Oracle11gRAC数据库巡检手册

1、数据库巡检数据库检测硬件机型HP DL580 G7是否集群是系统实际用户数10数据库进程进入操作系统，登陆 Oracle用户，命令：su - oracleProcess进程情况进入操作系统，登陆Oracle用户命令：ps -ef|grep ora_说明Oracle 10g后台进程SMON(System Monitor)用于执仃历程恢复、合并仝1可碎片并释放临时段。PMON(Process Monitor)用于监视服务器进程的执行，摒弃在服务器进程 失败时清除该服务器进程。DBWR(Database Writer)用于将数据库缓存的脏缓冲区数据写入到数据文 件中。LGWR(Log Writer)

2、用于将重做日志缓冲区所记载的全部内容写入到充作 日志文件中。CKPT(Checkpoint Process)用于发出检查点(Checkpoint),检查点会同步 数据库的数据文件、控制文件和重做日志，当发出检查点时，后台进程 CKPTW检查点时刻的 SCN(System Change Number)写入到控制文件和数 据文件头部，同时促使后台进程DBWR所有脏缓冲区写入到数据文件中。ARCH(Archive Process)用于将重做日志的事物变化复制到归档日志中， 该后台进程只启在 archivelog模式下/包国义。Oracle 11g新增的后台进程DBRM说明：数据库 资源管理进程(Th

3、e database resource manager process),负责设置资源计划和其他的资源管理的工作.DIAG说明：数据库诊断进程(The diagnosibility process),负责维护管理各种用丁诊断的转储文件，并执行oradebug命令。DIA0说明：另一个数据库诊断进程，负责检测Oracle数据库中的挂起(hang)和死锁的处理。PSP0说明：process spawner ,附于产生 oracle 进程 SMCO说明：space management coordinator ,该进程负责空间管理协调管理 工作，负责执行空间的分配和回收。Wnnn说明：命名为 W00

4、0 W001, W002.,由smcO动态产生执行上述相关任务。VKTM说明：virtual keeper of time,用 丁提供 wall-clock time ,(每秒钟更新一次)。提供每一十量秒更新一次的reference-time counter,看起来有点类似计时命的功能。GMON说明：用于维护asm磁盘组的磁盘之间的关系。KATE说明：当ASM的磁盘离线的时候，该进程负责asm的元文件的io读写。MARK说明：如果有向asm离线磁盘的 missed与请求，该进程将ASM分配的 单元的状态标记为 staleFBDA说明：涉及到 flashback-data-archive新特性的

5、 一个进程，Theflashback data archiver proces。用丁斗哥"轨表"(trackedtables)的历史数据进行归档。当"轨表”上的事务提交以后，fbda进程负责将数据的前镜像保存到flashback archive 区域。该进程还负责 flashback的数据归档的空间管理、分配、保留，跟踪tracked transactions。什么是"轨表"(tracked tables):是指启用了 flashback archive 特性的表。RMSn说明：The Oracle RAC management process

6、es,负责执行 Oracle RAC的管理任务，比如 RACf关资源的创建和集群中新实例的添加。DSKM说明：The slave diskmon process ,负责 oracle 实例、asm实例和 磁盘的管理进程之间的io fencing信息的交换。如果使用SAGE勺存储，该进程还负责 SAG用储的一些信息的管理。RAC进程GSD global services daemon全局服务守护进程lock process(LCK)锁管理进程DIAG:DIAGNOSABILITY DAEMO陕败进程的诊断信息捕获进程Operating System-Dependent(OSD)操作系统资源访问进

7、程LMS - Gobal Cache Service Process全局缓存服务进程LMD - Global Enqueue Service Daemon全局查询服务守护进程LMON-Global Enqueue Service Monitor全局查询服务监视进程LCK0 - Instance Enqueue Process实例查询进程进程状态Space使用情况df -hListener监听状态命令：lsnrctl status 配置正常运行情况命令：sqlplus "/as sysdba ”查看日志命令：More /u01/app/oracle/product/11.2.0/db_

8、1/network/log/sqlnet.log宿错误才有日志Tnsname配置情况标准配置位直：/u01/app/oracle/product/11.2.0/db_1/network/admin运行情况正常查看状态Tnsping tnsname（数据库实例名）Alert配置情况标准配置/u01/app/oracle/diag/rdbms/<SID>/<INSTANCE>/trace/alert_sid.log运行情况正常More alert_sid.logCRS服务运行 情况crsctl check crs检查crs的健康情况资源运行 情况crs_stat t用来查看

