JPH0895841A - Data base processing system - Google Patents

Abstract

PURPOSE: To perform memory access for obtaining information required for simultaneous execution control by the same method irrespective of the number of nodes by enabling a server to make a request to refer to data in the same data base. CONSTITUTION: The nodes A and B are connected by a communication line 3 and a common memory device S which can be shared, is coupled with the respective nodes through a line 1. A common disk device H is coupled with the respective nodes by a common disk coupling means 2. Further, a node C is a computer node which functions as a data base client and connected to the nodes A and B by a line 4. Then a data is constructed on the common disk device H and the control information required for the simultaneous access control is written in the common memory device S from the server to enable the both to refer to them, thereby allowing the server to make a request to refer to the data in the same data base. Consequently, the memory access for obtaining the information required for the simultaneous execution control can be performed by the same method irrespective of the number of the nodes.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database processing system in a system composed of a plurality of computers each including a server.

[0002]

2. Description of the Related Art Conventionally, as a database processing method for simultaneously referring to one database from a plurality of computer sites each including a server, for example, "The ORACLE" of ORACLE company is used.
Paracrl Server-white pap
er (Part No. 3001003-0291) "
There is a system using a computer that uses memory devices each having a mutual communication function between nodes, such as a product called parallel server of ORACLE company described in. This allows the control information of the server of its own node written in each memory to be referred to from other nodes by using the memory communication function.

The reason why the control information of the server of another node is necessary will be explained in a little more detail. Generally, when referring to the same data from a plurality of servers or a plurality of users at the same time, it is necessary to avoid simultaneous writing and ensure data consistency. For this the server
A method called two-phase lock / two-phase commit (a typical commitment control method) is used.

In the two-phase lock, when a node updates data, it is checked whether the target data has already been locked in order to avoid simultaneous writing from another node.
If it is not locked, lock it and make it inaccessible to others before starting the update process.
The update process in the transaction does not write the change to the database every time the update process ends, but writes only the change content to the transaction log file. After all update processing is completed, the changed data is written to the disk using this transaction log file, the database is updated, and the lock is released. This is called commit. If the lock request is already issued but the lock request is not executed, the simultaneous update of the same data is avoided by waiting for the lock release or canceling the lock request.

Next, the commit is performed by the method of two-phase commit as described above. First, issue a commit request, confirm that no other commit is in progress, and then execute the commit. Wait for the end of the commit while others are committing. This avoids simultaneous writing to the database.

From the above, when performing update processing on the same database from a plurality of nodes, each individual node manages control information such as lock information and commit information in the entire system and mutually controls execution of lock / commit. There is a need to. ORACLE, which was mentioned earlier as a conventional example
In the case of the parallel server of the company, the control information of the other nodes necessary for this control is mutually notified by the communication function of the memory and is realized.

FIG. 12 shows, for example, January 20, 1994.
At the "First Oracle Open World" held on the 21st, the above-mentioned seminar about parallel servers by ORACLE "Solution Seminar / UNIX / Parallel Server"
YRAMID Valcncc Clustcr & O
It is a figure which showed an example of the system configuration which implement | achieves the parallel server of ORACLE company as it was described in "RACE7 parallel server seminar." The system has a memory device m that can be synchronized between the nodes and a disk device h that can be shared between the nodes on the nodes a and b including the servers. The shared disk device h is connected to the nodes a and b by a line 81, and the shared memory device m is connected to the nodes a and b by a line 82 such as a dedicated memory bus different from the line 81. The node c is a computer including a database client, and is connected to the nodes a and b by a communication line 4 such as Ethernet. The server of each node writes the necessary control information of its own node on the memory device m, and reads the control information of the other node by using the above-mentioned inter-memory communication function. Simultaneous execution control is performed by combining these two pieces of control information.

For example, when the lock is executed, the inter-memory communication function is used to notify the other node of the lock execution. In this way, each node obtains the lock information of other nodes,
It realizes lock control of its own node.

Further, by using the above inter-memory communication function,
Periodically notify other nodes of status information. Each node checks the status information sent from the other node to detect the failure of the other node. In addition, the node that detects the failure of another node by this function
The transaction log information of the failed node is referenced using the inter-memory communication function to perform database recovery processing.

