KR100753565B1 - High availability system and his task devide takeover method - Google Patents

Abstract

The present invention relates to a high availability system and a task partitioning method thereof, wherein when a task event occurs for a node, the task is partitioned and handed over according to the node's resource utilization rate, so as to achieve high system availability. This is to minimize the performance degradation of the service for the request. To this end, the present invention consists of 2 to N nodes (server systems), and in a high availability system for monitoring and recovering from a failure of a service process, the node includes an HA master for activating a manager for HA operation. ; A node manager which collects its own state information and multicasts it to another node, determines necessary tasks based on the collected information of the information manager, transmits them to the task manager, monitors the task manager, and re-executes in case of failure; A task manager based on the information transmitted from the node manager, monitoring the node manager, the load manager, and the information manager to re-execute in the event of a failure to activate the node manager, the load manager, and the information manager; An information manager which collects and manages the multicasted information from other nodes through the heartbeat, and transmits the information to the node manager and the load manager; It calculates its own resource usage rate, and configures the load manager that ranks the resource utilization rate of other nodes received from the information manager with its own.

Description

HIGH AVAILABILITY SYSTEM AND HIS TASK DEVIDE TAKEOVER METHOD}

1 is a schematic diagram illustrating a configuration of a node of a conventional high availability system.

Figure 2 is a schematic diagram showing the configuration of a typical high availability system.

Figure 3 is a schematic diagram showing the configuration of a node of the high availability system of the present invention.

4 is a schematic diagram showing the configuration of a load manager in FIG.

Fig. 5 is a schematic diagram showing the configuration of SRC and SC in Fig. 4;

Figure 6 shows the flow of the overall operation of the load manager in Figure 4;

*** Explanation of symbols for main parts of drawing ***

NM: Node Manager TM: Task Manager

IM: Information Manager LM: Load Manager

MEM: Local memory

The present invention relates to a high availability system, and more particularly, to a high availability system and a method for splitting a task, in which a task is divided and handed over according to a node's resource utilization rate when a task event occurs for the node.

In general, HA system is composed of two or more N server systems, and the system operation status is continuously checked between the systems to determine whether there are problems in the working tasks or in the operating system. Then, the service node is transferred to the designated node that has a problem or the task running on the node, so that continuous service can be provided in response to an external client service request.

At this time, the system nodes largely use three managers for information exchange between each node. There are a node manager, a task manager, and an information manager.

Referring to FIG. 1, the operation of each manager in each node will be described. The HA master starts HA to activate the node manager NM, the task manager TM, and the information manager IM (S1).

Then, after transmitting the status information of the task to the local memory (MEM) (S2), the node information transmitted from another connected node is collected in the information manager (IM) (S3).

Next, after the status information is transferred from the information manager IM to the local memory MEM (S4), the node manager NM refers to its task status information stored in the local memory MEM (NM). ).

 Next, the state information of another node stored in the local memory MEM is read, and multicasting of the entire node information thereof based on the read information is performed to the external node (S7).

Then, the task is managed (executed, stopped, monitored) according to the information transmitted from the node manager NM, and the node manager NM refers to the information of the information manager IM to perform a decision (S9). When a failure occurs, the task manager TM is rerun by the node manager NM (S10).

FIG. 2 is a schematic diagram showing a general high availability system configuration. HA configuration of each node is the same as that of FIG. 1, and the three managers operate by starting HA, and in the system configuration file before starting HA to take over when a problem occurs. The takeover order is assigned artificially.

In other words, if a problem occurs in node A_, take over at node C_, a problem occurs in node C_, take over at node B_, and if a problem occurs in node B_, at node N_, or at node N_ Then, in order to take over all nodes to take over from N-1_ nodes, manually save the configuration and configuration files in the configuration file before starting HA, and read each file to prepare for failure.

As shown in Fig. 2, if no problem occurs in the state where each node performs each of the assigned tasks a to n, no handover occurs between each node.                         

That is, for example, when an external client makes a request to task a to receive a related service, node A_ recognizes that it needs to process and responds appropriately to the service request.

However, if there is no response to the service request, the task (IP, FS, AP) of the A_node is mounted on the shared disk and handed over in the order specified in the configuration file.

Before the takeover situation occurs, the A_Node's TM checks the status of its task killed and restarts it.

If there is no abnormality by performing the above restart, no handover occurs, but if a situation such as a network failure that cannot be solved occurs, the handover proceeds in order.

At this time, the task manager TM always monitors the task even before there is a service request from the outside.

