007.如何进行hang分析？
0.环境
- 主题：RAC某节点v$asm_disk查询hang分析处理
- 环境：Oracle 11.2.0.3 RAC
- 故障描述：RAC环境2个节点，节点1查询v$asm_disk正常返回结果，节点2查询v$asm_disk就会一直hang，查询会话对应event是ASM file metadata operation.
1.初步判断
首先连接节点1，查询v$asm_disk没有问题： --节点1查询v$asm_disk都没有问题
PATH
---------------------------------------------------------
/dev/oracleasm/disks/DATADISK
/dev/oracleasm/disks/FRADISK
/dev/oracleasm/disks/CRSDISK1
/dev/oracleasm/disks/CRSDISK3
/dev/oracleasm/disks/CRSDISK2
然后连接节点2，查询v$asm_disk就一直hang
hang住一直无结果返回.. 故障重现，确认客户描述故障现场属实。
既然是hang住，自然去查该会话的等待事件是什么？是否有阻塞会话？
select sid, serial#, inst_id, sql_id, event, p1,p2,p3, machine, username, blocking_session from gv$session where wait_class# <> 6;
SID SERIAL# INST_ID SQL_ID EVENT P1 P2 P3 MACHINE USERNAME BLOCKING_SESSION
---- ------- ------- ------------- --------------------------- ----- --- --- ------- -------- ----------------
289 33911 2 b5tshv0auyqm0 ASM file metadata operation 6229 9 0 newdb2 SYS
此时查了下MOS，初步怀疑是asm_diskstring配置有问题，或是对应的磁盘权限有什么异常。但实际排查并未发现异常。 两个节点asm实例的参数设置都一样；对应目录下的磁盘等权限等也一致。
两个节点asm实例的参数设置都一样 --节点1：
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
asm_diskgroups string DATADG, FRADG, CRSDG_NEW
asm_diskstring string /dev/oracleasm/disks/*
asm_power_limit integer 1
asm_preferred_read_failure_groups string
--节点2 一样的：
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
asm_diskgroups string DATADG, FRADG, CRSDG_NEW
asm_diskstring string /dev/oracleasm/disks/*
asm_power_limit integer 1
asm_preferred_read_failure_groups string
对应目录下的磁盘等权限等也一致
total 0
brw-rw---- 1 grid asmadmin 253, 11 Jan 29 15:05 CRSDISK1
brw-rw---- 1 grid asmadmin 253, 9 Jan 29 15:05 CRSDISK2
brw-rw---- 1 grid asmadmin 253, 10 Jan 29 15:05 CRSDISK3
brw-rw---- 1 grid asmadmin 253, 14 Jan 29 15:05 DATADISK
brw-rw---- 1 grid asmadmin 253, 13 Jan 29 15:05 FRADISK
total 0
brw-rw---- 1 grid asmadmin 253, 21 Jan 29 15:05 CRSDISK1
brw-rw---- 1 grid asmadmin 253, 18 Jan 29 15:05 CRSDISK2
brw-rw---- 1 grid asmadmin 253, 19 Jan 29 15:05 CRSDISK3
brw-rw---- 1 grid asmadmin 253, 22 Jan 29 15:05 DATADISK
brw-rw---- 1 grid asmadmin 253, 20 Jan 29 15:05 FRADISK
2.深入分析
在故障节点2上使用oradebug获取short_stack以及SSD 266，然后进行深入分析。 首先获取上面289,33911会话的spid：
select p.spid
from v$process p, v$session s
where p.addr = s.paddr
and s.sid = &sid
and s.serial# = &serial;
SQL> select p.spid
2 from v$process p, v$session s
3 where p.addr = s.paddr
4 and s.sid = &sid
5 and s.serial# = &serial;
Enter value for sid: 289
old 4: and s.sid = &sid
new 4: and s.sid = 289
Enter value for serial: 33911
old 5: and s.serial# = &serial
new 5: and s.serial# = 33911
oracle 8763 8762 0 16:31 ? 00:00:00 oraclenewdb2 (DESCRIPTION=(LOCAL=YES)(ADDRESS=(PROTOCOL=beq)))
grid 9639 1 0 16:35 ? 00:00:00 oracle+ASM2_user8763_newdb2 (DESCRIPTION=(LOCAL=YES)(ADDRESS=(PROTOCOL=beq)))
oracle 12022 9750 0 16:43 pts/2 00:00:00 /bin/bash -c ps -ef|grep 8763
oracle 12024 12022 0 16:43 pts/2 00:00:00 grep 8763
然后使用oradebug 跟踪8763进程，获取short_stack和SSD 266：
SQL> oradebug setospid 8763
Oracle pid: 68, Unix process pid: 8763, image: oracle@newdb2 (TNS V1-V3)
SQL> oradebug unlimit
Statement processed.
