[转帖]记一次数据库优化_MySQL, Oracle及数据库讨论区_Weblogic技术|Tuxedo技术|中间件技术|Oracle论坛|JAVA论坛|Linux/Unix技术|hadoop论坛_联动北方技术论坛  
网站首页 | 关于我们 | 服务中心 | 经验交流 | 公司荣誉 | 成功案例 | 合作伙伴 | 联系我们 |
联动北方-国内领先的云技术服务提供商
»  游客             当前位置:  论坛首页 »  自由讨论区 »  MySQL, Oracle及数据库讨论区 »
总帖数
2
每页帖数
101/1页1
返回列表
0
发起投票  发起投票 发新帖子
查看: 2958 | 回复: 1   主题: [转帖]记一次数据库优化        下一篇 
d87025284
注册用户
等级:新兵
经验:62
发帖:59
精华:0
注册:2011-10-31
状态:离线
发送短消息息给d87025284 加好友    发送短消息息给d87025284 发消息
发表于: IP:您无权察看 2015-1-14 17:10:37 | [全部帖] [楼主帖] 楼主

客户反映数据库时快时慢,数据库版本11.2.0.4.0,操作系统RHEL5U5,查看数据库的活动会话,发现基本100%的SQL都在同一条SQL,而且大多都是latch: row cache objects等待。

01    SQL> select inst_id,username,sql_id,event from gv$session where username is not null and status='ACTIVE';
02
03       INST_ID USERNAME                       SQL_ID        EVENT
04    ---------- ------------------------------ ------------- ---------------------------------
05             1 SYS                            8s2qkvk056ugr PX Deq: Execution Msg
06             1 SYS                            8s2qkvk056ugr PX Deq: Execute Reply
07             1 PRESSO                         9rwzwamtgv6m6 gc cr request
08             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
09             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
10             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
11             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
12             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
13             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
14             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
15             2 PRESSO                         4z9c5071vvaz5 resmgr:cpu quantum
16             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
17             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
18             2 SYS                            8s2qkvk056ugr PX Deq: Execution Msg
19             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
20             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
21             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
22             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
23             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
24             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
25             2 PRESSO                         dfftdnm7cu76f enq: RC - Result Cache: Contention
26
27    21 rows selected.


latch: row cache objects等待事件通常是在修改数据字典相关的等待,通常由于硬解析过高导致,查看这条SQL的SQL语句。

01    SQL> select sql_text from v$sqltext where sql_id='dfftdnm7cu76f' order by piece
02
03    SQL_TEXT
04    ---------------------------------------------------------------------------------------------------------
05    select *
06      from (select t.apply_id,
07                   bb.cn_name,
08                   o.cn_name as oname,
09                   UPPER(bb.card_code),
10                   (select count(bd.BAD_RECORD_ID) as n
11                      from biz_badrecoed_info bd wherebd.BAD_RECORD_ID = 4 and bd.PUNISH_ORG = o.org_id) || '-' ||
12                   (select count(bd.BAD_RECORD_ID) as nll
13                      from biz_badrecoed_info bd
14                     where bd.BAD_RECORD_ID = 1
15                       and bd.ID_CARD = bb.card_code) || '-' ||
16                   (select count(bd.BAD_RECORD_ID) as nlllfrom biz_badrecoed_info bd where bd.BAD_RECORD_ID = 2 and bd.ID_CARD = bb.card_code) || '-' ||
17                   (select count(bd.BAD_RECORD_ID) as nllll
18                      from biz_badrecoed_info bd
19                     where bd.BAD_RECORD_ID = 3
20                       and bd.ID_CARD = bb.card_code) bdn,
21                   o.type_id,
22                   bcode.code_name,
23                   trim(ps.cn_name) as psname,
24                   t.apply_status as st,
25                   lead(t.apply_id, 1) over(order by t.create_time, t.apply_id desc) as nextID,
26                   lag(t.apply_id, 1) over(order by t.create_time, t.apply_id desc) as prevID,
27                   t.create_time as create_time
28              from biz_presscard_application t,
29                   biz_org_info              o,
30                   biz_reporter_base_info    bb,
31                   base_code_info            bcode,
32                   biz_pressstation_info     ps
33             where t.org_id = o.org_id
34               and bcode.code_sort_id = '1'
35               and t.reporter_uuid = bb.reporter_uuid
36               and to_char(t.apply_status) = bcode.code_id
37             andt.sat_id = ps.sat_id(+)
38               and t.apply_status = '2'
39               and t.next_orgid = '1'
40             ORDER BY T.create_time, t.apply_id DESC)
41     where rownum <= :1
42
43    24 rows selected.


