JP2015035182A - Distributed database system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a data loss when a data management server configuring a cluster breaks down, and to specify the range of the hash value of the key of the lost data.SOLUTION: A lost data specification device 3 disposed in a distributed database system 1000 is configured to generate dummy data making a pair of each of data management servers in which data are stored, and to store the dummy data in each data management server via a distribution server 1. The lost data specification device 3 is configured to, when receiving the change notification of distributed hash information for managing the increase/decrease of a data management device, compare the distributed hash information before and after change, and to determine the increase/decrease of the data management server. Then, the lost data specification device 3 is configured to, when the data management server is deleted, transmit the deletion request of the dummy data via the distribution server 1 to the data management server, and to, when the registered dummy data are undeletable, detect the loss of data, and to specify the range of the key of the lost data on the basis of the distributed hash information before change.

Description

  The present invention relates to a clustered distributed database system in which a plurality of servers cooperate to perform data processing.

  A cloud computing storage is required to retain a large amount of data and efficiently process it, but to have no data loss even when a failure or the like occurs. In this regard, a distributed database system has been proposed that includes a front-end server and a cluster that includes a plurality of servers and that stores and manages a large amount of data in a distributed manner for each server.

The front-end server accepts requests for data registration, deletion, reference, etc., and distributes these requests to the server that holds and manages the data. The server that holds and manages the data processes the distributed request.
In addition, the server that holds the data generates one or more copies (replicated data) of the data, and the server other than the server that generated the replicated data transmits and receives the replicated data between the servers. Hold. This is to avoid data loss due to the presence of duplicate data even when a failure or the like occurs in one server.
In the following, a server serving as a front end is referred to as a distribution server, and each server that constitutes a cluster and stores data in a distributed manner is referred to as a data management server.

  To distribute and hold data, the distribution server manages the combination of which data should be distributed to which data management server, and forwards requests for data registration, deletion, reference, etc. to the correct distribution destination There is a need. In addition, when a data management server failure occurs, the cluster configuration of the data management server changes, so the combination of which data should be assigned to which data management server is changed according to the change in configuration. There is a need to. As a technique for realizing these, a technique using a consistent hash method is known (for example, see Patent Document 1 and Non-Patent Document 1).

In the consistent hash method, the concept of a virtual node may be introduced as a distribution destination. The virtual node is a logical data management server set in association with one data management server (physical node) in the distribution algorithm, and has a many-to-one relationship with the physical data management server. This virtual node is introduced in order to reduce the variation in the load of the physical data management server constituting the cluster.
In the consistent hash method, information that becomes a key (identifier unique to the data) for distributing the data is set in a request for registration or deletion of the data. For example, in a data registration request, a key set in the data registration request is subjected to a predetermined hash function to obtain a hash value of the key. Then, the data management server (virtual node) to which the request is distributed is obtained from the hash value of the key and the distributed hash table, the request is transferred to the data management server (virtual node), and the registration process is performed. Executed. In the following description, it is assumed that a virtual node is set in the data management server.

  FIG. 1 is a diagram for explaining an algorithm of the consistent hash method. In the consistent hash method, a set of hash values, which is an output of a hash function, is represented by a line, and the maximum value and the minimum value are connected and expressed as a ring (consistent hash ring) as shown in FIG. To do. On the consistent hash ring, hash values (“a”, “b”,..., “Y”) corresponding to the virtual node IDs (“A”, “B”,..., “Y”) of the virtual nodes. Is placed. In the following description and drawings, for example, a virtual node whose virtual node ID is “A” is a virtual node “A”, a virtual node whose virtual node ID is “B” is a virtual node “B”, and the like. May be expressed.

  Then, a certain virtual node (for example, virtual node “B”) has a hash value of a virtual node (virtual node “A”) adjacent in the counterclockwise direction from its own hash value “b” on the consistent hash ring. A range of hash values from the hash value next to “a” to the hash value “b” of the virtual node (virtual node “B”) itself (a <hash value ≦ b handled by the virtual node “B”) Scope of charge.

  FIG. 1B is a distributed hash table, which represents the consistent hash ring shown in FIG. In the distributed hash table, records are sorted in the order of the hash value of the virtual node, and the range from the next value of the virtual node hash value of the record on one line to the hash value of its own virtual node is This is the scope of the assigned hash value. In FIG. 1B, for example, a record on one line of the record of the virtual node “B” is referred to, and from the hash value “a” next to the virtual node “A” indicated in the record, The range of hash values up to the hash value “b” of the virtual node (virtual node “B”) itself (a <hash value that the virtual node “B” is responsible ≦ b) is the hash value that the virtual node “B” is responsible for It becomes the range (charge range).

