KR100492167B1 - Shared-nothing database cluster server system and On-line scaling method - Google Patents

Abstract

The present invention relates to a non-shared database cluster server and an online extension technique, including: a client; A basic query processing component; A database management component; Global meta manager; Central controller; And a database; the first step of placing the original data in the non-shared database cluster server and the original node, the first node N1, and the duplicate data of the original data in the second node N2, the replication node. ; The predetermined data of the first node is transmitted to the third node, which is an expansion node, and the predetermined data of the second node (where the predetermined data of the second node is different from the predetermined data of the first node). Data) transmitting to a fourth node which is an expansion node; And a third step of, when data transmission is completed in the second step, transmitting data of different parts previously transmitted from each expansion node to the other expansion node to generate a complete data. An online expansion method of a cluster server system.

Description

Shared-nothing database cluster server system and on-line scaling method

The present invention relates to a non-shared database cluster server system and an online expansion method. More specifically, the present invention relates to a large capacity in a dynamic environment in which a pattern is irregular by a large number of users and the number of queries may increase temporarily. This paper relates to a database cluster server system and an online expansion method for stably and continuously processing queries.

In general, the technology of a database cluster can be divided into a shared disk structure, a shared memory structure, and a non-shared structure.

The shared disk structure and the shared memory structure have a high system configuration cost, communication load, and management cost, while the non-shared structure has high scalability and high cost efficiency. For this reason, a nonshared structure is being adopted by several research and commercial database clusters.

A non-shared database cluster system includes systems that operate independently as nodes of a cluster and connects them to a high speed network. The entire data stored in the database is stored in the form of replication or fragmentation on each cluster node for availability. Based on this storage type, if a service fails due to a failure in any node, the service is provided instead of the data of the node having the replica.

The non-shared database cluster will return the result to the client first when the update operation of the original node and the update node completes the update of the original data and the replica at the same time according to the method of updating the replica during query processing. It can be divided into a lazy update method that performs an operation.

In addition, depending on the update area, it can be divided into the whole update method and the master update method. In the case of the total update, the integrity can be guaranteed, but the processing of the update operation is difficult, and the master update is efficient for the update operation. But there is a problem with some integrity.

In addition, depending on the type of data storage, it is divided into Full Replication and Partially Replication using Partition and Replication together. Full replication provides simple storage structure and management, and provides high availability by immediately using a given replication node in the event of a failure. However, in the case of update query processing, expensive operations are performed to maintain concurrency of all replica nodes and to process the same update operation, thereby greatly increasing query processing time. On the other hand, partial replication has more complicated storage structure and failure handling scheme than full replication, and the performance is different depending on the characteristics and storage structure of data.

Non-shared database clusters generally have higher scalability than systems with shared memory and shared disk structures. Unlike shared memory and disk structures, which require the server to be stopped and restarted for hardware or software expansion, the non-shared structure allows each node to be configured independently, freely expanding and contracting nodes through data reorganization and expansion. Do.

Leverage these features to ensure on-line reorganization and on-line when a problem occurs where the throughput of an existing system reaches its limit due to load concentration issues, failures, or increased users, even when the service is provided on-line. The solution can be solved through the line extension technique.

On-line reorganization technique continuously changes the existing database configuration in the cluster while continuously providing services, and reorganizes the database by creating, deleting, moving, and merging existing data based on the changed configuration. Technique. In order to process the reorganization process on-line, a separate transaction processing log is placed in the case of data movement to maintain the integrity of the data being moved, and the update area is determined simultaneously with transaction processing, and whether the determined update area is the transmission area or not. As a result, the method can be divided into a technique for determining whether to transmit real-time update data.

The on-line extension technique is a special technique that uses the on-line reorganization technique only in the case of node expansion through idle nodes. An idle node is an inactive node that is included in the cluster server but does not participate in service processing. This technique enables real-time server expansion by having idle nodes always waiting in case the user needs to increase the server throughput. It may seem more efficient to pre-scale the server's throughput by involving all available nodes, including idle ones, in the server. These problems are caused by heavy load concentration, and when dealt with through on-line reorganization techniques, they cause more problems by placing a greater load on the server's throughput. On-line scaling, on the other hand, minimizes the additional load in the problem-prone areas that can be caused by data movement through simple data movement and processing, while at the same time, idle nodes do not provide resources for the entire node. Assigning to on-line extensions allows for faster processing than on-line reorganization techniques. Therefore, an on-line scaling technique that can handle the load concentration problem in a short time while minimizing the load of the server through the idle node can be usefully used.

