JP2010257335A - Computer system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a rollback system in which active regular resources are reduced, and intensive control is not used. <P>SOLUTION: A computer system includes a first computer, a second computer and a third computer, and the respective computers are connected through a network. The first computer, the second computer and the third computer are configured to successively transmit and receive jobs for processing those jobs, and the first computer is configured to manage dependency showing a relation between the processing of the second computer and the processing of the other computer, and to determine whether or not the second computer is normally operating, and to, when determining that the second computer is not normally operating, obtain the identifier of the latest job processed by the second computer from the third computer, and to perform rollback based on the obtained identifier of the job. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a computer system, and more particularly to a computer system in which a plurality of hosts perform processing by cooperation through data transmission / reception.

  A system that cooperates by transferring data between hosts provides various functions including functions such as a database and network routing. However, when one host goes down in the system, other hosts linked to the down host cannot complete the traffic, and traffic saturation occurs one after another. As a result, a down in one host causes a large-scale failure that brings down the entire system. A method for avoiding a large-scale failure caused by a local failure, such as a large-scale failure that occurs when one host goes down, has been required and studied so far.

  There are two main methods that have been studied so far.

  The first method is to periodically acquire a static backup of the data held by the host in the system to the alternative host, and if the host that has acquired the backup detects that the host has gone down, the alternative host continues to operate. There is a hot standby to. As a technique for effectively implementing the first method, a rollback point specifying method, an alternative host specifying method, and the like have been conventionally proposed (for example, see Patent Document 1).

  The second method is a method for managing jobs by centralized control (see, for example, Patent Document 2). Each host successively notifies (messaging) its processing status to a central control host such as a resource manager, whereby the central control host centrally manages the job status. When a failure occurs in the host, the centralized control host determines a rollback point based on the status of the job processed in the host in which the failure has occurred.

JP 2001-43105 A JP-A-7-262073

  When the first method described above is used, each host needs to multiplex resources and data areas and always operate hardware corresponding to the multiplexed data areas. Therefore, a system using a hot standby has a very high running cost.

  On the other hand, when using the second method described above, the system does not need to always operate the resources required for taking a backup periodically, so compared with the first method, The running cost of the second method is low. However, when using the second method, the availability of the resource manager directly affects the availability of the system, so the second method essentially does not increase the overall system availability.

  Accordingly, an object of the present invention is to provide a rollback system that reduces resources that are always in operation and does not perform centralized control.

  According to a typical embodiment of the present invention, a computer system comprising a first computer, a second computer, and a third computer, wherein the computers are connected by a network, the first computer, The second computer and the third computer process the job by transmitting and receiving jobs in sequence, and the first computer performs processing of the second computer and processing of the other computers. If it is determined that the second computer is not operating normally, it is determined whether or not the second computer is operating normally. The identifier of the latest job processed in the computer is acquired from the third computer, and rollback is performed based on the acquired identifier of the job.

  According to an embodiment of the present invention, a rollback system with high availability can be provided.

It is explanatory drawing which shows the example of the process of embodiment of this invention. 1 is a block diagram illustrating an overall configuration of a system according to a first embodiment of this invention. It is a sequence diagram which shows the process of the 1st Embodiment of this invention. It is a flowchart which shows the dependence relationship information delivery process of the 1st Embodiment of this invention. It is a flowchart which shows the rollback process of the 1st Embodiment of this invention. It is explanatory drawing which shows the relationship between the hosts of the 1st Embodiment of this invention. It is explanatory drawing which shows the relationship between the hosts of the 2nd Embodiment of this invention. It is explanatory drawing which shows the structure of the whole system of the 2nd Embodiment of this invention. It is explanatory drawing which shows the structure of the whole system of the 3rd Embodiment of this invention. It is a sequence diagram which shows the process of the 3rd Embodiment of this invention. It is explanatory drawing which shows the structure of the whole system of the 4th Embodiment of this invention. It is a sequence diagram which shows the process of the 4th Embodiment of this invention. It is explanatory drawing which shows the combination of the flow of the process of embodiment of this invention.

  FIG. 1 is an explanatory diagram illustrating a specific example of processing according to the embodiment of this invention.

  The process shown in FIG. 1 is a process for obtaining approval necessary for ordering materials and transmitting data for ordering to a dealer in the business of placing orders for materials.

  The host A 1100 illustrated in FIG. 1 executes process A 1114, the host B 1200 processes process B 1205, the host C 1300 processes process C 1305, and the host D 1600 processes process D 1601. In process A 1114, purchase information is input from the outside, and an approval request sheet for managers is output to process B. In process B1205, the approval request sheet for managers is input from process A1114, and the approval request sheet for center managers is output to process C1305. In process C1305, the approval request sheet for the center manager is input from process B1205, and the order sheet for the dealer is output to process D1601. In process D1601, the order sheet for dealer is input from process C1305, and the input order sheet is processed.

