JP2010182017A - Distributed computer system, manager succession method and manager succession program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To select an active-system management program suitable for a situation in which the status of a node or a network dynamically changes even in such a situation with a distributed computer system comprising a plurality of nodes. <P>SOLUTION: When failure occurs in an active-system manager 3, each node 1 constituting this distributed computer system selects a candidate of the active-system manager 3 from a plurality of standby-system managers 3, and transmits a selection result to the selected standby-system manager 3. Each standby-system manager 3 receives the selection result, decides whether or not itself becomes the active-system manager 3 based on the received selection result, and changes into the active-system manager 3 according to a decision result thereof. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a distributed computer system that performs distributed processing at a plurality of nodes, and more specifically, a distributed computer system that takes over a manager when a failure occurs in a manager (management program) that manages the plurality of nodes (computers). The present invention relates to a manager takeover method and a manager takeover program.

  In order to improve processing speed, a distributed computer system is known in which processing to be performed is distributed to a plurality of nodes and executed. In this type of distributed computer system, the load on each node is monitored, and appropriate processing is assigned to each node. If a failure occurs on one node, the processing executed on that node is It is possible to flexibly cope with changes in the system configuration and system state, for example, by delegating to a node (see, for example, Patent Document 1).

  In the distributed computer system as described above, one manager exists on the system in order to assign an appropriate process to each node. If a failure occurs in the manager or the node on which the manager exists, the manager will be absent and the entire system will become inoperable. Therefore, the standby manager for taking over the manager processing (currently managing the standby manager) (The manager in charge is called the working manager). In order to ensure higher reliability, a plurality of standby managers may be prepared. In that case, a method of statically predetermining the order of becoming active managers is often employed.

JP 2000-207374 A

  In order to ensure high reliability, when a plurality of standby managers are prepared and a manager to be an active manager is determined according to a static order, there are the following problems.

The first problem is that other processing programs are also dynamically allocated on the node where the standby manager is allocated, and an appropriate manager is always selected when the load on the node changes from moment to moment. That is not necessarily the case.
The reason is that when the active manager is absent, the standby manager placed on the node where the important processing program with high load is running may be selected as the active manager according to the static order. In this case, if the manager on the node starts to operate as the active manager, the processing program that has been operating may not be able to produce sufficient performance.

The second problem is that an appropriate manager is not always selected in a mobile network or an ad hoc network environment.
The reason is that the standby manager selected to become the active manager according to the static order exists in a position far from the other nodes in the network, or the network performance with the other nodes is poor. Because there are cases.

  The present invention has been considered in view of the above circumstances. When a manager that manages a plurality of nodes fails, the state of each node or network changes dynamically when the manager takes over. Even if there is a standby computer manager that is suitable for the situation, it can be selected as the active manager, so that the performance of the processing program and the communication efficiency between nodes can be avoided. For the purpose of provision.

  In order to achieve the above object, a distributed computer system according to the present invention comprises a plurality of nodes constituting a distributed computer system, one node provided with an active manager that assigns processing to each of these nodes, and the active manager And a plurality of nodes provided with standby managers operable as active managers when a failure occurs, wherein each node has a plurality of standbys when a failure occurs in the active manager Manager selection means for selecting a candidate for the active system manager from among the system managers, and selection result transmission means for transmitting the selection result to the selected standby system manager, wherein each standby system manager displays the selection result. Based on the selection result receiving means to be received and the received selection result, it is determined whether or not it is an active manager. A selection result determining means that, depending on the determination result, a structure comprising a state changing unit that changes the working system manager, a.

  In order to achieve the above object, the manager takeover method of the present invention provides a plurality of nodes constituting a distributed computer system, wherein one node provided with an active manager that assigns processing to each of these nodes, and the active And a plurality of nodes provided with standby managers operable as active managers when a failure occurs in the system manager, and each node includes a plurality of nodes when a failure occurs in the active manager. A candidate for the active manager is selected from the standby managers, and the selection result is transmitted to the selected standby manager. Each standby manager receives the selection result and is based on the received selection result. And determine whether or not to become the active manager, and change to the active manager according to the determination result. A.

