KR100363523B1 - Method for controlling distributed processing in cluster severs - Google Patents

Abstract

The present invention allows each server to automatically determine one main server for distributed processing of Internet traffic, and to monitor the occurrence of errors in a specific server through mutual monitoring between each server, thereby enabling efficient distributed processing of traffic and cluster operation. The present invention relates to a distributed processing control scheme in a clustering server to realize stability. To this end, the present invention relates to the aforementioned conventional method for monitoring and distributedly controlling the activity of all slave servers participating in a clustering. Unlike the master server, all servers broadcast a status check packet at regular intervals to indicate that they are operating normally on the network. Each server by checking the operation of other servers It monitors each other's operation status and decides which server among the multiple servers is the temporary master server for error recovery control according to the set protocol when an error occurs in any server. By controlling, it is possible to realize efficient distributed processing, error recovery, and stability of cluster operation in a clustering server environment.

Description

Distributed processing control method in clustering server {METHOD FOR CONTROLLING DISTRIBUTED PROCESSING IN CLUSTER SEVERS}

The present invention relates to a method for controlling a clustering server in a distributed processing environment, and more particularly, a clustering server suitable for controlling distributed processing and failure response processing among a plurality of clustering servers in a distributed processing environment of Internet traffic. It relates to a distributed process control method in.

In recent years, as the bandwidth of the Internet increases and the number of users of the Internet increases exponentially, the demand for distributed processing of Internet servers, especially web servers, is increasing. By performing distributed processing, a server control method has been proposed that efficiently processes a large number of user requests and ensures stability by causing another server to take over when a specific server to which it belongs malfunctions.

In particular, the biggest issue in server clusters where multiple servers (eg computers) share the work and handle them is when one computer fails (i.e. the server dies or the service process dies). In other words, the failure should be discovered by other computers and set up the packet corresponding to the dead server to be accepted by the rest of the servers in operation.

To do this, how each server will find out if other servers are currently up or down, and how to deal with the situation, the configuration must be applied to each server at the same time (i.e. configuration mismatch between each server). In this case, the setting must be applied at the same time since the operation is not performed properly.) At this time, it is necessary to decide how to synchronize the setting time.

Accordingly, in view of the above, conventionally, a method of distributing Internet traffic by selecting one of a plurality of servers as a master server and selecting other servers as a slave server has been proposed and actually applied.

4 is a configuration diagram of a conventional clustering server for applying a method of controlling distributed processing in a clustering server according to a conventional method, and includes one master server 402 and a plurality of slave servers 404/1 to 404 / connected thereto. n).

Referring to FIG. 4, the master server 402 performs a function of distributing and supervising information among respective slave servers connected to the master server 402. It performs a function of distributing to one slave server among a plurality of managed slave servers 404/1 to 404 / n. That is, the master server 402 manages and controls work distribution for each slave server while checking the load state of each slave server 404/1-404 / n.

In addition, the master server 402 performs a function of frequently checking whether each slave server 404/1 to 404 / n managed by the master is currently operating normally, and for this purpose, a network request message to each slave server. Each slave server 404/1-404 / n returns a response message to the network request message to the master server 402 within a given time.

Therefore, the master server 402 checks whether or not each slave server has failed based on whether a response message from each slave server 404/1-404n is received. That is, if a response message does not arrive from any slave server within a given time period, the master server 402 considers that the slave server has failed (or has an error). By distributing to the remaining slave servers and controlling them, the stability of data processing is guaranteed.

That is, in the conventional method, one selected master server monitors the activity of a plurality of slave servers managed by itself (i.e., checks the current state of each slave server by using information to query and respond to each slave server). When a problem occurs (i.e., a failure of a specific slave server), a method of coping with a failure situation (distributing work of a failed slave server to the remaining slave servers in normal operation) is used.

However, as described above, the conventional control method of performing distributed processing of Internet traffic using a pre-selected master-slave structure among servers has the following problems.

First, when a selected master server manages multiple slave servers, if a master server fails, there is no object that can recognize it in real time. The process of doing this is very complicated and also takes a long time.

Second, if one server fails, if the timing of the reset does not match when the other servers reset, some of the range of Internet traffic may overlap with each other, resulting in a malfunction, thus synchronizing the timing of the reset (ie It is imperative that the servers have a batch execution environment, and it is difficult to match the timing of operation because one master server manages all servers directly, that is, the master server communicates with each slave server. Since the time factor is necessary during the process, it is difficult to apply to server clustering, which is an environment that changes in real time.