  Process A1114, process B1205, process C1305, and process D1601 are sequentially executed in series. In the first, third, and fourth embodiments to be described later, processing up to processing C1305 is described, and in the second embodiment, processing up to processing D1601 is described.

  FIG. 2 is a block diagram showing the overall configuration of the system according to the first embodiment of this invention.

  In the system shown in FIG. 2, a host A 1100, a host B 1200, a host B ′ 1400, and a host C 1300 are connected by a network 1500. The host A 1100, host B 1200, host B ′ 1400, and host C 1300 are computers including a processor, a memory, an input device, an output device, and a network interface.

  The host A 1100 includes a storage device 1101 and a ring buffer 1103 in the memory. The storage device 1101 includes middleware A1102, rollback A1112, buffer push A1113, processing A1114, scheduled execution daemon 1115, and dependency relationship table 1116.

  The middleware A1102, the rollback A1112, the buffer push A1113, the process A1114, and the scheduled execution daemon 1115 are programs that are referenced and executed by the processor. The dependency relationship table 1116 is a set of data, and is transmitted from the host B described later and includes data indicating the dependency relationship between the hosts.

  The middleware A 1102 includes a dependency relationship table update 1110, a diagnosis 1111, and a determination 1119. The dependency table update 1110, diagnosis 1111, and determination 1119 are programs.

  The ring buffer 1103 includes an area 1117 for storing a head pointer and an area 1118 for storing data.

  The host B 1200 includes a storage device 1201 in the memory. The storage device 1201 includes middleware B 1202, processing B 1205, and processing B dependency table 1206. The middleware B 1202 includes dependency relationship information distribution 1203 and a consultation 1204. The dependency relationship information distribution 1203 and the consultation 1204 are programs. Processing B1205 is also a program. The process B dependency table 1206 is a set of data. The host that inputs data to the process B 1205, that is, the host that is the dependency source of the process B 1205, and the host that receives data from the process B 1205, that is, the dependency destination of the process B 1205 Indicates a host that is

  The host C 1300 includes a storage device 1301 in the memory. The storage device 1301 includes middleware C1302, processing C1305, and latest job list 1306. The middleware C1302 and the process C1305 are programs. The middleware C1302 includes a list update 1303 and a latest job search 1304.

  The host B ′ 1400 replaces the host B 1200 when a failure occurs in the host B 1200. Host B′1400 includes the same program as host B1200, but the dynamic data of host B1200 is not synchronized with host B′1400. In addition, the host B ′ 1400 is not operating when the host B 1200 is operating normally. The host B ′ 1400 starts when requested to start from another host.

  FIG. 3 is a sequence diagram showing processing of the first exemplary embodiment of the present invention.

  FIG. 3 shows processing executed by the computer system shown in FIG.

  With the processing illustrated in FIG. 3, the computer system according to the embodiment of this invention causes the host B ′ 1400 to execute the processing that has been executed in the host B 1200 when a failure occurs in the host B 1200.

  After the host B 1200 is activated (step 2001), the host B 1200 transmits the dependency relationship of the process B 1205 in the host B 1200 to the host A 1100 that is the dependency source of the process B 1205 by the dependency relationship information distribution 1203. Specifically, the host B 1200 transmits a process B dependency table 1206 including a dependency relationship [B-> C] indicating that the output of the host B 1200 is input to the host C 1300 to the host A 1100.

  When the host A 1100 receives the process B dependency table 1206, the host A 1100 depends on the dependency table update 1110 for the received dependency [B-> C] for the process B 1205 and the type ID uniquely assigned to the dependency relationship. Input to the relationship table 1116. The type ID is an arbitrary alphanumeric character. The dependency relationship table 1116 illustrated in FIG. 2 indicates the dependency relationship table 1116 after the dependency relationship update 1110 is executed, the type column indicates “T1”, and the data column indicates the dependency relationship [B-> C].

  Details of the dependency relationship information distribution 1203 will be described later with reference to FIG.

  After the host A 1100 updates the dependency table 1116 by the dependency table update 1110, the host A 1100 executes the process A 1114. Subsequent to the process A 1114, the host B 1200 executes the process B 1205. Subsequent to the process B1205, the host C1300 executes a process C1305.

