JP2015022353A - Snapshot creation system of data management server, and snapshot creation system of server cluster - Google Patents

Abstract

PROBLEM TO BE SOLVED: To release data locked due to snapshot creation in a short time.SOLUTION: A cluster system 1 includes data management servers 3-1 to 3-3, and a distribution server 2. The distribution server 2 includes a distribution unit 21 and a distribution table 22. Each of the data management server 3 includes: a table management unit 32 which manages a data management table 341; a data management unit 33 which manages data 343; a snapshot creation unit 35 which write-locks the distribution table 22 and the data management table 341 to create a snapshot, releases the locked state, and creates a snapshot without write-locking the data 343; a data write reference processing unit 31; and a storage unit 34 which stores the data 343, the data management table 341, and snapshot data 344.

Description

  The present invention relates to a system for creating a snapshot at the same time in a data management server or a server cluster including a plurality of data management servers for the purpose of holding and managing data.

The amount of data handled by public services and the amount of data handled by specific organizations such as companies and universities are increasing year by year. In the past, there were many cases where data was centrally managed on a single server, and multiple servers individually managed data. When the data to be managed needs reliability, there are many cases where the data to be managed is duplicated.
However, in the second half of the 2000s, a cluster system composed of a cluster of a plurality of data management servers and one or a plurality of distribution servers has been introduced and used. This cluster system behaves like a single data management server that manages a large amount of data for external clients, and is internally configured as a plurality of data management servers that divide and manage a large amount of data.

  In this cluster system, the distribution server is a front end to the client. The distribution server accepts the data write request and data reference request transmitted by the client on behalf of the client and transfers them to the appropriate data management server. With this mechanism, the cluster system behaves like a single data management server for clients. Even if one of the data management servers is broken, this cluster system can prevent the loss of the data by a technique of copying individual data and holding it in a plurality of data management servers.

  In a database composed of a distribution server and a data management server, a management method divided into information that is a key of each data and information (value) that is each data itself is often used. These databases realize scalability by directly connecting the hash of the key value and the storage position of the value value, and managing a table that connects the hash of the key value and the storage position of the value value.

  These data management servers often manage each data with a non-volatile medium such as a hard disk. In addition, the data management server generally manages each data with a volatile memory such as a RAM (Random Access Memory) when the response is important. In such a case, the data management server may back up the data on the volatile memory to a nonvolatile medium.

The cluster system can increase the number of data management servers as the amount of data handled increases. When the cluster system increases the number of data management servers, it is desirable that data can be moved to a new data management server in order to level the load on each data management server. When such high data fluidity is required, the cluster system may not back up the data on the volatile memory to a nonvolatile medium. In such a cluster system, replicated data is stored in a plurality of servers, and therefore, if only one data management server is broken, no data loss occurs. However, in this cluster system, data loss occurs when a plurality of data management servers are broken at a time due to a catastrophic disaster or the like. Therefore, in the cluster system, data backup is necessary to avoid data loss.
In a cluster system, data backup is performed by creating a snapshot of data at an arbitrary timing.

  Non-Patent Document 1 is for creating a snapshot of a cluster system. Non-Patent Document 1 states that “The database staticization function is a function for executing batch processing such as daily / monthly and database backup acquisition in parallel with online processing. By creating a static (consistent state with complete transaction processing) database, the online system will be supported 24 hours a day. "

  TMS-4V / SP described in Non-Patent Document 1 manages each data by duplicating the data by duplex writing by a pair device. When creating a snapshot, TMS-4V / SP temporarily stops receiving data update requests, stops changing the data, cancels duplication, and performs a single operation to staticize the data. . After the quiescence, TMS-4V / SP continues to operate with one of the duplicated data and generates a snapshot with the other data. Thereby, TMS-4V / SP can back up the data of arbitrary time.

Hitachi, "TMS-4V / SP: Database static function", [online], [Search June 19, 2013] Internet <URL: http://www.hitachi.co.jp/Prod/comp/ soft1 / 4vsp / products / staticizer1.html>

A general method of duplicating data is a method of managing data twice on a nonvolatile medium and a method of managing data twice on a volatile memory and a nonvolatile medium. The method of double data management on the volatile memory is not common because all servers require a large amount of RAM and are expensive.
In a system with high data fluidity, when viewed from a single server, each data is only a single data on a volatile memory. In this system, since each data is not duplicated, it is difficult to staticize each data in order to create a snapshot.

  The data managed by unification can be collected by taking a snapshot by locking them so that they cannot be updated. However, according to this method, the server must lock data updates while collecting snapshots of all data. Data snapshot collection takes a long time. Therefore, this method cannot be applied to a system that needs to update data at any time or a system that has high real-time characteristics and cannot stop updating data.

  The present invention provides a snapshot creation system for a data management server and a snapshot creation system for a server cluster capable of solving the above-described problems and releasing the lock of data for creating a snapshot in a short time. The task is to do.

In order to solve the above-mentioned problem, in the invention according to claim 1, the server cluster snapshot creation system includes a plurality of data management servers and a distribution server. The distribution server includes a distribution unit that distributes external requests to the data management server, and a distribution table that indicates correspondence between external requests and the data management server. Each of the data management servers writes a table management unit that manages a data management table including an accommodation position of each data, a data management unit that manages each data, the distribution table, and the data management table After creating a snapshot by locking, the lock is released, and the snapshot creation unit that creates a snapshot without writing and locking each data, and writing or referencing data based on an external request A data writing reference processing unit to be performed; and a storage unit for storing each data, the data management table, and snapshot data created by the snapshot creation unit.
If the data write reference processing unit receives a write request related to non-updated data after the start of snapshot creation during the snapshot creation of the snapshot creation unit, the data write reference processing unit relates to the write request. Copy unupdated data. The data write reference processing unit further makes a write request for the copied data and sets the updated attribute to be the updated data, and the unupdated attribute related to the write request is the attribute before the update. Is set as pre-update data.

  In this way, according to the present invention, data lock for creating a snapshot can be released in a short time.

  In the second aspect of the present invention, if the data write reference processing unit receives a data reference request from outside during the snapshot creation of the snapshot creation unit, based on the data management table, 2. The server cluster snapshot creation system according to claim 1, wherein each data is referred to.