  In order to achieve the above object, the manager takeover program of the present invention provides a plurality of nodes constituting a distributed computer system, one node provided with an active manager for assigning processing to each of these nodes, and the active And a plurality of nodes with standby managers that can operate as active managers when a failure occurs in the active manager, and the manager takeover program for each node generates a fault in the active manager A process for selecting an active manager candidate from the plurality of standby managers and a process for transmitting the selection result to the selected standby manager, and for each standby manager. The manager takeover program for receiving the selection result and the received selection result. Based on the results, the treatment to which it is determined whether becomes active system manager, according to the determination result, a process of changing the current system manager, as configured to be executed by a computer.

  According to the present invention, even when the load of a node changes from moment to moment, a standby manager suitable for the situation at that time can be selected as an active manager, and even in a mobile network or an ad hoc network environment, A standby manager suitable for the situation can be selected as the active manager.

It is a block diagram which shows the structure of the distributed computer system which concerns on embodiment of this invention. 5 is a flowchart showing an operation of a manager selection device in the distributed computer system according to the embodiment of the present invention. 5 is a flowchart showing an operation of a manager in the distributed computer system according to the embodiment of the present invention. It is a block diagram which shows the specific structure of the distributed computer system which concerns on embodiment of this invention. 5 is a flowchart showing a specific operation of the manager selection device in the distributed computer system according to the embodiment of the present invention. 5 is a flowchart showing a specific operation of a manager in the distributed computer system according to the embodiment of the present invention. 1 is a block diagram showing a specific system configuration of a distributed computer system according to an embodiment of the present invention. It is explanatory drawing which shows the specific example of the information recorded on a node information recording part in the distributed computer system which concerns on embodiment of this invention. It is explanatory drawing which shows the specific example of the information recorded on the selection result recording part in the distributed computer system which concerns on embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[Outline of Distributed Computer System According to Embodiment of the Present Invention]
First, an overview of a distributed computer system according to an embodiment of the present invention will be described with reference to FIGS.
FIG. 1 is a block diagram showing a configuration of a distributed computer system according to an embodiment of the present invention.

As shown in this figure, the distributed computer system according to the embodiment of the present invention is composed of a plurality of nodes 1, and each node 1 is provided with a manager selection device 2 and a manager 3 as necessary. ing.
Here, when a failure occurs in one node 1 provided with an active manager 3 that assigns processing to each of these nodes 1 and the active manager 3 in a plurality of nodes 1 constituting the distributed computer system, A plurality of nodes 1 provided with the standby manager 3 operable as the active manager 3 and one or a plurality of nodes 1 not including the manager 3 are included.
FIG. 1 shows a configuration of two nodes 1 provided with a standby manager 3.

The manager selection device 2 is a device configured on a computer (computer) by a manager takeover program for each node, and includes a manager selection means 21 and a selection result transmission means 22.
The manager selection means 21 selects a candidate for the active manager 3 from the plurality of standby managers 3 when a failure occurs in the active manager 3. Manager selection conditions and selection procedures by the manager selection means 21 may be different among the nodes 1.
The selection result transmission unit 22 transmits the result selected by the manager selection unit 21 to the selected standby manager 3.

The manager 3 is an apparatus configured on the computer by a management program and a manager takeover program for the standby manager, and includes a selection result receiving unit 31, a selection result recording unit 32, a selection result determining unit 33, and a state change Means 34 are included.
The selection result receiving unit 31 receives the selection result transmitted from the selection result transmitting unit 22 of the manager selection device 2 and records it in the selection result recording unit 32.
The selection result recording unit 32 records the selection result received by the selection result receiving unit 31.
The selection result determination means 33 reads the selection result recorded in the selection result recording unit 32 when it is determined that the transmission of the selection result from each manager selection device 2 has been completed. It is determined whether or not to become a system manager 3. The judgment criteria of the selection result judgment means 33 need to be the same for all nodes 1.
The state changing unit 34 starts the operation as the active manager 3 when the selection result determining unit 33 determines that it becomes the active manager.

