Systemtap(2.4) fixed BUG(1.8) : delete from statistics(aggregates) type stored in array elements

背景

上一篇BLOG中在测试使用delete删除数组元素中存储的统计信息时出现未删除的BUG.

http://blog.163.com/digoal@126/blog/static/163877040201391234023230/

本文从git上下载最新的2.4版本进行同样场景的测试, 发现这个bug已经不存在了. (建议使用CentOS 6.x以上版本测试, 否则可能因为elfutils版本太低无法安装stap 2.4)  
测试过程如下 :   
环境准备 :   
yum install -y systemtap-devel systemtap-client systemtap-runtime systemtap-sdt-devel systemtap  
其中systemtap-sdt-devel 是PostgreSQL --enable-dtrace需要的包.  
  
安装git, CentOS 6.x 可以直接从yum源中下载.  
或者从google code中下载.  

http://blog.163.com/digoal@126/blog/static/163877040201242512825860/

从sourceware 站点下载和安装stap :   
git clone git://sourceware.org/git/systemtap.git  
cd systemtap/  
./configure --prefix=/opt/systemtap && gmake all && gmake install  
[root@db-172-16-3-150 ~]# /opt/systemtap/bin/stap -V  
Systemtap translator/driver (version 2.4/0.152, commit release-2.3-133-g335e342 + changes)  
Copyright (C) 2005-2013 Red Hat, Inc. and others  
This is free software; see the source for copying conditions.  
enabled features: LIBSQLITE3 BOOST_SHARED_PTR TR1_UNORDERED_MAP NLS JAVA  
  
测试 :   
脚本如下, 与1.8版本相同 :   
[root@db-172-16-3-150 ~]# cat test.stp  
global var1[50000], var2[50000], var3[50000]  
  
probe process("/home/pg93/pgsql9.3.1/bin/postgres").mark("query__start") {  
  pid=pid()  
  delete var1[pid]  
  delete var2[pid]  
  delete var3[pid]  
  var3[pid] = user_string($arg1)  
  printf("pid:%u, query start: %s, read: %u,%u, write: %u,%u\n", pid, var3[pid], @sum(var1[pid]), @count(var1[pid]), @sum(var2[pid]), @count(var2[pid]))  
}  
  
probe process("/home/pg93/pgsql9.3.1/bin/postgres").mark("query__done") {  
  pid=pid()  
  printf("pid:%u, query done: %s, read: %u,%u, write: %u,%u\n", pid, var3[pid], @sum(var1[pid]), @count(var1[pid]), @sum(var2[pid]), @count(var2[pid]))  
  delete var1[pid]  
  delete var2[pid]  
  delete var3[pid]  
}  
  
probe syscall.read.return {  
  if (execname()=="postgres") {  
    pid=pid()  
    var1[pid] <<< $count  
  }  
}  
  
probe syscall.write.return {  
  if (execname()=="postgres") {  
    pid=pid()  
    var2[pid] <<< $count  
  }  
}  
  
probe end {  
  delete var1  
  delete var2  
  delete var3  
}  
执行stap脚本, 测试场景与1.8版本相同.  
[root@db-172-16-3-150 ~]# /opt/systemtap/bin/stap -D MAXSKIPPED=1 test.stp   
[root@db-172-16-3-150 ~]# su - pg93  
pg93@db-172-16-3-150-> psql  
psql (9.3.1)  
Type "help" for help.  
digoal=# drop table t;  
DROP TABLE  
digoal=# create table t(id int, info text, crt_time timestamp);  
CREATE TABLE  
digoal=# insert into t select generate_series(1,1000000),md5(random()::text),now();  
INSERT 0 1000000  
digoal=# select count(*) from t;  
  count    
---------  
 1000000  
(1 row)  
digoal=# select count(*) from t;  
  count    
---------  
 1000000  
(1 row)  
digoal=# create index idx_t_1 on t(id);  
CREATE INDEX  
digoal=# select count(*) from t;  
  count    
---------  
 1000000  
(1 row)  
digoal=# select count(*) from t;  
  count    
---------  
 1000000  
(1 row)  
digoal=# select count(*) from t;  
  count    
---------  
 1000000  
(1 row)  
digoal=# checkpoint;  
CHECKPOINT  
digoal=# checkpoint;  
CHECKPOINT  
stap输出结果如下, delete var1[pid]和delete var2[pid]已经可以正常清楚statistic类型的数据了.  
pid:17397, query start: drop table t;, read: 0,0, write: 0,0  
pid:17397, query done: drop table t;, read: 0,0, write: 65536,1  
pid:17397, query start: create table t(id int, info text, crt_time timestamp);, read: 0,0, write: 0,0  
pid:17397, query done: create table t(id int, info text, crt_time timestamp);, read: 8192,1, write: 16384,2  
pid:17397, query start: insert into t select generate_series(1,1000000),md5(random()::text),now();, read: 0,0, write: 0,0  
pid:17397, query done: insert into t select generate_series(1,1000000),md5(random()::text),now();, read: 32768,4, write: 76603392,9351  
pid:17397, query start: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query done: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query start: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query done: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query start: create index idx_t_1 on t(id);, read: 0,0, write: 0,0  
pid:17397, query done: create index idx_t_1 on t(id);, read: 0,0, write: 24158208,2853  
pid:17397, query start: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query done: select count(*) from t;, read: 8192,1, write: 0,0  
pid:17397, query start: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query done: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query start: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query done: select count(*) from t;, read: 0,0, write: 0,0  
pid:17397, query start: checkpoint;, read: 0,0, write: 0,0  
pid:17397, query done: checkpoint;, read: 0,0, write: 0,0  
pid:17397, query start: checkpoint;, read: 0,0, write: 0,0  
pid:17397, query done: checkpoint;, read: 0,0, write: 0,0  

参考

1. http://blog.163.com/digoal@126/blog/static/163877040201391234023230/

2. http://blog.163.com/digoal@126/blog/static/16387704020137140265557/

3. http://blog.163.com/digoal@126/blog/static/163877040201242512825860/

digoal’s 大量PostgreSQL文章入口