In the case of a conventional non-shared database cluster, on-line reorganization and on-line expansion during service operation were greatly limited for the integrity of the database. In order for the user's query processing to continue to be serviced during the reorganization and expansion process, it is necessary to change the configuration of a separate database, change the log processing method, and additional network, memory, and storage space to maintain the integrity of the data included in the current reorganization area. Resource allocation is required. As a result, the processing performance of the currently provided database service for the on-line reorganization is greatly reduced. Various techniques have been applied to solve this problem, but when most of the techniques are performed, the performance of the system is greatly degraded.

Accordingly, the technical problem to be achieved by the present invention is to minimize the impact on system performance, to be suitable for dynamic environments, and to minimize response in processing a large volume of queries without violating integrity through efficient on-line reorganization and on-line expansion. To provide a non-shared database cluster server system that maintains time and an online expansion method.

Another technical problem of the present invention is to design a database cluster that does not cause performance degradation while maximizing availability through a replication technique, and to configure a cost-effective, non-shared, common desktop node into a cluster, even if one or more servers are stopped. It is to provide a non-shared database cluster server system and an online expansion method that provide fault-tolerant services.

Another technical problem of the present invention is a non-shared database cluster server system and on-line which enables stable real-time expansion while maintaining query throughput by minimizing additional server load with an on-line expansion technique that copes with rapid user growth in real time. To provide an extension method.

In order to achieve the above technical problem, the non-shared database cluster server system according to the present invention includes a GUI-based query processing tool, an administrator program for managing the cluster system, a monitoring tool for monitoring the resources of each node, the interface library A client for transmitting a query to the basic query component through; A basic query processing component that analyzes the query received from the client and transmits the query to the database management component and provides the query processing result to the client; A database management component for determining an operation order, reorganization and expansion of a corresponding query, and a degree of allocating transaction resources according to an analysis result of the query statement; A global meta manager which provides information of data accessed by the query statement to the basic query processing component and manages to maintain the same meta information of all nodes at any time based on a consistency maintaining protocol; A central controller controlling server operation including starting, stopping, expanding, reducing, and changing settings of the server; And a database storing predetermined data.

In order to achieve the above, another technical problem, the online expansion method of the database cluster server system according to the present invention is the original data to the first node (N1), which is the original node, the original data to the second node (N2), which is a replication node. A first step of placing duplicate data of a second step of performing an expansion operation from Ta1 'to Ta2 at the second node; And transmitting, by the second node, the same copy to Ta2 'of the third node, which is an idle node, using Ta2, which is the divided data of Ta1', as the original data of the corresponding node. .

In order to achieve the above, another technical problem, the online expansion method of the database cluster server system according to the present invention is the original data to the first node (N1), which is the original node, the original data to the second node (N2), which is a replication node. A first step of placing duplicate data of the; The predetermined data of the first node is transmitted to the third node, which is an expansion node, and the predetermined data of the second node (where the predetermined data of the second node is different from the predetermined data of the first node). Data) transmitting to a fourth node which is an expansion node; And a third step of generating complete data by transmitting data of different parts previously transmitted from each expansion node to the other expansion node when data transmission is completed in the second step.

Hereinafter, with reference to a preferred embodiment according to the present invention will be described in detail. In the following description, only parts necessary for understanding the operation according to the present invention will be described, it should be noted that the description of other parts will be omitted so as not to distract from the gist of the present invention.

1 is a diagram illustrating a configuration environment of a database cluster server.

Referring to FIG. 1, a database cluster server according to an exemplary embodiment of the present invention includes a web browser 100 that users generally use as an access client, a web server cluster 200 that is responsible for web-based access, and an application application service. It can be operated in an environment including a database cluster 300 that provides a basis for.