  As described above, the data write reference processing unit can refer to each data by referring to the data management table in the same manner as in the normal time even during the creation of the snapshot. Thereby, the overhead of the time calculation amount accompanying snapshot creation can be eliminated.

  According to a third aspect of the present invention, the snapshot creation unit collects each non-updated data and each pre-update data after the snapshot creation start time to create a snapshot. The server cluster snapshot creation system described in 1.

  Thereby, the snapshot creation part can back up all the data at the time of starting snapshot creation.

  The invention according to claim 4 is characterized in that, when the snapshot creation is completed, the snapshot creation unit deletes each pre-update data and sets an unupdated attribute to each post-update data. The server cluster snapshot creation system according to claim 3.

  As a result, each data management server 3 can reduce the amount of space calculation during normal times and allocate more resources for data management.

  In the invention according to claim 5, if the data write reference processing unit receives a request for deleting unupdated data from the outside during the snapshot creation by the snapshot creation unit, the data write reference processing unit 2. The server cluster snapshot creation system according to claim 1, wherein an attribute before update is set in the update data.

  As described above, the data write reference processing unit moves the data to be deleted to the pre-update set when the snapshot is being created and the attribute of the data to be deleted is not updated. As a result, the data management server can remove the data to be deleted from the target of the subsequent write update processing while keeping the target of snapshot creation.

  In the invention according to claim 6, if the data writing reference processing unit receives a new data creation request from the outside during the snapshot creation of the snapshot creation unit, the data related to the new creation request The server cluster snapshot creation system according to claim 1, wherein the updated attribute is set in.

  As described above, the data write reference processing unit creates newly created data only in the updated set during snapshot creation, and does not create data in the pre-update set. Thereby, the data management server can exclude the data newly created during the creation of the snapshot from the target of the creation of the snapshot.

  The invention according to claim 7, wherein the data management unit manages attributes of the data based on a second data management table stored in the storage unit. Server cluster snapshot creation system.

  As described above, the data management server manages the data at the snapshot creation start time by the second data management table. Therefore, the data management server can create a snapshot at an arbitrary timing by only duplicating updated data without duplicating the entire data.

In the invention according to claim 8, a data management server that is a snapshot creation system includes a table management unit that manages a data management table including an accommodation position of each data, a data management unit that manages each data, The data management table is write-locked to create a snapshot, then the lock is released, and the snapshot creation unit creates a snapshot without writing-locking each of the data, and data based on an external request A data writing reference processing unit for writing or referencing, and a storage unit for storing the respective data, the data management table, and snapshot data created by the snapshot creating unit.
If the data write reference processing unit receives a write request related to non-updated data after the start of snapshot creation during the snapshot creation of the snapshot creation unit, the data write reference processing unit relates to the write request. Copy unupdated data. The data write reference processing unit further makes a write request for the copied data and sets the updated attribute to be the updated data, and the unupdated attribute related to the write request is the attribute before the update. Is set as pre-update data.

  In this way, according to the present invention, data lock for creating a snapshot can be released in a short time.

  ADVANTAGE OF THE INVENTION According to this invention, the snapshot creation system of the data management server which can cancel | release the update lock of the data for snapshot creation in a short time, and the snapshot creation system of a server cluster can be provided.

1 is a schematic configuration diagram showing a cluster system in a first embodiment. It is a sequence diagram which shows the data processing at the normal time. It is a sequence diagram which shows a snapshot creation process (part 1). It is a sequence diagram which shows a snapshot creation process (part 2). It is a figure which shows a data management table and a 2nd data management table. It is a figure which shows snapshot creation operation | movement and data update operation | movement. It is a flowchart which shows a data writing reference process. It is a flowchart which shows a data update process. It is a flowchart which shows a new creation process. It is a flowchart which shows a data deletion process. It is a flowchart which shows a snapshot creation process. It is a flowchart which shows a snapshot creation post-process. It is a figure which shows each set and each data during snapshot creation. It is a schematic block diagram which shows the data management server in 2nd Embodiment.

Next, modes for carrying out the present invention (referred to as “embodiments”) will be described in detail with reference to the drawings as appropriate.
(First embodiment)
FIG. 1 is a schematic configuration diagram showing a cluster system 1 in the first embodiment.
The cluster system 1 is a server cluster snapshot creation system including a distribution server 2, a plurality of data management servers 3-1 to 3-3, and a snapshot creation management server 4. Hereinafter, when the data management servers 3-1 to 3-3 are not particularly distinguished, the data management server 3 may be simply described. In each figure, the data management server 3-1 is “data management server # 1”, the data management server 3-2 is “data management server # 2”, and the data management server 3-3 is “data management server # 3”. May be indicated.

  When the cluster system 1 receives a data write request signal or a data reference request signal from the outside, the cluster system 1 performs data write processing or data reference processing based on the request. The cluster system 1 behaves like a single data management server 3 that manages a large amount of data to the outside. Further, when receiving a snapshot creation request signal from the outside, the cluster system 1 creates a snapshot of the data based on the request.

  The distribution server 2 is a front end for an external device (not shown), accepts a data write request signal or a data reference request signal from the external device (not shown), and receives data management servers 3-1 to 3. -3. The distribution server 2 distributes each request signal to any one of the data management servers 3-1 to 3-3, a distribution table indicating correspondence between each request signal and the distribution destination of the request signal. 22.

The data management servers 3-1 to 3-3 each include a data write reference processing unit 31, a table management unit 32, a data management unit 33, a storage unit 34, and a snapshot creation unit 35. ing. The data management servers 3-1 to 3-3 perform write processing and reference processing of data to be stored based on the request signals distributed by the distribution server 2.
The data write reference processing unit 31 processes each received data write request signal and data reference request signal. If the data write reference processing unit 31 receives a write request related to the non-updated data after the snapshot creation start from the outside during the snapshot creation, the data write reference processing unit 31 copies the non-updated data related to the write request. Then, a write request is made to the copied data and the updated attribute is set as the updated data. Further, the data write reference processing unit 31 sets the pre-update attribute to the non-updated data related to this write request and sets it as pre-update data.

  The table management unit 32 manages a data management table 341 and a second data management table 342 which will be described later. The data management table 341 and the second data management table 342 include information on the accommodation position of each data. Based on the key name of the processing target data included in each request signal, the table management unit 32 responds with the accommodation position of the processing target data.