If there is a problem with the node, not with a task (with or without a service request), i.e., always check the node's active state between the nodes, first through the Hert bit, and first through the Hert bit. If confirmed, the other node is active and no handover occurs.

However, if the signal cannot be received by the heartbeat (node killed, HB cable short-circuit problem, etc.), a second Ping check is performed at the node to be pre-set in the configuration file.

At this time, when the response of the normal state is received, the counterpart node is in an active state and does not take over.

 In other words, the internal network has failed, but the service to external clients is normal.

If it is not confirmed by ping, it checks the file system lastly. If there is action in this file system, it simply means that the node cannot check by HB and Ping. .

However, even if the file system check does not confirm the action, the node is killed and the target node takes over in the order specified in the configuration file.

At this time, in the HA system environment in which N nodes are operated, in the process of taking over a problem in which a node has a problem or a problem in a running task, the specified contents are inputted in advance for each node. All takeovers take place with reference to the recorded configuration file.

At this time, when the task is handed over, the progress node unconditionally performs the handover regardless of how large or small the load of its own node currently executing the task is.

In the conventional high availability system as described above, when a takeover event occurs in a multi-node HA system, task takeover for any node is performed to take over to a specific node that has been designated in advance. Tasks may be concentrated and the service speed of the node may be lowered, and there is a problem that it is not efficient in terms of evenly using the system resources.

The present invention has been made to solve the above problems, and when a task event for a node occurs, the task is divided and handed over according to the resource utilization rate of the node, so as to achieve high system availability, and to request the client. An object of the present invention is to provide a high availability system and a method for splitting a task to minimize a performance degradation of a service.

The present invention for achieving the above object is composed of 2 to N nodes (server system), in the high availability system for monitoring and recovering from the failure of the service process, the node, for the HA operation An HA master for activating the manager; A node manager which collects its own state information and multicasts it to another node, determines necessary tasks based on the collected information of the information manager, transmits them to the task manager, monitors the task manager, and re-executes in case of failure; A task manager based on the information transmitted from the node manager, monitoring the node manager, the load manager, and the information manager to re-execute in the event of a failure to activate the node manager, the load manager, and the information manager; An information manager which collects and manages the multicasted information from other nodes through the heartbeat, and transmits the information to the node manager and the load manager; It is characterized by comprising a load manager that calculates its own resource usage rate, and compares the resource usage rate of other nodes received from the information manager with the ranking.

The present invention for achieving the above object, the HA master, the first step of operating the load manager, node manager, information manager, task manager; A second process of transmitting the task state information to the local memory and collecting corresponding node information from another node through a heartbeat bit; Transmitting the state information of the other node to the load manager, and then storing the takeover order information and the state information of the corresponding node by comparing the resource use rate of the other node with the resource use rate of the other node; A fourth process of multicasting own state information and a takeover order to an external node; A fifth step of the node manager determining a task according to the information of the information manager and managing the task by the task manager; Detecting a failure of a task operation for each node, if a failure occurs in a predetermined node, and performs a sixth process of taking over the task of the node to the node according to the takeover order information.

Hereinafter, with reference to the accompanying drawings, the operation and effects of the high availability system according to the present invention and the task splitting method thereof will be described in detail.

Figure 3 is a schematic diagram showing a configuration of a node of the high availability system of the present invention. As shown in the drawing, each node includes: an HA master for activating a manager for HA operation; It collects its own status information and multicasts it to other nodes. Based on the collected information of the information manager (IM), it decides the necessary work and sends it to the task manager (TM). It monitors the task manager (TM) and fails. A node manager (NM) which is redone at time; Task management is performed based on the information transmitted from the node manager (NM), and the node manager (NM), the load manager (LM), and the information manager (IM) are monitored and re-executed in case of failure, and the node manager (NM) and the load manager are executed. A task manager TM for activating the LM and the information manager IM; An information manager IM that collects and manages the multicasted information from other nodes through the heartbeat, and transmits the information to the node manager NM and the load manager LM; It is composed of a load manager (LM) that calculates its own resource utilization rate and ranks the resource utilization rate of other nodes received from the information manager (IM).

The load manager LM includes a self resource checker (SRC) for calculating its own resource utilization rate, as shown in FIG. 4; SC (Sequence) which determines the takeover order by comparing the resource use rate of another node received from the information manager (IM) with the own resource use rate, and stores the takeover order information and state information of each node in the local memory (MEM). Comparator: hereafter referred to as SC).