Third, while the master server checks for errors and recovers from errors, if the failed server is automatically recovered (for example, a bad contact on a network line), the processing is complicated, i.e., a short contact. There is a problem that it is complicated and inefficient to exclude a server that is recognized as bad and recognize a server, distribute the work to the remaining servers, and process the error server when it is automatically restored to normal.

Accordingly, the present invention is to solve the above problems of the prior art, each server automatically determines one main server for distributed processing of Internet traffic, and the error in a specific server through mutual monitoring between each server It is an object of the present invention to provide a distributed processing control method in a clustering server capable of realizing efficient distributed processing of traffic and stability of cluster operation by monitoring occurrence.

In order to achieve the above object, the present invention provides a method for controlling distributed processing and error monitoring and recovery between servers in a clustering server environment having a plurality of subs interconnected through a network. A first step of continuously broadcasting a status check packet carrying the current status on the network at a predetermined time interval; A second step of checking each server's current operation state except for itself by receiving and analyzing the status check packet; A third step of automatically determining any one of the remaining servers currently operating as a temporary master server according to a predetermined protocol when an error is detected in one of the plurality of servers; A fourth step of the temporary master server generating a configuration change packet including the start time information of the configuration change and broadcasting it on the network; And a fifth process in which the temporary master server and the remaining servers currently in operation perform a synchronized configuration change based on configuration change start time information included in the configuration change packet and state information of each server. Provide a process control method.

1 is a configuration diagram of a clustering server suitable for applying a method for controlling distributed processing in a clustering server according to the present invention;

2 is a flowchart illustrating a process of controlling distributed processing in a clustering server according to a preferred embodiment of the present invention;

3 is a packet structure diagram illustrating an example of a status check or configuration change packet structure broadcasted by each server;

4 is a configuration diagram of a conventional clustering server for applying a method of controlling distributed processing in a clustering server according to a conventional method.

<Description of the code | symbol about the principal part of drawing>

102: user 104/1-1-4 / 3: server

The above and other objects and various advantages of the present invention will become more apparent from the preferred embodiments of the present invention described below with reference to the accompanying drawings by those skilled in the art.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

First, a key technical aspect of the present invention is that unlike the aforementioned conventional method of selecting one of a plurality of servers as a master server and selecting one master server to monitor and distribute the activity of all slave servers, When not selected, all servers have a certain amount of time (e.g. 1 second, 2 seconds, etc.) to indicate that they are alive (i.e. performing normal operations) on the network (i.e. health check packets). At the interval, each server analyzes each broadcast packet and checks the operation status of each server other than itself (normal operation or error). In the event of a failure, any one of the multiple servers (for example, the currently running server with the smallest IP address) acts as a temporary master server to recover from the failure. (That is, an error is generated in the distribution server operations to other sub) that that controls the setting change for, by these technical means can easily attain the bar for the purpose of the present invention.

1 is a block diagram of a clustering server suitable for applying a method for controlling distributed processing in a clustering server according to the present invention.

Referring to FIG. 1, the clustering server is composed of a plurality of servers interconnected through a network, for example, three servers 104/1, 104/2, 104/3, wherein the master server is preselected. Unlike the conventional method described above, here the master server is not preselected and only one of the multiple servers (e.g., the server currently running at the time of a special situation (i.e. an error on any server)) occurs. Servers with small IP addresses) are automatically determined as temporary master servers to control configuration changes, synchronization, and failover, and these temporary master servers are variable depending on the failure situation, that is, if a failure occurs at the temporary master server, The server with the smallest IP address among the running servers is again determined as the new temporary master server.

Thus, each server (104/1, 104/2, 104/3) broadcasts its own status check packet over the network, and when an error occurs in any server and a particular server is determined to be a temporary master server, each server is temporary. The configuration change is performed while synchronizing with each other based on the configuration change packet broadcast from the master server.

At this time, the status check packet or the setting change packet broadcasted by each server on the network is, for example, as shown in FIG. 3, the transmission cluster number information 302, the transmission IP address information 304, and the reception cluster number information 306. And control data information 308, and the configuration change packet broadcasted by the temporary master server to the network for error recovery includes time information on a configuration change start time. Here, the control data information 308 area in the status check or setting change packet is inserted with status information (operation check information or setting change information) of the corresponding server and information related to execution of other applications.

