TWI721355B - High-availability database system - Google Patents

Abstract

A high-availability database system includes at least a pair of subordinate servers, each of which is built with a database management system (DBMS); a main storage carrier, the main storage carrier and a pair of subordinate servers use the ISCSI protocol, To enable the connection and communication between a pair of slave servers as a DBMS and the main storage medium as a database (DB), and to be independent and separate; and a first master server, according to a preset weight value Assign services to the pair of slave servers, that is, the first master server directs all database connections to the slave server with a high weight value by default, and another slave server with a low weight value The server is on standby and becomes a backup for database management services to provide uninterrupted high availability of system services.

Description

High-availability database system

The invention relates to a database system, in particular to a database system architecture that provides a high-availability mechanism to ensure that the service of the database is uninterrupted and the stored data are consistent and complete.

The so-called "High Availability System" (High Availability System) is to provide a mechanism for operating equipment to be immediately converted to backup equipment in the event of a disaster or failure to maintain uninterrupted services, so as to reduce the failure of a single device or power outage in the system And the incidence of service suspension. In order to maintain the "high availability" (HA) of the equipment, there are at least two switchable equipment in the system. The equipment in use is called the "operating equipment", which has a destination IP address and the other The station is called a "standby device", which does not have a destination IP address; when the so-called "running device" fails or fails, the "standby device" will automatically switch to the running device. And take over the destination IP address to provide services.

What the present invention is about is a database system used to store business operating data. Since it is related to the lifeblood of the company’s survival, how to maintain the "high availability" of the system so that the management and services it provides will not be interrupted. It is extremely important. Generally speaking, the most common reason for database system damages and failures is that sudden crashes or power outages cause disasters during data transactions. In particular, general database management systems have an option to store data in temporary memory. The mechanism of writing to the hard disk to improve the operating performance of the system. Therefore, if the database system crashes or loses power, even if a "high availability" backup mechanism is built in the system, failures or disasters will occur The transfer (failover) is still unavoidable because the data is too late to be written to the hard drive and the loss is caused. The original standby device will take over operation without obtaining the complete and correct storage data, which will eventually cause the database in the hard drive to be confused and difficult to remedy. .

In order to improve the prior art, the present invention provides a novel "high-availability database system", which separates the database management system (DBMS) from the database (DB) under the establishment of a high-availability mechanism. Through The ISCSI protocol establishes communication and connection with each other, and can flexibly plan the capacity of the configuration file for data storage, and write it into the storage carrier during data storage. Even if any DBMS crashes or is damaged, it will be transferred through a failure or disaster. The main purpose of the present invention is to avoid data loss due to the data being stored in the temporary memory and too late to be stored in the hard disk, and to ensure the consistency and integrity of the data.

To achieve this objective, the database system disclosed in the present invention includes a pair of subordinate servers, each of which is built with a database management system (DBMS); two storage carriers, respectively, a primary storage carrier and a second storage The carrier, the main storage carrier and the pair of slave servers are connected and communicated between the pair of slave servers as the DBMS and the main storage carrier as the database (DB) through the ISCSI protocol, and Each is independent and separate; and, a first master server distributes services to the pair of slave servers (DBMS) based on a preset weight value instead of load balancing, that is, the first master server will Direct database connections to the slave server with a high weight value by default, and another slave server with a low weight value as standby, which becomes a backup mechanism for database management services; at the same time, on any slave server (DBMS) When each piece of data is written into the database (DB), it will be pre-recorded in the main storage medium In a log file file, whether any slave server fails, another slave server that takes over can obtain data from the log file file that has not been written to the database (DB) to maintain data consistency . On the other hand, the main storage carrier and the second storage carrier have a DRBD mirroring mechanism built through the Internet. When the main storage carrier fails, a real-time backup data can be obtained for disaster recovery. The database system constructs a highly available environment, taking into account future scalability and the efficiency of data backup.

10‧‧‧The first master server MA

11‧‧‧Second Main Server MB

20‧‧‧The first slave server SA

21‧‧‧Second slave server SB

30‧‧‧The main storage carrier LUNA

31‧‧‧The second storage carrier LUNB

4‧‧‧ISCSI

5‧‧‧DRBD

6: Heartbeat

7: log file file

Figure 1 is the architecture diagram of the database system built by the present invention.