Further, each node obtains data on the load condition and places it on the memory device m, and refers to it between the nodes using the above-mentioned inter-memory communication function to detect a low load server. As a result, the client can automatically connect to a node having a low load when issuing a connection request without being aware of the position of the node.

[0011]

In the conventional method using a memory device having a communication function, as the number of nodes increases, the number of memories each node has to communicate with increases. Then, the number of communication lines in each memory increases, which causes a problem that management becomes complicated. Also,
If a failure occurs in the connected node while the client is referencing data, and the server goes down without notifying the client of any abnormality, other nodes will not notify the client of the abnormality. So
The client itself continues to wait for a response from the server, and processing stops. It is possible to set a timeout process that terminates the program on the client side when there is no response from the server within a fixed time, but nothing can be done until the set fixed time elapses.

The present invention has been made in order to solve the above-mentioned problems, and even if the number of nodes increases, it is not necessary to change the memory access method, and the problem that the memory management becomes complicated by the addition of nodes. In addition to providing a database processing method that solves the problem, even when a server on that node shuts down without being able to notify the connected client due to a failure on another node, unlike other nodes, It is an object of the present invention to provide a database processing method having a server abnormality detecting means which is notified from any one of the above to a client that an abnormality of the server is being connected.

[0013]

According to a first aspect of the present invention, there is provided a database processing system comprising a computer node having a server, a shared disk device which can be shared by the node, and a shared memory device coupled to the node. A database is constructed on the shared disk device, control information necessary for simultaneous access control is written from the server to the shared memory device and cross-referenced, and a data reference request to the same database can be made from the server. .

A database processing system according to a second aspect comprises a computer node having a server, a shared disk device sharable by this node, and a shared memory device coupled to the node, and the node periodically. Write information about the operating status of its own node in the shared memory device, and refer to this information to determine whether there is a failure in another node. If it is determined that there is a failure, the node other than the failure has occurred. To recover the transaction that was being executed in the failed node.

A database processing method according to a third aspect of the present invention comprises a computer node having a server, a shared disk device which can be shared by this node, and a shared memory device which is coupled to the node, and is provided for a node in which a failure has occurred. If the server itself fails to notify the connected client that the server is down, the server other than the node in which the failure has occurred is referring to the information written in the shared memory device regarding the client connection and It detects a client, notifies the client of the occurrence of a server abnormality, and the client receives the notification and recognizes the server abnormality.

A database processing method according to a fourth aspect of the present invention comprises a computer node having a server, a shared disk device that can be shared by this node, and a shared memory device coupled to the node, and the shared memory device is provided on the shared memory device. Load information of the server is set, and when a connection request is sent from the client to the server, a low-load server is determined by referring to the load information on the shared memory device, and this server is automatically selected and connected. Is.

According to a fifth aspect of the present invention, there is provided a database processing system, comprising a computer node having a server, a shared disk device which can be shared by the node, and a shared memory device which is coupled to the node. While executing the data reference process, the load on the currently connected server will increase from the time the connection is opened, and the load on any other server will decrease. Judgment is made and the connection is automatically switched to this server at a predetermined timing.

[0018]

According to the database processing method of the first aspect, regardless of the number of nodes, the memory access for obtaining information necessary for concurrency control can be performed by the same method.

According to the database processing method of the second aspect, it is possible to detect an abnormality in each node and perform a recovery process from another node.

According to another aspect of the database processing method of the present invention, the client is notified of the abnormality, the client performs the abnormality processing, and detects the failed node without waiting for a certain period of time to start the abnormality processing.

According to the database processing method of the fourth aspect, load distribution can be achieved by automatic selection and connection switching of low-load servers.

According to the database processing method of the fifth aspect, the server of the node having a low load is always detected without regard for the position of the connection destination node, and the server is automatically connected and processed.

[0023]

【Example】

Example 1. FIG. 1 is a schematic diagram showing an example of a system configuration according to a first embodiment of the present invention. In this case, 3 of A, B and C are shown.
It consists of two nodes. Node A, B is communication line 3
The shared memory device S, which is connected to each other and is sharable, is connected to each node through the line 1. Shared disk device H
Are connected to each node by the shared disk connecting means 2. Further, it is possible to configure a database on the shared disk device H and use it as a database that allows servers on each node to simultaneously refer to data. The node C is a computer node having a function as a database client, and is connected to the nodes A and B by the line 4.