The data management unit 33 manages each data. The data management unit 33 performs writing processing and reference processing of the processing target data based on the storage location of the processing target data.
The storage unit 34 stores a data management table 341, a second data management table 342, data 343, and snapshot data 344. Since the storage unit 34 is composed of a volatile memory, writing and reference to each data and each table can be performed in a short time.

  The data management table 341 stores the relationship between the key value of each data and information such as the accommodation position of each data. The second data management table 342 stores attribute information of each data in addition to the same information as the data management table 341. Data 343 is each data body to be processed. The snapshot data 344 is a snapshot of the data management table 341 and the data 343 at an arbitrary time.

  The snapshot creation unit 35 creates a snapshot of the data management table 341 and data 343 stored in the data management server 3 in response to a snapshot creation request signal received from the outside. The snapshot creation unit 35 locks the allocation table 22 and the data management table 341 in writing and creates a snapshot. Then, the snapshot creation unit 35 releases the writing lock and snapshots the data 343 without writing lock. Create

  The snapshot creation management server 4 includes a snapshot creation management unit 41. The snapshot creation management server 4 is a front end for the outside. The snapshot creation management server 4 receives a snapshot creation request signal from the outside by the snapshot creation management unit 41 and transfers it to the data management servers 3-1 to 3-3.

FIG. 2 is a sequence diagram showing normal data processing.
(Processing for data reference request signal)
If a data reference request signal is transmitted from the external device (not shown) to the distribution server 2, the processes shown in sequences Q10 to Q19 are started. In the following sequence diagrams, the data reference request signal is abbreviated as a data reference request, and the data write request signal is abbreviated as a data write request.
In sequence Q10, the distribution unit 21 of the distribution server 2 receives a data reference request signal from an external device (not shown). Further, the allocating unit 21 determines which data management server 3 stores the data stored in the data management servers # 1 to # 3. If the already distributed data is a processing target, the distribution unit 21 specifies the data management server 3 to be distributed based on the distribution table 22.
In sequence Q11, the distribution unit 21 of the distribution server 2 distributes the data reference request signal to any of the data management servers # 1 to # 3. The distributed data reference request signal is received by the data write reference processing unit 31 of the corresponding data management server 3.

In sequence Q12, the data write reference processing unit 31 inquires the table management unit 32 about the accommodation position of the processing target data corresponding to the key value included in the data reference request signal. This inquiry is input to the table management unit 32.
In sequence Q13, the table management unit 32 searches the data management table 341 based on the input key value, and acquires the data storage position corresponding to the key value.
In sequence Q14, the table management unit 32 responds to the data write reference processing unit 31 with the data storage position corresponding to the key value included in the data reference request signal. The response of the accommodation position is input to the data writing reference processing unit 31.

In sequence Q15, the data write reference processing unit 31 notifies the data management unit 33 of the data reference request signal and the accommodation position of the processing target data. This notification is input to the data management unit 33.
In sequence Q16, the data management unit 33 refers to the processing target data based on the accommodation position of the processing target data.

In sequence Q17, the data management unit 33 responds to the data write reference processing unit 31 with the reference result of the processing target data. The response of the reference result is input to the data write reference processing unit 31.
In sequence Q <b> 18, the data write reference processing unit 31 returns the reference result of the processing target data to the distribution unit 21 of the distribution server 2. The reference result response is received by the distribution unit 21.
In sequence Q19, the allocating unit 21 of the allocating server 2 returns a reference result of the processing target data to an external device (not shown) that is the transmission source of the data write request signal. Thereby, the process for the data reference request signal is completed.

(Processing for data write request signal)
When a data write request signal is transmitted from the external device (not shown) to distribution server 2, the processes shown in sequences Q20 to Q29 are started. The processing of sequences Q20 to Q29 is different from the processing of sequences Q10 to Q19 described above, and the processing type is data writing instead of data reference. Note that data writing is one of three processes: data update, new data creation, and data deletion.
In sequence Q20, the sorting unit 21 of the sorting server 2 receives a data write request signal from an external device (not shown). The allocating unit 21 determines which data management server 3 among the data management servers # 1 to # 3 is to write to the data stored. If the already distributed data is a processing target, the distribution unit 21 specifies the data management server 3 to be distributed based on the distribution table 22. If the data that has not yet been distributed is to be processed, the distribution unit 21 specifies the data management server 3 to be distributed by a hash function, and records the relationship between the specified data management server 3 and the data in the distribution table 22. To do.

  In sequence Q21, the distribution unit 21 of the distribution server 2 distributes the data write request signal to any one of the data management servers # 1 to # 3. The distributed data write request signal is received by the data write reference processing unit 31 of the corresponding data management server 3.

In sequence Q22, the data write reference processing unit 31 inquires of the table management unit 32 about the accommodation position of the processing target data corresponding to the key value included in the data write request signal. This inquiry is input to the table management unit 32.
In sequence Q23, the table management unit 32 searches the data management table 341 based on the key value, and acquires the data storage position corresponding to the key value.
In sequence Q24, the table management unit 32 responds to the data write reference processing unit 31 with the data storage position corresponding to the key value included in the data write request signal. The response of the accommodation position is input to the data writing reference processing unit 31.

In sequence Q25, the data write reference processing unit 31 notifies the data management unit 33 of the data write request signal and the storage location of the processing target data. This notification is input to the data management unit 33.
In sequence Q26, the data management unit 33 writes the processing target data based on the accommodation position of the processing target data.

In sequence Q27, the data management unit 33 responds to the data write reference processing unit 31 with the result of writing the processing target data. This write result response is input to the data write reference processor 31.
In sequence Q28, the data write reference processing unit 31 responds to the distribution unit 21 of the distribution server 2 with the result of writing the processing target data. The response of the writing result is received by the distribution unit 21.
In sequence Q29, the allocating unit 21 of the allocating server 2 returns a result of writing the processing target data to an external device (not shown). This write result response is received by an external device (not shown) which is the transmission source of the data write request signal. Thereby, the process for the data write request signal is completed.

3 and 4 are sequence diagrams showing snapshot creation processing. FIG. 3 shows a sequence Q40 to Q61 of the snapshot creation process. FIG. 4 shows a sequence Q62 to Q66 of snapshot creation processing following FIG.
If a snapshot creation request signal is transmitted from an external terminal (not shown) to the snapshot creation management server 4, the snapshot creation processing shown in sequences Q40 to Q66 starts.