As shown in Figure 1, the database system of the present invention includes a first master server MA (10), at least a pair of slave servers SA, SB (20, 21), and at least two independent storage carriers LUNA ( 30) and LUNB (31), through the Internet, the pair of slave servers (20, 21) and the two storage carriers (30, 31) are connected to the first master server (10) and receive its control .

More specifically, the pair of slave servers includes a first slave server SA(20), and a second slave server SB(21). The first slave server SA(20) and the second slave server SB(21) A database management system (DBMS) is built in each of the two slave servers SB (21). The database management system (DBMS) is a set of computer programs that can control a database (database, Below are referred to as DB) classification, calculation and data access.

The two storage carriers include a main storage carrier LUNA (30), and a second storage carrier LUNB (31), the main storage carrier LUNA (30) and the first slave server SA (20), the second slave server The SB(21) is a small computer system interface (Internet Small Computer System Interface, hereinafter referred to as ISCSI) built on a network, based on the storage technology under the Internet and the SCSI protocol, the main storage carrier LUNA (30) is set as the target of ISCSI (4). The first slave server SA (20) and the second slave server SB (21) are set as initiators, so that they can be used here as The first slave server SA (20) of the database management system (DBMS), the second slave server SB (21) and the main storage carrier LUNA (30) as the database (DB) obtain connection and communication , And generate the database management The system (DBMS) and the database (DB) are independent and separate.

A DRBD (Distributed Replicated Block Device) mechanism is established at the bottom of the main storage carrier LUNA (30) and the second storage carrier LUNB (31) through the Internet. This DRBD (5) is a distributed storage system on the Linux platform , RAID 1 mirroring technology similar to disk arrays synchronizes the data of the two. In the design of the present invention, the DRBD (5) of the storage carrier adopts a single-master mode, which means that any resource at any specific time, this two-node cluster has only one master node, that is, as the first slave of the DBMS The server SA (20) or the second slave server SB (21) will both write data into the main storage carrier LUNA (30), and the second storage carrier LUNB (31) is in the standby state and is responsible for The main storage carrier LUNA (30) performs real-time backup to avoid data loss. In other words, when the main storage carrier LUNA (30) accidentally fails, it will still generate a real-time backup data for database (DB) Recovery.

The first master server MA (10) controls the operation status of the first slave server SA (20) and the second slave server SB (21), in particular, in the database system disclosed in the present invention , The first master server MA (10) is based on the preset weight value to the first slave server SA (20), and the second slave The subordinate server SB (21) performs service distribution instead of load balancing. In more detail, through an automatically controlled program, the running master server MA (10) directs all database connections to the first slave server SA ( 20), and the second slave server SB(21) with another low weight value (w=0) is on standby; if the first slave server SA(20) cannot provide services normally, the first master server MA(10) will increase the weight value of the second slave server SB(21) from 0 to 1, and direct all database connections to the second slave server SB(21) as a database service Backup. For example, the system will first generate a virtual IP address (VIP) as the IP representative location of the subordinate server cluster (SA and SB) in the intranet, and then set this VIP to the first master server MA ( 10) Above, the first master server MA (10) mainly uses the two Linux packages ipvsadm and keepalived to realize each other's "HA" function, and also uses these two packages to control the weight value. Achieve the user connection function assigned to the subordinate server clusters (SA and SB). The settings of this backup mechanism are in the keepalived.conf configuration file in the keepalivd package. The TCP or udp port of the database is detected through keepalive. When the SA (that is, the first slave server) cannot provide services, the MA(10 ) Will trigger the notify_down to control the weight value, and set the SB (that is, the second slave servo The weight value of the database management system (DBMS) is set from 0 to 1. At the same time, the connection guide SB will continue to provide users with database management system (DBMS) services. If the SA database management system (DBMS) returns to normal, MA(10) will Use notify up to trigger the SB weight value from 1 to 0, and lead the connection back to SA again.

In addition, in order to avoid data loss when the first slave server SA (20) and the second slave server SB (21) fail or failover (failover), the first slave server SA (20) or the second slave server SB (21) When the slave server SB (21) writes each piece of data into the database (DB), it will use WAL (Write-Ahead-Logging) technology, leading the database (DB) to record the data in the main storage medium LUNA (30 ) In a log file file (7), and use DRBD (5) to synchronize the database (DB) and the log file file (7) to the second storage carrier LUNB (31), so that whether it is the first After the slave server SA (20) or the second slave server SB (21) fails, the other database management system (DBMS) that takes over can obtain from the log file (7) that has not been written to the database ( DB) data to maintain a consistent service.