  A series of processes of process A 1114, process B 1205, and process C 1305 cooperate by communicating data between hosts. In the embodiment of the present invention, a series of data units communicated between hosts is called a job. A series of jobs from the process A 1114 to the process C 1305 is assigned a unique identifier by the process A 1114. The identifier assigned by the process A 1114 is referred to as a job ID.

  The job ID includes information indicating in which host the data is processed. Therefore, the host A 1100, the host B 1200, and the host C 1300 always include information indicating in which host the data has been processed in the past.

  After the process A1114 is completed and the output of the process A1114 is transmitted to the process B1205, the host A1100 creates rollback data for restoring the process A1114, and the created rollback data is transferred to the ring buffer 1103 by the buffer push A1113. To enter. Rollback data is indicated by a job ID. Further, the buffer push A 1113 inputs the rollback data at the address indicated by the head pointer 1117. Then, the buffer push A 1113 updates the address of the head pointer 1117 to the address where the oldest job ID is input.

  Note that the oldest job ID in the ring buffer 1103 is the job ID of the address one more than the address indicated by the head pointer when the job IDs are input in the ascending order of the addresses in the ring buffer 1103. When the job IDs are input to the ring buffer 1103 in descending address order, the oldest job ID in the ring buffer 1103 is the job ID of the address one less than the address indicated by the head pointer. In this embodiment, the ring buffer 1103 receives job IDs in ascending order of addresses.

  The host C 1300 is a job in which “A” indicating the host A 1100 is displayed in the host column of the latest job list 1306 and “A” is displayed in the job ID column after the processing B 1205 in the host B 1200 is completed. Is recorded. Since the latest job list 1306 indicates the latest job ID, the list update 1303 overwrites the past job ID of the latest job list 1306 with the latest job ID after the processing B1205 in the host B 1200 is completed.

  The host A 1100 periodically diagnoses 1111 whether or not the host B 1200 is operating normally. The host A 1100 periodically executes the diagnosis 1111 using the scheduled execution daemon 1115. The diagnosis 1111 includes confirmation of the operation status of the network interface by the ping command and whether or not processing can be normally executed. After receiving the diagnosis 1111 from the host A 1100, the host B 1200 returns the operating status to the host A 1100 through the consultation 1204.

  When the host A 1100 receives the operation status from the host B 1200, the host B 1200 operates normally based on the execution status of the process included in the operation status, the time until the operation status is returned, and the content of the operation status according to the determination 1119. It is determined whether or not.

  If it is determined by the determination 1119 that the host B 1200 is operating normally, the host A 1100 repeats the diagnosis 1111 according to the scheduled execution daemon 1115 because there is no need to roll back.

  If it is determined by the determination 1119 that the host B 1200 is abnormal, the host A 1100 requests the host C 1300 to transmit the latest job ID and starts up the host B ′ 1400 in order to roll back. The host C 1300 may activate the host B ′ 1400 after the host C 1300 receives a request to transmit the latest job ID to the host A 1100.

  The host C 1300 executes the latest job search 1304 after receiving the latest job ID from the host A 1100. The host C 1300 searches the latest job search 1304 for the latest job ID among the jobs executed by the host A 1100, and returns the latest job ID as a result of the search to the host A 1100. When executing the latest job search 1304, the host C 1300 returns a job ID (hereinafter referred to as the latest job ID) in which the host column of the latest job list 1306 indicates “A” to the host A 1100.

  When the host A1100 receives the latest job ID, the host A1100 executes rollback A1112. Details of the rollback A 1112 will be described later with reference to FIG.

  The host B ′ 1400 is activated in accordance with the activation instruction transmitted from the host A 1100 (step 2002). Thereafter, the host B ′ 1400 transmits dependency information related to the process B 1205 executed in the host B ′ 1400 to the host A 1100 (dependency relationship information distribution 2003). The host A 1100 changes the dependence destination of the process A 1114 from the process B 1205 in the host B 1200 to the process B 1205 in the host B ′ 1400, and after determining in the determination 1119 that the host B 1200 is abnormal, until the process A 1114 is executed. The timing may be changed at any timing.

  Further, in the determination 1119, the host C 1300 is notified from the host A 1100 that the host B ′ 1400 has been activated instead of the host B 1200, and the dependence source of the process C 1305 is changed from the process B 1205 in the host B 1200 to the process in the host B ′ 1400. You may change to B1205. Alternatively, when the process B1205 in the host B′1400 is completed, the host C1300 is notified that the host B′1400 has started up instead of the host B1200, and the host of the process C1305 is changed from the process B1205 to the host B1200. You may change to process B1205 in B'1400.

  FIG. 4 is a flowchart showing the dependency relationship information distribution 1203 according to the first embodiment of this invention.