That is, the consistent hash ring in FIG. 1A and the distributed hash table in FIG. 1B both represent the assigned range (hash value range) of each virtual node.
In the consistent hash method, a hash value of a key for a registration request or the like is obtained by a hash function, and depending on which virtual node is in charge of the range of hash values (the range in charge), the registration request or the like A virtual node to be a distribution destination is determined.

JP 2013-025497 A

D. Karger, et al., “Consistent Hashing and Random Trees: Distributed Caching Protocols for Relieving Hot Spots on the World Wide Web”, in Proceedings of the 29th ACM Symposium on Theory of Computing (STOC'97), May 1997, pp .654-663, [online], [Search July 27, 2013], Internet <URL: http: //thor.cs.ucsb.edu/~ravenben/papers/coreos/KLL+97.pdf>

  In a distributed database system using the consistent hash method, if the number of replicas + 1 (redundancy number) or more of data management servers fails at a time, all data in the assigned range of a specific virtual node may be lost. .

FIG. 2 is a diagram for explaining a method of storing original data and duplicate data of the data management server in the consistent hash method.
When the number of replicas is n, the data management server corresponding to the virtual node uses the original data that is the data in the range of responsibility on the consistent hash ring of the virtual node and the counter clock from the virtual node of the consistent hash ring. The replica data of n virtual nodes continuous in the rotation direction are stored. For example, when the number of replicas is “2” (n = 2), as shown in FIG. 2, the data management server corresponding to the virtual node “C” (virtual node ID: “C”) is the virtual node “C”. ”And duplicate data obtained by copying the original data of the virtual nodes“ A ”and“ B ”are stored. In other words, the replicated data of a certain virtual node is stored in the corresponding data management server of n virtual nodes that continue in the clockwise direction from the virtual node. In FIG. 2, the replicated data of the virtual node “A” is stored in the data management server corresponding to the virtual nodes “B” and “C”.

  Based on the above, along the consistent hash ring, if the data management server corresponding to these virtual nodes has a failure, etc. at once for the continuous virtual nodes of the number of replicas n + 1 (redundancy number) or more, the missing data Occur. Note that the missing data in this embodiment means data in which both original data and duplicated data are missing in all data management servers constituting the cluster. For example, when the number of replicas is “2” and the data management server corresponding to the virtual nodes “A”, “B”, and “C” fails at a time, the virtual node “A” has a consistent hash ring. The data in the area of charge is missing data.

  When data is lost, it is necessary to restore each piece of missing data included in the assigned range of the failed virtual node using backup data or the like. However, since the virtual node disappears from the distributed hash table or the consistent hash ring when the corresponding data management server fails, it is difficult to specify which virtual node data is missing.

FIG. 3 is a diagram for explaining the change of the assigned range of the virtual node when the data management server fails.
When the data management server fails, the corresponding virtual node disappears. Specifically, the record of the corresponding virtual node is deleted from the distributed hash table. As a result, the assigned range of the corresponding virtual node is taken over by the virtual node that becomes the next record in the distributed hash table. The example illustrated in FIG. 3A indicates that the record of the corresponding virtual node “B” is deleted because the data management server corresponding to the virtual node “B” has failed. As a result, the assigned range of the virtual node “B” is taken over by the virtual node “C” of the next record.
In the consistent hash ring, when the virtual node disappears, the hash value point of the virtual node is deleted from the consistent hash ring. Therefore, the assigned range of the virtual node is merged with the assigned range of the virtual node adjacent in the clockwise direction. In the example illustrated in FIG. 3B, the hash value “b” of the failed virtual node “B” is deleted, and the range from the next value after the hash value “a” to the hash value “c” disappears. This indicates that the assigned range of the virtual node “C” adjacent to the node “B” in the clockwise direction is included.

  As described above, the assigned range of the virtual node when data is lost becomes a part of the assigned range of other virtual nodes, and the distributed hash table at the time of failure is changed. For this reason, it is difficult to obtain the range of missing data from the information after the failure. Further, in a situation where a plurality of units are out of order, the order of failure is also involved, so it is more difficult to determine the range of missing data. Therefore, with the conventional technology, it is impossible to specify the key range of the missing data, so the data management server's cluster data is checked against the backup data, and the missing data is identified and restored. It was. Therefore, when there is a large amount of data, there is a problem that it takes a long time to recover.