可见这条SQL没有使用绑定变量,修改cursor_sharing参数,在数据库层打开绑定变量。

01    SQL> show parameter cursor
02
03    NAME                                 TYPE        VALUE
04    ------------------------------------ ----------- ------------------------------
05    cursor_bind_capture_destination      string      memory+disk
06    cursor_sharing                       string      SIMILAR
07    cursor_space_for_time                boolean     FALSE
08    open_cursors                         integer     3000
09    session_cached_cursors               integer     50
10    SQL> alter system set cursor_sharing=SIMILAR sid='*';
11
12    System altered.


再次查询这个等待已经很少,但出现了大量的latch: cache buffers chains等待事件。

01    SQL> select inst_id,username,sql_id,event from gv$session where username is not null and status='ACTIVE' and username !='SYS';
02
03       INST_ID USERNAME                       SQL_ID        EVENT
04    ---------- ------------------------------ ------------- ------------------------------------
05             1 PRESSO                         1fsdcuajuxncg enq: RC - Result Cache: Contention
06             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
07             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
08             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
09             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
10             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
11             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
12             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
13             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
14             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
15             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
16             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
17             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
18             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
19             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
20             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
21             2 PRESSO                         dfftdnm7cu76f latch: cache buffers chains
22             2 PRESSO                         dfftdnm7cu76f resmgr:cpu quantum
23             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
24             2 PRESSO                         dfftdnm7cu76f latch: row cache objects
25             2 PRESSO                         dfftdnm7cu76f enq: RC - Result Cache: Contention
26
27    21 rows selected.


这套应用代码写的相当那啥,SQL还不能动,只能从数据库角度去解决问题。对这套系统来说,热点块一直都是个问题,为了解决这个问题,数据库的block_size已经修改为4K。

1    SQL> show parameter block
2
3    NAME                                 TYPE        VALUE
4    ------------------------------------ ----------- ------------------------------
5    db_block_buffers                     integer     0
6    db_block_checking                    string      FALSE
7    db_block_checksum                    string      TYPICAL
8    db_block_size                        integer     4096
9    db_file_multiblock_read_count        integer     136


现在热点块依旧很严重,那么就只能通过调整pctfree来减少热点块了,先查出热点块严重的表。

01    SQL> SELECT *
02      2    FROM (SELECT O.OWNER, O.OBJECT_NAME, O.OBJECT_TYPE, SUM(TCH) TOUCHTIME
03      3            FROM X$BH B, DBA_OBJECTS O
04      4           WHERE B.OBJ = O.DATA_OBJECT_ID
05      5             AND B.TS# > 0
06      6           GROUP BY O.OWNER, O.OBJECT_NAME, O.OBJECT_TYPE
07      7           ORDER BY SUM(TCH) DESC)
08      8   WHERE ROWNUM <= 10;
09
10
11    OWNER                          OBJECT_NAME                              OBJECT_TYPE          TOUCHTIME
12    ------------------------------ ---------------------------------------- ------------------- ----------
13    PRESSO                         BIZ_REPORTER_BASE_INFO                   TABLE                  3691280
14    PRESSO                         BIZ_REPORTER_EDU_INFO                    TABLE                  3547004
15    PRESSO                         BIZ_REPORTER_ORG_INFO                    TABLE                  2273524
16    PRESSO                         BIZ_PRESSCARD_LOGOUT                     TABLE                  2099499
17    PRESSO                         BIZ_REPORTER_INFO                        TABLE                  1619598
18    PRESSO                         BIZ_PRESSCARD_APPLICATION                TABLE                  1191751
19    PRESSO                         BIZ_SYSTEM_MESSAGE_INFO                  TABLE                   730829
20    PRESSO                         BIZ_REPORTER_EXTEND_INFO                 TABLE                   610540
21    SYS                            WRH$_SQLSTAT                             TABLE PARTITION         193465
22    PRESSO                         IDX_REPORTER_UUID                        INDEX                   190901
23
24    10 rows selected.