  The host B 1200 refers to the process B dependency table 1206 of the process B 1205 (step 11001). Based on the result of the reference in step 11001, the host B 1200 transmits information indicating the dependency relationship to the dependency source of the process B (step 11002). Specifically, the host B 1200 transmits an identifier indicating the host C 1300 that outputs the result of the process B 1205 to the host A 1100 that inputs data to the process B 1205. In step 11002 in the first embodiment, the host B 1200 transmits a dependency [B-> C] indicating that “the result of the process B in the host B 1200 is passed to the host C 1300” to the host A 1100.

  FIG. 5 is a flowchart showing rollback A1112 according to the first embodiment of this invention.

  After receiving the latest job ID from the host C 1300 (step 13001), the host A 1100 searches the job ID column of the ring buffer 1103 for an entry that matches the received latest job ID, and matches the received latest job ID. It is determined whether or not an entry exists (step 13002).

  The ring buffer 1103 deletes old job IDs when more jobs are input than a predetermined amount. For this reason, the ring buffer 1103 does not necessarily include the received latest job ID. In step 13002, if the received latest job ID does not exist in the ring buffer 1103, the host A 1100 selects the oldest job ID, that is, the job ID of the address indicated by the head pointer 1117 (step 13003).

  When the received latest job ID exists in the ring buffer 1103, the host A 1100 sets the job ID to be executed next to the received latest job ID as a rollback point (step 13004).

  After step 1303 or step 1304, the host A 1100 determines whether or not processing is possible in the host B ′ 1400 (step 13005). When the host B′1400 cannot process, the host A1100 waits for a predetermined time (13006) and waits until the host B′1400 can process.

  If it is determined in step 13005 that processing can be performed in the host B ′ 1400, the host A 1100 determines from the rollback point selected in step 13004 to the latest job in the host A 1100, that is, the head pointer in the ring buffer 1103. Up to a job having an address one less than the address indicated by 1117 is executed (rollback: step 13007), and rollback A1112 is terminated.

  In the first embodiment, the processes are coordinated in the order of host A1100, host B1200, and host C1300. When host B1200 fails, rollback by host A1100, host B1200, host C1300, and host B′1400 Showed cooperation. In the first embodiment, the host B ′ 1400 is a cool standby of the host B 1200.

  As described above, according to the first embodiment, the backup resource can be set as a cool standby, and rollback can be performed without using a resource manager that performs centralized control.

  FIG. 6 is an explanatory diagram showing the processing relationship of the first embodiment of the present invention.

  A4001, B4002, C4003, and B′4004 shown in FIG. 6 correspond to the host A1100, host B1200, host C1300, and host B′1400 in the first embodiment. Play an alternative role. If a failure occurs in B4002, B′4004 becomes an alternative host for B4002.

  When four or more hosts cooperate, a plurality of hosts may be replaced by a single cool standby host without changing the number of cool standby hosts. When three or more hosts cooperate with each other, the first embodiment may be implemented in a multiplex manner. The second embodiment is a computer system when the first embodiment is implemented in a multiple manner.

  FIG. 7 is an explanatory diagram showing the relationship of processing according to the second embodiment of the present invention.

  In the second embodiment, A5001, B5002, C5003, and D5004 shown in FIG. 7 are a combination of the relationships of A4001, B4002, and C4003 shown in FIG. A5001, B5002, and C5003 and B5002, C5003, and D5004 play upstream, central, and downstream roles, respectively. In addition, in the relationship between A5001, B5002, and C5003, B'5005 is an alternative to B5002. Further, in the relationship between B5002, C5003, and D5004, the alternative of C5003 is also B′5005.

  If a failure occurs in either B5002 or C5003, B′5005 can share the resources of B5002 or C5003. B′5005 plays the role of a cool standby of B5002 or C5003 by separately providing static software and settings provided in B5002 and C5003.

  FIG. 8 is an explanatory diagram showing the configuration of the entire system according to the second embodiment of this invention.

  The host A1100, host B1200, host C1300, host D1600, and host B′1400 shown in FIG. 8 play the same role as A5001, B5002, C5003, D5004, and B′5005 shown in FIG. When a failure occurs in either the host B 1200 or the host C 1300, the host B ′ 1400 becomes an alternative host for the host in which the failure has occurred.

  The host A 1100 shown in FIG. 8 is the same as the host A 1100 shown in FIG. Further, to the host B 1200 and the host C 1300 shown in FIG. 8, a set of software and data for performing the upstream and central roles shown in FIG. 7 is added to the host B 1200 and the host C 1300 shown in FIG. .