9、 RAC中各下点上 resources 的运行状况Resources的属性等OCR运行情况ocrcheck验证OCR勺状态以及空间使用情况Voting Disk运行情况crsctl query css votediskVotingdisk状态查询ASM运行情况crs_stat -t | grep asm使用情况# su - grid$ asmcmdASMCMD> lsDATA/ORA_DATA/ASMCMD> lsdg ora_dataASMCMD> ls -s DATA.354.774196981数据库su - oraclesqlplus"/as sysdba ”

10、诊断结果及建议DBstatus数据库状 态select status from v$instance;配置情况Select * from v$parameter;运行情况正常察看参数隐se含参数t linesize 132column name format a30column value format a25 selectx.ksppinm name, y.ksppstvl value,y.ksppstdf isdefault, decode(bitand(y.ksppstvf,7),1,'MOD)IFIED',4,'SYSTEM_MOD','FALS

11、E') ismod,decode(bitand(y.ksppstvf,2),2,'TRUE','FALSE') isadjfromsys.x$ksppi x, sys.x$ksppcv ywherex.inst_id = userenv('Instance') andy.inst_id = userenv('Instance') and x.indx = y.indx and x.ksppinm like '%_&par%'order bytranslate(x.ksppinm, ' _&#

12、39;,'')/)使用资源 情况select * from v$resource_limit；ASM使用情况select group_number,name,total_mb,free_mb from v$asm_diskgroup;DBfile数据文件 状态select name,status from v$datafile;控制文件 状态select status,name from v$controlfile;日志文件 状态select group#,members,archived,status from v$log;表空间使 用率set pagesize 50colu

13、mn "Tablespace" format a13column "Used MB" format 99,999,999column "Free MB" format 99,999,999column "Total MB" format 99,999,999selectfs.tablespace_name "Tablespace", (df.totalspace - fs.freespace) "Used MB", fs.freespace "Free MB&quo

14、t;, df.totalspace "Total MB",round(100 * (fs.freespace / df.totalspace) "Pct Free" from(selecttablespace_name,round(sum(bytes)/1048576) TotalSpace fromdba_data_files group bytablespace_name )df, (selecttablespace_name,round(sum(bytes)/1048576) FreeSpace fromdba_free_spacegroup by

15、 tablespace_name)fswhere df.tablespace_name=fs.tablespace_name order by "Pct Free"/运行情况正常Sessionsmmselect count(*) from v$session where status='ACTIVE'Redo使用情况selectb.THREAD#,a.GROUP#,a.STATUS,a.MEMBER,b.BYTES,b.ARCHIVED,b.STATUSfrom v$logfile a,v$log b where a.GROUP#=b.GROUP#;Perf

16、orman配置情况ce察看数据 库锁表锁表有时候是瞬间的，长时间锁定的表才可能是死锁。select l.*, s.OSUSER, s.ACTION, o.OBJECT_NAME from gv$locked_object l, gv_$session s, all_objects o where l.SESSION_ID=s.SID and o.OBJECT_ID=l.OBJECT_ID死锁set linesize 200column oracle_username for a16column os_user_name for a12column object_name for a30SELE

17、CTl.object_id,l.oracle_username,l.os_user_name,cess, l.session_id,s.serial#, l.locked_mode,o.object_name FROM v$locked_object l,dba_objects o,v$session s where l.object_id = o.object_id and s.sid = l.session_id; selectt2.username|''|t2.sid|''|t2.serial#|'2.logon_time|'&#

18、39;|t3.sql_textfrom v$locked_object t1,v$session t2,v$sqltext t3 where t1.session_id=t2.sid and t2.sql_address=t3.address order by t2.logon_time;l.xidusn,'|tlatch1、从 V$LATCH,V$LATCH_CHILDREN,V$LATCH_MIS笳看有关 LATC用统 计信息例：SELECT latch#, name, gets, misses, sleepsFROM v$latchWHERE sleeps。ORDER BY sl

