pg ash自制版 pg_active_session_history

一、 实现功能

        由于pgsentinel插件存在严重的内存占用问题,本篇改为自行实现,但其语句仍可以参考pgsentinel插件。PostgreSQL ash —— pgsentinel插件 学习与踩坑记录_CSDN博客

        v1.0 根据pg 14版本设计及测试,仅支持收集主库信息。默认每10秒收集一次 active与idle in transaction 状态会话信息,保留两个月。

二、 历史会话与阻塞信息

       参考 pgsentinel插件的pg_active_session_history视图及pg pg_stat_activity视图,根据不同版本,其中部分字段的值可能为空。

pg_ash表

列名

数据类型

字段含义

ash_time timestamp with time zone 采样时间
datid oid 会话连接的dbid
datname name 会话连接的DB名
pid integer 会话进程ID
leader_pid integer 并行进程leader id,pg 13新增
usesysid oid user id
usename name 用户名
application_name text 应用程序名
client_addr inet 客户端ip
client_hostname text 客户端主机名
client_port integer 客户端端口
backend_start timestamp with time zone 会话连接到服务器的时间
xact_start timestamp with time zone 当前事务开始的时间,若无活跃事务则为 NULL
query_start timestamp with time zone 当前活跃查询的开始时间。如果state不为active,则表示上个查询的开始时间
state_change timestamp with time zone state上次更改的时间
wait_event_type text 正在等待的事件类型(如果有)
wait_event text 正在等待的事件名(如果有)
state text 当前会话状态
backend_xid xid 该会话的顶层事务id(如果有)
backend_xmin xid 该会话的xmin horizon
query text

active状态下,为当前正在执行的查询;其他状态下,表示最后执行的查询。

默认情况下,查询文本被截断为 1024 字节(由参数track_activity_query_size控制)

query_id bigint 查询id,类似Oracle的sql_id,pg 14新增
backend_type text 当前会话类型,例如client backend, checkpointer, startup, walreceiver... pg 10新增
blockers integer 阻塞者数量
blockerpid integer 阻塞者进程id
blocker_state text 阻塞者状态

三、 表结构创建

1. 按月进行分区

CREATE TABLE public.pg_ash (

    ash_time timestamp with time zone,

    datid oid,

    datname name,

    pid integer,

    leader_pid integer,

    usesysid oid,

    usename name,

    application_name text,

    client_addr inet,

    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,

    query text,

    query_id bigint,

    backend_type text,

    blockers integer,

    blockerpid integer,

    blocker_state text

) PARTITION BY RANGE(ash_time);

-- 索引创建

CREATE INDEX idx_pg_ash_n1 ON pg_ash(ash_time);

-- 分区创建,超出最大范围的值会落入默认的final分区

CREATE TABLE pg_ash_history PARTITION OF pg_ash DEFAULT;

CREATE TABLE pg_ash_202310 PARTITION OF pg_ash FOR VALUES FROM ('2023-10-01'TO ('2023-11-01');

CREATE TABLE pg_ash_202311 PARTITION OF pg_ash FOR VALUES FROM ('2023-11-01'TO ('2023-12-01');

CREATE TABLE pg_ash_202312 PARTITION OF pg_ash FOR VALUES FROM ('2023-12-01'TO ('2024-01-01');

2. 定期自动新增分区

3. 自动清理旧分区

四、 数据插入

vi pg_ash.sh

#!/bin/bash

. ~/profile << EOF

5432

EOF

psql << EOF

insert into pg_ash

select now(), act.datid, act.datname, act.pid, act.leader_pid, act.usesysid, act.usename,

 act.application_name, act.client_addr, act.client_hostname,

 act.client_port, act.backend_start, act.xact_start, act.query_start,

 act.state_change, act.wait_event_type, act.wait_event, act.state, act.backend_xid,

 act.backend_xmin,act.query,act.query_id,act.backend_type,cardinality(pg_blocking_pids(act.pid))

 as blockers,(pg_blocking_pids(act.pid))[1] as blockerpid ,blk.state as blocker_state

 from pg_stat_activity act left join pg_stat_activity blk

 on (pg_blocking_pids(act.pid))[1] = blk.pid

 where act.state in ('active''idle in transaction') and act.pid != pg_backend_pid();

EOF

五、 设置定时执行

由于crontab最小只能按分钟执行,这里利用while true+sleep实现每十秒执行。

vi run.sh   与pg_ash.sh放在相同目录

#!/bin/bash

source .bash_profile

while true ]

do

sh ./pg_ash.sh

sleep 10

done

后台运行run.sh

nohup ./run.sh &

六、 测试运行效果

1. pgbench压测

  • 初始化数据

-bash-4.2$ createdb pgbench       

-bash-4.2$ pgbench -i pgbench

dropping old tables...