  A function corresponding to the function provided for the host A 1100 shown in FIG. 2 to play the upstream role shown in FIG. 6 is added to the host B 1200. That is, middleware A1102, rollback B3001, buffer push B3002, scheduled execution daemon 1115, dependency table 1116, and ring buffer 1103 are added to the host B1200 shown in FIG.

  Further, the host C1300 is added with a function corresponding to the function provided for the B1200 shown in FIG. 2 to play the central role shown in FIG. That is, the middleware B 1202 and the processing C dependency table 3306 are added to the host C 1300 illustrated in FIG.

  Further, the host D1600 has a function corresponding to the function provided for the C1300 shown in FIG. 2 to play the downstream role shown in FIG. The host D1600 is a computer including a processor, a memory, an input device, an output device, and a network interface. The host D1600 includes a storage device 1601 in a memory, and the storage device 1601 includes a middleware C1302, a process D1601, and a latest job list 1602.

  The computer system according to the second embodiment executes processing A 1114, processing B 1205, processing C 1305, and processing D 1601 illustrated in FIG. 1 by the host A 1100, host B 1200, host C 1300, and host D 1600 illustrated in FIG. When a failure occurs in the host B 1200 shown in FIG. 8, the system of the second embodiment rolls back by the same processing flow as the processing flow shown in FIG. When a failure occurs in the host C1300, the system of the second embodiment, in the processing flow shown in FIG. 3, the host B1200 performs the processing of the host A1, the host C1300 performs the processing of the host B1200, and the processing of the host C1300. When D1600 executes, it rolls back.

  According to the second embodiment, even when processing is performed by a plurality of hosts, it is possible to roll back using the first embodiment.

  In the first embodiment, the grasping of the dependency relationship and the diagnosis 1111 of the host B1200 are executed by the host A1100 which is the upstream role. In the third embodiment, the grasping of the dependency relationship and the diagnosis of the host B1200 are performed downstream. Executed by the host C1300.

  FIG. 9 is an explanatory diagram showing the configuration of the entire system according to the third embodiment of this invention.

  A host A 1100 illustrated in FIG. 9 includes middleware AA6001 instead of the middleware A1102 illustrated in FIG. The middleware AA 6001 includes a rollback daemon 6002.

  The host B 1200 shown in FIG. 9 includes middleware BB6003 instead of the middleware B1202 shown in FIG. The middleware BB 6003 includes dependency relationship information distribution 6004.

  A host C 1300 illustrated in FIG. 9 includes middleware CC6005 instead of the middleware C1302 illustrated in FIG. The scheduled execution daemon 1115 is provided in the host A 1100 in FIG. 2, but is provided in the host C 1300 in FIG. 2 is the same as the dependency relationship table 6007 provided in the host C 1300 in FIG. The middleware CC 6005 includes a dependency table update 6006, a diagnosis 1111, a determination 1119, a list update 1303, and a latest job search 1304.

  FIG. 10 is a sequence diagram showing processing of the third exemplary embodiment of the present invention.

  In the third embodiment, the host B 1200 transmits dependency relationship information distribution 6004 to the host C 1300. The host B 1200 transmits the dependency relationship to the host C 1300 that is the dependency destination of the process B 1205 based on the process B dependency table 1206 of the process B 1205. Specifically, the host B 1200 transmits the identifier of the host A 1100 that is the dependency source of the process B 1205 to the host C 1300 that is the dependency destination of the process B 1205. That is, the host B 1200 transmits a dependency relationship [A-> B] representing the meaning that “the result of the host A 1100 is passed to the host B 1200” to the host C 1300.

  After the dependency relationship is transmitted by the dependency relationship information distribution 6004, the host C 1300 updates the dependency relationship table 6007 using the received dependency relationship [A-> B] (dependency relationship update 6006).

  The host A 1100, the host B 1200, and the host C 1300 execute the process A 1114, the process B 1205, and the process C 1305, respectively.

  The host C 1300 periodically diagnoses whether the host B 1200 is operating normally. The host C 1300 periodically executes the diagnosis 1111 using the scheduled execution daemon 1115. The diagnosis 1111 is the same as the diagnosis 1111 in the first embodiment. After receiving the diagnosis 1111 from the host C 1300, the host B 1200 returns the operating status to the host C 1300 through the consultation 1204.

  Upon receiving the operating status from the host B 1200, the host C 1300 determines whether or not the host B 1200 is operating normally by a determination 1119.

  If it is determined by the determination 1119 that the host B 1200 is operating normally, the host C 1300 repeats the diagnosis 1111 according to the scheduled execution daemon 1115 because it is not necessary to roll back.