19、eeps;2、在PEAK TIME,查看当前的 LATCH争例：select count(*), namelatchname from v$session_wait, v$latchname where event='latch free' and state='WAITING' and p2=latch#group by name order by 1 desc;3、确定影响系统的主要LATCHCONTENTION并针对具体LATCH争采取不同的ACTION例：如果LATCHFREE的等待事件在 WAIT EVENT占了比较大的比重，需要检查具体的LATCHB

20、争的情况。最常见的 LATCH争是LIBRARY CACHE SHARED POOLCACHE BUFFER CHAINS CACHE BUFFERS LRU CHAINS 前两者是关十 SHAREPOOL(SHARED_POOL_SIZE勺，是 SHARED BUFFER (DB_BLOCK_BUFFERS勺/* 显本系统范围内的latch 统计信息*/column name format A32 truncate heading "LATCH NAME" column pid heading "HOLDER PID"select ,a.a

21、ddr,a.gets,a.misses,a.sleeps,a.immediate_gets,a.immediate_misses,b.pidfrom v$latch a,v$latchholderb, v$latchname c wherea.addr = b.laddr(+)and a.latch# = c.latch#order bya.latch#;/* 给出一个 latch 地址，查出相应的 latch 名称*/column nameformat a64heading 'Name'selectname fromv$latchnamea, v$latch bwhereb.

22、addr = '&addr' and b.latch#=a.latch#;/* 显示latch 统计信息*/columnname formata32 heading'LATCH NAME'columnpid heading 'HOLDER PID'select ,a.addr,a.gets,a.misses,a.sleeps,a.immediate_gets, a.immediate_misses,b.pidfromv$latch a, v$latchholderb, v$latchname c wherea.addr= b.

23、laddr(+)and a.latch# = c.latch# and like '&latch_name%' order by a.latch#;enqueue等待Select eq_type "lock",total_req# "gets",total_wait#"waits",cum_wait_time from v$enqueue_stat where total_wait#>0;等待事件set linesize 200column username for a12column pr

24、ogram for a30column event for a28column p1text for a15column p1 for 999,999,999,999,999selects.username,gram,sw.event,sw.p1text,sw.p1fromv$session s,v$session_wait sw where s.sid=sw.sid and s.status='ACTIVE'order by sw.p1;select event,p1 "File #",p2 "Block #",p3 &quo

25、t;Reason Code" from v$session_wait order by event;where event = 'buffer busy waits'select owner,segment_name,segment_type,file_id,block_idfromdba_extentswhere file_id = &P1 and &P2 between block_id and block_id + blocks -1;column event for a35;column p1text for a40;select sid,ev

26、ent,p1,p1text from v$session_wait order by event;RAC全局 等待事 件select * from v$event_name where NAME like 'gc%' and WAIT_CLASS='Cluster'数据文件IOselect name,f.phyblkrd pbr,f.phyblkwrtpbw,f.phyblkrdpyr,f.readtim,f.writetimfrom v$filestat f, v$dbfile fswhere f.file# = fs.file# order

27、by 2,3,4 desc;表空间IOselecttablespace_name,sum(f.phyblkrd)pbr,sum(f.phyblkwrt)pbw,sum(f.phyblkrd) pyr,sum(f.readtim),sum(f.writetim) from v$filestat f, dba_data_files fswhere f.file# = fs.file_idgroup by tablespace_nameorder by 2,3,4 desc;长事务set linesize 200 column name for a16 column username for a10

28、select ,b.xacts,c.sid,c.serial#,c.username,d.sql_textfrom v$rollnamea,v$rollstatb,v$sessionc,v$sqltextd,v$transaction e where a.usn=b.usn and b.usn=e.XIDUSN and c.taddr=e.addrand c.sql_address=d.ADDRESSand c.sql_hash_value=d.hash_value order by ,c.sid,d.piece;大事务select sid,serial#,to_cha

29、r(start_time,'yyyy-mm-dd hh24:mi:ss')start_time,sofar,totalwork,(sofar/decode(totalwork,0,1,totalwork)*100 ratio,message from v$session_longopswhere message like '%RMAN%'select sid,serial#,to_char(start_time,'yyyy-mm-dd hh24:mi:ss') start_time,sofar,totalwork,(sofar/decode(to

30、talwork,0,1,totalwor k)*100 ratio,message from v$session_longopswhere sofar <> totalwork;where (sofar/totalwork)*100 < 100;察看用户 session 数量Select count(*) from v$session;如果session 过多，察看 select * from v$session,察看是什么程序察看用户 session 使用内 存大小SELECT username,sess.sid,sum(trunc(value/1024) MEM_USED

