1 Introduction
Why RAC. --HA, Scalability, Cost.
each instance see the recent version of block.(by internode sync).
GRD - read & write policy. (s)
Service : application group?
listener, service, instance 关系。
2 Install and config
VIP, client can use multiple public IP, but when one node fail, timeout will wait. Public IP(real and physical), VIP is logical , can float if node fail, so .... good solution
RAC software installation.
3 RAC database creation.
post work: srvctl config database -d RAC
Single instance -->convert to RAC. dbca, rcconfig.
4 RAC database administration.
Redo Log Files and RAC. thread.
Automatic Undo Management: -->one instance one undo space, but others can read/update in recovery,
Autostart/Manual, srvctl modify database -d xxx -y MANUAL/AUTOMATIC
RAC parameter, alter system set xxx sid=, remove entry reset.
Parameters: CLUSTER_DATABASE, CLUSTER_DATABASE_INSTANCES, CLUSTER_INTERCONNECTS, thread, spfile. undo_management一致,但undo_tablespace不一样。
Quiescing RAC database - db can not be opened by other instance.
impact: switch logfile, archive log current, alter system checkpoint (local/global), recover --database not instance.
ASM : parameter, cluster_database=true, asm_diskstring, group,
srvctl start/stop/status asm xxx
5 管理rac的备份和恢复
Archived log should be access by all instance.
Flash recovery area should be on shared storage.
archive log naming %t. raman autobackup switch on -->
Parallel can from different instance, but instance open status must be same(mount or open). or channel select one least workload.
During backup & restore -->autolocation from which node to recover.
5 RAC performance tuning (skip most of all)
instance recovery-->2 phase.
GC wait event.
6 Services.
Perferred service fail,then connection to another available service.
create service with srvctl, -r xx -a (optional). TAF --> none, basic, and preconnect.
8 HA connections
Connection: client side load balance(listner's LB)/failover/server side load balance.--实例级负载均衡。
run-time connection load balance.(中间件 连接池)
TAP type - session, select. Method=basic-->reconnect when fail.
常用管理命令:
http://trailblizer.blog.163.com/blog/static/5963036420100752639263/ -gooda
http://wenku.baidu.com/view/6436a61dfc4ffe473368ab5a.html
http://hi.baidu.com/lee_eva/blog/item/8c8911a8c899fdf61f17a2f3.html