In sequence Q40, the snapshot creation managing unit 41 of the snapshot creation management server 4 receives a snapshot creation request signal from an external terminal (not shown).
In sequence Q <b> 41, the snapshot creation managing unit 41 of the snapshot creation management server 4 transmits a lock instruction for the distribution table 22 to the distribution server 2. The lock instruction of the distribution table 22 is received by the distribution unit 21 of the distribution server 2.
In sequence Q42, the distribution unit 21 locks the writing of the distribution table 22.
In sequence Q43, the allocating unit 21 returns a lock completion response indicating that the writing table 22 has been locked to the snapshot creation management server 4. This lock completion response is received by the snapshot creation management unit 41 of the snapshot creation management server 4.

  In sequence Q44, the snapshot creation management unit 41 transmits a table lock instruction to each data management server 3 constituting the cluster system 1. These table lock instructions are received by the snapshot creation unit 35 of each data management server 3. This process is repeated for the number of data management servers 3.

In sequence Q45, the snapshot creation unit 35 notifies the table management unit 32 of a lock instruction for the data management table 341. This lock instruction is input to each table management unit 32.
In sequence Q46, the table management unit 32 locks the writing of the data management table 341.
In sequence Q47, the table management unit 32 returns a lock completion response of the data management table 341 to the snapshot creation unit 35. This lock completion response is returned to the snapshot creation unit 35. This sequence Q47 is for the case where the write lock of the table management unit 32 is successful.

In the operation example shown in FIGS. 3 and 4, the entire data management table 341 is write-locked. However, the present invention is not limited to this, and the data management server 3 may perform a partial write lock or read / write lock of the data management table 341. Furthermore, the data management server 3 may select what kind of locking is to be performed according to the parameter of the snapshot creation request signal received from the snapshot creation management server 4.
In sequence Q48, the snapshot creation unit 35 returns a table lock completion response to the snapshot creation management server 4. This process is repeated for the number of data management servers 3. This table lock completion response is received by the snapshot creation control unit 41.
In sequence Q49, the snapshot creation managing unit 41 checks whether a table lock completion response is returned from all the data management servers 3 of the cluster system 1 (response check). If table lock completion responses have been returned from all the data management servers 3, the distribution unit 21 of the distribution server 2 performs the processing of sequence Q50 and subsequent steps.

In sequence Q <b> 50, the snapshot creation supervision unit 41 transmits a copy instruction for the distribution table 22 to the distribution server 2. This copy instruction is received by the distribution unit 21. Here, copying of the distribution table 22 means creating a snapshot by duplicating the distribution table 22 and making it stationary.
In sequence Q51, the distribution unit 21 creates a snapshot by executing a copy of the distribution table 22.
In sequence Q52, the allocating unit 21 releases the write lock of the allocating table 22.
In sequence Q53, the distribution unit 21 returns a copy completion response to the snapshot creation management server 4. This copy completion response is received by the snapshot creation control unit 41.

In sequence Q54, the snapshot creation supervising unit 41 transmits a snapshot creation instruction to each data management server 3. This process is repeated for the number of data management servers 3. This snapshot creation instruction is received by the snapshot creation unit 35.
In sequence Q55, the snapshot creation unit 35 notifies the table management unit 32 of a copy instruction for the data management table 341. This copy instruction is input to the table management unit 32.
In sequence Q56, the table management unit 32 creates a snapshot by copying the data management table 341. The following sequence is for a case where copying by the table management unit 32 is successful.

In sequence Q57, the table management unit 32 creates the second data management table 342 based on the data management table 341. That is, the table management unit 32 creates the second data management table 342 by copying the data management table 341 and adding an attribute column 342n (see FIG. 5B).
In sequence Q58, the table management unit 32 releases the write lock of the data management table 341.
In sequence Q59, the table management unit 32 returns a copy completion response to the snapshot creation unit 35. This copy completion response is input to the snapshot creation unit 35.

In sequence Q60, the snapshot creation unit 35 returns a copy instruction reception response to the snapshot creation management server 4. This copy instruction reception response is received by the snapshot creation control unit 41 for the number of copies. This copy instruction reception response is a primary response, and each data management server 3 continuously creates snapshots.
In sequence Q61, the snapshot creation management unit 41 checks whether the copy instruction reception responses have been returned from all the data management servers 3. When the process of sequence Q61 ends, the process of sequence Q62 of FIG. 4 is performed.

As shown in FIG. 4, in sequence Q62, the snapshot creation unit 35 performs a snapshot creation process in cooperation with the table management unit 32, and creates a snapshot. The snapshot creation in-process will be described in detail with reference to FIG.
In sequence Q63, the snapshot creation unit 35 performs post-snapshot creation processing in cooperation with the table management unit 32. The snapshot creation post-process will be described in detail with reference to FIG.

In sequence Q <b> 64, the snapshot creation unit 35 returns a snapshot creation response to the snapshot creation management server 4. This snapshot creation response is received by the snapshot creation supervision unit 41 for the number of data management servers 3.
In sequence Q65, the snapshot creation management unit 41 checks whether a snapshot creation response is returned from all the data management servers 3 in the cluster system 1.
In sequence Q66, the snapshot creation supervising unit 41 returns a snapshot creation response to an external terminal (not shown). This external terminal (not shown) is the same as the one that has transmitted the snapshot creation request signal in sequence Q40.

  Through the processing of these sequences Q40 to Q66, the cluster system 1 can create a snapshot of the distribution table 22, the data management table 341, and the data 343 by only locking the distribution table 22 and the data management table 341. it can. Since the data amounts of the distribution table 22 and the data management table 341 are extremely small compared to the data 343, these copies are completed in a short time. Thereby, the cluster system 1 can create a snapshot of a large amount of data with a short-time lock.

5A and 5B are diagrams showing the data management table 341 and the second data management table 342. FIG.
FIG. 5A shows the data management table 341.
The data management table 341 includes a key value column 341a, a storage location column 341b, and other columns.

The key value column 341a is a column for storing key values for accessing data, and stores key values “#A”, “#B”,. The corresponding data A can be accessed by the key value “#A”. These key values are stored in the data write request signal and the data reference request signal, indicating that the data corresponding to the key value is processed.
The accommodation position column 341b is a column for storing the accommodation position for accessing data, and stores the address of each data. Each data body can be accessed by these accommodation positions.