FIG. 2 is a diagram showing an example of the internal configuration of the nodes A and B. One node includes a CPU 6, a memory 7, and an input / output control device 8 that controls input / output to / from the shared disk device H. A line 3 management unit 9 that manages the line 3 and a shared memory device S management unit 10 that manages the line 1 are included. The server 5 exists on the memory 7 as a server process. The server 5 is configured as shown in FIG.
There are roughly a database management unit 11 and a database communication control unit 12, and the database management unit 11 controls database processing performed on its own node such as general question processing and database reference control. Further, the database management unit 11 includes a control information management unit 13 and a data information management unit 14.
The control information management unit 13 further includes, as part of its functions, a lock request processing function 15, a node operation state transmission function 16, a recovery function 17, an abnormality detection function 18,
It is composed of various control functions including exchange of control information on the shared memory, such as a client abnormality notification function 19 and a load detection function 20. The database communication control unit 12 performs lock management in a portion that manages communication with the outside of the node, and according to a request from the database management unit or the like, the input / output control device 8, the line 3 management unit 9, and the shared memory device management unit 10
Interact with.

In general, when a database management system performs an update process, the unit of a table or a record
Lock the target object so that no one else can access it, and then perform the update process. This lock is released after the update process is completed and the change writing (commit) to the disk is completed. With this function, data consistency can be maintained even if data update requests are issued from multiple servers. FIG. 4 shows the node A as an example.
When the update processing request for the record x is issued from the node B before the update processing for the record x has been completed, each update request is processed. In step 21, node A first issues a lock request, and in step 22, it is confirmed whether or not it is already locked. If it is not locked, it is locked in step 23, the update process of step 24 is executed, and the lock is released in the commit process of step 25.
Period from lock execution to commit end (Step 23
25 to 25), even if a lock request for the record x is issued from another node, the lock cannot be performed. Even in this case, the lock request in step 26 is determined to be already locked to the node A in step 23 by checking the presence / absence of lock in step 27, and the lock release waits in step 28. After the end of the commit of the node A in step 25, the lock of the node B is executed (step 29), the update process (step 30) is all finished, and the lock is released after the end of the commit (step 31).

Each lock request made at this time is processed by the procedure shown in FIG. Node B in FIG.
The case of the lock request issued by will be described. First, in step 32, the node B tries to update the record x
However, the lock request is rejected because it is already locked by the node A in step 33 by checking whether or not there is a lock from another node.
Then, the lock request of the node B is entered in the lock waiting queue of the record x. Further, in step 35, the number of entries is incremented by one each time the queue is entered. In addition, as the processing of the lock request for the data already locked by others, as in step 34, the method of putting the lock queue in the lock wait queue and waiting for the lock release before executing the lock, and the method of canceling the lock request. There is. In FIG. 5, the method of putting in the lock waiting queue is adopted. When the previously executed lock of the node A is released in step 37, it is checked in step 38 whether or not there is an entry in the lock wait queue for the data x, and in step 39, the highest priority is given to the queue prioritized by some algorithm. Perform high priority lock requests. Since one lock in the queue has been executed, the number of queue entries is decremented by one in step 40. If the data requested by the node B in step 32 for locking is not locked, step 3
The lock execution of No. 6 is performed, and information such as a node identifier, an object identifier, and a record identifier is written on the shared memory device S as lock information. This information is erased at step 37 when the commit is performed.

According to this embodiment, by placing the control information of the server on the shared memory device S, the memory access for obtaining the information necessary for the simultaneous execution control can be performed by the same method regardless of the number of nodes. , Easy to manage.

