JP6838334B2 - Cluster system, server, server operation method, and program - Google Patents

Description

The present invention relates to a cluster system, and more particularly to a cluster system including a plurality of servers.

The present invention also relates to a server used in a cluster system, an operation method thereof, and a program.

A cluster system is known in which a plurality of computers are bundled and operated as one. Of the cluster systems, those built to increase the availability of the system are called high availability cluster systems. A highly available cluster system includes a plurality of nodes (servers) connected to each other via a network.

In a highly available cluster system, each node monitors each other alive via a network. In a high availability cluster system, a task is executed on only one node, and whether or not the task is executed normally is monitored. The high availability cluster system has a function to take over the business to another node when it detects that the node on which the business was being executed has stopped, and a function to take over the business to another node when the business monitoring detects an abnormality. Have. In a highly available cluster system, when it is determined that a certain node in the cluster cannot communicate, it is determined that the node has stopped, and the business running on that node is started on the other node.

Regarding a cluster system, Patent Document 1 discloses a system in which a normal node continues to serve. In Patent Document 1, definition information including an allotted time in which a node can execute a service process and a priority of the allotted time set for each node is used. In Patent Document 1, when communication between two servers (nodes) constituting the cluster system is interrupted, each node calculates the start time of the next allocation time based on the definition information. Each node delays the start of a predetermined service (business) process until the calculated start time. At the start time of the allocated time, each node starts the business process, continues the business process if the business can be executed within the allocated time, and stops the operation if the business cannot be executed.

Further, regarding a cluster system, Patent Document 2 discloses that a master node that provides a service to a client is selected when a split brain occurs. In Patent Document 2, the cluster management unit of each node detects a node failure by performing heartbeat communication with each other. The weighting processing unit of each node checks the state related to the start of the service of the node, and updates the weight of the own node in the weight information stored in the shared storage device according to the checked state. The tiebreaker mechanism of each node determines whether or not the priority of the own node is the highest based on the weight indicated by the updated weight information, and if the priority is the highest, the own node is selected as the master node.

Japanese Unexamined Patent Publication No. 2006-48477 JP-A-2009-223519

In Patent Documents 1 and 2, each node monitors the life and death of other nodes. Therefore, as the number of nodes increases, the amount of communication for alive monitoring increases accordingly. For example, when N is an integer of 2 or more and the number of nodes constituting the cluster system is N, the number of communications for alive monitoring in the cluster system is N (N-1) / 2. In Patent Documents 1 and 2, when the number of nodes increases, the communication load increases and communication management becomes complicated.

In view of the above circumstances, an object of the present invention is to provide a cluster system capable of suppressing an increase in communication for alive monitoring even if the number of nodes increases.

Another object of the present invention is to provide a server used in the cluster system, an operation method of the server, and a program.

In order to achieve the above object, the present invention is a server used in a cluster system including a plurality of servers, and a leader selection unit that selects one of the plurality of servers as a leader and its own server. If is a reader, it sends a keep alive to other servers included in the cluster system to monitor other servers, and if its own server is not a leader, it receives the keep alive from the server that is the leader. It includes a keep alive communication control unit that transmits a response to the keep alive to the server that is the reader, and a node state recording unit that records the monitoring result in the keep alive communication control unit when the own server is the reader. Provide a server.

The present invention is also a cluster system including a plurality of servers, in which one of the plurality of servers is elected as a leader, and the server selected as the leader transmits a keep alive to another server. It provides a cluster system that monitors other servers and sends a response to keep-alive to the server elected as the leader.

Further, the present invention is a method of operating a server that operates as a leader in a cluster system including a plurality of servers, in which a step of transmitting a keep alive to another server included in the cluster system and a keep from the other server. Provided is an operation method of a server having a step of receiving a response to an alive and a step of recording a monitoring result indicating whether or not the response could be received in a node state recording unit.

The present invention is a step of transmitting a keep alive to a server operating as a leader in a cluster system including a plurality of servers to another server included in the cluster system, and a step of receiving a response to the keep alive from the other server. And a program for executing the step of recording the monitoring result indicating whether or not the response could be received in the node state recording unit.

The cluster system, server, server operation method, and program of the present invention can suppress an increase in communication for alive monitoring even if the number of nodes in the cluster system increases.

The block diagram which shows the server used for the cluster system of this invention. The block diagram which shows the cluster system which concerns on one Embodiment of this invention. A block diagram showing the server configuration. A flowchart showing an operation procedure when a leader is selected. A flowchart showing the operation procedure of the leader node processing. Flowchart showing the operation procedure of non-leader node processing

Prior to the description of embodiments of the present invention, an outline of the present invention will be described. FIG. 1 shows a server used in the cluster system of the present invention. The server 10 has a leader selection unit 11, a keep-alive communication control unit 12, and a node status recording unit 13. The cluster system includes a plurality of servers 10.