  If it is determined by the determination 1119 that the host B 1200 is abnormal, the host C 1300 activates the host B ′ 1400 (step 2001). Then, the host C 1300 searches the latest job list 1306 to search for the latest job ID (latest job search 1304), and transmits the latest job ID to the host A 1100.

  Note that the communication corresponding to the request for transmitting the latest job ID from the host A 1100 to the host C 1300 as executed in the first embodiment is unnecessary in the third embodiment. However, since the host A 1100 cannot use the diagnosis result of the host B 1200 as a rollback trigger in the third embodiment, the host A 1100 always monitors its network port by the rollback daemon 6002. When the host C 1300 transmits a rollback point that is the latest job ID by the latest job search 1304, the host A 1100 executes the rollback 1112 based on the received rollback point.

  The host B ′ 1400 receives the activation from the host C 1300 and activates (step 2002). Thereafter, the dependency relationship information regarding the process B 1205 is transmitted to the host C 1300. The host C1300 changes the dependency source of the process C1305 from the process B1205 in the host B1200 to the process B1205 in the host B′1400, and after determining in the determination 1119 that the host B1200 is abnormal, until the process C1305 is executed. The timing may be changed at any timing.

  Further, when the latest job ID is transmitted from the host C 1300 in the latest job search 1304, the host A 1100 may transmit that the host B ′ 1400 is activated instead of the host B 1200. The host A 1100 may change the dependence destination of the process A 1114 to the process B 1205 in the host B ′ 1400.

  In the third embodiment, the host A 1100 does not request the host C 1300 to send the latest job ID. For this reason, the third embodiment requires less communication from the host A 1100 to the host C 1300 than the first embodiment.

  On the other hand, since the host A 1100 continues to execute the rollback daemon 6002 in the third embodiment, the load on the host A 1100 in the third embodiment is higher than the load on the host A 1100 in the first embodiment. Get higher.

  In the fourth embodiment, the dependence source and dependence destination of the host B 1200 recognize the end of the processing B 1206 in the host B 1200, and the dependence source manages the unprocessed job in the host B 1200 by the list buffer instead of the ring buffer. Indicates the system.

  FIG. 11 is an explanatory diagram showing the configuration of the entire system according to the fourth embodiment of this invention.

  A host A 1100 illustrated in FIG. 11 includes a roll back A 8112, a buffer push A 8113, a buffer update 8114, and a list buffer 8103 instead of the roll back A 1112, the buffer push A 1113, and the ring buffer 1103 provided in the host A 1100 illustrated in FIG. . The buffer push A 8113 adds an entry to the list buffer 8103 and updates “o” to the flag of the added entry. In the buffer update, the flag of the job ID indicating the job processed by the host B 1200 is set to “x”. The list buffer 8103 includes an address column, a flag column, a job ID column, and a pointer column.

  A host B 1200 illustrated in FIG. 11 includes a dependency relationship information distribution 8203 instead of the dependency relationship information distribution 1203 illustrated in FIG.

  A host C 1300 illustrated in FIG. 11 includes a dependency table update 5110, a token back 8305, and a dependency table 5116 in addition to the functions of the host C 1300 illustrated in FIG.

  FIG. 12 is a sequence diagram showing processing of the fourth exemplary embodiment of the present invention.

  The host B 1200 transmits the dependency relationship information of the process B 1205 to both the host A 1100 and the host C 1300 (dependency relationship information distribution 8203). In the dependency relationship information distribution 8203, the dependency relationship information transmitted to the host A 1100 is the same as the dependency relationship transmitted to the host A 1100 in the dependency relationship information distribution 1203 of the first embodiment. Further, the dependency relationship information transmitted to the host C 1300 is the same as the dependency relationship transmitted to the host C 1300 in the dependency relationship information distribution 6004 of the third embodiment.

  The host A 1100 and the host C 1300 include a dependency relationship table 1116 and a dependency relationship table 5116, respectively. In addition, the host A 1100, the host B 1200, and the host C 1300 execute the process A 1114, the process B 1205, and the process C 1305. At this time, the process A 1114 transmits the job ID to the process B 1205. After the processing B1205 is completed and before the processing C1305 is executed, the host C1300 transmits to the host A1100 that the job coming from the host A1100 has been processed by the host B1200 (tokenback 8305). When the host A 1100 receives information that the job has been processed from the host C 1300, the host A 1100 deletes the job from the list buffer 8103 by the buffer update 8114 because the job has been processed by the host B 1200.

  The list buffer 8103 is a buffer having a list structure, and each entry includes a flag, a job ID, and a pointer to the next entry. The list buffer 8103 is updated at two timings.