Next, the operation of the distributed computer system according to the embodiment of the present invention will be described with reference to FIGS.
FIG. 2 is a flowchart showing the operation of the manager selection device in the distributed computer system according to the embodiment of the present invention. FIG. 3 is a flowchart showing the operation of the manager in the distributed computer system according to the embodiment of the present invention.

  As shown in FIG. 2, the manager selection device 2 of each node 1 selects an appropriate standby manager 3 according to a predetermined criterion at the timing when the absence of the active manager 3 is detected (step A1: The manager selection means 21) transmits the selection result to the selected standby manager 3 (step A2: selection result transmission means 22).

On the other hand, as shown in FIG. 3, the standby manager 3 of each node 1 waits for transmission from the manager selection device 2 (steps B1 and B2), and if there is transmission, receives the selection result and selects the selection result recording unit 32. (Step B3: Selection result receiving means 31).
Next, it is determined whether or not the transmission of the selection results from all the manager selection devices 2 has been completed (step B4). If it is determined that the selection results have not yet been transmitted, the process returns to step B1 and the other manager selection devices 2 Wait for transmission.
If it is determined in step B4 that the transmission from the manager selection device 2 has been completed, or if it is determined that the predetermined waiting time has passed in steps B1 and B2, the selection result is recorded from the selection result recording unit 32. Reading (step B5), it is determined from the contents whether or not it should become the active system (step B6: selection result determination means 33).
If the standby manager 3 determines that it should become the active system, it operates as the active manager (step B7: state change means 34), and if it determines that it should not become the active system, The operation is continued (step B8).

  According to the distributed computer system of the present embodiment configured as described above, the plurality of nodes 1 constituting the distributed computer system are provided with the active manager 3 that assigns processing to each of these nodes 1. A node 1 and a plurality of nodes 1 provided with a standby manager 3 that can operate as the active manager 3 when a failure occurs in the active manager 3. When a failure occurs, a candidate for the active manager 3 is selected from the plurality of standby managers 3 and the selection result is transmitted to the selected standby manager 3, and each standby manager 3 receives the selection result. Then, based on the received selection result, it is determined whether or not it becomes the active manager 3 and changes to the active manager 3 according to this determination result. Compared to the system to keep determines the order to be active system manager 3, it is possible to select the active system manager 3 which is suitable for conditions at that time.

[Details of Distributed Computer System According to Embodiment of the Present Invention]
Next, a specific configuration and operation of the distributed computer system according to the embodiment of the present invention will be described with reference to FIGS.
FIG. 4 is a block diagram showing a specific configuration of the distributed computer system according to the embodiment of the present invention.

  Compared with the distributed computer system shown in FIG. 1, the distributed computer system shown in this figure includes node information transmission means 35 and arbitration means 36 in the standby manager 3, and node information reception means in the manager selection device 2. 23 and a node information recording unit 24 are added.

The node information transmitting unit 35 transmits node information that is used as a judgment material for the manager selecting unit 21 to select the standby manager 3, to the node information receiving unit 23 of each node 1, and the node information receiving unit 23 receives the received node information. Node information is recorded in the node information recording unit 24 (see broken line arrows in FIG. 4).
The arbitration unit 36 determines that the number of votes transmitted from the selection result transmission unit 22 is less than a predetermined ratio compared to the total number of standby system managers 3 and cannot determine whether or not to become the active system manager 3. It is a means for determining one standby manager 3 to be changed to the manager 3, for example, asks the arbitration means 36 of another standby manager 3 to send information on the number of votes, and more It is determined whether or not there is a standby manager 3 that has obtained the number of votes.