Embodiment 2 FIG. FIG. 6 is an example of a process performed by each node in the second embodiment of the present invention to detect the occurrence of an abnormality in the server of another node. The nodes A and B sequentially perform the next operation (step 4) at regular intervals.
1, 42 and 43, 44). First, data regarding the operating state of the server of its own node is acquired and written in the shared memory device S as operating state information. As a method of acquiring this data, for example, a daemon process that monitors the server process communicates with the server process, determines the response result from some predetermined operating states, and determines, for example, in FIG. The shared memory device S is written in the format as shown. In this case, the time at the time of writing is added in addition to the node identifier and the operating state, and the information written previously is overwritten. There are three operating states: "server stopped", "normal operation", and "abnormality". Next, by the same procedure, the operating state information of the server of the other node which has already been written is read and it is confirmed whether or not it is operating normally. As a result of the operation state check in step 44, for example, when it is determined that the value of the operation state is'abnormal occurrence 'as shown in FIG. 7 or the operation state information has not been updated from the time of the previous reading, an abnormality occurs. Therefore, the server abnormality of the node A is detected.

For example, if a failure occurs in the node A in step 45 and a server abnormality occurs, and the result of the determination of the operating state of the node A performed by the node B in step 44 is abnormal, the node B determines that the node B is a node in step 46. Upon detecting the abnormality of A, the node B starts the recovery process of the process executed by the node A in step 47. When each server performs an update process in advance, the updated content is written in the shared memory device S as transaction log information each time one update process is completed, and is retained until the commit is completed. Recovery processing can be performed using this transaction log.
When the update process is committed, the changed contents are written to the disk and the transaction log information on the shared memory device S is cleared. The transaction log information includes a node identifier, a record identifier, update contents, time and the like. FIG. 8 shows an operation example of a recovery process performed by a server of a node that has detected an abnormality of another server for the data reference process performed by the node that has caused the abnormality. First, when the server is in the process of updating and an error occurs in a state where it is not committed, the nodes other than the faulty node start the rollback process of step 48, that is, the recovery process of canceling the update and unlocking the target record.
As its contents, in order to confirm the presence or absence of uncommitted update processing at the time of failure at step 49, the transaction log information of each node on the shared memory device H is read, and at step 50 the transaction log of the node in which an error has occurred is read. Check for any. Then, if the corresponding transaction log remains, the corresponding transaction log information is cleared in step 51, the update record is unlocked in step 52, and the recovery in step 53 ends. Information necessary for unlocking in step 52 is referred to the lock information on the shared memory device H described in the first embodiment.

According to this embodiment, the reliability is improved by the abnormality detection between the nodes and the recovery processing from the other nodes.

Example 3. FIG. 9 is a flow chart showing the operation of the third embodiment of the present invention. When each server opens a connection with a client in advance, it is assumed that the server identifier, the client identifier, and the like are written in the shared memory device H as the connection client information. It also clears this when the connection is released.
First, in steps 54 to 57, the client of the node C connects to the server of the node A and executes the application that refers to the data. At this time, it is assumed that a failure occurs in the node A in step 58 and the server goes down without notifying the client of anything. Here, the node B detects the server down due to the failure of the node A in step 59, detects the connected client, and notifies the client of the abnormal occurrence of the server A. This confirms the presence or absence of the above-mentioned connected client information issued by the node A from the shared memory device H, and if there is, detects the client identifier. Here, it detects that the node C is a client and notifies the node C that the connected server is down. Further, the client receiving this notification in step 60 detects an abnormality and abnormally terminates the application being executed. This process
For example, if a daemon process other than the client process is prepared on the client side, an abnormal notification message from the node B can be recognized, and when the message is received, abnormal processing such as abnormal termination of the client process can be performed. realizable.

According to this embodiment, the client is notified of the abnormality, the client performs the abnormality processing, and the failure node is quickly detected and the abnormality processing is performed without waiting for the elapse of a fixed time such as timeout. You can start.