The leader selection unit 11 selects one of a plurality of servers 10 included in the cluster system as a leader. When the own server is the leader, the keep-alive communication control unit 12 transmits the keep-alive to the other server 10 included in the cluster system and monitors the other server. If the own server is not the reader, the keep-alive communication control unit 12 receives the keep-alive from the server 10 which is the reader, and transmits a response to the received keep-alive to the server 10 which is the leader. When the local server is the leader, the node state recording unit 13 records the monitoring result in the keep-alive communication control unit.

In the present invention, one of the plurality of servers constituting the cluster system is elected as the leader. The server elected as the leader sends a keepalive to the remaining servers and receives a response to the keepalive. The server elected as the leader monitors the operating status of the remaining servers based on whether or not there is a response to keepalives. In the present invention, keepalive and response communication may be performed between the server selected as the leader and each of the remaining servers, so that communication for life-and-death monitoring is performed even if the number of nodes in the cluster system increases. Can be suppressed.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 2 shows a cluster system according to an embodiment of the present invention. The cluster system 100 has servers 110-1 to 110-N, where N is an integer of 3 or more. These servers 110-1 to 110-N are typically computer devices having a processor, memory, auxiliary storage, and the like. The servers 110-1 to 110-N are connected to each other via a network. Servers 110-1 to 110-N form, for example, a high availability cluster system. In the following description, when it is not necessary to distinguish a plurality of servers, it is also referred to as a server 110.

In the present embodiment, in the cluster system 100, one of the plurality of servers 110 is dynamically elected as a leader. The server 110 elected as the leader periodically transmits a keepalive to the remaining servers 110 and receives a response to the keep-alive to monitor the life and death of the remaining servers 110. The remaining servers 110 that are not leaders monitor the life and death of the servers 110 selected as leaders by using keepalives transmitted from the readers.

Here, the servers that do not respond to the keepalive include a server that has stopped operating and a server that has been disconnected from the network for some reason and cannot communicate with the reader. In the following, a server that has stopped operating is also referred to as a stopped node, and a server that has lost communication with the reader is also referred to as a non-response node. In order to distinguish between the non-responding node and the stopped node, each server 110 shall send a stop notification to the server 110 selected as the leader in the stop processing. A server that has not sent a stop notification and does not respond to keepalives is treated as a non-responding node.

In the present embodiment, each of the plurality of servers 110 has a function of receiving a forced stop instruction from the outside. When the server 110 elected as the leader detects the non-responding node, the non-responding node can be forcibly stopped. For example, if the server 110 of the non-responding node is a physical machine, the server 110 elected as the leader executes power off from the outside using BMC (Baseboard Management Controller) compliant with IPMI (Intelligent Platform Management Interface). To do. If the server 110 of the non-responding node is a virtual machine in the virtual environment, the server 110 elected as the leader executes a forced stop request of the virtual machine to the host machine.

When the server 110 elected as the leader is stopped, the leader election is executed again if there is another operating server 110. The server 110 elected as the leader executes the stop process after the server 110 newly elected as the leader starts the operation as the leader. If there is no other operating server 110, the server 110 elected as the leader executes the stop process as it is.

In the cluster system 100, the server 110 selected as the leader may not be able to communicate with other servers 110 for some reason. In order to deal with such a case, the server 110 elected as the leader can maintain the leader when it can communicate with more than half of the servers 110 in the cluster.

FIG. 3 shows the configuration of the server 110. The server 110 includes a leader selection unit 111, a keep-alive communication control unit 112, a node state recording unit 113, a communication unit 114, and a power supply control unit 115. In the server 110, at least a part of the functions of the leader selection unit 111, the keep-alive communication control unit 112, the node status recording unit 113, the communication unit 114, and the power supply control unit 115 is that the server executes processing according to a program. It is realized by.

The communication unit 114 communicates with at least one of the other server 110 and a client terminal (not shown). The leader selection unit 111 selects one of the plurality of servers 110-1 to 110-N as a leader. If the own server is the leader, the keep-alive communication control unit 112 transmits the keep-alive to another server. If the local server is not the reader, the keep-alive communication control unit 112 receives the keep-alive from the server that is the reader, and transmits a response to the keep-alive.

The node state recording unit 113 includes a storage device and stores the monitoring result in the keep-alive communication control unit 112. The node status recording unit 113 records, for example, the server 110 that has not responded to the keepalive and has not received the stop notification as a non-responding node. The node status recording unit 113 records the server that has received the stop notification as a stop node. Further, the node state recording unit 113 stores information indicating which server is the leader.