  The present invention has been made in view of such a background, and in the present invention, when a data management server constituting a cluster fails, a data loss is detected, and a range of hash values of keys of the lost data It is an object of the present invention to provide a distributed database system that can specify a database.

  In order to solve the above-described problem, the invention according to claim 1 is based on a distribution server that distributes a request signal received from a client device to a plurality of data management servers and a request signal received from the distribution server. A distributed database system comprising: a plurality of the data management servers that execute processing on stored data; and a missing data specifying device that detects missing data stored in the data management server, wherein the distribution server includes: On the consistent hash ring, a hash value calculated by applying an identifier unique to the data management server to a predetermined hash function is arranged in ascending order, and a key unique to the data attached to the received request signal is assigned to the predetermined hash function. By placing the hash value calculated over the hash function, In addition to adopting a distribution method to determine the data management server to which the received request signal is distributed, each of the data management servers stores data corresponding to the request signal distributed to itself as original data The replica data of the original data is generated and transmitted to another data management server having a predetermined number of replicas adjacent on the consistent hash ring, and the replicated data is stored in another data management server. A distribution method in which the distribution server stores a range of hash values of the key of the data indicating the assigned range to which each of the data management servers is distributed on the consistent hash ring. A storage unit in which hash information is stored and attached to the received request signal The hash value of the key is calculated, the distribution processing unit that transfers the request signal to the data management server that is the distribution destination determined based on the distributed hash information, and the life and death of the plurality of data management servers are monitored, and the life and death An alive monitoring unit that changes the assigned range of the distributed hash information when there is a change, a change notification unit that transmits a distributed hash information change notification signal including the changed distributed hash information to the missing data specifying device, And the missing data specifying device uses the distributed hash information included in the received distributed hash information change notification signal as the new generation distributed hash information, and the distributed hash information received before the reception is the old generation. Compare the new generation of distributed hash information with the old generation of distributed hash information The increase / decrease detection unit for determining the increase / decrease of the data management server, and if the data management server has increased as a result of the determination, or the old generation distributed hash information is not stored, Dummy data is generated in association with each data management server, and a dummy data registration request signal including the generated dummy data is transmitted to the data management server paired with the dummy data via the distribution server. And dummy data deletion requesting deletion of dummy data associated with the decreased data management server when the data management server has decreased as a result of the determination. A request signal is transmitted to the reduced data management server via the distribution server, and the decrease By not being able to delete the dummy data associated with the data management server, the dummy data deletion unit that detects the loss of data, and the assigned range on the consistent hash ring of the reduced data management server, Alert notification that extracts the range of the hash value of the key of the data based on the old-generation distributed hash information and identifies the range of the key of the missing data in which the original data and the predetermined number of copies of the duplicated data are lost A distributed database system.

  In this way, when a plurality of data management servers constituting the cluster fail, it is possible to detect data loss due to loss of original data and a predetermined number of replicas. In addition, since the lost data specifying device stores the old generation of distributed hash information, the hash value range of the missing data key can be specified based on the reduced range of the data management device.

  The invention according to claim 2 is the data management in which the missing data specifying device randomly generates a key, and a hash value calculated by applying the key to the predetermined hash function is indicated in the new generation distributed hash information. The generation of the key and the calculation of the hash value of the key are repeated until the key is included in the assigned range of the data management server extracted from the server, and the key is the distribution destination of the extracted data management server The key generation part which determines the said key as a key of the said dummy data matched with the said extracted data management server when it is contained in the charge range used as described in Claim 1 further characterized by the above-mentioned. A distributed database system was adopted.

  By doing in this way, the missing data identification device can use the key included in the assigned range on the consistent hash ring of the data management server as the key of the dummy data corresponding to the data management server. Therefore, the distribution server can securely register the dummy data in the corresponding data management server by executing the distribution process using the key attached to the dummy data.

  According to the present invention, there is provided a distributed database system capable of detecting a missing data and specifying a range of hash values of keys of the missing data when a data management server constituting the cluster fails.