31、 FROM v$session sess, v$sesstat stat, v$statname name WHERE sess.sid = stat.sidAND stat.statistic# = name.statistic#AND in ('session uga memory' ,'session pga memory') group by username,sess.sid order by mem_used归档的生 成频率set linesize 120column begin_time for a26column end_ti

32、me for a26select a.recid,to_char(a.first_time,'yyyy-mm-ddhh24:mi:ss')begin_time, b.recid,to_char(b.first_time,'yyyy-mm-ddhh24:mi:ss')end_time, round(b.first_time - a.first_time)*24*60,2) minutes from v$log_history a,v$log_history b where b.recid = a.recid+1;缓冲区命 中率SELECT (1 - (SUM(DE

33、CODE(NAME, 'physical reads', VALUE, 0) / (SUM(DECODE(NAME, 'db block gets', VALUE, 0) + SUM(DECODE(NAME, 'consistent gets', VALUE, 0) * 100 "缓冲区命中率"FROM V$SYSSTAT;重做日志 命中率SELECT name, gets, misses, immediate_gets, immediate_misses,100 - Decode(gets,0,0,misses/(gets+

34、misses)*100 ratio1, 100-Decode(immediate_gets+immediate_misses,0,0,immediate_misses/( immediate_gets+immediate_misses)*100 ratio2FROM v$latch WHERE name IN ('redo allocation', 'redo copy');低效SQLSELECT EXECUTIONS , DISK_READS, BUFFER_GETS,ROUND(BUFFER_GETS-DISK_READS)/BUFFER_GETS,2) H

35、it_radio,ROUND(DISK_READS/EXECUTIONS,2) Reads_per_run,SQL_TEXTFROM V$SQLAREAWHERE EXECUTIONS。AND BUFFER_GETS > 0AND (BUFFER_GETS-DISK_READS)/BUFFER_GETS < 0.8ORDER BY 4 DESC;TOP SQLselect sql_text,buffer_gets,executions,buffer_gets/executionsfrom v$sqlarea where buffer_gets>1000000 order by

36、 4 desc;查看oracle sga使用 情况，及 分析select round(sga,0) sga, round(cache_size,0) cache_size, round(coun/cache_size*100,0)|'%' cache_used, round(shared_size,0) shared_size, round(freemb/shared_size*100,0)' free_shared, round(java_size,0) java,round(large_pool_size,0) large from(select bytes/102

37、4/1024 sga from v$sgainfo where name='Maximum SGA Size'), (select bytes/1024/1024 shared_size from v$sgainfo where name='Shared Pool Size'), (select bytes/1024/1024 cache_size from v$sgainfo where name='Buffer Cache Size'), (selectbytes/1024/1024java_size from v$sgainfo where

38、name='Java Pool Size'), (select bytes/1024/1024 large_pool_size from v$sgainfo where name='Large Pool Size'), (selectbytes/1024/1024 freemb from v$sgastat s wheres.pool='shared pool' and ='free memory'), (select sum(count(*)*8/1024 coun from v$bh where status<

39、;>'free' group by status );注:sga:系统全局区大小 cache_size:数据缓冲区大小 shared_size: 共享缓冲区大小 cache_used:数据缓冲区使用率 free_shared: 共享缓冲区空闲率 上面的均以MB为单位分析：上面的 数据缓冲区使用率统计可能会不太精确，你会发现当没有 v$bh视图里状态没有free的时候使用率，也就是全被用的时候，使用率也不是 100%.如果看更详细的数据缓冲区使用情况可以用卜面的sql,我们可以主要看是否有free,关于v$bh视图可以去网上找找相关文章.select count（*）, s

40、tatus from v$bh group by status ;没有free不一te说明需要增加，还要结合当前cache_size的大小，缓冲区 命中率，等待事件（如:free buffer waits'）来判断是否需要增加大小，但多数的时候添加缓冲区的大小不会解决质的问题，而真正的问题主要是对sql语句的优化（如：是否存在大量的全表扫描等）,.我在生产中遇到过类似的问题，200M的cache_size,命中率很低21%,但通过对sql语句的优化（添加索引，避免全表扫描）,命中率增加到96%,程 序运行时间由原来的2小时减少到/、到10分钟.关于数据缓冲区的优化可以参考我的另外一篇文

