Oracle RAC/Clusterware 多种心跳heartbeat机制介绍 RAC超时机制分析

ORACLE RAC中最主要存在2种clusterware集群件心跳 &  RAC超时机制分析:

1、Network Heartbeat 网络心跳 每秒发生一次; 10.2.0.4以后网络心跳超时misscount为60s,;11.2以后网络心跳超时misscount为30s。

2、Disk Heartbeat 磁盘心跳  每秒发生一次; 10.2.0.4以后 磁盘心跳超时DiskTimeout为200s。

注意不管是磁盘心跳还是网络心跳都依赖于cssd.bin进程来实施这些操作,在真实世界中任何造成cssd.bin这个普通用户进程无法正常工作的原因均可能造成上述2种心跳超时, 原因包括但不局限于 CPU无法分配足够的时间片、内存不足、SWAP、网络问题、Votedisk IO问题、本次磁盘IO问题等等(askmac.cn)。

 

此外在使用ASM的情况下,DB作为ASM实例的Client客户; ASM实例会对DB实例的ASMB等进程进行监控, 以保证DB与ASM之间通信正常。 若DB的ASMB进程长期无响应(大约为200s)则ASM实例将考虑KILL DB的ASMB进程,由于ASMB是关键后台进程所以将导致DB实例重启。

也存在其他可能的情况,例如由于ASMB 被某些latch block, 会阻塞其他进程,导致PMON进行强制清理。

 

综上所述不管是Clusterware的 cssd.bin进程还是ASMB进程,他们都是OS上的普通用户进程,OS本身出现的问题、超时、延迟均可能造成它们无法正常工作导致。建议在确认对造成OS长时间的网络、IO延时的维护操作,考虑先停止节点上的Clusterware后再实施。

另可以考虑修改misscount、Disktimeout等 心跳超时机制为更大值,但修改这些值并不能保证就可以不触发Node Evication。

 

关于RAC /CRS对于本地盘的问题,详见如下的SR回复:

Does RAC/CRS monitor Local Disk IO ?

 

Oracle software use local ORACLE_HOME / GRID_HOME library files for main process operations.

 

 

There are some socket files under /tmp or /var/tmp needed for CRS communication.

 

Also, the init processes are all depending on the /etc directory to spawn the child processes.

 

Again, this is a complicated design for a cluster software which mainly rely on the OS stability including local file system.

 

Any changes to storage / OS are all recommended to stop CRS services since those are out of our release Q/A tests.

 

由于10.2的环境已经超出我们开发的支持服务期限,建议考虑升级到11.2.0.3来获得更全面的技术支持。


Posted

in

by

Tags:

Comments

One response to “Oracle RAC/Clusterware 多种心跳heartbeat机制介绍 RAC超时机制分析”

  1. Ask_Maclean_liu_Oracle Avatar

    于我们希望分析的对象 ASMB 会有很多种hang住的可能,这里只能针对部分情况给予参考ASM Technical Best Practices For 10g and 11gR1 Release (Doc ID 265633.1)ASMB – This process contacts CSS using the diskgroup name, and acquires the associated ASM connect string. This connect string is then used to connect into ASM instance. Using this persistent connection, periodic messages are exchanged to update statistics and provide a heartbeat mechanism. During operations that require ASM intervention, such as a file creation by a database foreground, the database foreground connects directly to the ASM instance to perform the operation. Upon successful completion of file creation, database file extent maps are sent by ASM to ASMB. Additionally, ASMB also sends database IO statistics to ASM instance.参考1. Bug 7340341 : ASM INSTANCE CRASH WITH ORA-15082 AFTER DROP DISKThere is a hang before the crash. CKPT process terminates the ASM instance after waiting 5 minutes.2. Dianostics for error ORA-600 [kfnmClientsRegister12] [ID 886616.1]* As a result of this process, a connection to the ASM instance was requested and a shadow process was generated on the ASM instance. That process is known as a NFG (Network Foreground)Function kfnmsg_SendtoUFGs, which indicates this process is sending messages to the UFG process, which is the process on the ASM instance that talks to the ASMB process* This UFG is the process talking to the process ASMB on the database instance. There will be one UFG for each database client of the ASM instance.* Most of the cases, the UFG process is showing the same call stack, where the functions belong to the network layer and wait event is sqlnet*message from client.In other words, UFG on the ASM instance can not talk to ASMB on the db instance. Processes on the ASM instance trying to talk to UFG, also can not do it, generating the internal error ORA-600 3. 其他情况下,由于ASMB 被某些latch block, 会阻塞其他进程,导致PMON进行强制清理。综合以上,我们只能给予猜测给予部分解答。并且10.2的环境已经超出我们开发的支持服务期限,建议您考虑升级到11.2来获得更全面的技术支持。如果您能找出具体测试案例来重现某一个具体现象,我们可以根据该现象搜集信息来提交新的SR 进行具体分析该情况下的超时机制。

Leave a Reply

Your email address will not be published. Required fields are marked *