It is a figure for demonstrating the algorithm of a consistent hash method. It is a figure for demonstrating the preservation | save method of the original data and replication data of a data management server in a consistent hash method. It is a figure for demonstrating the change of the charge range of a virtual node when a data management server fails. It is a figure for demonstrating the whole structure of the distributed database system which concerns on this embodiment, and the functional block of each apparatus. It is a figure which shows the data structure of the distributed hash table which concerns on this embodiment. It is a figure which shows the data structure of the dummy data registration request signal which concerns on this embodiment. It is a figure which shows the data structure of the dummy data deletion request signal which concerns on this embodiment. It is a figure which shows the data structure of the response signal which concerns on this embodiment. It is a figure which shows the data structure of the dummy data key table which concerns on this embodiment. It is a flowchart for demonstrating the key generation process of the dummy data which concerns on this embodiment. It is a sequence diagram which shows the dummy data registration process with respect to all the virtual nodes at the time of cluster setup. It is a figure for demonstrating the division | segmentation of the charge range of the existing virtual node at the time of addition of a virtual node. It is a figure for demonstrating the change of a distributed hash table at the time of the addition of a virtual node. It is a sequence diagram for demonstrating the dummy data registration process at the time of addition of a virtual node. It is a sequence diagram for demonstrating the dummy data registration process at the time of addition of a virtual node. It is a sequence diagram for demonstrating the dummy data deletion process at the time of extinction of a virtual node. It is a sequence diagram for explaining detection of data loss and output processing of alarm information when a virtual node disappears.

  Next, a distributed database system 1000 and the like in a mode for carrying out the present invention (hereinafter referred to as “the present embodiment”) will be described.

<System configuration and processing overview>
FIG. 4 is a diagram for explaining the overall configuration of the distributed database system 1000 according to the present embodiment and the functional blocks of each device.
As shown in FIG. 4, the distributed database system 1000 includes a distribution server 1 connected to a plurality of client devices 5, and a plurality of data management servers (here, “virtual node 2”) constituting a cluster. And the missing data specifying device 3 connected to the alarm monitoring device 4 and the like.

The distribution server 1 receives the request signal from each client device 5, determines the virtual node 2 to which the received request signal is distributed based on the consistent hash method, and determines the determined virtual node 2 of the distribution destination In response to this, a request signal is transmitted.
In addition, the distribution server 1 is connected to the missing data specifying device 3, and requests the dummy data registration request signal for requesting registration of dummy data (details will be described later) and the deletion of the dummy data from the missing data specifying device 3. Receives a dummy data deletion request signal or the like. Then, the distribution server 1 determines the virtual node 2 that is the distribution destination based on the same method as the distribution method for the request signal received from the client device 5, that is, the consistent hash method, A request signal for dummy data is transmitted.
Further, the distribution server 1 receives response information, which is a processing result for the request information, from the virtual node 2 and transmits the response information to the client device 5 and the missing data specifying device 3.
In addition, the distribution server 1 performs alive monitoring of the virtual node 2 (data management server), and the virtual node 2 is added due to an increase in the number of data management servers in the cluster (addition of the virtual node 2) or a failure of the data management server. When 2 disappears, the distributed hash table (distributed hash information) 100 that manages the virtual node 2 (data management server) that is the distribution destination is changed (addition or deletion of records). Then, the distribution server 1 transmits a distributed hash table change notification signal (distributed hash information change notification signal) including the change contents of the distributed hash table 100 to the missing data specifying device 3.

The virtual node 2 is a logical data management server set on the data management server (physical node). In the present embodiment, as shown in FIG. 4, the virtual node 2 whose virtual node ID is “A” is represented as a virtual node “A”, and the virtual node 2 whose virtual node ID is “B” is represented as a virtual node “A”. B ”etc.
Each virtual node 2 stores original data that is data determined as a distribution destination by the distribution server 1 and duplicate data in which the other virtual node 2 is the storage destination of the original data. In the above-described consistent hash ring, the duplicate data is stored in the virtual nodes 2 corresponding to the number of replicas clockwise from the virtual node 2 that stores the original data. In addition, the virtual node 2 uses the dummy data generated by the missing data identification device 3 as dummy data (original data) in the same area as the normal data (“original data” and “replicated data”). ) And the dummy data included in the assigned range of the other virtual node 2 is stored as dummy data (replication).