41、章：/htm_data/586/0904/46110.html共享缓冲区空闲率也是一样的道理.（如:sql语句是否没有绑定变量等）临时表空 问情况SELECT se.username 用户名, se.sid, se.serial#,se.sql_address, se.machine, gram, su.tablespace, su.blocks*8192/1024/1024 "Used Space(M)”, su.segtype, su.contentsFROM v$session se, v$sort_usage su WHERE se.saddr=su

42、.session_addr;在线日志 情况SELECTB.RECID,B.FIRST_TIME,A.FIRST_TIME,ROUND(A.FIRST_TIME-B.FIRST _TIME)*24*60,2) MINATESFROM V$LOG_HISTORY A,V$LOG_HISTORY BWHEREA.RECID=B.RECID +1 AND A.FIRST_TIME>SYSDATE - 20 AND ROUND(A.FIRST_TIME-B.FIRST_TIME)*24*60,2)<30ORDER BY A.FIRST_TIME DESC;在线日志 切换频 率select s

43、equence#,First_change#,to_char(first_time,'yyyy-mm-dd hh24:mi:ss') First_time,switch_change# fromv$loghistwherefirst_time>to_date('2013-03-0400:00:00','YYYY-MM-DDHH24:MI:SS')andfirst_time<to_date('2013-03-0508:00:00','YYYY-MM-DDHH24:MI:SS')order by 1;察看统

44、计信息SELECT * FROMDBA_TAB_STATISTICS where owner= 'HRWTM'SELECT * FROMDBA_IND_STATISTICS where owner= 'HRWTM'收集统计 信息-收橐数裾库信息DBMS STATS. gait her d&tebage stats;-收集耳:h之信息lXlC DEMS STATS . gather schema stats (1 zCT' );-收集表信息一一EXEC DBM5_5TAT5lgather_table_3tat3 (' SCOT"1

45、, 1 EMPLOYEE：1);-收集Index信息EXE'' DEMS STATS.gather index stats(1SCQTT1f ' EKPLOYEE5 PK');- 删除收集信息二父支 DBMS STATS.delete database stac.3；EXEC DEH523TAT3 - delete2scliema_stats ('5C07T1 );EXEC DBHSSTATS. de letetab le_stats (1 SCOT" 'EMPLOYEES');,XiC DBMS-STATS.clelete-i

46、tidex-stat3 (11 S JOT：1. EMPLOYEES PR1);- 备份收集蓿息-_BEGINdb如s 9tars.export tstole stats (USEft tatoiiotie =1 TEI ;1 ,statteb END J- 导入收集信息BEGINdbTtis_stat3. INPOT_TABLE_STATS (U3EP, FEI_" ,stattab => 1 STAT_TAE1END；-""自动? 方法一:exec dbms_scheduler.disable('SYS.GATHER_STATS_JOB'

47、);exec dbms_scheduler.enable('SYS.GATHER_STATS_JOB'); ? 方法二： alter system set "_optimizer_autostats_job"=false scope=spfile; alter system set "_optimizer_autostats_job"=true scope=spfile;(STiTABLE1E)；修改记录 最多的 对象（产 生归档 过快）selectsnap_id,obj#,db_block_changes_totalfromdba_hi

48、st_seg_stat order bydb_block_changes_total;select owner,object_name,object_idfrom dba_objects whereobject_id=obj#;统计Oracle读块的IO能力比如一个表有1万个块，每个块8K得到公司的IO能力每秒285M; 36480 块.总觉得 ALTER SYSTEMFLUSHBUFFER_CACHE SHARED_POO睛空 内存总是无效样.在家里做相应的实验启动了作用.用一个 384 个块的表 SELET COUNT(*) FROM T_TESTFLUSH BUFFER_CACHE SH

49、ARED_POOL 波动范围 0.329 ,0.297 0.266秒flush shared_pool; -0.187flush buffer_cache; -0.172 0.156秒那么 IO 能力是：384/0.172*8/1024=17.44M使用并行提示 /*+parallel(t,4)*/*+parallel(t,3)*/*+parallel(t,2)*/分别得到时间：0.64 0.375 0.296看来单CPU的并行能力很耗时间的.内存读是多少了 .只要重里执行卜就便晓得了而得到时间是0.031和0.0151 内存 IO: 96.77M 和 200M行迁移行 链接SELECT na