  One timing is when the process A 1114 ends in the host A 1100. When the processing A 1114 ends in the host A 1100 and the resulting data is transmitted to the host B 1200, the host A 1100 adds a new job ID to the list buffer 8103 by the buffer push A 8113. The host A 1100 adds a job ID transmitted from the host A 1100 and processed in the host B 1200 to the end of the pointer of the list buffer 8103 by the buffer push A 8113, and sets the flag of the added job ID to “◯”.

  Another timing is when the processing B 1205 ends in the host B 1200, the job processed by the processing B 1205 arrives at the host C 1300, and the host C 1300 transmits the job ID of the arrived job to the host A 1100. . When the host B 1200 transmits from the host C 1300 that the job has been processed by the process B 1205, the host A 1100 updates the job ID flag indicating the completed job to “×” by the buffer update 8114.

  When the host A 1100 adds a job ID to the list buffer 8103 by the buffer push A 8113, the host A 1100 may add a job ID to an entry whose flag is “x”.

  After the buffer update 8114 is completed, the host A 1100 diagnoses the host B 1200 to determine whether or not the host B 1200 is operating normally. The host B 1200 receives the diagnosis 1111 from the host A 1100 and returns whether or not it is operating normally. The host A 1100 determines 1119 whether or not the host B 1200 is operating normally. If it is operating normally, the host A 1100 executes the next diagnosis 1111 according to the scheduled execution daemon 1115.

  When it is determined in the determination 1119 that the host B 1200 is abnormal, the host A 1100 executes rollback A 8112. The rollback A 8112 is a process in which steps 13001 to 13004 in the flowchart shown in FIG. 5 are replaced with a step of selecting a job ID with a flag “◯” from the list buffer 8103 and using it as a rollback point.

  If it is determined in the determination 1119 that the host B 1200 is abnormal, the host A 1100 activates the host B ′ 1400 2002.

  After the host B′1400 is activated 2002 according to the host A1100, the dependency relationship information distribution 2003 transmits the dependency relationship of the process B1205 in the host B′1400 to the host A1100.

  The timing at which the host A 1100 and the host C 1300 change the dependency destination of the process A 1114 and the dependency source of the process C 1305 when the host B ′ 1400 is activated instead of the host B 1200 is the same as in the first embodiment. .

  According to the fourth embodiment, the process A 1114 can recognize whether or not the process B 1205 has finished the process. Unlike the ring buffer 1103, the list buffer 8103 can manage individual jobs. Therefore, according to the fourth embodiment, in a system having a processing flow (Out-Of-Order) in which the job order changes, it is possible to roll back the jobs without duplication.

  Note that the third embodiment and the fourth embodiment can be applied to a system including four or more hosts as in the second embodiment described above. In this case, by selecting a set of three hosts that successively execute jab processing, and causing each selected host set to perform the operation of the third embodiment or the fourth embodiment, The effect of this embodiment can be obtained.

  As described above, the process A 1114, the process B 1205, the process C 1305, and the process D 1601 in the embodiment of the present invention are each executed by physically separated hosts. However, the above-described hosts do not need to be physically separated, and may be a plurality of virtual machines created in one computer.

  FIG. 13 is an explanatory diagram showing a process flow pattern according to the embodiment of this invention.

  Pattern 1 and pattern 2 shown in FIG. 13 are processing flow patterns in the first embodiment. Further, pattern 3 and pattern 4 shown in FIG. 13 are processing flow patterns in the third embodiment. The dependency relationship shown in FIG. 13 is shown as the relationship between host A, host B, host C, and host D.

  Pattern 1 is a pattern processed from host A to host B and from host B to host C and host D. In pattern 1, the dependence destination of host B is host C and host D. The host B in the pattern 1 transmits the dependency [B-> (C, D)] to the host A. When the host B stops operating, the host A sends a request for the latest job search to both the host C and the host D. Then, the rollback point is determined according to the latest job search result by the host C and the host D.

  Pattern 2 is a pattern in which processing flows from host A to host B and from host B to host C or host D. In the pattern 2, the dependence destination of the host B is either the host C or the host D according to the result of processing executed in the host B. The host B in the pattern 2 also transmits the dependency [B-> (C, D)] to the host A. When the host A of the pattern 2 stops operating, the latest job search request is sent to both the host C and the host D, and rollback is performed according to the latest job search result by the host C and the host D. Determine points.

  In pattern 3, processing flows from host A or host B to host C, and from host C to host D. In Pattern 3, the dependency source of the host C is the host A or the host B. The host C in the pattern 3 transmits the dependency [A-> C] or the dependency [B-> C] to the host D. When the host C stops operating, the host D searches for the latest job, and if the latest job is a job that flows from the host A, the latest job ID is transmitted to the host A, and the latest job flows from the host B. If it is a received job, the latest job ID is transmitted to the host B. Host A or host B determines a rollback point according to the latest job ID from host B.