Users connect to a web server through their web browser to use Internet services. Web servers include networks such as Open Database Connectivity (ODBC), Java Database Connectivity (JDBC), Internet Server Application Programming Interface (ISAPI), Web-based Common Gateway Interface (CGI), Call Level Interface (CLI), and Winsock. It communicates with the database cluster server through an application written using a function-based interface library. The database cluster 300 includes an active node that participates in a current service and performs a query processing, and a spare node that is waiting for on-line expansion, and each node has an independent database.

2 is a configuration diagram of a database cluster server according to an embodiment of the present invention.

Referring to FIG. 2, a database cluster server system includes a client 400, a global meta manager 500, a central controller 600, a basic query processing component 700, a database management component 800, and a database 900. do.

The client 400 is an interface library, a GUI (Graphical User Interface) based tool, a server-specific tool for directly inquiring database requests, an administrator program for managing a cluster system in the form of a general database management system, and each. It includes a monitoring tool for monitoring the resource usage of the node.

Unlike the existing single database system, the global meta manager 500 manages meta information of cluster nodes connected through a network. That is, it manages information related to query processing such as information of active nodes and idle nodes included in the cluster, partition and replication information of relations, configuration information of master and replicas, and user information. In addition, it manages the meta information unalterably in special cases where the configuration of the database such as on-line reorganization, on-line expansion, and recovery processing must be changed temporarily.

The global meta manager 500 is equally distributed to each node, and the meta information of each node can be changed at any time through the coherence protocol included in the internal network connection handler 740 when there are some database schema changes. It has a structure of maintaining consistency so that meta information of nodes is the same.

The central controller 600 is a component that controls functions related to server operation such as starting, stopping, expanding, reducing, and changing settings of the server. That is, based on the cluster node information registered in the meta information, it manages the control of the system operation and database structure such as start, stop, expand, shrink and move nodes.

In addition, it handles interfaces for processing database reconfiguration requests, node control requests, cluster node addition and deletion requests, and service start and stop requests except general query processing. It is also a component that includes a message transceiving process with the basic query processing component 700 or the database management component 800.

The manager program controls the server through communication with the central controller 600.

The basic query processing component 700 includes an external connection processor 710, a query processor 720, a transaction processor 730, an internal network connection processor 740, and a storage manager 750. The basic query processing component 700 is a component that is responsible for the overall execution of the query processing process, and is almost similar to the structure of a single database management system. However, the process for query processing is designed in consideration of the cluster.

When the query is input from the external connection processor 710, the basic query processing component 700 analyzes the query in the query processor 720 to obtain information on the data accessed by the query from the global meta manager 500. . The query is passed back to the transaction handler 730 and is assigned a session for the transaction. The transaction processor 730 is responsible for transaction interworking and management of each node in the cluster environment.

If the data included in the access area does not exist in the current node, the query is forwarded through the internal network connection processor 740. At this time, the session information and transaction information are transmitted together to receive a separate session including the forwarding information from the corresponding node. The transaction will run.

The basic query processing component 700 returns the result to the node that originally received the query when the query processing is completed, and delivers the result to the client. The storage manager 750 stores and manages the processing log, forwarding log, recovery log, reorganization log, and extension log of the cluster server in addition to the general data and query processing log.

The database management component 800 includes an on-line expansion processor 810, a recovery processor 820 and an on-line reorganization processor 830.

It performs functions related to database reorganization and recovery processing, such as on-line reorganization and on-line expansion. That is, the central controller 500, the global meta manager 600, and the basic query processing component 700 are in charge of reconstruction of the database, recovery of nodes, and expansion processes.

The on-line scalability processor 810 has idle nodes in the database cluster, includes queries as active nodes and includes data as idle nodes when real-time processing is needed because queries are concentrated on arbitrary nodes or when resource limits are reached. Splitting and moving to an idle node.

If the update method of the database cluster uses a delayed update method, the on-line extension processor 810 performs on-line expansion on a replication node without master data, and stores newly created partitioned master data on the corresponding replication node. To reduce data movement.

When the update method of the database cluster uses an immediate update method, the on-line extension processor 810 performs data movement in parallel to reduce the data movement area at the original node and shorten the expansion time.

FIG. 3 is a diagram illustrating an operation flow of the system shown in FIG. 2.