Example 4. FIG. 10 shows a fourth embodiment of the present invention.
5 is a flowchart showing an operation example of FIG. When APP is started from the client of the node C in step 61, the client first determines a connection destination server, so in step 62, a message is sent to one of the servers (here, it is assumed to be node A), and in step 63, the load is the highest. Detect and notify the server with low power. In step 64, a connection request is sent to the notified server in accordance with the notification result (here, node A), and after the connection in step 65 is completed, the data reference process in step 66 is started. As a method of realizing this, there is a method of using a daemon process that manages connection with a server, in addition to a client process that executes each application on the client. This will be described in more detail. In this method, step 62
Then, the daemon process of the node C receives the connection request from the client process to the server (the node does not specify the node on the client process side), and the server process detects and notifies the server with the lowest load. Send a message requesting execution of. It should be noted that which of the nodes A and B the client daemon process sends a message to is determined beforehand by default settings or the like, and according to this, the node A is used here. Next, in step 63, the server process of the node A determines the server with the lowest load using the load detection function 20 of FIG. 3 (here, it is assumed to be the node A), and notifies the result to the daemon process of the client. The daemon process that receives this notification notifies the client process that issued the connection request of the node (node A here) with the lowest load. The client process receiving this in step 64 issues a connection request by designating the server of that node, and when the connection in step 65 is completed, the data reference process given by the application in step 66 is started.

The load detecting function of each server obtains the load information of the server of its own node and writes it in the shared memory device S, or refers to the load information of another node already written in the same way, According to the standard, each node periodically performs the three processes of determining the server with the lowest load and notifying the result to the daemon process on the client.

For load detection performed by each server, the load information is, for example, the congestion degree of the lines 1 and 2 of each node, C
Information such as the PU utilization rate, the memory free area, and the number of connected clients is acquired, some determination formula using these is prepared, and the server with the lowest load is determined from the calculation result.

According to this embodiment, the load can be distributed and the processing capacity can be improved by automatically selecting and connecting the low load server.

Example 5. FIG. 11 shows an operation example of the fifth embodiment of the present invention. The client of node C is
It is assumed that the function of the fourth embodiment is used to connect to the server of the node A having the lowest load at that time to perform data reference (steps 67 to 70). Then nodes A and B
, The node B detected the low-load node in step 71, and determined that the node B had a lower load than the node A, and notified the client of the node C.
On the client side that has received the notification, for example, a daemon process that performs connection management different from the client process receives this, and further notifies the client process that node B currently has a lower load. When the notified client process switches the connection server, the data reference process of step 72 is interrupted at an appropriate timing set in advance such as a transaction break, and the connection with node B is further released in step 73. It issues a request, waits for the disconnection with node A to end in step 74, issues a connection request to node B in step 75, connects in step 76, and then suspends the data reference processing suspended in step 77. Resume. In this example, no reversal of the load status is detected or notified from the restart of the process of step 77 to the end of the process of step 78, so the connection with the node B is released in steps 79 and 80, and then in step 81. APP has ended. However, if the reverse of the load condition is detected before the end of the data reference processing of steps 77 and 78, the same processing as steps 72 to 77 is repeated.

According to this embodiment, it is possible to always detect a server of a node having a low load and automatically connect it to perform processing without paying attention to the position of the connection destination node.

[0039]

According to the first aspect of the present invention, there is provided a database processing system comprising: a computer node having a server; a shared disk device sharable by this node; and a shared memory device coupled to the node. Since a database is constructed on the above, control information necessary for simultaneous access control is written from the server to the shared memory device and cross-referenced, and a data reference request to the same database can be made from the server, the number of nodes Regardless of this, memory access for obtaining information necessary for concurrency control can be performed by the same method, so management becomes easy.

A database processing method according to a second aspect of the present invention comprises a computer node having a server, a shared disk device that can be shared by this node, and a shared memory device coupled to the node, and the node periodically. Write information about the operating status of its own node in the shared memory device, and refer to this information to determine whether there is a failure in another node. If it is determined that there is a failure, the node other than the failure has occurred. Therefore, the recovery processing of the transaction executed in the failed node is performed, so that the reliability is improved by the abnormality detection between the nodes and the recovery processing from the other nodes.

A database processing method according to a third aspect of the present invention comprises a computer node having a server, a shared disk device which can be shared by this node, and a shared memory device which is coupled to the node. If the server itself fails to notify the connected client that the server is down, the server other than the node in which the failure has occurred is referring to the information written in the shared memory device regarding the client connection and Since the client is detected and the server is notified of the abnormality of the server, and the client receives the notification and recognizes the abnormality of the server, the client is notified of the abnormality, the client performs the abnormality processing, and the fixed time elapses. It is possible to detect the faulty node and start the abnormality processing without waiting for.