  Pattern 4 starts processing at host C on condition that processing by host A and host B ends, and processing flows from host C to host D. In pattern 4, the dependency sources of the host C are the host A and the host B. The host C in the pattern 4 transmits the dependency [(A, B)-> C] to the host D. When the host C stops operating, the host D searches for the latest job and transmits the latest job ID to both the host A and the host B. Then, host A and host B determine a rollback point according to the latest received job ID.

  According to Pattern 1 to Pattern 4 shown in FIG. 13, the system of the above-described embodiment can be applied even when an input or output to one process is derived from a plurality of processes.

  The computer system proposed in the above embodiment includes a plurality of hosts, has high availability, and is low in cost.

1100, 1200, 1300, 1400 Host 1500 Network 1101, 1201, 1301, 1601 Storage device 1103 Ring buffer 8103 List buffer

Claims (12)

A computer system comprising a first computer, a second computer, and a third computer, wherein the computers are connected by a network,
The first computer, the second computer, and the third computer process the job by transmitting and receiving the job in order,
The first calculator is:
Managing the dependency indicating the relationship between the processing of the second computer and the processing of the other computer;
Determining whether the second computer is operating normally;
If it is determined that the second computer is not operating normally, the latest identifier of the job processed in the second computer is acquired from the third computer,
A computer system that performs rollback based on the acquired identifier of the job. The first computer requests an identifier of the latest job processed in the second computer from the third computer,
The computer system according to claim 1, wherein the third computer transmits the requested latest job to the first computer. The computer system further includes a fourth computer,
When it is determined that the second computer is not operating normally, the first computer starts the fourth computer to execute processing instead of the second computer,
The said 4th computer transmits the dependency which shows the relationship between the process of the said 4th computer, and the process of the said other computer to the said 1st computer, The Claim 1 or 2 characterized by the above-mentioned. Computer system.   The computer system according to claim 1, wherein the first computer includes a ring buffer for managing an identifier of the job. The computer system further includes a fifth computer,
The first computer, the second computer, the third computer, and the fifth computer process the job by transmitting and receiving the job in order,
Roll back the first computer in a combination of the first computer, the second computer and the third computer;
5. The computer system according to claim 1, wherein the second computer is rolled back in a combination of the second computer, the third computer, and the fifth computer. 6. . A computer system comprising a first computer, a second computer, and a third computer, wherein the computers are connected by a network,
The first computer, the second computer, and the third computer process the job by transmitting and receiving the job in order,
The third computer is
Managing the dependency indicating the relationship between the processing of the second computer and the processing of the other computer;
Determining whether the second computer is operating normally;
When it is determined that the second computer is not operating normally, the latest identifier of the job processed in the second computer is transmitted to the first computer,
The computer system is characterized in that the first computer rolls back based on the identifier of the transmitted job. The computer system further includes a fourth computer,
When it is determined that the second computer is not operating normally, the third computer starts up the fourth computer to execute processing instead of the second computer,
The computer according to claim 6, wherein the fourth computer transmits a dependency relationship indicating a relationship between the processing of the fourth computer and the processing of the other computer to the third computer. system.   The computer system according to claim 6 or 7, wherein the third computer includes a ring buffer for managing an identifier of the job. The computer system further includes a fifth computer,
The first computer, the second computer, the third computer, and the fifth computer process the job by transmitting and receiving the job in order,
Roll back the first computer in a combination of the first computer, the second computer and the third computer;
9. The computer system according to claim 6, wherein the second computer is rolled back in a combination of the second computer, the third computer, and the fifth computer. . A computer system comprising a first computer, a second computer, and a third computer, wherein the computers are connected by a network,
The first computer, the second computer, and the third computer process the job by transmitting and receiving the job in order,
The first computer and the third computer manage a dependency indicating a relationship between the processing of the second computer and the processing of the other computer,
The third computer sends an end status indicating that the job has been processed in the second computer to the first computer;
The first calculator is:
Determining whether the second computer is operating normally;
When it is determined that the second computer is not operating normally, rollback is performed based on the transmitted end status. The computer system further includes a fourth computer,
When it is determined that the second computer is not operating normally, the first computer starts the fourth computer to execute processing instead of the second computer,
The computer system according to claim 10, wherein the fourth computer transmits a dependency relationship of processing in the fourth computer to the third computer.   The computer system according to claim 10 or 11, wherein the first computer includes a list buffer for managing an identifier of the job.

Images