In addition, in order to improve the high availability "HA" of the database system architecture of the present invention, a second master server MB (11) and the at least one pair of slave servers SA, SB (20) are further built in the present invention. ,21), and at least two storage carriers LUNA and LUNB (30,31) are connected through the Internet and control them. When the first master server MA (10) is defaulted as the running device, the second master server MB (11) will act as On standby, a Heartbeat (6) detection mechanism is established between the first main server MA (10) and the second main server MB (11), which periodically send heartbeat packets to each other within a predetermined time , If the second main server MB (11) does not receive the heartbeat packet from the first main server MA (10), the second main server (11) will automatically switch to the operating device and take over the resources to maintain The normal operation of database services has established a highly available operating environment.

From the above description, it can be understood that the database system of the present invention completes the separation of the DBMS and the DB using the ISCSI protocol, and can flexibly plan the capacity of the configuration file for data storage, and directly write the data to the main storage carrier LUNA ( In 30), even if any DBMS (SA or SB) crashes or is damaged, it will not be too late to be stored in the hard disk due to the data still in the memory during the failure or disaster transfer (failover) through the weight value adjustment of MA(10). The loss of data generated by the disk (LUNA) can better ensure the integrity and correctness of the data; at the same time, the DRBD mirroring mechanism built through the main storage carrier LUNA (30) and the second storage carrier LUNB (31), when the main storage Carrier LUNA (30) no In the event of a failure, you can still get an instant backup data for disaster recovery. It not only constructs a high-availability mechanism for the database system, but also takes into account future scalability and the efficiency of data backup.

Although the present invention has been described in detail through the above-mentioned preferred embodiments, those familiar with the relevant art should know that modifications to the types and details of the present invention should not deviate from the spirit and scope of the patent application described later in the present invention. Chen Ming first.

10‧‧‧The first master server MA

11‧‧‧Second Main Server MB

20‧‧‧The first slave server SA

21‧‧‧Second slave server SB

30‧‧‧The main storage carrier LUNA

31‧‧‧The second storage carrier LUNB

4‧‧‧ISCSI

5‧‧‧DRBD

6‧‧‧Heartbeat

7‧‧‧log file file

Claims (4)

A high-availability database system includes a first master server, at least a pair of slave servers, and a master storage carrier. Through the Internet, the pair of slave servers, the master storage carrier, and the first master The server is connected to and receives its control; wherein, the pair of slave servers includes a first slave server and a second slave server, each of which is built in the first slave server and the second slave server Database management system (DBMS); the primary storage carrier is set as the destination of ISCSI through the ISCSI protocol between the first subordinate server and the second subordinate server, and the first subordinate server The server and the second slave server are set as the initiator, so that the first slave server and the second slave server as the database management system (DBMS) and the main storage medium as the database (DB) Obtain connection and communication, and generate that the database management system (DBMS) and the database (DB) are independent and separate; in addition, when any slave server (DBMS) writes each piece of data into the database (DB), The data will be recorded in a log file of the main storage medium in the database (DB) In the file file, after any one of the slave servers fails, another slave server that takes over can obtain data from the log file file that has not been written to the database (DB) to maintain data consistency; The first master server allocates services to the first slave server and the second slave server according to the preset weight value, and directs the connection of the database to the first slave server preset to the high weight value The second slave server with a low weight value is on standby. The high-availability database system according to claim 1, which further includes a second storage carrier. The main storage carrier and the second storage carrier establish a DRBD mechanism through the Internet, and the DRBD mechanism adopts a single master Mode refers to any resource at any specific time, this two-node cluster has only one master node, that is, the first slave server or the second slave server as the DBMS will write data to the main storage In the carrier, the second storage carrier is in a standby state and is responsible for real-time backup of the main storage carrier. The high-availability database system according to claim 2, wherein a second master server, the at least a pair of slave servers and two storage carriers are connected and connected via the Internet in the framework. Control over it, in the first When the main server is preset to run the device, the second main server is on standby. The high-availability database system according to claim 3, wherein a heartbeat detection mechanism is established between the first host server and the second host server, and heartbeat packets are periodically sent to each other. If the second main server does not receive the heartbeat packet from the first main server within a predetermined time, the second main server will automatically switch to the operating device and take over the resources to maintain the normal operation of the database service.

Images