As shown in Fig. 5, the SRC includes a CUC for calculating its own CUI resource utilization rate, a MUC for calculating a memory (MEM) resource utilization rate, a DUC for calculating disk I / O resource utilization rate, and a network I / O. O consists of NUCs for calculating resource utilization.

As shown in Fig. 5, the SC includes: a UC for comparing resource utilization of each node with its own resource usage; After determining the takeover order using the comparison result of the UC, it is configured as a DTS that stores the result in a local memory (MEM), and the operation of the present invention will be described.

First, the operation of any node will be described.

The HA master activates four managers managing internal operations, and each manager starts transmitting information and obtaining information for performing HA.                     

That is, the HA master activates the node manager NM, the task manager TM, the information manager IM, and the load manager LM for HA operation.

Accordingly, the node manager NM collects its own state information, multicasts it to another node, determines the required work based on the collected information of the information manager IM, and transmits the required job to the task manager TM, and the task Manager (TM) is monitored and re-run in case of failure.

The task manager TM manages a task based on the information transmitted from the node manager NM, monitors the node manager NM, the load manager LM, and the information manager IM, and re-executes when a failure occurs. The manager NM, the load manager LM and the information manager IM are activated.

The information manager IM collects and manages multicasted information from other nodes through the heartbeat, and transmits the information to the node manager NM and the load manager LM.

The load manager LM calculates its own resource utilization rate, compares the resource utilization rate of other nodes received from the information manager IM with itself, and determines a task takeover priority, and stores it in the local memory MEM.

That is, the SRC of the load manager LM calculates its own resource utilization rate and transmits it to the SC. Then, the SC compares the resource utilization rate of the other node received from the information manager IM with the resource utilization rate of the other node. After determining the takeover order, the takeover order information and the state information of each node are stored in the local memory MEM.

At this time, the SRC is performed four functions, that is, CPC Usage (CUC) calculates its own seed oil resource utilization, MEM Usage (MUC) calculates the memory (MEM) resource usage, Disk I / O Calculate disk I / O resource utilization, and N / WI / O calculates network I / O resource utilization.

The UC of the SC compares resource utilization rates of other nodes with its own resource utilization rate, and the DTS determines a takeover order using the comparison result of the UC, and then stores it in the local memory MEM.

6 shows an internal operation of the load manager LM. When the internal SRC is started, four routines are performed by the SR master to determine the usage of its own resources, and the output value is the Cul of the SC. It is sent to the calculator using a Trans function.

The SC in standby sends the node information collected from the IM to the Cul calculator to calculate the value.

As a result of the calculation, the takeover priority is the result, which is stored in its local memory (MEM) as a W2M function. When a node takes over the stored contents, the memory (MEM) area in which the priority is stored is invalidated and reused for future events.

Referring to FIG. 3, the task partitioning method of the high availability system of the present invention will be described by the HA master as the load manager (LM), the node manager (NM), the information manager (IM), and the task manager (TM). After operating (S1), and transmits the task status information to the local memory (MEM) (S2), and collects the corresponding node information from the other node through the heartbeat (S3).

 Then, after transmitting the status information of the other node to the load manager (LM) (S4), the load manager (LM) compares the resource utilization rate of its own resource with the resource utilization rate of the other node, and the takeover order information and the state of the node. The information is stored (S5, S6).

Next, it multicasts its own state information and takeover order to external nodes (S7 ~ S9).

Next, the node manager NM determines a task according to the information of the information manager IM, and manages the task by the task manager TM (S10 and S11).

Then, if a failure of a task operation for each node is detected, and a failure occurs in a predetermined node, the task of the corresponding node is transferred to the node according to the takeover order information (S12). Take over a task to one node that owns it, or several nodes distribute and take over the task.

In other words, as shown in FIG. 2, when a task is performed to each node a to m, and a request to receive service through the service network enters the node A_, if the a-task is operating normally, the service continues. If a problem occurs, task takeover is performed.

Here, the condition that the takeover occurs regardless of whether there is a service request from the outside is when the task is killed and re-executed by the task manager (TM), or when the service is not activated due to a network problem. Checking the turnover order determined by the manager LM, the task is handed over to the node with the least resource utilization.

If there is an error in the node itself, all the tasks performed in that node are killed, so check the resource usage rate of each node and each node stored in the local memory (MEM) of each node. In this case, the node with the smallest resource utilization may take over all tasks, and several nodes distribute the tasks to the tasks.

