RAC Architecture Evolution

From a single instance to RAC, the architecture is also built by the RAC cluster and clusterware cluster.

In essence, RAC databases and single-instance databases are not much different. data files and control files are shared, but each instance has its own independent redo log group; for the automatic undo management mode, each instance has an independent undo tablespace. There is no essential difference between the RAC system and the single-instance system in the format of data files and control files. Therefore, the single-instance system can be easily upgraded to the RAC system.

(I) SGA changes

A significant change in SGA is the addition of GRD.
Save GRD:
① PCM lock Information
② Bitmap of node health status

(Ii) background process changes
Only the changes of background processes on the RAC cluster are described here.
① Lmon Process
Lmon provides the CGS and NM services to maintain the status of the RAC cluster.
(1) nm
Is the communication channel between the RAC cluster and the clusterware cluster.
Register the resources of the current node to the local clusterware through nm, and then pass the local clusterware to other nodes.
At the same time, Nm also obtains their resource status from clusterware of other nodes.

All processes of each RAC instance are registered to clusterware as a NM group.
The lmon process is the leader and obtains the member ID. Other processes are registered with the same ID.

(2) CGS
Cluster Group Services
This service is mainly responsible:
The bitmap in GRD records the health status of the node. 0 indicates that the node is disabled, and 1 indicates that the node is running normally.
Lmon scheduled communication between nodes to ensure GRD bitmap consistency
In addition, lmon can work with lower-level clusterware or independently
The RAC cluster does not always assume that the clusterware cluster can handle the problem.
If the wait times out, lmon will automatically trigger IMR (instance membership recovery)

② Lmsn Process
Transfers data blocks between instances

③ LMD Process
Coordinates the access sequence of data blocks among multiple instances.

④ LCK Process
Responsible for concurrent access to non-Cache fusion Resources

⑤ Diag Process
Errors During instance running and recorded in alert. Log

⑥ GSD Process
Receives user commands from client tools

(Iii) file changes
① Redo thread
An instance corresponds to a redo thread
Redo is shared, but belongs to various instances.
Because in crash recovery, the node that executes the recover must be able to access the online logs of the faulty node.
In RAC, each instance is written to its own redo log, but it can read the redo log of other instances.
Each instance has a thread #, which can be different from the redo of other instances.
When the instance is restored, it reads its own thread # redo.
Read the global Rodo when the media is restored

② Undo tablespace
Each instance requires a separate undo
Use the SID. undo_tablespace parameter to configure

③ Spfile
To be accessed by all nodes, it should be placed on the shared Disk

 

④ Archived log

First, there is a concept that archive logs on any node can be seen on any node in the RAC environment. So there is no v $ or GV $ problem.

For RAC, archive logs are a series of redo logs. The most important deciding factor is the name of the archive log. If you can find this name, all nodes are required. The name contains the full path, this is why we have an NFS solution for RMAN backup.

SQL> select THREAD#,count(*) from v$archived_log group by THREAD#;   THREAD#   COUNT(*)---------- ----------         1        280         2        256