Referring to Figure 3 describes the flow of each step as follows. In step 310, the client 400 sends a query to the basic query processing component 700. In step 320, the query processor 720 of the basic query processing component 700 analyzes the query statement. In step 330, the basic query processing component 700 transmits the analyzed query statement to the database management component 800. In step 340, the database management component 800 determines the operation order of the query. In step 345, the database management component 800 allocates transactional resources. In step 360, the database management component 800 sends the query statement to the transaction processor 730. In step 370, the basic query processing component 700 sends a query to the target node if the target node is different from the current node. Depending on the analysis result, the query sent to other nodes has a different format. In step 380, the basic query processing component 700 physically reflects the results of the query to the database 900 on disk as a commit step. In step 390, the basic query processing component 700 returns the query processing results to the client 400.

Referring to FIG. 3, the query received from the client 400 is analyzed by the query processor 720 through the external connection processor 710. The analyzed results are used by the database management component 800 to determine whether to reorganize and expand, the order of operations, and the degree of transaction resource allocation.

Once the database reorganization is determined by the database management component 800, the on-line reorganization handler 830 may then be configured to the global meta manager 500, the central controller 600, and the basic query processing component of all nodes currently included in the database cluster 300. In conjunction with 700 to perform an on-line reorganization process. In other cases, the query is sent to the current node or another node for processing based on the consistency protocol, and the result is returned to the client 400.

4 is a block diagram of the on-line reorganization processor shown in FIG.

Referring to FIG. 4, the on-line reorganization processor 830 includes a resource manager 831, a data processing manager 832, an optimization processor 833, a log transfer processor 834, and a data movement processor 835. .

The on-line reorganization processor 830 works with the global meta manager 500 and the central controller 600 to perform data generation, deletion, moving, merging, and partitioning of data in a node for reconstruction of the database.

The on-line reorganization process is divided into two processes: a change query process of a database configuration executed by a database management component and a reorganization process performed automatically by a resource manager 831 and a data processing manager 832.

The on-line reorganization process executed by the database management component 800 is handled by the log transfer processor 834 and the data movement processor 835 without any separate optimization process. The on-line reorganization process performed by the resource manager 831 and the data processing manager 832 is performed by all nodes in the optimization processor 833 based on the information of the nodes obtained from the resource manager 831 and the data processing manager 832. The process proceeds to finding a modified configuration of the database that is optimized to distribute the load evenly. The optimization algorithm used in this process is operated by receiving the numerical values such as the update rate of the entire query, the number of query processing per table, the response time per node, the query forwarding rate, and the network usage rate from the resource manager.

The resource manager 831 is a component that continuously monitors resources of each node included in the cluster to determine whether to reorganize on-line. If the resource usage of a particular node exceeds the threshold, the resource manager 831 obtains the information of the node from the data processing manager 832 to determine whether to perform on-line reorganization processing, and invokes the reorganization operation of the optimization processor 833. Done.

The data processing manager 832 manages statistical information on user query processing, such as a query update area, an access node, a forwarding frequency, and a disk access frequency. Based on this information, the data processing manager 832 obtains resource information of the node from the resource manager 831 like the resource manager 831 and determines whether to process on-line reorganization. Here, the resource and statistics values of the node used as reference values may be changed by the database administrator.

The optimization processor 833 performs a function of creating, deleting, dividing, merging, and moving relations together with the log transfer processor 834 and the data movement processor 835 for on-line reorganization processing. As a protocol used for maintaining integrity in a network environment, a coherence protocol is used that manages all nodes to perform transaction processing in the same order at any time. Based on the consistency protocol, integrity can be maintained without performing separate network management in the optimization processor 833, the log transfer processor 834, and the data movement processor 835 to perform the on-line reorganization process. .

Log transfer processor 834 and data movement processor 835 perform the operations required for the on-line reorganization, and include components responsible for log transfer to maintain the integrity of the data.

When an update query occurs during the reorganization process, the log transfer processor 834 compares the update region of the query with the data region transmitted up to now, selects the updated region after transmission, and transmits the corresponding data to the target node. .

FIG. 5 is a diagram illustrating an operation flow of the on-line reorganization processor illustrated in FIG. 4. Referring to FIG. 5, the on-line reorganization process starts with a query form and determines whether the operation is correct to drive the on-line reorganization only in the correct case.

Referring to Figure 5, it will be described the operation flow of the on-line reorganization processor 830 according to the steps as follows.