NOTICE:  table "pgbench_accounts" does not exist, skipping

NOTICE:  table "pgbench_branches" does not exist, skipping

NOTICE:  table "pgbench_history" does not exist, skipping

NOTICE:  table "pgbench_tellers" does not exist, skipping

creating tables...

generating data (client-side)...

100000 of 100000 tuples (100%) done (elapsed 0.23 s, remaining 0.00 s)

vacuuming...

creating primary keys...

done in 0.49 s (drop tables 0.00 s, create tables 0.01 s, client-side generate 0.27 s, vacuum 0.11 s, primary keys 0.10 s).

-bash-4.2$

-bash-4.2$ psql

psql (14.0)

Type "help" for help.

postgres=# \l

                                 List of databases

   Name    |  Owner   | Encoding |  Collate   |   Ctype    |   Access privileges  

-----------+----------+----------+------------+------------+-----------------------

 pgbench   | postgres | UTF8     | en_US.utf8 | en_US.utf8 |

 postgres  | postgres | UTF8     | en_US.utf8 | en_US.utf8 |

 template0 | postgres | UTF8     | en_US.utf8 | en_US.utf8 | =c/postgres          +

           |          |          |            |            | postgres=CTc/postgres

 template1 | postgres | UTF8     | en_US.utf8 | en_US.utf8 | =c/postgres          +

           |          |          |            |            | postgres=CTc/postgres

 testdb    | postgres | UTF8     | en_US.utf8 | en_US.utf8 |

(5 rows)

postgres=# \c pgbench

You are now connected to database "pgbench" as user "postgres".

pgbench=# \d

              List of relations

 Schema |       Name       | Type  |  Owner  

--------+------------------+-------+----------

 public | pgbench_accounts | table | postgres

 public | pgbench_branches | table | postgres

 public | pgbench_history  | table | postgres

 public | pgbench_tellers  | table | postgres

(4 rows)

  • 压测脚本

vi test.sql

\set aid random(1, 100000 * :scale)

\set bid random(1, 1 * :scale)

\set tid random(1, 10 * :scale)

\set delta random(-5000, 5000)

BEGIN;

UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;

SELECT abalance FROM pgbench_accounts WHERE aid = :aid;

UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;

UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;

INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);

END;

  • 执行压测

pgbench -c 4 -t 30000 pgbench -r -f test.sql

2. 运行ash脚本

nohup ./run.sh &

3. 查询ash数据

postgres=# select from pg_ash;       

-[ RECORD 1 ]----+-----------------------------------------------------------------------

ash_time         | 2023-10-12 05:44:07.152751+08

datid            | 41585

datname          | pgbench

pid              | 1530

leader_pid       |

usesysid         | 10

usename          | postgres

application_name | pgbench

client_addr      |

client_hostname  |

client_port      | -1

backend_start    | 2023-10-12 05:44:04.461672+08

xact_start       | 2023-10-12 05:44:07.144351+08

query_start      | 2023-10-12 05:44:07.145214+08

state_change     | 2023-10-12 05:44:07.145215+08

wait_event_type  | Lock

wait_event       | transactionid

state            | active

backend_xid      | 677819

backend_xmin     | 677814

query            | UPDATE pgbench_branches SET bbalance = bbalance + 3177 WHERE bid = 1;

query_id         | -6995838559535145041

backend_type     | client backend

blockers         | 1

blockerpid       | 1533

blocker_state    | active

-[ RECORD 2 ]----+-----------------------------------------------------------------------

ash_time         | 2023-10-12 05:44:07.152751+08

datid            | 41585

datname          | pgbench

pid              | 1531

leader_pid       |

usesysid         | 10

usename          | postgres

application_name | pgbench

client_addr      |

client_hostname  |

client_port      | -1

backend_start    | 2023-10-12 05:44:04.463697+08

xact_start       | 2023-10-12 05:44:07.151628+08

query_start      | 2023-10-12 05:44:07.152311+08

state_change     | 2023-10-12 05:44:07.152312+08

wait_event_type  | Lock

wait_event       | transactionid

state            | active

backend_xid      | 677821

backend_xmin     | 677817

query            | UPDATE pgbench_tellers SET tbalance = tbalance + 1637 WHERE tid = 8;

query_id         | -9151069917332221911

backend_type     | client backend

blockers         | 1

blockerpid       | 1530

blocker_state    | active

...

参考:

PostgreSQL Observability

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

你可能感兴趣的:(PostgreSQL,性能,监控,postgresql,ash,awr,pgsentinel,自制)