FIG. 5B shows the second data management table 342.
The second data management table 342 includes a key value column 342a, a storage location column 342b, and other columns similar to those of the data management table 341, and further includes an attribute column 342n. The second data management table 342 is created by copying the data management table 341 at the snapshot creation start time.
The attribute column 342n is a column in which attributes of each data are stored, and attributes of “not updated”, “before update”, and “after update” are stored.

If the attribute in the attribute column 342n is “unupdated”, it indicates that the corresponding data body has not been updated since the snapshot creation start time.
If the attribute in the attribute column 342n is “before update”, it indicates that the corresponding data body is the one at the time of starting snapshot creation and that the update process for the data has been performed.
If the attribute in the attribute column 342n is “after update”, it indicates that the corresponding data body is after the update and corresponding pre-update data exists.
As described above, the data management server 3 manages the data at the snapshot creation start time by the second data management table 342. Therefore, the data management server 3 can create a snapshot at an arbitrary time without performing a duplication process.

FIG. 6 is a diagram illustrating a snapshot creation operation and a data update operation.
As shown in FIG. 6, each piece of data 343 includes “updated data” copied and updated during snapshot creation, “unupdated data” that has not been updated since the start of snapshot creation, It is logically divided into three sets of “pre-update data” that is a copy source related to the post-update data.
Data A, C, and E are “unupdated data” that has not been updated since the start of snapshot creation. The data B and D are the original data and “pre-update data” at the start of the snapshot creation among the data received from the outside during the snapshot creation. Data B2 and D2 are “updated data”, which are data B and D updated at the start of snapshot creation.

  Each data A, B, C,... Is managed by the attribute column 342n (see FIG. 5B) of the second data management table 342. The second data management table 342 is used for a snapshot creation operation by the snapshot creation unit 35. The snapshot creation unit 35 can access each data of the data 343 by the table management unit 32 and the data management unit 33. Dotted arrows extending from the snapshot creation unit 35 to the pre-update data and the non-update data indicate access to the pre-update data and the non-update data related to the snapshot creation operation.

On the other hand, the data management table 341 is used for a data reference operation by the data write reference processing unit 31. The data write reference processing unit 31 can access each data of the data 343 by the table management unit 32 and the data management unit 33. Dotted arrows extending from the data write reference processing unit 31 to updated data and unupdated data indicate access to the updated data and the unupdated data related to the data reference operation.
As described above, the data write reference processing unit 31 and the snapshot creation unit 35 access each data by referring to different tables. Therefore, in parallel with the data reference by the data write reference processing unit 31, the snapshot creation unit 35 can perform a snapshot creation operation.

FIG. 7 is a flowchart showing the data writing reference process.
When the data write reference processing unit 31 of each data management server 3 receives an external request signal, the data write reference process shown in FIG. 7 is started.
In step S10, the data write reference processing unit 31 determines the processing type of the received request signal.

If the process type is a reference, the data write reference processing unit 31 performs the process of step S11.
In step S11, the data write reference processing unit 31 refers to the corresponding data based on the data management table 341 that is the original. Based on the key value included in the request signal, the data write reference processing unit 31 acquires the accommodation position of the corresponding data by the table management unit 32 and refers to the corresponding data by the data management unit 33. When the process of step S11 ends, the data write reference processing unit 31 ends the process of FIG.

  What is managed by the data management table 341 is data included in the pre-update set and the post-update set. Accordingly, the accommodation position of the corresponding data can be acquired by referring to the data management table 341 based on the key value included in the reference request from the outside. The snapshot creation unit 35 creates a snapshot by referring to the second data management table 342 which is a different table. As a result, the snapshot creation unit 35 can create a snapshot without writing and locking the data 343.

If the process type is update, the data write reference processing unit 31 performs the data update process of step S12 and ends the process of FIG. The data update process will be described in detail with reference to FIG.
If the process type is a new creation, the data write reference processing unit 31 performs a new creation process in step S13 and ends the process in FIG. The new creation process will be described in detail with reference to FIG.
If the processing type is deletion, the data write reference processing unit 31 performs the data deletion processing in step S14 and ends the processing in FIG. The data deletion process will be described in detail with reference to FIG.

FIG. 8 is a flowchart showing data update processing.
When the data write reference processing unit 31 of each data management server 3 receives an external data update request signal, the data update process shown in FIG. 8 is started.
In step S30, the data write reference processing unit 31 determines whether a snapshot is being created. The data writing reference processing unit 31 performs the process of step S31 if the determination condition is not satisfied (No), and performs the process of step S32 if the determination condition is satisfied (Yes).

  In step S31, the data write reference processing unit 31 updates the corresponding data. Based on the key value included in the request signal, the data write reference processing unit 31 acquires the accommodation position of the corresponding data by the table management unit 32 and accesses the corresponding data by the data management unit 33. The data write reference processing unit 31 ends the process of FIG. 8 when the process of step S31 ends.

  In step S32, the data writing reference processing unit 31 determines whether or not the attribute of the corresponding data is “unupdated”. Based on the key value included in the request signal, the data write reference processing unit 31 acquires the attribute of the corresponding data by the table management unit 32 and determines whether or not it is “unupdated”. The data write reference processing unit 31 performs the process of step S33 if the determination condition is not satisfied (No), and performs the process of step S34 if the determination condition is satisfied (Yes).

  In step S <b> 33, the data write reference processing unit 31 updates the corresponding data of the “updated” attribute. Based on the key value included in the request signal, the data write reference processing unit 31 refers to the data management table 341 by the table management unit 32 to obtain the storage location of the updated data, and the data management unit 33 updates the updated data Update. The data write reference processing unit 31 ends the process of FIG. 8 when the process of step S33 ends.

The processing in steps S34 to S39 is processing when a snapshot is being created and the attribute of the corresponding data is not updated.
In step S34, the data write reference processing unit 31 copies the corresponding data. Based on the key value included in the request signal, the data write reference processing unit 31 refers to the data management table 341 by the table management unit 32 to acquire the accommodation position, and the data write reference processing unit 31 further acquires Based on the storage position, the data management unit 33 acquires an area of the corresponding data size in the storage unit 34, and then copies the corresponding data to the acquired area.
In step S35, the data write reference processing unit 31 registers the reference information of the copied data in the second data management table 342. That is, the data write reference processing unit 31 creates a record of the copied data in the second data management table 342 by the table management unit 32. Further, the data write reference processing unit 31 sets the key value column 342a of this record to the key value of the copy source data, and rewrites the storage location column 342b with the address of the copied data.