调整热点块较高的表的pctfree。

01    SQL> ALTER TABLE PRESSO.BIZ_REPORTER_BASE_INFO PCTFREE 30;
02
03    Table altered.
04
05    SQL> ALTER TABLE PRESSO.BIZ_REPORTER_EDU_INFO PCTFREE 30;
06
07    Table altered.
08
09    SQL> ALTER TABLE PRESSO.BIZ_REPORTER_ORG_INFO PCTFREE 30;
10
11    Table altered.


这条SQL同样伴随着resmgr:cpu quantum等待事件,这显然是和资源管理器相关的等待事件,告警日志也可以看到相关的信息。

1    Setting Resource Manager plan SCHEDULER[0x32DB]:DEFAULT_MAINTENANCE_PLAN via scheduler window
2    Setting Resource Manager plan DEFAULT_MAINTENANCE_PLAN via parameter
3    Begin automatic SQL Tuning Advisor run for special tuning task  "SYS_AUTO_SQL_TUNING_TASK"
4    End automatic SQL Tuning Advisor run for special tuning task  "SYS_AUTO_SQL_TUNING_TASK"


当前服务器CPU使用率在50%左右,对PC服务器来讲,CPU超过50%通常是很危险的,就当前系统而言,AUTO_SQL_TUNING和资源管理器是不想看到的,禁用这些。

01    SQL> alter system set resource_manager_plan='' sid='*';
02
03    System altered.
04
05    SQL> execute dbms_scheduler.set_attribute('WEEKNIGHT_WINDOW','RESOURCE_PLAN','');
06
07    PL/SQL procedure successfully completed.
08
09    SQL> execute dbms_scheduler.set_attribute('WEEKEND_WINDOW','RESOURCE_PLAN','');
10
11    PL/SQL procedure successfully completed.
12
13    SQL> execute dbms_scheduler.set_attribute('MONDAY_WINDOW','RESOURCE_PLAN','');
14
15    PL/SQL procedure successfully completed.
16
17    SQL> execute dbms_scheduler.set_attribute('TUESDAY_WINDOW','RESOURCE_PLAN','');
18
19    PL/SQL procedure successfully completed.
20
21    SQL> execute dbms_scheduler.set_attribute('THURSDAY_WINDOW','RESOURCE_PLAN','');
22
23    PL/SQL procedure successfully completed.
24
25    SQL> execute dbms_scheduler.set_attribute('FRIDAY_WINDOW','RESOURCE_PLAN','');
26
27    PL/SQL procedure successfully completed.
28
29    SQL> execute dbms_scheduler.set_attribute('SATURDAY_WINDOW','RESOURCE_PLAN','');
30
31    PL/SQL procedure successfully completed.
32
33    SQL> execute dbms_scheduler.set_attribute('SUNDAY_WINDOW','RESOURCE_PLAN','');
34
35    PL/SQL procedure successfully completed.
36
37    SQL> BEGIN
38      2  DBMS_AUTO_TASK_ADMIN.DISABLE(
39      3  client_name => 'auto space advisor',
40      4  operation => NULL,
41      5  window_name => NULL);
42      6  END;
43      7  /
44
45    PL/SQL procedure successfully completed.


过断时间再次查看,发现这条SQL的等待事件又出现了enq: RC – Result Cache: Contention。这受隐含参数_result_cache_timeout的影响,在11.2.0.4.0版本默认是10秒,在10g版本是60秒。