Next, specific operations of the distributed computer system according to the embodiment of the present invention will be described with reference to FIGS.
FIG. 5 is a flowchart showing a specific operation of the manager selection device in the distributed computer system according to the embodiment of the present invention. FIG. 6 is a specific operation of the manager in the distributed computer system according to the embodiment of the present invention. FIG. 7 is a block diagram showing a specific system configuration of the distributed computer system according to the embodiment of the present invention, and FIG. 8 is a node information recording unit in the distributed computer system according to the embodiment of the present invention. FIG. 9 is an explanatory diagram showing a specific example of information recorded in the selection result recording unit in the distributed computer system according to the embodiment of the present invention.
FIG. 6 is a flowchart specifically showing a portion corresponding to step B6 in FIG. 3 in the operation of the manager 3 in the present embodiment.

  The distributed computer system shown in FIG. 7 has four nodes X11 to X14. In each of the nodes X11 to X14, the manager selection device 2 and the manager 3 shown in FIG. 4 are arranged, and it is assumed that the manager 3 arranged in the node X11 operates as an active system. The active manager 3 of the node X11 periodically transmits heartbeat packets to the nodes X12 to X14. Thereby, each of the nodes X12 to X14 can recognize that the manager 3 of the node X11 is operating normally as the active manager 3.

The node information transmission means 35 of each standby manager 3 of the nodes X12 to X14 periodically sends the information on the free memory capacity and the free disk capacity as node information every 5 seconds, for example, as the node information receiving means of each node X12 to X14. 23.
The node information receiving means 23 of each manager selecting device 2 records the free memory capacity and free disk capacity received as node information from each standby manager 3 in the node information recording unit 24 in the format shown in FIG.
When a failure occurs in the node X11 and the heartbeat packet is not received for a certain period, the manager selection device 2 of each of the nodes X12 to X14 detects the absence of the active manager 3. When the absence of the active manager 3 is detected, the manager selection means 21 of each manager selection device 2 reads the node information recorded in the node information recording unit 24 (step A0 in FIG. 5).

It is assumed that the manager selection means 21 of the nodes X12 and X13 implements, for example, a procedure for selecting the active manager 3 with a free memory capacity as the manager selection procedure in step A1. As a result, the manager selection means 21 of the nodes X12 and X13 selects the standby manager 3 of the node X12, and transmits the selection result from the selection result transmission means 22 to the standby manager 3 of the node X12 (step A2).
Assume that the manager selection means 21 of the node X14 implements, for example, a procedure for selecting the active manager 3 with a free disk capacity as the manager selection procedure in step A1. As a result, the manager selection unit 21 of the node X14 selects the standby manager 3 of the node X14, and transmits the selection result from the selection result transmission unit 22 to the standby manager 3 of the node X14 (step A2).

  The selection result receiving means 31 of each standby manager 3 records the number of selection results received by itself (referred to as the number of votes) in the selection result recording unit 32. The number of votes recorded in the standby manager 3 of each of the nodes X12 to X14 is as shown in FIG.

  The selection result judging means 33 of the standby manager 3 of the node X12 reads information “2” from the selection result recording unit 32 (step B5), and obtains a majority of votes with respect to the total number “3” of the manager selection devices 3. (Step B61), the operation is started as the active manager 3 (step B7).

  The selection result determination unit 33 of the standby manager 3 of the node X13 reads information “0” from the selection result recording unit 32 (step B5). As a result, it is determined that there is a standby manager 3 with more votes than itself (step B62), and the operation as the standby manager 3 is continued (step B8).

  The selection result judging means 33 of the standby manager 3 of the node X14 reads information “1” from the selection result recording unit 32 (step B5). With this information alone, it cannot be determined whether or not there is a standby manager 3 with more votes than itself, so the arbitrating means 36 requests the standby manager 3 of the nodes X12 and X13 to tell the number of votes. (Step B63). As a result, it is determined that the node X12 has more votes than itself (step B64), and the operation as the standby manager 3 is continued (step B8).

  According to the distributed computer system of the present embodiment configured as described above, each standby manager 3 transmits information of the node 1 provided with the standby manager 3 to each node 1, and each node 1 The node information is received, the received node information is recorded in the node information recording unit 24, and the manager selection unit 21 determines whether the plurality of standby managers 3 are based on the node information recorded in the node information recording unit 24. Therefore, the candidate for the active manager 3 can be selected based on the information of each node 1 in which the standby manager 3 is provided.