In step 510, the reorganization query is input to the on-line reorganization processor 830. In step 520, the on-line reorganization processor 830 obtains information of data necessary for the query through the global meta manager 500. In step 530, the on-line reorganization processor 830 determines whether the planning is correct, and if it is correct drive the online reorganization, and if it is not correct returns the result. Criteria for judgment include the existence and defect of the node to be reorganized, the contents conflicting with the reorganization policy by the administrator, the existence of the original data, and the existence of sufficient space for the reorganization operation on the target node. In operation 540, the resource manager 831 determines an appropriate operation time and priority of operation. In operation 550, the data processing manager 832 finds a source node and a target node of the corresponding data, and generates unit processes necessary for performing an operation. In operation 560, the optimization processor 833 sets a setting for a disk area and an index of a relation stored in the database 900, and performs operations on data through the data movement processor 835. In step 570, it is determined whether there is a query being performed on the data. If there is a query running on the data, the result is returned. In step 580, when there is no query being performed on the data, the log transmission processor 834 determines whether or not the query is being performed on the data and performs log transmission. In step 590, the result is returned and terminated.

6 is a diagram illustrating an expansion process in a replication node according to an embodiment of the present invention.

Referring to FIG. 6, nodes from N1 to N5 are present, and boxes of white and gray backgrounds represent storage spaces in which original data and duplicate data are stored. If any table Ta1 is currently duplicated in N1 and N2, and there is original data in N1 and its duplicate data in N2, on-line expansion in N1 and on- again in N1 with the previously extended state. Indicates how line expansion occurs. In the initial state, the on-line extension performs an extension operation from Ta1 'of N2, which is a replication node, generates the divided data of Ta1' as original data of the node, and transmits the same copy to an idle node. When the on-line expansion is re-occurred in the previously expanded state, the on-line expansion processor 810 selects two idle nodes and transmits the divided replicas to the original node and the replication node, respectively.

The on-line expansion process according to FIG. 6 is all made from the replication node having a copy of the corresponding data. This extension process performs on-line extension process in the replication node regardless of query processing of the original node when using delayed update. This reduces the load due to expansion and enables faster processing than traditional on-line expansion techniques. The on-line expansion processor 810 also uses an algorithm to select nodes that do not contain possible master data in selecting duplicate nodes. This node selection method reduces the load due to data movement by increasing the possibility of local copy when performing on-line expansion.

7 is a diagram illustrating a parallel expansion process according to an embodiment of the present invention.

According to FIG. 7, the on-line expansion process is performed from all the replication nodes without distinguishing between the original node and the replication node, and each node shares the transmission area with each other. In addition, when data transfer is completed in each node, the data is transferred between the expansion nodes to the remaining expansion nodes to complete the on-line expansion by collecting data that makes a complete copy so that all the expansion nodes have identical copies. .

For example, if there is x data in node 10, y data in node 11, and z data in node 12, x data is stored in node 11 and node 12, y data is stored in node 10 and node 12, and z data is stored in node 10. To node 11. This on-line extension is used in case of full update.

This on-line expansion reduces the data transmission area of each node to reduce the load caused by the expansion, and by collecting data from the expansion nodes, all original nodes that are currently providing services log only their own transmission area after data transmission. By reducing the load on the source node, the data transfer and aggregation process is performed in parallel, which differs depending on the number of replicas, but has a faster processing time than the existing on-line expansion scheme.

FIG. 8 is a diagram illustrating an operation flow of the on-line expansion processor illustrated in FIG. 2.

Referring to FIG. 8, the operation flow of the on-line expansion processor 810 according to the steps will be described below.