In step S <b> 36, the data write reference processing unit 31 sets an “after update” attribute to the copied data. The data write reference processing unit 31 refers to the record of the second data management table 342 related to the copied data by the table management unit 32, and sets the attribute column 342n to “after update”.
In step S <b> 37, the data write reference processing unit 31 updates the copied data by the data management unit 33.

In step S <b> 38, the data write reference processing unit 31 rewrites the accommodation position on the data management table 341 related to the corresponding data to the address of the copied data by the table management unit 32. That is, the data writing reference processing unit 31 refers to the record of the data management table 341 related to the corresponding data by the table management unit 32, and sets the address of the copied data in the accommodation position column 341b.
In step S39, the data write reference processing unit 31 sets the “before update” attribute in the copy source data of the corresponding data. The data writing reference processing unit 31 refers to the record related to the copy source data of the corresponding data in the second data management table 342 by the table management unit 32, and sets the attribute column 342n to “before update”. When the process of step S39 is completed, the data write reference processing unit 31 ends the process of FIG.

  As described above, the data write reference processing unit 31 duplicates the data to be processed into pre-update data and post-update data only when updating the data of the “unupdated” attribute during snapshot creation. Thereby, the data management server 3 can reduce the time calculation amount and the space calculation amount due to the duplication of data.

FIG. 9 is a flowchart showing a new data creation process.
When the data writing reference processing unit 31 of each data management server 3 receives a new data creation request signal from the outside, a new data creation process shown in FIG. 9 is started.
In step S40, the data write reference processing unit 31 newly creates data. That is, the data write reference processing unit 31 newly creates data and a key value of the data by the data management unit 33.
In step S <b> 41, the data write reference processing unit 31 registers reference information for newly created data in the data management table 341 and the second data management table 342. That is, the data writing reference processing unit 31 creates a record related to newly created data in the data management table 341 and the second data management table 342 by the table management unit 32. The data write reference processing unit 31 sets the newly created key value in the key value column 341a of the record, and sets the address of the newly created data in the accommodation position column 341b of the record.
In step S42, the data write reference processing unit 31 determines whether a snapshot is being created. The data write reference processing unit 31 performs the process of step S43 if the determination condition is not satisfied (No), and performs the process of step S44 if the determination condition is satisfied (Yes).

In step S43, the data write reference processing unit 31 sets the “unupdated” attribute to the newly created data. That is, the data writing reference processing unit 31 refers to the record of the second data management table 342 related to the newly created data by the table management unit 32 and sets the attribute column 342n to “unupdated”. The data write reference processing unit 31 ends the process of FIG. 9 when the process of step S43 ends.
In step S44, the data writing reference processing unit 31 sets the “after update” attribute to the newly created data. That is, the data write reference processing unit 31 refers to the record of the second data management table 342 related to the newly created data by the table management unit 32, and sets the attribute column 342n to “after update”. The data write reference processing unit 31 ends the process of FIG. 9 when the process of step S44 ends.

  In this way, the data write reference processing unit 31 creates new creation data only in the updated set during snapshot creation, and does not create data in the pre-update set. As a result, the data management server 3 can exclude data newly created during the creation of the snapshot from the scope of this snapshot creation.

FIG. 10 is a flowchart showing data deletion processing.
When the data write reference processing unit 31 of each data management server 3 receives an external data deletion request signal, the data deletion process shown in FIG. 10 is started.
In step S50, the data write reference processing unit 31 determines whether a snapshot is being created. The data writing reference processing unit 31 performs the process of step S51 if the determination condition is not satisfied (No), and performs the process of step S52 if the determination condition is satisfied (Yes).

  In step S51, the data write reference processing unit 31 deletes the corresponding data. Based on the key value included in the data deletion request signal, the data write reference processing unit 31 acquires the storage position of the processing target data by the table management unit 32 and deletes the processing target data by the data management unit 33. The data write reference processing unit 31 performs the process of step S54 when the process of step S51 is completed.

  In step S52, the data writing reference processing unit 31 determines whether or not the attribute of the corresponding data is “unupdated”. Based on the key value included in the data deletion request signal, the data write reference processing unit 31 acquires the attribute of the data to be processed by the table management unit 32 and determines whether or not it is “unupdated”. The data write reference processing unit 31 performs the process of step S53 if the determination condition is not satisfied (No), and performs the process of step S55 if the determination condition is satisfied (Yes).

  In step S <b> 53, the data write reference processing unit 31 deletes the corresponding data with the “after update” attribute. The data writing reference processing unit 31 acquires the storage location of the updated data based on the data management table 341 by the table management unit 32, and releases the storage area of the processing target data by the data management unit 33. Delete the target data. The data write reference processing unit 31 performs the process of step S54 when the process of step S53 is completed.

In step S54, the data write reference processing unit 31 deletes the reference information (record) of the corresponding data in the second data management table 342. The data write reference processing unit 31 performs the process of step S56 when the process of step S54 is completed.
In step S55, the data writing reference processing unit 31 sets the “before update” attribute to the corresponding data. The data writing reference processing unit 31 uses the table management unit 32 to identify the record related to the corresponding data in the second data management table 342, and sets the “before update” attribute in the attribute column 342n. The data writing reference processing unit 31 performs the process of step S56 when the process of step S55 is completed.
In step S56, the data writing reference processing unit 31 deletes the reference information (record) of the corresponding data in the data management table 341. When the process of step S56 is completed, the data write reference processing unit 31 ends the process of FIG.

In this way, the data writing reference processing unit 31 moves the deletion target data to the pre-update set and creates the deletion target data in the data management table 341 if the snapshot is being created and the attribute of the deletion target data is not updated. Delete the reference information (record). Thereby, the data management server 3 can exclude the deletion target data from the target of the subsequent write update processing while keeping the target of snapshot creation.
If the snapshot is being created and the attribute of the deletion target data is not unupdated, the data write reference processing unit 31 deletes only the updated data of the deletion target data and leaves the pre-update data. Further, the data write reference processing unit 31 deletes the reference information (record) of the deletion target data from the data management table 341 during the creation of the snapshot. Thereby, the data management server 3 can exclude the deletion target data from the target of the subsequent write update processing.