  Each standby manager 3 determines one standby manager 3 to be changed to the active manager 3 when the number of received selection results is below a predetermined ratio compared to the total number of standby managers 3. Therefore, in any situation, one standby system manager 3 is reliably selected to prevent the occurrence of a state where the active system manager 3 is absent or a state where a plurality of active system managers 3 exist. be able to.

  In addition, the arbitrating unit 36 of the present embodiment, when the number of received selection results is less than a predetermined ratio compared to the total number of standby managers 3, the other standby managers 3 3 inquires about the number of selection results received, and when there is no standby manager 3 having more selection results received than itself, it is determined that the manager 3 becomes the active manager 3, so the exchange between the standby managers 3 Only one standby manager 3 can be selected with certainty.

[Second Embodiment]
Next, a distributed computer system according to the second embodiment of the present invention will be described.
In the distributed computer system according to the second embodiment, when the number of the selection results received by the arbitrating unit 36 is lower than a predetermined ratio compared to the total number of standby managers 3, each node 1 has an active manager 3. The point that the candidates are selected again and the selection result is transmitted to the selected standby manager 3 is different from the above embodiment.

  For example, in the above-described embodiment, the selection result determination unit 33 determines that the standby system manager 3 with the largest number of votes should be the active system, but only those who have obtained a majority of the votes will be the active system. When a rule is defined and there is no standby system manager 3 that obtains a majority, the arbitration means 36 requests each manager selection device 2 to vote again. Then, the standby manager 3 that has obtained a majority vote is changed to the active manager 3.

  According to the distributed computer system of the second embodiment configured as described above, when the arbitration unit 36 receives the number of selection results below a predetermined ratio as compared to the total number of standby managers 3, Since each node 1 is made to select a candidate for the active manager 3 again and the selection result is transmitted to the selected standby manager 3, the standby manager 3 suitable for the situation can be selected carefully.

[Third embodiment]
Next, a distributed computer system according to the third embodiment of the present invention will be described.
In the distributed computer system according to the third embodiment, the selection result transmitting unit 22 transmits the selection result to all the standby managers 3, and the selection result determining unit 33 totals the received selection results and selects itself. This is different from the above-described embodiment in that it determines that it becomes the active manager 3 when the number obtained is the maximum.

  According to the distributed computer system of the third embodiment configured as described above, the selection result transmission unit 22 of the manager selection device 2 does not transmit the selection result only to the selected standby manager 3, but all the standbys. The selection result is transmitted to the system manager 3, and the selection result determination means 33 of the standby system manager 3 totals the received selection results. Therefore, although the network traffic increases, it is determined whether or not it should become the active manager 3 even when its own number of votes falls below a predetermined ratio compared to the total number of standby managers 3. As a result, it is possible to omit the step (step B64) of inquiring other standby managers 3 for the number-of-voting information by the arbitrating means 36.

[Fourth embodiment]
Next, a distributed computer system according to the fourth embodiment of the present invention will be described.
In the distributed computer system of the fourth embodiment, the manager selection means 21 is configured such that the load on each standby manager 3, the number of network hops to each standby manager 3, and / or the network traffic to each standby manager 3. The point which selects the candidate of the active system manager 3 according to quantity differs from the said embodiment.

  For example, in the embodiment, the node information transmission unit 34 transmits the free memory capacity and the free disk capacity as the node information. However, as the node information other than the free memory capacity and the free disk capacity, the CPU usage rate and Send the transmission time. When the transmission time is transmitted as the node information, the node information receiving unit 23 of the manager selection device 2 records the time taken for the node information to reach from the difference from the reception time of the node information in the node information recording unit 24. Therefore, it can be used as information for selecting the standby manager 3 in the manager selection means 21.

  In the above embodiment, the node information transmitted by the node information transmission unit 35 is recorded as it is as the information to be recorded in the node information recording unit 24. In addition, the information is acquired from the transmission packet and recorded. be able to. Examples of information that can be acquired from the transmission packet include the number of hops between the nodes 1 and the network line thickness. In this way, the manager selection means 21 selects the active manager according to the load of each standby manager 3, the number of network hops to each standby manager 3, the amount of network traffic to each standby manager 3, and the like. Can be selected.