The power supply control unit 115 controls the power supply of the server 110. When the local server is the leader, the power control unit 115 stops the power supply of the node that does not respond to the keepalive. The power supply control unit 115 corresponds to a node stop control unit that controls the stop of a node. When the own server is the leader, the power supply control unit 115 transmits a power cutoff request to the power supply control unit 115 of the server 110 to be stopped. When the power control unit 115 receives a power cutoff request from the reader server 110, the power control unit 115 stops the power supply of its own server.

The power control unit 115 of the server 110 elected as the leader forcibly stops the power supply of the node (server) recorded as a non-response node in the node status recording unit 113, for example. The power control unit 115 may perform a power stop of the server when the number of servers recorded as a non-response node is less than half of all the servers. When the number of servers recorded as a non-response node is half or more of all the servers, the leader selection unit 111 may select a new leader.

Subsequently, the operation procedure will be described. FIG. 4 shows an operation procedure when a leader is selected. Each server 110 starts the leader election process, for example, when it is started. In each server 110, the leader selection unit 111 selects one of the plurality of servers 110 constituting the cluster system as a leader (step A1). The leader selection unit 111 selects one server (leader node) to operate as a leader by using, for example, a distributed consensus algorithm.

The leader election unit 111 records the leader node in the node status recording unit 113. Each server determines whether or not its own server is the leader (step A3). When the own server is a leader, each server 110 executes a leader node process (step A4), and when the server 110 is not a leader, executes a non-leader node process (step A5).

FIG. 5 shows the operation procedure of the leader node processing. The keep-alive communication control unit 112 of the server 110 selected as the leader transmits the keep-alive to each of the servers 110 other than the reader (step B1). The keep-alive communication control unit 112 waits for a response to the keep-alive (step B2). The keep-alive communication control unit 112 determines whether or not the response wait time-out has occurred (step B3).

If the keep-alive communication control unit 112 can receive the response before the time-out occurs, the keep-alive communication control unit 112 records that fact in the node state recording unit 113 (step B4). In step B4, the keep-alive communication control unit 112 deletes the record of the non-response node when the response to the keep-alive can be received from the server recorded as the non-response node. After that, the keep-alive communication control unit 112 returns to step B1 after waiting for a certain period of time, and transmits the keep-alive periodically, for example, at predetermined time intervals.

When it is determined that the time-out has occurred in step B3, the node state recording unit 113 records the node whose response has timed out as a non-response node (step B5). The keep-alive communication control unit 112 determines whether or not the number of non-response nodes is more than half of all the servers (step B6). If it is determined in step B6 that the number of non-responding nodes is more than half of all the servers, the leader selection unit 111 determines that the leader cannot be maintained and redoes the leader selection process (step B7).

When it is determined in step B6 that the number of non-response nodes is not more than half of all the servers, the power supply control unit 115 forcibly stops the server 110 recorded as the non-response nodes (step B8). In step B8, for example, the keep-alive communication control unit 112 of the server 110, which is the leader, transmits a power stop request to the power control unit 115 of the server 110 recorded as a non-response node. The power control unit 115 receives the power stop request and executes the power stop to stop the own server recorded as the non-response node.

FIG. 6 shows an operation procedure of non-leader node processing. The keep-alive communication control unit 112 of the remaining servers 110 that are not elected as leaders waits for the keep-alives transmitted from the servers 110 that are elected as leaders (step C1). The keep-alive communication control unit 112 determines whether or not the keep-alive has timed out (step C2). If the keep-alive communication control unit 112 determines in step C2 that the keep-alive has not timed out, the keep-alive communication control unit 112 transmits a response to the keep-alive to the server 110 selected as the leader (step C3). After that, the keep-alive communication control unit 112 returns to step C1 and waits for the next keep-alive.

When the keep-alive communication control unit 112 determines that the keep-alive has timed out in step C2, it considers that the leader has disappeared and instructs the leader selection unit 111 to redo the leader selection. The leader selection unit 111 executes the leader selection process according to the instruction (step C4).

It should be noted that each server 110 constituting the cluster system independently determines to start the leader selection process again after the leader is selected. If the original leader starts the leader selection process for some reason, the other server 110 stops receiving the keepalive, so that the other server 110 also starts the leader selection process. As a result, the leader election process is executed on all the servers 110. On the other hand, when the server 110, which cannot receive the keepalive from the reader, starts the leader election process, a new leader is elected if the majority of the servers 110 start the leader election process. Otherwise, the original reader will continue to act as the leader, and the server 110 that did not receive the keepalives sent by the reader, that is, the server 110 that performed the leader election process but could not elect the leader, is the leader. It is stopped by the instruction from.

In the present embodiment, one server 110 is elected as the leader among the plurality of servers 110-1 to 110-N constituting the cluster system. The server 110 selected as the leader performs life-and-death monitoring using keep-alive communication on the remaining servers 110. The servers 110 that are not elected as leaders may carry out keep-alive communication with the servers 110 that are elected as leaders. In the present embodiment, for example, when the number of servers 110 is N, the number of keep-alive communications is N-1. In the present embodiment, since it is not necessary to perform keep-alive communication between the servers 110 that are not selected as leaders, it is possible to suppress an increase in the communication load when the number of servers constituting the cluster system increases. Communication management can also be simplified.