In operation 910, the on-line expansion processor 810 is driven while receiving an expansion event, and includes a resource manager 831, a data processing manager 832, and an optimization processor 833 of the on-line reorganization processor 830. In conjunction, it searches for idle nodes to be extended and creates new global meta information. In step 920, the on-line extension processor 810 locks the meta information so that the schema of the database 900 cannot be changed. In step 930 and 935, the on-line extension processor 810 transmits the on-line extension event and new meta information to all nodes included in the database cluster. In operation 940, the meta information of the target node may not be changed. In step 945, the nodes identified as the extension targets through the meta information execute the receiver and the inserter process. In step 950, after preparing for expansion, the target node sends a standby message to the source node. In steps 960 and 965, the source node starts transmitting data and logs from the point where the on-line expansion preparation is completed. In step 970, the expansion node maintains the integrity of the data by reflecting the log again after the data is reflected. The on-line extension processor 810 changes the system to an extension change mode while an extension operation is being performed. In step 975, when the movement of the data is completed, the data transmission completion message is transmitted to the original node, and the process proceeds to the completion stage of the on-line expansion. In step 980, the source node sends the query input termination to the target node. In steps 985 and 987, the source node and the target node wait until the existing transaction ends. In step 990, the end of all operations are transmitted to the source node and the target node. In step 993, the original node sends an extension termination message to the on-line extension processor 810. In step 996, the original node ends the expansion.

From the completion phase of the on-line extension, all transactions are temporarily put into a waiting state, and the service is restarted with the new meta information applied from the point when log transmission and reflection are completed.

9 is a configuration diagram of the recovery processor shown in FIG. 2.

Referring to FIG. 9, the recovery processor 820 may include a system recovery processor 821, a data recovery processor 822, a recovery log recorder 823, a system statistics recorder 824, an error detector 825, and statistical information 826. ), Recovery processing setting information 827 is included.

The system recovery processor 821 is a component that is operated when a failure occurs in the system and recovery is impossible in a short time.

The data recovery processor 822 is a component that is in charge of recovery processing in the case of recovering within a short time such as network disconnection, disk error, and execution error.

The recovery log recorder 823 is a component that manages logs that are stored from a time point when a failure occurs in any node.

The system statistics recorder 824 is a component that records error occurrence statistics of each node and provides processing information on nodes which frequently occur after recovery.

The error detector 825 is a component that detects error information such as whether the node is disconnected from the network or whether a system error occurs.

The recovery process in the present invention is based on a copy of the data and on-line reorganization processor.

The recovery processor 820 includes a process of performing recovery processing using on-line reorganization or on-line extension when a failure occurs in a short time and recoverable in a short time. In addition, when a failure occurs, data that was used as a replication node before failure occurs includes a process of being used as master data during the failure period.

The recovery processor 820 recovers an on-line state by reflecting the recovery log stored during the failure period after recovering the node again when a short-term error occurs.

If it is determined that the failure is not recoverable or the failure recovery time has passed the reference time, the idle node is included as an active node and the same data as the failed node is received from the replication nodes to provide a service. In this process, modules such as relation generation, deletion, movement, merging, and division of the on-line reorganization processor 830 are used. At this time, the data that was used as a replication node before the error occurred is used as the master data during the error period.

FIG. 10 is a diagram illustrating an operation flow of the recovery processor illustrated in FIG. 9.

Referring to FIG. 10, the operation flow of the recovery processor 820 will be described as follows.

In operation 1100, when an error occurs, the operation of the recovery processor 820 is started. In operation 1110, the error detector 824 detects the error information and delivers the error information to the active nodes included in the database cluster. In step 1120, the on-line reorganization processor 830 determines whether recovery is possible. If recovery is not possible, return a result. In operation 1130, the recovery processor 820 checks the data designated for the node in which the error occurred. In step 1140, the on-line reorganization processor 830 performs an operation of changing the role of the main data or the replica of the relation. In operation 1150, the recovery log recorder 823 records a recovery log in which the replica is converted into master data due to a failure, to each node. In operation 1160, the recovery processor 820 determines whether the recovery is completely performed. If the recovery was not completed, the result is returned. In step 1170, the recovery log recorder 823 reflects the recovery log stored in each node so far when the recovery is completely performed. In operation 1180, the recovery log recorder 823 returns a result. The recovery processor 820 ends the recovery process after the synchronization process is completed.

As described above, according to the present invention, when an existing database configuration significantly degrades service processing performance or causes a failure due to an increase in users or a change in a query pattern, on-line reorganization, on-line expansion, and recovery process By using replicated data to process, the database cluster server can be processed faster and with less resources than the existing server.

Also, in case of failure, on-line recovery is possible without interruption of service, and when the load of the server increases, the system can be expanded to an idle node to increase the throughput of the system and minimize user response time, thereby providing high performance, high availability, and high scalability. Will have

1 is a diagram illustrating a configuration environment of a database cluster server;

2 is a configuration diagram of a database cluster server according to an embodiment of the present invention;

3 is a flow diagram illustrating the operation of the system shown in FIG. 2;

4 is a block diagram of the on-line reorganization processor shown in FIG.

