某客户的一个rac 节点CPU使用率为100%，导致整体系统运行缓慢。 通过mpstat命令检查发现系统CPU使用率为100%，系统的CPU idle几乎为0. 导致系统整体负载极高，如下:
[oracle@ecdbs1 ~]$ mpstat 1 10
Linux 2.6.9-89.5AXS2largesmp (ecdbs1) 2014年10月04日
09时52分39秒 CPU %user %nice %system %iowait %irq %soft %idle intr/s
09时52分40秒 all 82.42 0.00 17.52 0.00 0.00 0.06 0.00 5321.00
09时52分41秒 all 82.63 0.00 17.30 0.00 0.00 0.06 0.00 3562.00
09时52分42秒 all 81.52 0.00 18.36 0.00 0.00 0.13 0.00 6107.07
09时52分43秒 all 82.36 0.00 17.45 0.00 0.06 0.13 0.00 4194.06
09时52分44秒 all 82.47 0.00 17.53 0.00 0.00 0.00 0.00 3551.52
09时52分45秒 all 82.63 0.00 17.24 0.00 0.00 0.13 0.00 3553.54
09时52分46秒 all 83.29 0.00 16.71 0.00 0.00 0.00 0.00 3526.73
09时52分47秒 all 82.48 0.00 17.40 0.00 0.00 0.13 0.00 3553.54
09时52分48秒 all 82.47 0.00 17.35 0.00 0.06 0.13 0.00 6888.00
09时52分49秒 all 82.15 0.00 17.78 0.00 0.00 0.06 0.00 5739.60
Average: all 82.44 0.00 17.46 0.00 0.01 0.08 0.00 4601.00
由于这个主机节点上跑了2个实例，其中一个是rac的实例，另外一个是单实例。通过top 检查发现cpu基本上都是被该节点上的
单实例的进程所消耗，如下：
top - 16:34:59 up 747 days, 20:49, 9 users, load average: 139.68, 140.21, 140.37
Tasks: 861 total, 141 running, 720 sleeping, 0 stopped, 0 zombie
Cpu(s): 82.4% us, 17.5% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.1% si
Mem: 66002524k total, 63895780k used, 2106744k free, 165428k buffers
Swap: 12289716k total, 503928k used, 11785788k free, 4302948k cached
根据进程号，定位到相关的user process：
SQL> oradebug setospid 20377
Oracle pid: 115, Unix process pid: 20377, image: oracle@ecdbs1
SQL> select s.sid,s.serial#,p.spid from v$process p,v$session s where p.addr=s.paddr and p.spid=17000;
SID SERIAL# SPID
---------- ---------- ------------------------------------
2586 959 17000
SQL> !ps -ef|grep 17000
oracle 5438 3305 0 17:11 pts/7 00:00:00 /bin/bash -c ps -ef|grep 17000
oracle 5440 5438 0 17:11 pts/7 00:00:00 grep 17000
oracle 17000 1 14 Sep27 ? 17:18:09 oracleewms (LOCAL=NO)
SQL> oradebug setospid 17000
Oracle pid: 40, Unix process pid: 17000, image: oracle@ecdbs1
SQL> show parameter instance_name
NAME TYPE VALUE
------------------------------------ --------------------------------- ------------------------------
instance_name string ewms
SQL> oradebug dump processstate 10
Statement processed.
SQL> oradebug close_trace
oradStatement processed.
SQL> ebug tracefile_name
/opt/oracle/admin/ewms/udump/ewms_ora_17000.trc
SQL> !
通过检查发现，存在大量的I/O salve wait 等待事件，如下：
SQL> select distinct sql_id,event,count(1) from v$session where event like 'i/o%' group by sql_id,event order by 3;
更要命的是，我从top 命令的结果中挑选了几个cpu消耗高的进程，检查发现其对应的等待事件都是这个。
从上面的查询来看，这部分会话执行的SQL都类似，开始我怀疑难道是会话执行的SQL有问题 ？ 通过dump定位到SQL语句和执行计划，发现执行计划是ok，效率还挺高。
============
Plan Table
============
-------------------------------------------------------------------------+-----------------------------------+
| Id | Operation | Name | Rows | Bytes | Cost | Time |
-------------------------------------------------------------------------+-----------------------------------+
| 0 | SELECT STATEMENT | | | | 708 | |
| 1 | TEMP TABLE TRANSFORMATION | | | | | |
| 2 | LOAD AS SELECT | | | | | |
| 3 | VIEW | | 1 | 154 | 681 | 00:00:09 |
| 4 | NESTED LOOPS | | 1 | 137 | 681 | 00:00:09 |
| 5 | VIEW | | 1468 | 126K | 681 | 00:00:09 |
| 6 | MERGE JOIN | | 1468 | 100K | 681 | 00:00:09 |