The missing data specifying device 3 is connected to the distribution server 1 and the alarm monitoring device 4.
The missing data specifying device 3 generates dummy data for each virtual node 2. The dummy data is data that is paired with the virtual node 2, that is, data included in the assigned range of the virtual node 2 in the consistent hash ring. This dummy data is not used for providing services to the client device 5 but is data stored in each virtual node 2 for the purpose of detecting missing data. The missing data specifying device 3 stores the dummy data (original) and dummy data (replicated) in each virtual node 2 via the distribution server 1.
Further, upon receiving the distributed hash table change notification signal from the distribution server 1, the missing data specifying device 3 stores the received new (new generation) distributed hash table (distributed hash information) 100 a and its own storage unit 32. The addition or disappearance of the virtual node 2 is detected by comparing the distributed hash table (distributed hash information) 100b. When the missing data specifying device 3 detects the disappearance of the virtual node 2, it transmits a dummy data deletion request (dummy data deletion request signal) paired with the disappeared virtual node 2 to the distribution server 1, and the dummy data , It is detected that the data in the assigned range of the corresponding virtual node 2 is missing. Then, the assigned range of the virtual node 2 paired with the dummy data is specified by referring to the old-generation distributed hash table 100b. Furthermore, the missing data specifying device 3 transmits alarm information (data missing detection signal) including the occurrence of missing data and the key range of the missing data (the assigned range of the virtual node) to the alarm monitoring device 4.

As described above, the distributed database system 1000 according to the present embodiment has dummy data paired with the virtual node 2 in advance, the virtual node 2 that stores the dummy data (original), and the virtual data that stores the dummy data (replica). By registering with the node 2 and detecting that the registered dummy data cannot be deleted, it is detected that the data of the virtual node 2 has been lost. The distribution server 1 can also specify the key range of the missing data by referring to the old-generation distributed hash table 100b for the assigned range of the virtual node 2 paired with the dummy data.
Therefore, when the data management server constituting the cluster fails, it is possible to detect data loss and specify the range of the hash value of the key of the lost data.
Since the missing data can be extracted from the backup (a client device holding the missing data, a backup device connected to the data management server, etc.) based on the range of the hash value, the extracted data is assigned to the distribution server 1. The data can be restored by inputting the data to the data management server 2 via the network.

<Configuration of each device>
Hereinafter, a configuration example of each device configuring the distributed database system 1000 according to the present embodiment will be specifically described with reference to FIG.

≪Distribution server≫
The distribution server 1 is a virtual node 2 (data management server) serving as a distribution destination such as a request signal received from each client device 5 or a dummy data registration request signal and a dummy data deletion request signal received from the missing data specifying device 3. ) Is determined based on the consistent hash method, a response signal that is a processing result for the request signal or the like is received from the determined virtual node 2, and the client device 5 or the missing data specifying device 3 that has transmitted the request signal or the like. Send to.
Furthermore, the distribution server 1 monitors the addition (extension) and disappearance (removal) of each virtual node 2 (data management server) constituting the cluster. The distributed hash table 100 stored in the storage unit (not shown) is changed. Then, the distribution server 1 transmits the change information of the distributed hash table 100 to each virtual node 2 (data management server) and also to the missing data specifying device 3.

  The distribution server 1 according to the present embodiment includes a control unit, an input / output unit, and a storage unit, which are not shown.

  The input / output unit inputs / outputs information between the client device 5, each virtual node 2 (data management server), and the missing data specifying device 3. This input / output unit includes a communication interface that transmits and receives information via a communication line, and an input / output interface that performs input / output between an input unit such as a keyboard (not shown) and an output unit such as a monitor. Is done.

  The storage unit includes storage means such as a hard disk, flash memory, and RAM (Random Access Memory), and stores the distributed hash table 100 and the like described above.

  As shown in FIG. 4, the control unit includes a distribution processing unit 10, an alive monitoring unit 11, and a distributed hash table change notification unit (change notification unit) 12. In addition, this control part is implement | achieved when CPU (Central Processing Unit) which abbreviate | omitted illustration expand | deployed and executed the program stored in the memory | storage part, for example.

Upon receiving a request signal, a dummy data registration request signal, a dummy data deletion request signal, or the like from the client device 5 or the missing data specifying device 3, the distribution processing unit 10 extracts keys included in these request signals, By applying the hash function, the hash value of the data targeted by the request signal is calculated. Then, the distribution processing unit 10 refers to the distributed hash table 100 stored in the storage unit (not shown) based on the calculated hash value, and determines the virtual node 2 (data management server) that is the distribution destination of the request signal. decide.
Here, the distributed hash table (distributed hash information) 100 stored in the storage unit, as shown in FIG. 5, is an identifier (here, “virtual node ID”) unique to the data management server and the virtual node 2. And a hash value obtained by applying the virtual node ID to the hash function, for example, a table sorted in ascending order of hash values. The distributed hash table 100 indicates a range of hash values (in charge range) that the virtual node 2 indicated by the virtual node ID is in charge of.