5 is a view illustrating an operation flow of the on-line reorganization processor shown in FIG. 4;

6 is a diagram illustrating an expansion process in a replication node according to an embodiment of the present invention;

7 is a diagram illustrating a parallel expansion process according to an embodiment of the present invention;

8 is a flowchart illustrating an operation of the on-line expansion processor shown in FIG. 2;

9 is a configuration diagram of the recovery processor shown in FIG.

FIG. 10 is a diagram illustrating an operation flow of the recovery processor illustrated in FIG. 9.

Claims (17)

For a database cluster server system: A client including a GUI-based query processing tool, an administrator program for managing a cluster system, and a monitoring tool for monitoring resources of each node, and transmitting a query to a basic query component through an interface library; A basic query processing component that analyzes the query received from the client and transmits the query to the database management component and provides the query processing result to the client; A database management component for determining an operation order, reorganization and expansion of a corresponding query, and a degree of allocating transaction resources according to an analysis result of the query statement; A global meta manager which provides information of data accessed by the query statement to the basic query processing component and manages to maintain the same meta information of all nodes at any time based on a consistency maintaining protocol; A central controller controlling server operation including starting, stopping, expanding, reducing, and changing settings of the server; And And a database storing predetermined data. The method of claim 1, The interface library A non-shared database cluster server system including ODBC, JDBC database communication protocols, ISAPI's Web-based CGI and CLI, and Winsock's underlying engine. The method of claim 1, The global meta manager It has a consistency structure that manages the meta information of all nodes to be the same at any time through the consistency protocol when each node changes some database schema. Manages information related to query processing such as information on active and idle nodes in a cluster, partition and replication information on relations, configuration information on master and replicas, and user information. A non-shared structured database cluster server system, characterized in that the meta information is unmodifiably managed in a special case in which the configuration of the database is temporarily changed, such as on-line reorganization, on-line expansion, and recovery processing. The method of claim 1, The central controller Based on the cluster node information registered in the meta information, it manages the control of system operation and database structure such as start, stop, expand, shrink and move nodes. In charge of message transmission and reception between the basic query processing component and the database management component, A non-shared database cluster server system including an interface for processing a database reconfiguration request, a node control request, a cluster node addition and deletion request, and a service start and stop request except general query processing. The method of claim 1, The basic query processing component When a query is input from the external connection processor, the query processor analyzes the query to obtain the information of the data accessed by the query from the global meta manager, and if the data included in the access area is not present in the current node, the internal network connection processor Forwards the query, returns the result back to the node that originally received the query, and delivers the result to the client when the query is completed, A non-shared database cluster server system including a storage manager that stores and manages processing logs, forwarding logs, recovery logs, reorganization logs, and extended logs of a cluster server in addition to general data and query processing logs. The method of claim 1, The database management component It consists of an on-line expansion processor, a recovery processor, and an on-line reorganization processor. And a reconstruction of the database, a node recovery, and an expansion process in association with the central controller, the global meta manager, and the basic query processing component. The method of claim 6, The on-line expansion processor If there is an idle node in the database cluster and a query is concentrated on a random node or a problem occurs that reaches a resource limit, real-time processing is required, and the idle node is included as an active node to split and move data to the corresponding idle node. Performing, If the update method of the database cluster uses a delayed update method, on-line expansion is performed on a replication node without master data, and data movement is reduced by storing newly created partitioned master data on the corresponding replication node. A non-shared database cluster server system characterized in that data movement is performed in parallel when an update method of a database cluster uses an immediate update method. The method of claim 6, The recovery processor System Statistics Recorder, Error Detector, System Recovery Processor, Data Recovery Processor, Recovery Log Recorder, and It includes modules for creating, deleting, moving, merging, and splitting relations of the on-line reorganization processor and extension modules to idle nodes of the on-line extension processor. In the event of a failure, recovery can be performed using on-line reorganization or on-line expansion, divided into recoverable and non-recoverable cases within a short period of time. When a failure occurs, data that was used as a replication node before the failure occurs is used as master data during the failure period. In the event of a short-term failure, the node is restored to the on-line state