Rac & DG

RAC Environment:

2. RAC version similarities and differences: [10R2,11R1 (and 10 similar)],11R2,12C:
   1. Directory:
      1. ASM 10.2 requires a separate directory (Oracle home): RDBMS Home,asm home, CRS Home
      2. 11.2:rdbms,grid (CRS+ASM)
   2. Ip:
      1. 10.2 has a private IP (public IP) and a virtual IP (VIP); Private IP is used to make an internal connection, and the common IP and VIP are external connections and are in the same network segment.
      2. 11.2 Add a Scan ip,scan IP to 3 on top of the external connection (public IP and virtual IP).
   3. Disk:
      1. Before 11.2, OCR and voting disks on bare devices, 11.2 were placed on ASM, and Oracle was not supporting bare devices.
   4. Specific configuration examples:
      1. 3-node Rac,os:sun solaris,version 11.2.3, ASM
      2. Common Cmd:crsctl,srvctl,crs_stat, CRSCTL status Resources
4. Services Service:
   1. Use service to reach load balance: Each node runs a specific application.
6. Asm:
2. Advantages and Disadvantages
1. Easy to manage: Add and subtract disks for ease of stripe and mirroring.
2. The file name is very strange, very difficult to copy, difficult to transfer files with FTP, sometimes very dangerous, sometimes SA does not understand ASM, can lead to disk corruption.
3. HA does not protect data.
4. Performance issues with RAC:
1. The main problem is the cache Fusion, a memory mechanism for RAC to maintain data consistency. Cache Fusion adds a lot of extra overhead, so the performance of two nodes is not necessarily better, so RAC does not improve performance. In practice, the service is used to assign functions to each node of the RAC, such as a node dedicated to DML, and a node responsible for reporting.
2. About patches: in Rolling patch mode: Each instance is turned off, Patch CRS and Oracle RDBMS home, using the local parameter in Opatch.
6. Load Balance:
1. On the client side, scan is used as the hostname in Tnsnames.ora, so that Oracle can dynamically allocate nodes, and scan is equivalent to encapsulating the real IP.
2. On the server side, when scan listener handles a call that comes in from the client, it puts it on node least load (from the CPU and the usage of the connection process).
3. So the remote listener should be set appropriately. For each node of the general RAC, we have two kinds of listener, local and scan. Can be seen from Listener.ora.
8. About init parameters:
1. In the Init.ora file, we have two parameter files: The local and Global,global parameters are in front of the * number.
2. Cluster_database must be true.
3. Other: Control_files,cluster_database_instances,instance parameters (starting with Instance name), Thread,instance_ Number,undo_tablespace,
4. Some parameters can be different on each node, or they can all be the same. such as memory parameters (Sga,pga,memory target).
10. Shared and unshared in RAC:
1. Non-shared: Binaries,alert log,redo,archive Logs,undo tablespace
2. Sharing: Control files, SPFile, temporary files
8. Case:
2. Performance Case 1: In OEM we observed: Global Cache buffer BUSY,GC current block 2-way/3-way and Cache buffer chains (hot blocks)
1. We have a lot of additions and deletions on the database, resulting in buffer cache synchronization problems. If DML operations are running on different nodes at the same time, global cache synchronization is required immediately, causing many systems to wait.
2. Workaround: Node feature segmentation. One node is used for additions and deletions, and the other for reporting.
4. Performance Case 2: On the OEM Performance table (log file sync), a commit high appears 7 minutes per hour.
1. First let the application group check the code to see the commit frequency.
2. Database, is there any bug related?
3. Storage group, to see if there are any problems, we put redo logs on the fast disk, still can not solve the problem.
4. Use the Bluestripe software (you can detect disk IO at the second level). Found during that time, there was a root process that consumed a lot of IO and competed against the database. We re-arrange the time period of the process running in crontab, solve the problem.
6. Reply case: SA scans the ASM disk, causing the database to crash.
1. HA does not guarantee storage security, there are many ora7445 and ora600 errors in alert log. Block bad error, the database cannot be restarted, or the database is up, but some data is not available.
2. We need incomplete RMAN recovery. (There is a cutoff time, after which the data will be lost)
3. Recover script Need-steps (restore datafiles and recover with incremental backups or archives)
8. About ASM: After restore, if recover fails, when you rerun the script, Oracle does not restore the data files under the Linux file system, only restore incremental backup pieces/archive Logs. In ASM, Oracle will restore the data file again, and it will take more time. So you need to skip the Restore datafile section in the script. Direct recover.
10. About Backup recovery: Incremental backups are really important.

Rac & DG