Further, in the present embodiment, each server 110 transmits a node stop notification to the server 110 selected as a leader at the time of its stop. When there is a server 110 that does not respond to the keepalive, the server 110 elected as the leader may return a response whether the server 110 is stopped or for some reason by checking for the presence or absence of the stop notification. It is possible to determine if it is not possible.

For example, consider a case where a server temporarily becomes inoperable and then resumes operation after a while. In a cluster system, if the server that provided the business temporarily becomes inoperable and it is determined that the server has stopped, the business is started on another server. If the original server resumes operation after the business is started on another server, there is a possibility that the business will be executed on multiple servers (hereinafter, also referred to as the active state of both systems).

In the present embodiment, the server 110 elected as the leader determines whether or not the stop notification has been received from the server 110 that has become temporarily inoperable. The server 110 elected as the leader can determine whether the server 110 has stopped or has no response based on the presence or absence of the stop notification. In the present embodiment, the server 110 selected as the leader forcibly stops the non-responsive server 110 by using an external power supply control using, for example, IPMI. By doing so, it is possible to prevent the non-responsive server 110 from resuming operation and becoming active in both systems.

Although the embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and changes and modifications are made to the above-described embodiments without departing from the spirit of the present invention. Also included in the present invention.

In the above embodiment, the program can be stored and supplied to a computer using various types of non-transitory computer readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-temporary computer-readable media are magnetic recording media such as flexible disks, magnetic tapes, or hard disks, such as magneto-optical recording media such as magneto-optical discs, CDs (compact discs), or DVDs (digital versatile disks). Includes optical disk media such as, and semiconductor memories such as mask ROM (read only memory), PROM (programmable ROM), EPROM (erasable PROM), flash ROM, or RAM (random access memory). The program may also be supplied to the computer using various types of transient computer readable media. Examples of temporary computer-readable media include electrical, optical, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

10: Server 11: Leader selection unit 12: Keepalive communication control unit 13: Node status recording unit 110: Server 111: Leader selection unit 112: Keepalive communication control unit 113: Node status recording unit 114: Communication unit 115: Power supply control Department

Claims (8)

A server used in a cluster system that contains multiple servers.
A leader election unit that selects one of the plurality of servers as a leader,
If the local server is the leader, it sends a keep alive to other servers included in the cluster system to monitor other servers, and if the local server is not the leader, it receives the keep alive from the server that is the leader. Then, the keep alive communication control unit that transmits the response to the keep alive to the server that is the reader,
When the local server is the reader, it is provided with a node state recording unit that records the monitoring result in the keepalive communication control unit .
Each of the other servers sends a node stop notification to the leader server when the own node is stopped.
Node state recording unit, when the local server is a leader, that records not respond to the keep-alive, and the server does not receive the stop notification as unresponsive node server. The server according to claim 1 , wherein the node status recording unit records the server that is the source of the stop notification as a stop node when the own server is a leader. The server according to claim 1 or 2 , further comprising a node stop control unit that stops a server recorded as a non-response node when the local server is a leader. Said node stop control section, wherein, when the number of recorded server as unresponsive node is less than half the number of servers included in the cluster system, according to claim 3 for stopping the recorded server as the unresponsive node The server described in. If the local server is the leader and the number of servers recorded as the non-response node exceeds half of the number of servers included in the cluster system, the leader election unit selects a new leader. The server according to claim 4 to be implemented. A cluster system that includes multiple servers
One of the multiple servers was elected as the leader and
Server elected by said reader, sending the keep-alive monitor other servers to other servers,
The other server sends a response to the keepalive to the server elected as the leader .
Each of the other servers sends a node stop notification to the leader server when the own node is stopped.
The server elected as the leader is a cluster system that records a server that does not respond to the keepalive and has not received the stop notification as a non-response node. A method of operating a server that operates as a leader in a cluster system that includes multiple servers.
The step of sending keepalives to other servers included in the cluster system,
The step of receiving the response to the keepalive from the other server,
And recording the monitoring result indicating whether or not received the response to the node state recording unit,
The step of receiving the node stop notification from the other server, and
A method of operating a server having a step of recording a server that does not respond to the keepalive and has not received the stop notification as a non-response node. For a server that acts as a leader in a cluster system that includes multiple servers
The step of sending keepalives to other servers included in the cluster system,
The step of receiving the response to the keepalive from the other server,
And recording the monitoring result indicating whether or not received the response to the node state recording unit,
The step of receiving the node stop notification from the other server, and
A program for executing a step of recording a server that does not respond to the keepalive and has not received the stop notification as a non-response node.

Images