In this case, the operation of the split-over for the multi-task will be described. After calculating the average resource utilization rate for a plurality of nodes, the sum of the self-use rate of a given node and the resource use rate of tasks to be taken over is greater than the average resource utilization value. If larger, the task to be handed over is given to the predetermined node. If the size is small, the task is divided and handed over in association with one or more other nodes.

That is, the resource usage values of N nodes are summed, and this value is divided by N again to obtain an average value.

If there are three tasks to be turned over and there are top five nodes lower than the average value, the top three nodes will take over one task by the takeover sequence.

If three tasks are taken over by the priority node, when the sum of their resource utilization rate and the resource utilization rate of the inherited task is higher than the average value, only one task having a larger task size is taken over to the priority node. The two tasks are checked at the second node.

In the same way, if the sum of the resource utilization rate of the own resource and the inherited task is lower than the average, both tasks are transferred to the second priority node. If the sum is higher than the average, the task having the larger size of the two tasks is removed. It takes over to the second rank node, and the other one task takes over to the third rank node.

By doing so, it is possible to utilize the node's resources sufficiently and efficiently, and to improve the performance of services to external clients.

Specific embodiments or examples made in the detailed description of the present invention are intended to clarify the technical contents of the present invention to the extent that they should not be construed as limited to these specific embodiments and should not be construed in consultation. Various changes can be made within the scope of.

As described in detail above, in the present invention, when a task event occurs for a node, the task is divided and handed over according to the resource utilization rate of the node to achieve high system availability, thereby degrading performance of a service for a client's request. By minimizing and allocating the resources of the high availability system, it is possible to secure the system operation stability, improve the work processing capacity, and improve the efficiency of system resource use.

Claims (7)

In a high availability system consisting of 2 to N nodes (server systems) to monitor and recover from failures of service processes, The node includes: an HA master for activating a manager for HA operation; A node manager which collects its own state information and multicasts it to another node, determines necessary tasks based on the collected information of the information manager, transmits them to the task manager, monitors the task manager, and re-executes in case of failure; A task manager based on the information transmitted from the node manager, monitoring the node manager, the load manager, and the information manager to re-execute in the event of a failure to activate the node manager, the load manager, and the information manager; An information manager which collects and manages the multicasted information from other nodes through the heartbeat, and transmits the information to the node manager and the load manager; A high availability system comprising a load manager that calculates its own resource utilization rate and ranks the resource utilization rate of the other nodes received from the information managers with its own. The method of claim 1, wherein the load manager, An SRC for calculating its own resource utilization rate; After determining the takeover order by comparing the resource usage rate of the other node received from the information manager and the resource utilization rate of the own node, and the SC to store the takeover order information and the status information of each node in the local memory (MEM) High availability system. The method of claim 2, wherein the SRC is A CUC that calculates your CPI resource utilization, A MUC for calculating a memory (MEM) resource utilization rate; A DUC for calculating disk I / O resource utilization; A high availability system comprising NUC for calculating network I / O resource utilization. The method of claim 2, wherein SC is UC for comparing the resource utilization rate of each node with its own resource utilization rate; The high availability system, characterized in that the DTS to determine the take-up order using the comparison result of the UC, and to store it in a local memory (MEM). A first step of operating, by the HA master, the load manager, the node manager, the information manager, and the task manager; A second process of transmitting the task state information to the local memory MEM and collecting corresponding node information from another node through a heartbeat bit; Transmitting the state information of the other node to the load manager, and then storing the takeover order information and the state information of the corresponding node by comparing the resource use rate of the other node with the resource use rate of the other node; A fourth process of multicasting own state information and a takeover order to an external node; A fifth step of the node manager determining a task according to the information of the information manager and managing the task by the task manager; Detecting a failure of a task operation for each node, if a failure occurs in a predetermined node, the task partitioning method of the high availability system, characterized in that performing the task of the node in the sixth process to take over to the node according to the takeover order information. The method of claim 5, wherein the sixth process is Checking the takeover order, the task is handed over to one node having the smallest resource utilization, or a plurality of nodes distribute the task to take over, the task partition takeover method of a high availability system. The method of claim 6, wherein the number of nodes distribute tasks. Calculating an average resource utilization rate for the node; If the sum of the self resource utilization rate of the given node and the resource utilization rate of the tasks to take over is greater than the average resource utilization value, the task to take over is transferred to the predetermined node; The task partitioning method of the high availability system, characterized in that consisting of.

Images