 idle in transaction 状态会话	off	on	reload生效
pgsentinel_pgssh.max_entries	pg_stat_statements_history 最大记录条数（占用ring buffer大小，单位为字节）	10000	视业务负载及需求而定	重启生效。设置过大可能内存不足，DB启动失败
pgsentinel_pgssh.enable	是否启用 pg_stat_statements_history	off	on	重启生效

这部分对应源码

从中也可以看到各参数含义、默认值、最小最大值，是否需重启生效等

static void
pgsentinel_load_params(void)
{

	DefineCustomIntVariable("pgsentinel_ash.sampling_period",
							"Duration between each pull (in seconds).",
							NULL,
							&ash_sampling_period,
							1,
							1,
							INT_MAX,
							PGC_SIGHUP,
							0,
							NULL,
							NULL,
							NULL);

	DefineCustomBoolVariable("pgsentinel_ash.track_idle_trans",
	                        "Track session in idle transaction state.",
							NULL,
							&ash_track_idle_trans,
							false,
							PGC_SIGHUP,
							0,
							NULL,
							NULL,
							NULL);

	if (!process_shared_preload_libraries_in_progress)
		return;

	/* can't define PGC_POSTMASTER variable after startup */
	DefineCustomIntVariable("pgsentinel_ash.max_entries",
							"Maximum number of ash entries.",
							NULL,
							&ash_max_entries,
							1000,
							1000,
							INT_MAX,
							PGC_POSTMASTER,
							0,
							NULL,
							NULL,
							NULL);

	EmitWarningsOnPlaceholders("pgsentinel_ash");

	DefineCustomIntVariable("pgsentinel_pgssh.max_entries",
							"Maximum number of pgssh entries.",
							NULL,
							&pgssh_max_entries,
							10000,
							10000,
							INT_MAX,
							PGC_POSTMASTER,
							0,
							NULL,
							NULL,
							NULL);

	DefineCustomBoolVariable("pgsentinel_pgssh.enable",
	                        "Enable pg_stat_statements_history.",
							NULL,
							&pgssh_enable,
							false,
							PGC_POSTMASTER,
							0,
							NULL,
							NULL,
							NULL);

	EmitWarningsOnPlaceholders("pgsentinel_pgssh");

	DefineCustomStringVariable("pgsentinel.db_name",
							gettext_noop("Database on which the worker connect."),
							NULL,
							&pgsentinelDbName,
							"postgres",
							PGC_POSTMASTER,
							GUC_SUPERUSER_ONLY,
							NULL, NULL, NULL);
}

• 其他相关参数

查询语句保留长度

# 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）

# 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

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

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

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

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

 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

/* 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

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，如果占用量过大，DB启动可能会失败。基本原理是：

• 每行占用内存 = 各字段占用内存之和
• 总占用内存 = 每行占用内存 * 最大行数 ash_max_entries

/* Estimate amount of shared memory needed for ash entry */
static Size
ash_entry_memsize(void)
{
	Size            size;

	/* AshEntryArray */
	size = mul_size(sizeof(ashEntry), ash_max_entries);
	/* AshEntryUsenameBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryDatnameBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryAppnameBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryClientaddrBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryWaitEventTypeBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryWaitEventBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryStateBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryClientHostnameBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryQueryBuffer */
	size = add_size(size, mul_size(pgstat_track_activity_query_size,
															ash_max_entries));
	/* AshEntryCmdTypeBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryTopLevelQueryBuffer */
	size = add_size(size, mul_size(pgstat_track_activity_query_size,
															ash_max_entries));
	/* AshEntryBackendTypeBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));
	/* AshEntryBlockerStateBuffer */
	size = add_size(size, mul_size(NAMEDATALEN, ash_max_entries));

	return size;
}

参考：

GitHub - pgsentinel/pgsentinel: postgresql extension providing Active session history

一种PostgreSQL数据库监控和溯源分析的方法和系统与流程

PostgreSQL 12.2官方手册学习( 第19章 运行时统计数据) - 墨天轮