50、me, value FROMV$sysstatWHERE name = 'table fetch continuedrow'历史等待 信息selectevent,to_char(sample_time,'yyyymmdd hh24mi')sample_time, count(*) event_countfrom v$active_session_historywhere sample_time between to_date('20130422 1245', 'yyyymmddhh24mi') and to_date('2

51、0130422 1315', 'yyyymmdd hh24mi')and event is not null and event not in('null event') group by event, to_char(sample_time, 'yyyymmdd hh24mi')having count(*) > 50order by 1,2;索引使用 情况col cl heading 'Object|Name' format a30col c2 heading 'Operation' format

52、 a15col c3 heading 'Option' format a15col c4 heading 'Index|Usage|Count' format 999,999break on c1 skip 2break on c2 skip 2select p.object_name c1, p.operation c2, p.options c3,count c4from dba_hist_sql_plan p, dba_hist_sqlstat swhere p.object_owner <> 'SYS'and p.operat

53、ion like '%INDEX%'and p.sql_id = s.sql_idgroup by p.object_name,p.operation, p.options order by 1,2,3;无效索引SELECT OWNER,INDEX_NAME,INDEX_TYPE,TABLE_NAME,STATUFROMDBA_INDEXESWHERE STATUS = 'UNUSABLE'SQL执行计划查看语句的历史执行信息，是否发生变化，何时发生了变化。如果发生了变化，找出以前的执行计划，与当前的执行计划进 行对比，有什么不同。使用如下sql可以发现某个sq

54、l的执行计划什么时候发生了变 化！selectdistinctSQL_ID,PLAN_HASH_VALUE,to_char(TIMESTAMP,'yyyymmdd hh24:mi:ss')TIMESTAMPfrom dba_hist_sql_plan where SQL_ID='68wnxdjxwwn2h' order by TIMESTAMP;查看出来执行计划的变化之后可以使用如下sql查看发生了那些变化！col options for a15col operation for a20col object_name for a20 selectplan_has

55、h_value,id,operation,options,object_name,dept h,cost,to_char(TIMESTAMP,'yyyymmdd hh24:mi:ss') from DBA_HIST_SQL_PLAN where sql_id ='68wnxdjxwwn2h'andplan_hash_valuein(1542630049,2754593971,2620382595) order by ID,TIMESTAMP;解析慢 sqlSELECT SQL_ID,AVG_HARD_PARSE_TIME_MSFROM(SELECTROUND(A

56、VG_HARD_PARSE_TIME / 1000, 2) AVG_HARD_PARSE_TIME_MS,SQL_ID,EXECUTIONSFROMV$SQLSTATSORDER BY AVG_HARD_PARSE_TIME DESC ) WHEREROWNUM <= 10;解析次数 过多的 语句SELECT SUBSTR(SQL_TEXT,1,180) "SQL", COUNT(*)"己录数", SUM(EXECUTIONS肤行次数"FROM V$SQLAREA-WHERE EXECUTIONS < 5GROUP BY SUBST

57、R(SQL_TEXT,1,180)HAVING COUNT(*) > 30ORDER BY 2 DESC;SELECT SUBSTR (sql_text, 1, 100), module, COUNT (*) total_numberFROM v$sqlGROUP BY module, SUBSTR (sql_text, 1, 100)HAVING COUNT (*) > 50ORDER BY 3 DESCselect substr(a.SQL_TEXT,1,70),count(*) from v$sqlarea agroup by substr(a.SQL_TEXT,1,70)

58、having count(*) > 10 ;select address, CHILD_ADDRESS from v$sql where address in(select addressfrom v$sqlgroup by addresshaving count(*) in (select max(count(*) from v$sql group by address);查看解析 语句未 能共享 原因selectaddress,auth_check_mismatch,translation_mismatch,opti mizer_mismatchfrom v$sql_shared_cursorwhere address in (select address from v$sqlwhere upper(sql_text)like 'SELECT /*OPT_DYN_SAMP7/*+ ALL_ROWIGNORE_WHERE_CLAUSE NO_PARALLEL(SAMPLESUB)')查看表碎 片SELECT table_name,ROUND(blocks * 8), 2)"高水位空间 k",ROUND(num_rows * avg_row_len / 1024), 2)"真实使用空间k",ROUN