by reflecting the recovery log stored during the failure period, and when the failure is determined to be unrecoverable or the failure recovery time passes the reference time, the idle node is moved to the active node A non-shared database cluster server system, comprising: providing a service by receiving the same data as a failed node from a replica node after inclusion. The method of claim 6, The on-line reorganization processor A resource manager, a data processing manager, an optimization processor, a log transfer processor, and a data movement handler, And a non-shared structured database cluster server system in association with the global meta manager and the central controller to generate, delete, move, merge, and divide data in a node for reconstruction of the database. The method of claim 9, The resource manager On-line reorganization is determined by continuously monitoring the resources of each node in the cluster. The non-shared database cluster server system of claim 1, wherein when the resource usage of a specific node exceeds a threshold, information about the node is obtained from the data processing manager to determine whether to perform on-line reorganization processing and to call the optimization processor. The method of claim 9, The data processing manager It manages statistical information such as update area, query node, forwarding frequency, disk access frequency by query, The non-shared structure of the database cluster server system, characterized in that for obtaining the resource information of the node from the resource manager based on this information whether or not on-line reorganization processing. The method of claim 9, The optimization processor It performs functions such as creating, deleting, dividing, merging, and moving relations with the log transfer processor and the data movement processor for on-line reorganization processing. The data movement processor and the log transfer processor are responsible for performing the operations necessary for the on-line reorganization and log transmission for maintaining the integrity of data. When the update query occurs during the reorganization process, the log transfer processor compares the update area of the query with the data area transmitted so far, selects the updated area after transmission, and transmits the corresponding data to the target node. Database cluster server system in structure. The first step of placing the original data in the first node (N1), the source node, and the duplicate data of the original data in the second node (N2), the replication node: A second step of performing an expansion operation from Ta1 'to Ta2 in the second node; And And transmitting, by the second node, the same copy to Ta2 'of the third node, which is an idle node, using Ta2, which is the divided data of Ta1', as the original data of the corresponding node. How to scale up a cluster server system online. The method of claim 13, The replica node of the first step is a node that does not contain master data. The method according to claim 13 or 14, If on-line expansion reoccurs The on-line expansion processor selects the fourth node and the fifth node as idle nodes, and transmits Ta1 'of the second node, which is a divided copy, to the source node of the fourth node and the replica node of the fifth node, respectively. The fourth step of the; further comprising the online expansion method of the database cluster server system. In online expansion of a database cluster server system, A first step of placing duplicate data of the original data in the first node N1, which is a source node, and a second node N2, which is a duplicate node; The predetermined data of the first node is transmitted to the third node, which is an expansion node, and the predetermined data of the second node (where the predetermined data of the second node is different from the predetermined data of the first node). Data) transmitting to a fourth node which is an expansion node; And a third step of, when data transmission is completed in the second step, transmitting data of different parts previously transmitted from each expansion node to the other expansion node to generate a complete data. How to scale up a database cluster server system online. In the online expansion method of the database cluster server system, The on-line extension processor 810 is driven while receiving an extension event, and works in conjunction with the resource manager 831, the data processing manager 832, and the optimization processor 833 of the on-line reorganization processor 830. Generating new global meta information by searching for a target idle node (step 910); The on-line extension processor 810 locks meta information to prevent the schema of the database 900 from being changed (step 920); The on-line extension processor 810 transmits an on-line extension event and new meta information to all nodes included in the database cluster (steps 930 and 935); Preventing change of meta information of the target node (step 940); The nodes which have been identified to be extended through the meta information may include executing a receiver and an inserter process (step 945); After the target node prepares for expansion, transmitting a standby message to the source node (step 950); Starting source and log transmissions from the time point at which preparation for on-line expansion is completed (steps 960 and 965); In step 970, the expansion node maintains the integrity of the data by reflecting the log again after the data is reflected; Transmitting a data transmission completion message to the original node when the movement of data is completed (step 975); The source node sends a query input termination to the target node (step 980); Waiting for the source node and the target node to terminate the existing transaction (steps 985 and 987); Transmitting the end of all operations to the source node and the target node (steps 985 and 987); And sending the extension end message to the on-line extension processor (810) (step 993).