SQL> oradebug short_stack
ksedsts()+461<-ksdxfstk()+32<-ksdxcb()+1876<-sspuser()+112<-__sighandler()<-read()+14<-ntpfprd()+115<-nsbasic_brc()+376<-nsbrecv()+69<-nioqrc()+485<-ttcdrv()+1461<-nioqwa()+61<-upirtrc()+1385<-kpurcsc()+98<-kpuexec()+10807<-OCIStmtExecute()+39<-kfdDskTableCb4Db()+4492<-kfdDskTableCbInternal()+301<-kfdDskTableCb()+56<-qerfxFetch()+2210<-rwsfcd()+103<-qerhjFetch()+3187<-opifch2()+2995<-kpoal8()+2939<-opiodr()+916<-ttcpip()+2242<-opitsk()+1673<-opiino()+966<-opiodr()+916<-opidrv()+570<-sou2o()+103<-opimai_real()+133<-ssthrdmain()+252<-main()+201<-__libc_start_main()+253<-_start()+36
SQL> oradebug dump systemstate 266
Statement processed.
SQL> oradebug tracefile_name
/home/app/oracle/diag/rdbms/newdg/newdb2/trace/newdb2_ora_8763.trc
SQL>
从获取到的trc文件中，并没有找到有用的线索。至此问题陷入僵局。
3.峰回路转
回过头来想之前的解决过程，发现虽然查询v$asm_disk这类操作是在db层，但是本质确实asm实例的管理资源，去查asm实例是否有阻塞呢？ --ASM INSTANCE
set lines 350 trimspool on pages 300
select sid, state, event, seconds_in_wait, blocking_session
from v$session
where blocking_session is not null
or sid in (select blocking_session
from v$session
where blocking_session is not null)
order by sid;
在故障节点2的asm实例上查询：
SQL> set lines 350 trimspool on pages 300
SQL>
SQL> select sid, state, event, seconds_in_wait, blocking_session
2 from v$session
3 where blocking_session is not null
4 or sid in (select blocking_session
5 from v$session
6 where blocking_session is not null)
7 order by sid;
SID STATE EVENT SECONDS_IN_WAIT BLOCKING_SESSION
---------- ------------------- --------------------------------------- --------------- ----------------
128 WAITING enq: DD - contention 542311 1702
380 WAITING enq: DD - contention 2159938 1702
442 WAITING enq: DD - contention 2159098 1702
506 WAITING enq: DD - contention 2158258 1702
569 WAITING enq: DD - contention 713509 1702
632 WAITING enq: DD - contention 2156578 1702
695 WAITING enq: DD - contention 2155738 1702
758 WAITING enq: DD - contention 1108924 1702
821 WAITING enq: DD - contention 2142990 1702
884 WAITING enq: DD - contention 2139901 1702
947 WAITING enq: DD - contention 2136286 1702
1010 WAITING enq: DD - contention 2078621 1702
1073 WAITING enq: DD - contention 1845020 1702
1135 WAITING GPnP Get Item 2161258
1136 WAITING enq: DD - contention 1844671 1702
1199 WAITING enq: DD - contention 1844394 1702
1263 WAITING enq: DD - contention 1844065 1702
1325 WAITING enq: DD - contention 1843549 1702
1388 WAITING enq: DD - contention 1471932 1702
1451 WAITING enq: DD - contention 1641694 1702
1514 WAITING enq: DD - contention 1641426 1702
1577 WAITING enq: DD - contention 1572189 1702
1640 WAITING enq: DD - contention 1501765 1702
1702 WAITING rdbms ipc reply 0 1135
1703 WAITING enq: DD - contention 896552 1702
1765 WAITING enq: DD - contention 2161018 1702
1766 WAITING enq: DD - contention 893474 1702
1829 WAITING enq: DD - contention 11782 1702
1892 WAITING enq: DD - contention 5041 1702
1955 WAITING enq: DD - contention 3357 1702
30 rows selected.