| 7 | TABLE ACCESS BY INDEX ROWID | MLOS_SM_INSTIINFO | 1 | 27 | 10 | 00:00:01 |
| 8 | INDEX FULL SCAN | PK_INSTIINFO | 46 | | 2 | 00:00:01 |
| 9 | SORT JOIN | | 16K | 697K | 671 | 00:00:09 |
| 10 | TABLE ACCESS FULL | EWMS_INV_TRANSACTION | 16K | 697K | 669 | 00:00:09 |
| 11 | TABLE ACCESS BY INDEX ROWID | EWMS_BASE_CARGO | 1 | 49 | 0 | |
| 12 | INDEX UNIQUE SCAN | PK_EWMS_BASE_CARGO | 1 | | 0 | |
| 13 | COUNT | | | | | |
| 14 | VIEW | | 1 | 398 | 27 | 00:00:01 |
| 15 | SORT ORDER BY | | 1 | 383 | 27 | 00:00:01 |
| 16 | HASH JOIN OUTER | | 1 | 383 | 26 | 00:00:01 |
| 17 | HASH JOIN OUTER | | 1 | 317 | 17 | 00:00:01 |
| 18 | HASH JOIN OUTER | | 1 | 251 | 14 | 00:00:01 |
| 19 | HASH JOIN OUTER | | 1 | 185 | 10 | 00:00:01 |
| 20 | HASH JOIN OUTER | | 1 | 119 | 7 | 00:00:01 |
| 21 | VIEW | | 1 | 53 | 3 | 00:00:01 |
| 22 | HASH GROUP BY | | 1 | 62 | 3 | 00:00:01 |
| 23 | VIEW | | 1 | 62 | 2 | 00:00:01 |
| 24 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6835_CC9CBC| 1 | 90 | 2 | 00:00:01 |
| 25 | VIEW | | 1 | 66 | 3 | 00:00:01 |
| 26 | HASH GROUP BY | | 1 | 78 | 3 | 00:00:01 |
| 27 | VIEW | | 1 | 78 | 2 | 00:00:01 |
| 28 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6835_CC9CBC| 1 | 90 | 2 | 00:00:01 |
| 29 | VIEW | | 1 | 66 | 3 | 00:00:01 |
| 30 | HASH GROUP BY | | 1 | 92 | 3 | 00:00:01 |
| 31 | FILTER | | | | | |
| 32 | VIEW | | 1 | 92 | 2 | 00:00:01 |
| 33 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6835_CC9CBC| 1 | 90 | 2 | 00:00:01 |
| 34 | VIEW | | 1 | 66 | 3 | 00:00:01 |
| 35 | HASH GROUP BY | | 1 | 92 | 3 | 00:00:01 |
| 36 | FILTER | | | | | |
| 37 | VIEW | | 1 | 92 | 2 | 00:00:01 |
| 38 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6835_CC9CBC| 1 | 90 | 2 | 00:00:01 |
| 39 | VIEW | | 1 | 66 | 3 | 00:00:01 |
| 40 | HASH GROUP BY | | 1 | 78 | 3 | 00:00:01 |
| 41 | VIEW | | 1 | 78 | 2 | 00:00:01 |
| 42 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6835_CC9CBC| 1 | 90 | 2 | 00:00:01 |
| 43 | VIEW | | 1 | 66 | 8 | 00:00:01 |
| 44 | HASH GROUP BY | | 1 | 109 | 8 | 00:00:01 |
| 45 | NESTED LOOPS | | 1 | 109 | 7 | 00:00:01 |
| 46 | NESTED LOOPS | | 1 | 82 | 6 | 00:00:01 |
| 47 | TABLE ACCESS FULL | EWMS_INV_LOGINVENTORY | 640 | 21K | 6 | 00:00:01 |
| 48 | TABLE ACCESS BY INDEX ROWID | EWMS_BASE_CARGO | 1 | 49 | 0 | |
| 49 | INDEX UNIQUE SCAN | PK_EWMS_BASE_CARGO | 1 | | 0 | |
| 50 | TABLE ACCESS BY INDEX ROWID | MLOS_SM_INSTIINFO | 1 | 27 | 1 | 00:00:01 |
| 51 | INDEX UNIQUE SCAN | PK_INSTIINFO | 1 | | 0 | |
-------------------------------------------------------------------------+-----------------------------------+
由此可以判断，应该跟SQL本身的效率是没有关系的。从dump的processstate strace还发现了一点有价值的信息：
WARNING:io_submit failed due to kernel limitations MAXAIO for process=0 pending aio=0
WARNING:asynch I/O kernel limits is set at AIO-MAX-NR=65536 AIO-NR=65529
WARNING:1 Oracle process running out of OS kernelI/O resources aiolimit=0
ksfdgo()+1488<-ksfdaio1()+9848<-kfkUfsIO()+594<-kfkDoIO()+631<-kfkIOPriv()+616<-
............
根据这个AIO的告警，mos上一搜，果然是跟AIO有关系。例如存在类似这样的bug：
Bug 9949948 Linux: Process spin under ksfdrwat0 if OS Async IO not configured high enough
最后建议调整Linux AIO参数之后目前一切正常。
简单记录一下，供大家参考！