FIG. 11 is a flowchart showing a snapshot creation process.
When a copy completion response is input from the table management unit 32, the snapshot creation unit 35 starts a snapshot creation process.
In steps S60 to S63, the snapshot creation unit 35 repeats the process for all the records in the second data management table 342.

  In step S61, the snapshot creation unit 35 refers to the attribute column 342n of the record by the table management unit 32, and determines whether it is “not updated” or “before update”. If the determination condition is satisfied (Yes), the snapshot creation unit 35 performs the process of step S62. If the determination condition is not satisfied (No), the snapshot creation unit 35 performs the process of step S63.

In step S <b> 62, the snapshot creation unit 35 causes the data management unit 33 to collect data relating to the record in the snapshot data 344.
In step S63, the snapshot creation unit 35 determines whether the processing has been repeated for all the records in the second data management table 342. If the determination condition is not satisfied, the snapshot creation unit 35 returns to the process of step S60, and if the determination condition is satisfied, the process of FIG. 11 ends.

The snapshot creation unit 35 acquires the data included in the unupdated set and the set before update among the data related to the second data management table 342 and collects the data in the snapshot data 344. The unupdated set is a set of data that has not been updated since the snapshot creation start time. The pre-update set is a set of copy source data related to data copied and updated during snapshot creation.
Thereby, the snapshot creation part 35 can back up all the data at the time of a snapshot creation start.

FIG. 12 is a flowchart showing post-snapshot creation processing.
When the snapshot creation process shown in FIG. 11 is completed, the snapshot creation unit 35 starts the post-snapshot creation process.
In steps S70 to S73, the snapshot creation unit 35 repeats the process for all records in the second data management table 342.

In step S71, the snapshot creation unit 35 determines the attribute column 342n of the record. The snapshot creation unit 35 refers to the attribute column 342n of the record by the table management unit 32. If the attribute column 342n is “before update”, the snapshot creation unit 35 performs the process of step S72. If the attribute column 342n is “after update”, the snapshot creation unit 35 performs the process of step S74. If the attribute column 342n is “unupdated”. Step S75 is performed.
In step S72, the snapshot creation unit 35 deletes the corresponding data. The snapshot creation unit 35 refers to the storage location column 342b of the record by the table management unit 32 and deletes the corresponding data stored in this storage location.

In step S73, the snapshot creation unit 35 deletes the reference information (record) of the corresponding data in the second data management table 342. When the process of step S73 ends, the snapshot creation unit 35 performs the process of step S75.
In step S74, the snapshot creation unit 35 sets an “unupdated” attribute for the corresponding data. The snapshot creation unit 35 uses the table management unit 32 to set an “unupdated” attribute in the attribute column 342n of the record.

  In step S75, the snapshot creation unit 35 determines whether the processing has been repeated for all the records in the second data management table 342. If the determination condition is not satisfied, the snapshot creating unit 35 returns to the process of step S70, and if the determination condition is satisfied, the process of FIG.

The snapshot creation unit 35 of the first embodiment deletes the pre-update set data in the post-snapshot creation process. As a result, the snapshot creation unit 35 can release the storage area for the pre-update data and reduce the amount of normal space calculation.
However, the present invention is not limited to this, and the snapshot creation unit 35 may finally delete the second data management table 342 in the post-snapshot creation process. Thereby, each data management server 3 can release the storage area in which the second data management table 342 was stored, and can further reduce the amount of space calculation at the normal time.

FIGS. 13A to 13G are diagrams showing each set and each data during snapshot creation.
FIG. 13 (a) to FIG. 13 (g) show the results of external write update processing performed in time series order. The unupdated set is a set of data in which the attribute column 342n of the second data management table 342 is “not updated”. The set before update is a set of data in which the attribute column 342n of the second data management table 342 is “before update”. The updated set is a set of data in which the attribute column 342n of the second data management table 342 is “after update”.

FIG. 13A shows each set and each data when the data management table 341 is copied to the snapshot data 344. This shows each data at the start of snapshot creation.
The unupdated set includes all data A to E. That is, the data A to E are the same as the snapshot creation start time and indicate that they have not been updated. The set before update and the set after update do not include data.

FIG. 13B shows each set and each data when data B having #B as a key value is updated.
Data B (see FIG. 13A) has moved to the pre-update set. Here, the movement of data is indicated by solid arrows. Data B (see FIG. 13A) is copied and updated to become data B2, and is moved to the updated set. Here, update and movement of data are indicated by broken-line arrows.

FIG. 13C shows each set and each data when the data D having #D as a key value is updated.
Data D (see FIG. 13B) has moved to the pre-update set. Further, the data D (see FIG. 13B) is copied and updated to become data D2, and is moved to the updated set.

FIG. 13D shows each set and each data when data B2 having #B as a key value is re-updated.
The updated set data B2 (see FIG. 13C) is re-updated to become data B3, and is also included in the updated set.

FIG. 13E shows each set and each data when data D2 having #D as a key value is deleted.
The updated set data D2 (see FIG. 13D) is deleted and disappears from the updated set.

FIG. 13 (f) shows each set and each data when data E having #E as a key value is deleted.
The data E of the unupdated set (see FIG. 13D) has moved to the pre-update set.

FIG. 13G shows each set and each data after the post-snapshot creation process is executed.
The updated set data B3 (see FIG. 13F) has moved to the unupdated set. Thereby, only the data A, B3, and C are included in the unupdated set. That is, this unupdated set reflects the deletion of the data D and E and the two updates of the data B.

  In all the periods shown in FIGS. 13A to 13F, the logical sum of the unupdated set and the set before update is data A to E, which matches the data A to E at the start of snapshot creation. Therefore, the snapshot creation unit 35 can collect data at the start of snapshot creation by collecting data of the unupdated set and the set before update.

(Effect of the invention)
The first embodiment described above has the following effects (A) to (G).

(A) The snapshot creation unit 35 does not lock the entire data 343 but locks the data management table 341 and releases the lock after creating this snapshot. Thereafter, the data management server 3 collects snapshots on the other hand while continuing to update data on the one hand. Thereby, the data management server 3 can create a snapshot of a large amount of data managed by this table by locking a small amount of table.