01    SQL> select
02      1   x.ksppinm name,
03      2   y.ksppstvl value,
04      3   y.ksppstdf isdefault
05      4  from
06      5   sys.x$ksppi x,
07      6   sys.x$ksppcv y
08      7  where
09      8   x.inst_id = userenv('Instance') and
10      9   y.inst_id = userenv('Instance') and
11     10   x.indx = y.indx and
12     11   x.ksppinm like '%result_cache%'
13     12  order by
14     13*  translate(x.ksppinm, ' _', ' ')
15    SQL> /
16
17    NAME                                                                             VALUE      ISDEFAULT
18    -------------------------------------------------------------------------------- ---------- ---------
19    _client_result_cache_bypass                                                      FALSE      TRUE
20    client_result_cache_lag                                                          3000       TRUE
21    client_result_cache_size                                                         0          TRUE
22    _optimizer_ads_use_result_cache                                                  TRUE       TRUE
23    _result_cache_auto_dml_monitoring_duration                                       15         TRUE
24    _result_cache_auto_dml_monitoring_slots                                          4          TRUE
25    _result_cache_auto_dml_threshold                                                 16         TRUE
26    _result_cache_auto_dml_trend_threshold                                           20         TRUE
27    _result_cache_auto_execution_threshold                                           1          TRUE
28    _result_cache_auto_size_threshold                                                100        TRUE
29    _result_cache_auto_time_distance                                                 300        TRUE
30    _result_cache_auto_time_threshold                                                1000       TRUE
31    _result_cache_block_size                                                         1024       TRUE
32    _result_cache_copy_block_count                                                   1          TRUE
33    _result_cache_deterministic_plsql                                                FALSE      TRUE
34    _result_cache_global                                                             TRUE       TRUE
35    result_cache_max_result                                                          100        TRUE
36    result_cache_max_size                                                            2147483648 TRUE
37    result_cache_mode                                                                FORCE      TRUE
38    result_cache_remote_expiration                                                   0          TRUE
39    _result_cache_timeout                                                            10         TRUE
40
41    21 rows selected.


降低_result_cache_timeout参数的值可以减少enq: RC – Result Cache: Contention等待事件。

经过查询,发现这些SQL运行时间已经超过2小时,在这期间应用程序重启过,这些进程消耗了大量的资源,并且有释放。经开发人员确认,超过15分钟以上的连接都是不正常的,要回收掉,通过修改profile,设定会话连接时间15分钟。

01    SQL> alter profile "DEFAULT" limit CONNECT_TIME 15;
02
03    Profile altered.
04
05    SQL> select * from dba_profiles;
06
07    PROFILE                        RESOURCE_NAME                    RESOURCE LIMIT
08    ------------------------------ -------------------------------- -------- ----------------------------------------
09    DEFAULT                        COMPOSITE_LIMIT                  KERNEL   UNLIMITED
10    DEFAULT                        SESSIONS_PER_USER                KERNEL   UNLIMITED
11    DEFAULT                        CPU_PER_SESSION                  KERNEL   UNLIMITED
12    DEFAULT                        CPU_PER_CALL                     KERNEL   UNLIMITED
13    DEFAULT                        LOGICAL_READS_PER_SESSION        KERNEL   UNLIMITED
14    DEFAULT                        LOGICAL_READS_PER_CALL           KERNEL   UNLIMITED
15    DEFAULT                        IDLE_TIME                        KERNEL   UNLIMITED
16    DEFAULT                        CONNECT_TIME                     KERNEL   15
17    DEFAULT                        PRIVATE_SGA                      KERNEL   UNLIMITED
18    DEFAULT                        FAILED_LOGIN_ATTEMPTS            PASSWORD UNLIMITED
19    DEFAULT                        PASSWORD_LIFE_TIME               PASSWORD UNLIMITED
20    DEFAULT                        PASSWORD_REUSE_TIME              PASSWORD UNLIMITED
21    DEFAULT                        PASSWORD_REUSE_MAX               PASSWORD UNLIMITED
22    DEFAULT                        PASSWORD_VERIFY_FUNCTION         PASSWORD NULL
23    DEFAULT                        PASSWORD_LOCK_TIME               PASSWORD UNLIMITED
24    DEFAULT                        PASSWORD_GRACE_TIME              PASSWORD UNLIMITED
25
26    16 rows selected.


杀掉这些进程后,问题解决,再次运行这些SQL,很快就有结果返回了。这个故障导致数据库响应时快时慢,原因是连到节点2的会话,由于服务器资源被这些没有回收的进程消耗和占用,速度就相对较慢,而节点1资源充足,分配到节点1的操作就相对较快。

--转自 北京联动北方科技有限公司




赞(0)    操作        顶端 
yaoxia.wu
注册用户
等级:中校
经验:2318
发帖:53
精华:1
注册:1970-1-1
状态:离线
发送短消息息给yaoxia.wu 加好友    发送短消息息给yaoxia.wu 发消息
发表于: IP:您无权察看 2015-1-14 22:32:49 | [全部帖] [楼主帖] 2  楼

值得参考.



赞(0)    操作        顶端 
总帖数
2
每页帖数
101/1页1
返回列表
发新帖子
请输入验证码: 点击刷新验证码
您需要登录后才可以回帖 登录 | 注册
技术讨论