  In the distributed computer system of the fourth embodiment configured as described above, the manager selection means 21 is configured so that the load on each standby system manager 3, the number of network hops to each standby system manager 3, and / or each standby system. Since the candidate of the active manager 3 is selected according to the traffic volume of the network up to the manager 3, even if the status of each node 1 or the network changes dynamically, the standby manager 3 suitable for the situation Can be selected as the working manager 3.

As described above, according to the embodiment of the present invention, the following effects can be obtained.
The first effect is that even when the load on the node changes from moment to moment, a standby manager suitable for the situation at that time can be selected as the active manager.
The reason for this is that if a standby manager that each manager selection means determines to be appropriate is selected according to the load information of each node where the standby manager is placed, it will be placed on the lightest node at that time according to the selection result. This is because the selected standby manager is selected as the active manager.

The second effect is that, even in a mobile network or an ad hoc network environment, a standby manager suitable for the situation at that time can be selected as an active manager.
The reason is that if the standby manager that each manager selection unit determines to be appropriate is selected according to the distance to each node where the standby manager is placed and the network performance, the most efficient network efficiency at that time can be obtained. This is because the standby manager placed in a good node is selected as the active manager.

The distributed computer system of the present invention has been described with reference to the embodiment. However, the distributed computer system according to the present invention is not limited to the above-described embodiment, and various modifications can be made within the scope of the claims. It goes without saying that is possible.
For example, in the embodiment, as shown in FIG. 7, the manager 3 is arranged in each node 1. However, in consideration of the balance between the reliability and the load of the node 1, the manager 3 is provided only in some nodes 1. May be a system configuration.

  In the above embodiment, the node information transmission unit 34 transmits the node information every 5 seconds. However, it is not necessarily 5 seconds. The balance between the accuracy of the information and the load of the network and the node is not necessarily required. You may decide any time. Further, it is not always necessary to send the node information regularly, and it is also possible to adopt a method of starting transmission of the node information when the absence of the active manager 3 is detected. In that case, when the manager selection means 21 reads node information in step A0 of FIG. 5, a procedure for waiting for completion of reception of all node information is required.

In the embodiment, only the latest node information is recorded as a method for recording the node information in the node information recording unit 24. However, a certain number of pieces of information or a certain number of information may be recorded as a history. .
In the above-described embodiment, the manager selection unit 21 selects only one standby manager 3, but it is not always necessary to select one, and a plurality may be selected. In this case, instead of giving the same number of votes to all the selected standby managers 3, a score representing the appropriateness may be given.

  In the embodiment, the selection method in the manager selection means 21 is determined by only one information. However, the selection is not limited to one information, and a plurality of information such as weighting and adding each information is added. It is also possible to judge in combination. When past history information is also recorded, it may be determined using a past fixed period or an average value of a certain number of pieces of information. There is also a method of selecting past history information that has little variation.

  In the above-described embodiment, a solution when there are a plurality of standby managers 3 of the same number in the number of votes obtained from the manager selection device 2 is not shown. It is possible to use a method in which a uniquely determined ranking table is stored in advance, and a method that uniquely determines the table, such as a method according to the table or a sequence in which IP addresses are younger.

  The present invention can be used in a distributed computer system that performs distributed processing with a plurality of nodes. In particular, when a failure occurs in a manager that manages a plurality of nodes, it can be suitably used in a distributed computer system that is required to ensure high reliability by automatically taking over the manager.