(B) The data to be backed up during the creation of the snapshot is managed separately into an updated set, an unupdated set, and a pre-update set. When updating the data in the unupdated set, the data write reference processing unit 31 copies the data, moves the data to the set before update and the set after update, and updates the data moved to the set after update. Processing is in progress. As a result, the data write reference processing unit 31 is duplexed only for updated data, so the time calculation amount and the space calculation amount for duplexing are reduced, and a snapshot can be created in a short time. Can do.

(C) The data write reference processing unit 31 creates data in the updated set when creating new data during snapshot creation. Thereby, the data write reference processing unit 31 can newly create data without affecting the snapshot creation processing.

(D) When deleting data during snapshot creation, the data write reference processing unit 31 moves to the set before update if the data is an unupdated set, and deletes if it is data after the update. Thereby, the data write reference processing unit 31 can delete the processing target data without affecting the snapshot creation processing.

(E) The data write reference processing unit 31 makes a data reference request during the creation of the snapshot based on the data management table 341 that is the original in which the data update status is reflected. The snapshot creation unit 35 copies the data management table 341 at the start of snapshot creation, and creates a snapshot based on the second data management table 342 that manages data included in the unupdated set and the set before update. . As described above, the data write reference processing unit 31 and the snapshot creation unit 35 refer to data and create a snapshot based on different tables. Therefore, each data management server 3 does not need to maintain consistency of data and the data management table 341 by exclusive control.

(F) After creating the snapshot, the snapshot creating unit 35 deletes each data in the set before update which is not normally required. As a result, the data management server 3 can secure sufficient memory resources at the normal time.

(G) The snapshot creation unit 35 manages each piece of data managed separately into an updated set and an unupdated data set during the creation of the snapshot, after the snapshot is created. Change to manage. Thereby, the data management server 3 can reduce the time calculation amount at the normal time.

(Second Embodiment)
FIG. 14 is a schematic configuration diagram illustrating a data management server according to the second embodiment.
The data management server 3 is configured in the same manner as each data management server 3 of the first embodiment shown in FIG.
Unlike the first embodiment, the data management server 3 transmits a data write request signal, a data reference request signal, and the like from the outside without going through the distribution server 2 (see FIG. 1).
Unlike the first embodiment, the data management server 3 transmits a snapshot creation request signal from an external terminal without going through the snapshot creation management server 4 (see FIG. 1). Other than that, it is configured similarly to each data management server 3 of the first embodiment and operates in the same manner.

  In this way, the snapshot creation unit 35 is not limited to the cluster system 1 (see FIG. 1) configured by a plurality of data management servers 3, and even a single data management server 3 has a small amount of data management table. With the lock 341, a snapshot of a large amount of data can be created.

(Modification)
The present invention is not limited to the above-described embodiment, and can be modified without departing from the spirit of the present invention. For example, there are the following (a) to (c).

(A) When deleting data during snapshot creation, the data management server 3 copies the data, moves the data to the set before update and the set after update, and deletes only the data in the set after update. May be.

(B) Upon receiving the snapshot creation instruction, the snapshot creation unit 35 may process the table management unit 32 to release the table lock.

(C) The snapshot creation unit 35 may lock the backup target part of the data management table 341, not the entire data management table 341, and release the lock after creating the backup. As a result, the snapshot creation unit 35 can create a snapshot of the limited backup target portion.

1 Cluster system (Server cluster snapshot creation system)
2 Distribution server 21 Distribution unit 22 Distribution table 3, 3-1 to 3-3 Data management server 31 Data write reference processing unit 32 Table management unit 33 Data management unit 34 Storage unit 341 Data management table 342 Second data Management table 343 Data 344 Snapshot data 35 Snapshot creation unit 4 Snapshot creation management server 41 Snapshot creation control unit

Claims (8)

Multiple data management servers in a cluster configuration,
A distribution server;
With
The distribution server
A distribution unit that distributes external requests to the data management server;
A distribution table showing correspondence between external requests and the data management server;
With
The data management servers are respectively
A table management unit for managing a data management table including a storage position of each data;
A data management unit for managing the data;
A snapshot creation unit that creates a snapshot without creating a snapshot after creating a snapshot by writing and locking the allocation table and the data management table; and
Data is written or referenced based on an external request, and during the snapshot creation of the snapshot creation unit, a write request for unupdated data after the start of snapshot creation is received from the outside If not, the unupdated data related to the write request is copied, the write request is made to the copied data, the updated attribute is set to the updated data, and the unupdated data related to the write request is set. A data writing reference processing unit that sets the pre-update attribute to the update data and sets the pre-update data;
A storage unit for storing each data, the data management table, and snapshot data created by the snapshot creation unit;
A server cluster snapshot creation system comprising: The data write reference processing unit
If a data reference request is received from outside during snapshot creation by the snapshot creation unit, each data is referenced based on the data management table.
The server cluster snapshot creation system according to claim 1. The snapshot creation unit
Create a snapshot by collecting each unupdated data and each pre-update data after the snapshot creation start,
The server cluster snapshot creation system according to claim 1. The snapshot creation unit
When the snapshot creation is completed, delete each pre-update data and set an unupdated attribute to each post-update data.
The server cluster snapshot creation system according to claim 3. The data write reference processing unit
If an unupdated data deletion request is received from the outside during snapshot creation by the snapshot creation unit, an attribute before update is set in the unupdated data related to the deletion request.
The server cluster snapshot creation system according to claim 1. The data write reference processing unit
If a new data creation request is received from the outside during the snapshot creation by the snapshot creation unit, an updated attribute is set in the data related to the new creation request.
The server cluster snapshot creation system according to claim 1. The data management unit manages attributes of each data based on a second data management table stored in the storage unit;
The server cluster snapshot creation system according to claim 1. A table management unit for managing a data management table including a storage position of each data;
A data management unit for managing the data;
A snapshot creation unit that creates a snapshot without creating a snapshot after creating a snapshot by writing-locking the data management table;
Data is written or referenced based on an external request, and during the snapshot creation of the snapshot creation unit, a write request for unupdated data after the start of snapshot creation is received from the outside If not, the unupdated data related to the write request is copied, the write request is made to the copied data, the updated attribute is set to the updated data, and the unupdated data related to the write request is set. A data writing reference processing unit that sets the pre-update attribute to the update data and sets the pre-update data;
A storage unit for storing each data, the data management table, and snapshot data created by the snapshot creation unit;
A snapshot creation system for a data management server, comprising:

Images