A database processing method according to a fourth aspect comprises a computer node having a server, a shared disk device which can be shared by this node, and a shared memory device coupled to the node, and the shared memory device is provided on the shared memory device. A configuration in which the load information of the server is set, and when a connection request is sent from the client to the server, the load information on the shared memory device is referred to, a low load server is determined, and this server is automatically selected and connected. Therefore, it is possible to improve load and load balancing by automatically selecting low load servers and switching connections.

A database processing system according to a fifth aspect comprises a computer node having a server, a shared disk device which can be shared by this node, and a shared memory device which is coupled to the node, and is connected to the server from the client. While executing the data reference process, the load on the currently connected server will increase from the time the connection is opened, and the load on any other server will decrease. Since it is configured to automatically judge the connection and switch the connection to this server at a specified timing, the server of the node with a low load is always detected without regard to the position of the connection destination node, and this is automatically detected. It can be connected and processed.

[Brief description of drawings]

FIG. 1 is a schematic diagram showing an overall configuration of a database system adopting a database processing method according to a first embodiment of the present invention.

FIG. 2 is a schematic diagram showing a configuration example of a single node A and a single node B according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a configuration of a database server process according to the first embodiment of the present invention.

FIG. 4 is a flowchart showing the operation of lock request processing performed in the operation of the first embodiment of the present invention.

FIG. 5 is a flowchart showing an operation of a lock request process performed by the operation of the first embodiment of the present invention.

FIG. 6 is a flowchart showing an operation of server abnormality detection processing according to the second embodiment of the present invention.

FIG. 7 is a diagram showing a configuration example of a message written on the shared memory device as status information of each server in the operation of the second embodiment of the present invention.

FIG. 8 is a flowchart showing an operation of recovery processing according to the second embodiment of the present invention.

FIG. 9 is a flowchart showing the operation of the third embodiment of the present invention.

FIG. 10 is a flow chart showing the operation of the fourth embodiment of the present invention.

FIG. 11 is a flowchart showing the operation of the fifth embodiment of the present invention.

FIG. 12 is a schematic diagram showing the entire configuration of a conventional system using a computer in which each memory has a synchronization function.

[Explanation of symbols]

1-4 lines, 5 database buses, 6 CPUs, 7
Memory, 8 input / output control device, 9 line management unit, 10
Shared memory device management unit, 11 database management unit,
12 database communication control unit, 15 control information management unit, 14 data reference management unit, 15 lock request processing function, 16 node status transmission function, 17 recovery processing function, 18 abnormality detection function, 19 client abnormality notification function, 20 load detection function .

Claims (5)

[Claims] 1. A computer node having a database server (hereinafter, referred to as a server), a shared disk device that can be shared by this node, and a shared memory device coupled to the node are provided on the shared disk device. A database processing method characterized in that a database is constructed, control information necessary for simultaneous access control is written from the server to the shared memory device and cross-referenced, and a data reference request to the same database can be made from the server. 2. A computer node having a server, a shared disk device sharable by this node, and a shared memory device coupled to the node, wherein the node regularly provides information on the operating state of its own node. On the shared memory device, the node refers to this information to determine whether or not there is a failure in another node, and if it is determined that there is a failure, the node other than the failure node will A database processing method characterized by performing recovery processing of the transaction that was executed in. 3. A computer node having a server, a shared disk device sharable by this node, and a shared memory device coupled to the node, wherein the server itself of the failed node is connected to If the server cannot be notified to the client, the server other than the faulty node refers to the information written in the shared memory device regarding the client connection, detects the connected client, and notifies the client. A database processing method characterized in that a server abnormality is notified, and the client receives the notification and recognizes the server abnormality. 4. A computer node having a server, a shared disk device sharable by this node, and a shared memory device coupled to the node, wherein load information of the server is placed on the shared memory device. When a connection request is sent from the client to the server, a low load server is determined by referring to the load information on the shared memory device, and this server is automatically selected and connected. . 5. A computer node having a server, a shared disk device sharable by this node, and a shared memory device coupled to the node, wherein a client connects to a server to execute a data reference process. , The load of the currently connected server increases from the time when the connection is opened, and conversely the load of one of the other servers decreases, the server in the appropriate load state is determined according to the change in the load state, and the determined timing A database processing method characterized by automatically switching the connection to this server.