果然看到很多阻塞，EVENT为"enq: DD - contention"的都是被1702会话阻塞，而1702会话又是被1135会话阻塞，1135会话的event是"GPnP Get Item"。 看两个会话的详细信息：
select sid,
SERIAL#,
sql_id,
event,
MACHINE,
PROGRAM,
username,
blocking_session
from v$session where sid = 1702;
select sid,
SERIAL#,
sql_id,
event,
MACHINE,
PROGRAM,
username,
blocking_session
from v$session where sid = 1135;
结果如下：
SQL> select sid,
2 SERIAL#,
3 sql_id,
4 event,
5 MACHINE,
6 PROGRAM,
7 username,
8 blocking_session
from v$session where sid = 1702;
SID SERIAL# SQL_ID EVENT MACHINE PROGRAM USERNAME BLOCKING_SESSION
----- ---------- ------------- ------------------ -------- ------------------------- -------- ----------------
1702 15127 b8a2pdbq3p2mj rdbms ipc reply newdb2 oracle@newdb2 (TNS V1-V3) SYS 1135
SQL> select sid,
2 SERIAL#,
3 sql_id,
4 event,
5 MACHINE,
6 PROGRAM,
7 username,
8 blocking_session from v$session where sid = 1135;
SID SERIAL# SQL_ID EVENT MACHINE PROGRAM USERNAME BLOCKING_SESSION
---- ---------- ------- ------------- --------- ---------------------- --------- ----------------
1135 1 GPnP Get Item newdb2 oracle@newdb2 (RBAL)
找到最终的原因：GPnP Get Item .
4.解决问题
有了上面的信息，再次查询MOS就可以匹配到 - Diskgroup Mount Hangs with RBAL Waiting on 'GPnP Get Item' and 'enq: DD - contention' (文档 ID 1375505.1) - "crsctl check cluster -all" command gives CRS-4404, CRS-4405 errors (文档 ID 1392934.1) 参照MOS解决方案：
#Use the following command as root on Compute Node 3 to identify the current gpnpd.bin process:
ps -ef | grep gpnpd.bin
#This will give you an output similar to:
ps -ef | grep pnp
oracle 31422 1 0 2011 ? 00:05:40 /u01/app/11.2.0.2/grid/bin/gpnpd.bin
#Please note the process ID number of the current gpnpd.bin process. In this example, = 31422
#Use the following command on Compute Node 3 to stop the current gpnpd.bin process:
kill -HUP
#for example, kill -HUP 31422
#After a few moments, gpnpd.bin should be automatically restarted by GI, run the command from step # 1 to verify that a new gpnpd.bin process exists
实际操作如下：
[root@newdb2 dev]# ps -ef | grep gpnpd.bin
grid 9835 1 0 2017 ? 00:23:19 /home/app/11.2.0/grid/bin/gpnpd.bin
root 26300 28772 0 17:43 pts/0 00:00:00 grep gpnpd.bin
[root@newdb2 dev]# kill -HUP 9835
You have mail in /var/spool/mail/root
[root@newdb2 dev]#
[root@newdb2 dev]# ps -ef | grep gpnpd.bin
grid 26740 1 1 17:45 ? 00:00:00 /home/app/11.2.0/grid/bin/gpnpd.bin
root 26762 28772 0 17:45 pts/0 00:00:00 grep gpnpd.bin
[root@newdb2 dev]# su - grid
[grid@newdb2 ~]$ sqlplus / as sysasm
SQL*Plus: Release 11.2.0.3.0 Production on Mon Jan 29 17:46:02 2018
Copyright (c) 1982, 2011, Oracle. All rights reserved.
Connected to:
Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production
With the Real Application Clusters and Automatic Storage Management options
SQL> set lines 350 trimspool on pages 300
SQL>
SQL> select sid, state, event, seconds_in_wait, blocking_session
2 from v$session
3 where blocking_session is not null
4 or sid in (select blocking_session
5 from v$session
6 where blocking_session is not null)
7 order by sid;
PATH
------------------------------------------
/dev/oracleasm/disks/FRADISK
/dev/oracleasm/disks/DATADISK
/dev/oracleasm/disks/CRSDISK3
/dev/oracleasm/disks/CRSDISK2
/dev/oracleasm/disks/CRSDISK1
至此，故障节点的ASM实例的阻塞消失，再次验证查询v$asm_disk已经恢复正常，故障算是完美解决。
后记：
- 实际上起初的解决思路出现偏差，ASM的问题就应该直接从ASM实例找原因，而之前都是在DB层分析，自然没有得到好的结果。这一点很值得自己反思。
- 最开始接到这个case是建议找停机时间直接重启节点2，这个方案简单粗暴且不需要深入分析问题，从最终解决方案看到，这种方法也是可以解决问题的。但恰恰由于客户坚持不能重启，反倒因此而学到了更多的技能点。