On the other hand, in the status check packet shown as an example in FIG. 3, the transmitting and receiving cluster numbers (i.e., virtual IP addresses) are separately configured, but not necessarily configured as such, and are not necessarily configured according to the situation or structure of the cluster. It can also be configured to have only an entry for the receiving cluster number.

In addition, in the present invention, when the power of the clustering server is set, any one of a plurality of participating servers is automatically determined as a response server responding to the information request based on the user IP or the port address TCP / IP that requested the information. That is, as shown in FIG. 1, the information request packet from the user 102 is transmitted to all servers participating in the clustering process through the network, that is, server 1 (104/1), server 2 (104/2), and server 3. (104/3) to all, for example, if Server 3 (104/3) is determined to be the answering server, then Server 1 (104/1) and Server 2 (104/2) respond to the request for information. Only the server 3 (104/3) will respond to the request for information and will send the user the requested information.

Next, a process of controlling distributed processing according to the present invention using the clustering server having the above-described configuration will be described.

2 is a flowchart illustrating a process of controlling distributed processing in a clustering server according to a preferred embodiment of the present invention.

Referring to FIG. 2, while the clustering server is set to power and performing an operation mode, each server generates status check information related to its current state and broadcasts on a network at a predetermined time interval, for example, one second interval. And each server monitors the current state (ie, operating state or error state) for all servers except itself based on the broadcasted status check packet (step 202). At this time, the broadcast status check packet includes, as an example, the transmission cluster number information 302, the transmission IP address information 304, the reception cluster number information 306, and the control data information 308 as shown in FIG. 3. It may include.

That is, each server checks a case where a status check packet is not broadcasted from an arbitrary server and detects an error occurrence of the corresponding server (step 204). As a result of the check here, a specific server, for example, server 2 ( If it is determined that an error has occurred (i.e., unbroadcasting of the acknowledgment packet) at 104/2), the other servers currently operating normally, that is, server 1 (104/1) and server 3 (104/3), return to zero. The count value C, which is in the cleared state, is increased by one (step 206), and then it is checked whether the current count value C has reached the preset count value n (step 208), and the check result shows that the current count value C is the preset count. If it is determined that the value n has not been reached, the process returns to step 202 described above and repeats the subsequent steps.

Here, the preset count value n can be set to, for example, 5, but setting the preset count value n to 5 indicates whether any server has not broadcasted the status check packet five times in a row. This means that if the status check packet is not delivered properly according to the traffic situation, or if the status check packet is not broadcasted once or twice due to bad contact, etc., it is not judged as an error occurrence. For that.

In other words, when one or two status check packets are not broadcasted or receive failures, all of them are regarded as an error occurrence and the error recovery process is performed. This is to prevent the work efficiency from being lowered.

On the other hand, as a result of the check in step 208, it is determined that the current count value C has reached the preset count value n. For example, the server 2 104/2 has not broadcasted the status check packet five times in a row. If it is determined that the server 1 (104/1) and the server 3 (104/3) determines that an error has occurred in the server 2 (104/2), to recover the error occurrence in the server 2 (104/2) Either Server 1 (104/1) or Server 3 (104/3) is responsible for performing configuration change control, which determines the temporary master server to oversee error recovery, e.g. the smallest IP The server with the address is determined to be a temporary master server (e.g., server 1 104/1) (step 210).

Therefore, the server 1 (104/1) determined as the temporary master server generates a configuration change packet including time information at the start of the configuration change (that is, time information indicating that the configuration change process starts after a few seconds), and then on the network. Broadcast (step 212), decrease the preset setting change control count value C by one (step 214), and then check whether the current setting change control count value has reached zero (0) ( In step 216, if it is determined that the current setting change control count value C has not reached the zero value, the process returns to step 212 described above and repeats the subsequent steps.

Here, the preset setting change control count value C may be set to, for example, 5, but setting the setting change control count value C to 5 means setting for setting change for error recovery in a server where an error has occurred. This means that the change control signal is broadcast five times in succession at a certain time interval (for example, 1 second, etc.), which means that other servers other than the temporary master server currently fail to receive the configuration change packet. To prepare for. That is, broadcasting the configuration change packet n times at regular time intervals is to more reliably perform the configuration change for error recovery.