1 Node 2 Manager selection device 3 Manager (standby manager, active manager)
21 Manager selection unit 22 Selection result transmission unit 23 Node information reception unit 24 Node information recording unit 31 Selection result reception unit 32 Selection result recording unit 33 Selection result determination unit 34 Status change unit 35 Node information transmission unit 36 Arbitration unit

Claims (11)

The multiple nodes that make up the distributed computer system
One node with an active manager that assigns processing to each of these nodes;
A plurality of nodes provided with standby managers operable as active managers when a failure occurs in the active managers, and
Each of the nodes
Manager selection means for selecting a candidate for the active manager from the plurality of standby managers when a failure occurs in the active manager;
Selection result transmission means for transmitting the selection result to the selected standby manager,
Each of the standby managers
Selection result receiving means for receiving the selection result;
A selection result judging means for judging whether or not itself becomes an active manager based on the received selection result;
A distributed computer system, comprising: state change means that changes to an active manager according to the determination result. Each of the standby managers
Node information transmission means for transmitting information of the node provided with the standby manager to each of the nodes;
Each of the nodes
Node information receiving means for receiving the node information;
A node information recording unit for recording the received node information,
The manager selecting means includes
The distributed computer system according to claim 1, wherein a candidate for the active manager is selected from the plurality of standby managers based on the node information recorded in the node information recording unit. Each of the standby managers
Arrangement means for determining one standby manager to be changed to an active manager when the number of received selection results is below a predetermined ratio compared to the total number of standby managers is provided. The distributed computer system according to claim 1 or 2. The mediation means is
When the number of the selection results received is lower than a predetermined ratio compared to the total number of standby managers, the other standby managers are inquired about the number of selection results received by the other standby managers. The distributed computer system according to claim 3, wherein when there is no standby manager that receives a larger number of selection results than the standby manager, the distributed computer system is determined to be the active manager. The mediation means is
When the number of received selection results falls below a predetermined ratio compared to the total number of standby managers, the nodes are selected again as candidates for the active manager, and the selection results are transmitted to the selected standby manager. The distributed computer system according to claim 3, wherein: The selection result transmission means includes:
Send the selection result to all the standby managers;
The selection result judging means includes
3. The distributed computer system according to claim 1, wherein the received selection results are totaled, and when the selected number is the maximum, it is determined that the selected result becomes an active manager. 4. The manager selecting means includes
A candidate for the active manager is selected according to the load of each standby manager, the number of network hops to each standby manager, and / or the amount of network traffic to each standby manager. The distributed computer system according to claim 1. The multiple nodes that make up the distributed computer system
One node with an active manager that assigns processing to each of these nodes;
A plurality of nodes provided with standby managers operable as active managers when a failure occurs in the active managers, and
Each of the nodes
When a failure occurs in the active manager, a candidate for the active manager is selected from the plurality of standby managers,
Send the selection result to the selected standby manager,
Each of the standby managers
Receiving the selection result;
Based on the received selection result, it is determined whether or not it becomes an active manager,
A manager takeover method, wherein the manager is changed to an active manager according to the determination result. Each of the standby managers
Send information of the node provided with the standby manager to each of the nodes,
Each of the nodes
Receiving the node information;
Record the received node information in a node information recording unit,
The manager takeover method according to claim 8, wherein a candidate for the active manager is selected from the plurality of standby managers based on the node information recorded in the node information recording unit. The multiple nodes that make up the distributed computer system
One node with an active manager that assigns processing to each of these nodes;
A plurality of nodes provided with standby managers operable as active managers when a failure occurs in the active managers, and
The manager takeover program for each node is:
When a failure occurs in the active manager, a process of selecting a candidate for the active manager from the plurality of standby managers;
Processing to send the selection result to the selected standby manager, and causing the computer to execute,
The manager takeover program for each standby manager is:
Receiving the selection result;
Based on the received selection result, a process for determining whether or not it becomes an active manager;
A manager takeover program that causes a computer to execute processing that changes to an active manager in accordance with the determination result. The manager takeover program for each standby manager is:
Causing the computer to execute node information transmission processing for transmitting information on the node provided with the standby manager to each of the nodes;
The manager takeover program for each node is:
Processing to receive the node information;
A process of recording the received node information in a node information recording unit;
The computer is configured to execute a process of selecting an active manager candidate from the plurality of standby managers based on the node information recorded in the node information recording unit. Manager takeover program.

Images