1. Oracle database architecture For
a single-node database, if the instance goes down, if a business is linked to the instance, the business will be interrupted. At this time, the system is not available, so the availability of a single node is very poor at this time.
For RAC, like a single instance, it is still a data file, which is the same storage with oracle files on it, but the same data file is shared by multiple instances. The advantage of this is redundancy between multiple instances.
Data Guard can realize physical backup, synchronize redo logs and then apply them, and can also realize logical backup by first converting redo logs into SQL and then executing them. Logical backups are highly portable on multiple platforms.
2. Rac system architecture
The storage can refer to this article (nfs+asm) oracle rac deployment
, among which clusterware is the software that oracle realizes clustering:
ORACLE Clusterware has the following components:
1. OCR Oracle Cluster Registry In order to solve the "forgetfulness" of the cluster problem, the simplest solution is to have only one configuration for the entire cluster, and all nodes share this configuration. The method adopted by ORACLE is to put this configuration file on the shared storage, and this file is OCR DISK. The node reads and writes OCR through the OCR Process. But only one node can read and write OCR DISK called OCR Master node, and other nodes submit requests to the OCR process of the master node through the OCR process of this node, and the master OCR Process completes the physical reading and writing, and synchronizes the contents in the OCR cache of all nodes. (The configuration is not synchronized)
2. Voting Disk This file is mainly used to record the status of the members in the node. When a split brain occurs, the Partition is arbitrated to obtain the control of the cluster
3. Cluster Synchronization Service (css): The CSS service monitors the health status of the cluster in real time through various heartbeat mechanisms, and provides cluster service functions such as split-brain protection. The CSS service
has two heartbeat mechanisms: the Network Heartbeat of the private network is used to monitor between nodes The health status (timeout threshold parameter MisCount MC), through the Disk Heartbeat of Voting Disk (timeout threshold parameter IO timeout IOT). It is used to monitor whether there is an IO failure accessing shared storage, such as HBA card failure, SAN switch or storage cable damage that causes I/O timeout.
These two heartbeat mechanisms decide whether a node should be evicted. The following describes the conditions for the eviction behavior to occur:
(1) No timeout, no eviction
(2) Network ping is completed within the MC time, disk ping exceeds MC, but within the IOT, there will be no eviction
(3) Network ping If it is completed within MC time and the disk ping exceeds IOT, the node will be expelled
(4) If the network ping exceeds the MC setting, the disk ping will be completed within MC, and the node will also be expelled.
4. Anything managed by Cluster Ready Service CRS is called For resources, they can be a database, an instance, a listener, a virtual IP (VIP) address, an application process, and so on. CRS manages these resources according to the resource configuration information stored in OCR. This includes startup, shutdown, monitoring, and failover (start, stop, monitor, and failover) operations. The CRS process generates an event when the state of a resource changes. When you install RAC, the CRS process monitors Oracle instances, listens, etc., and automatically starts these components when a failure occurs. By default, the CRS process will restart 5 times, and if the resource still cannot be started, no more attempts will be made. The high availability HA of the cluster depends on it
5. Event Manager Service is responsible for publishing various events generated by CRS.
6. Oracle Notification Service (ONS): publishing and subscription service of FAN Fast Application Notification events.
7. RACG expands the functions of clusterware to support Oracle's specific needs. It executes the server-side callout script (server callout script) when a FAN event occurs.
8. Process Monitor Daemon: monitors the cluster and provides I/O isolation (fencing)
II. rac network architecture
1. VIP IP
Let's talk about PUBLIC and VIP Oracle first Each node in RAC has a virtual IP, referred to as VIP, which is in the same network segment as PUBLIC IP. vip is attached to the public network interface. The main difference between VIP and PUBLIC IP is: VIP is floating, while PUBLIC IP is fixed. When all nodes are running normally, the VIP of each node will be assigned to the public NIC; check ifconfig under linux, there are 2 IP addresses on the public network card; if a node goes down, the VIP of this node will be transferred to still running on the node. That is, there will be 3 IP addresses on the public NIC of the surviving node.
2. Private IP
Let’s look at Private in RAC again. The Private private IP in RAC is used for heartbeat synchronization. This can be ignored directly for the user level. This IP is used to ensure that the two servers synchronize data, and it belongs to the internal communication between RAC. priv and public should not belong to the same interface.
3. SCAN IP
SCAN IP in RAC In 11gR2, SCAN IP appears as a new IP. Scan IP is actually a new connection layer added by oracle between the client and the database. When there is a client accessing , connect to the SCAN IP LISTENER, and when the SCAN IP LISTENER receives the connection request, it will forward the client’s connection request to the VIP LISTENER on the corresponding instance according to the LBA algorithm, thus completing the entire connection process between the client and the server. The simplification is as follows: client -> scan listener -> local listener -> local instance Scan can also be understood as a virtual host name, which corresponds to the entire RAC cluster. The client host can access any node of the database cluster only through this scan name. Of course, the nodes visited are random. Oracle strongly recommends configuring and resolving SCAN through DNS Server's round robin mode to achieve load balancing (that is, to rotate and connect to the IP address corresponding to SCAN). This is somewhat similar to the principle of load balancing through vip and listener loadbalance configuration.