On the other hand, if it is determined that the setting change control counter value C is zero as a result of the check in the above step 216, each server currently operating receives and stores the setting change time start time information contained in the setting change packet and previously stored. Synchronized configuration change is performed using the status information of each server. At this time, the configuration change is performed while finishing the temporary master server or synchronization (step 218).

Therefore, when the configuration change is completed, the temporary master server controls the error server to appropriately distribute and process the job of the server in error according to the job status, thereby recovering from the error. At this time, both the count value C incremented by each server and C counted by the temporary master server are reset to their original state (zero value and 5 value).

Meanwhile, in the above-described preferred embodiment, an error occurs in an arbitrary server participating in clustering as an example. However, even if an error occurs in a server determined as a temporary master server instead of an arbitrary server, the same process as described above is performed. Of course, you can perform error recovery on the server (temporary master server).

On the other hand, in the preferred embodiment of the present invention, the reception check of the status check packet is continuously performed five times at an interval of one second, and when the error occurs, the description has been made by broadcasting the setting change packet five times in succession. The present invention is not limited thereto, and a person of ordinary skill in the art may fully understand that it can be changed according to traffic conditions (or packet loss probability).

For example, in a network environment where faster failure detection is required, the reception check time can be reduced to less than 1 second. In a network environment where packet loss occurs relatively frequently, the number of status check packets and configuration change packets can be reduced. By increasing the number, the adaptability of the network environment can be more secured or expanded.

As described above, according to the present invention, unlike the above-described conventional method of monitoring and distributedly controlling the activity of all slave servers participating in clustering, the selected master server is not all the servers in the situation where the master server is not selected. Broadcasts a status check packet at regular intervals to indicate that it is operating normally on the network, and checks the operating status (normal operation or error) of other servers other than itself by analyzing each broadcast packet. Checks the operating status of each server in the same way, and if an error occurs in any server, it sets up for error recovery by determining one server among the multiple servers as a temporary master server for error recovery control according to the set protocol. By controlling change and error recovery, you It is possible to realize an efficient distributed processing, reliability of failover and cluster management traffic.

In addition, the distributed processing control method in the clustering server according to the present invention detects a predetermined n times of status check packet failure as an error occurrence and performs an error recovery process. It is possible to effectively suppress deterioration of efficiency, that is, to achieve distributed processing and control with environmental adaptability.

Claims (7)

A method for controlling distributed processing and error monitoring and recovery between servers in a clustering server environment having a plurality of subs interconnected through a network, A first step in which each of the plurality of servers participating in clustering continuously broadcasts a status check packet carrying its current status on the network at predetermined time intervals; A second step of checking each server's current operation state except for itself by receiving and analyzing the status check packet; A third step of automatically determining any one of the remaining servers currently operating as a temporary master server according to a predetermined protocol when an error is detected in one of the plurality of servers; A fourth step of the temporary master server generating a configuration change packet including the start time information of the configuration change and broadcasting it on the network; And Distributed processing in the clustering server comprising a fifth process of performing the configuration change synchronized with the temporary master server and the remaining servers currently operating based on the configuration change start time information included in the configuration change packet and state information of each server. Control method. 2. The clustering server of claim 1, wherein the method automatically determines a server having the smallest IP address among the remaining servers currently operating as the temporary master server when an error is detected at any server. Distributed processing control method. The method of claim 1 or 2, wherein the second process is: A twenty-first step of, if the status check packet for any server is not received, increasing the current count value by one; A twenty-second step of checking whether the current count value reaches a preset count value; A twenty-third step of repeating the first step if it is determined that the current count value does not reach the preset count value; And And a twenty-fourth step proceeding to the third step when the current count value reaches the preset count value. The method of claim 3, wherein the broadcast time interval of the status check packet is changeable according to a traffic environment of the network. 4. The distributed processing control method according to claim 3, wherein the check of the number of receptions of the status check packet is changeable according to a relative packet loss probability of the network environment. The method of claim 1 or 2, wherein the fourth process comprises: A 41st step of generating and broadcasting the configuration change packet and then reducing the predetermined configuration change control count value by one; A 42nd step of checking whether the reduced setting change control count value has reached a zero value; If it is determined that the reduced setting change control count value does not reach the zero value, step 43 of repeating steps 41 and 42; And And a forty-fourth step of proceeding to the fifth step when the reduced setting change control count value reaches the zero value. 7. The distributed processing control method according to claim 6, wherein the number of broadcasts of the configuration change packet can be changed according to a relative packet loss probability of the network environment.