  Returning to FIG. 4, the alive monitoring unit 11 monitors the addition (expansion) and disappearance (removal) of each virtual node 2 (data management server) constituting the cluster, and the cluster configuration has changed. The distributed hash table 100 stored in its own storage unit (not shown) is changed. Specifically, when the data management server is added and the virtual node 2 is added, the alive monitoring unit 11 adds the record of the corresponding virtual node 2 in the distributed hash table 100. In addition, when the data management server is removed due to failure or the like and the virtual node 2 disappears, the alive monitoring unit 11 deletes the record of the corresponding virtual node 2 in the distributed hash table 100.

Note that the alive monitoring unit 11 can acquire the increase / decrease information of each virtual node 2 (data management server) constituting the cluster, for example, as follows.
Each virtual node 2 (data management server), for example, transmits survival confirmation information at a predetermined time interval to the virtual node 2 (data management server) in the next row on the distributed hash table 100, and the survival confirmation When the information cannot be received within a predetermined time, the virtual node 2 (data management server) leaves the cluster, and the separation server including the ID of the detached virtual node 2 (data management server) is assigned to the distribution server. 1 to send. When a new virtual node 2 is added to the cluster, the virtual node 2 transmits information requesting that the virtual node 2 itself participates in the cluster to the distribution server 1. The information that a virtual node 2 (data management server) is added to the cluster is obtained.
In addition to this, for example, by sending survival confirmation information from the distribution server 1 to each virtual node 2 (data management server) at a predetermined time interval, information on the withdrawal of each virtual node 2 is obtained. Also good. Further, information indicating that the virtual node 2 (data management server) has been added to the system may be acquired from a network management server or the like of the entire system (not shown).

When the distributed hash table 100 is changed by the alive monitoring unit 11, the distributed hash table change notification unit (change notification unit) 12 transmits the changed distributed hash table 100 to the distributed hash table change notification signal (distributed hash information). The change notification signal is transmitted to each virtual node 2 and the missing data specifying device 3.
Note that the distributed hash table change notification unit 12 uses only the information of the changed portion indicating the record added or deleted in the distributed hash table 100 as the distributed hash table change notification signal, not the distributed hash table 100 itself. You may attach and transmit.

  In the present embodiment, as shown in FIG. 4, one distribution server 1 is described. However, a plurality of distribution servers 1 may be provided. In that case, one of the plurality of distribution servers 1 becomes a representative (master), and processing other than that executed by the distribution processing unit 10 described above, that is, processing for changing the distributed hash table 100 based on alive monitoring, Execute the process of sending the distributed hash table change notification.

≪Virtual node (data management server) ≫
The virtual node 2 is a logical data management server set on the data management server (physical node), and includes a data management unit 20, an input / output unit and a storage unit, which are not shown. .

The data management unit 20 stores the data determined by the distribution server 1 as a distribution destination in the storage unit as original data. Further, the data management unit 20 stores a copy (replicated data) of the original data received from the other virtual node 2 in the storage unit.
The storage unit (not shown) stores a distributed hash table 100 in which the same data as the distribution server 1 is stored in addition to the original data and the duplicated data. The distributed hash table 100 is stored in the distribution server 1 when each virtual node 2 receives the distributed hash table change notification from the distribution server 1 and changes the distributed hash table 100 stored in its own storage unit. Synchronized with the distributed hash table 100.
When the cluster is set up or when the distributed hash table 100 is changed, the data management unit 20 generates replica data for the original data stored by itself for the number n of replicas set in advance. Thus, the generated duplicated data is stored by being transmitted to n virtual nodes 2 in the clockwise direction (in the distributed hash table 100, up to n rows below its own row).

  Further, when the dummy data is received from the missing data specifying device 3 via the distribution server 1, the data management unit 20 stores the dummy data as an original in its own storage unit. In addition, the data management unit 20 refers to the distributed hash table 100 stored in its own storage unit, and the dummy data (original) stored by itself is the number n of preset replicas as in the case of normal data. Only dummy data (replicas) are generated and generated for n virtual nodes 2 in the clockwise direction on the consistent hash ring (up to n rows below its own row in the distributed hash table 100). Save by sending data (duplicate).

  An input / output unit (not shown) inputs / outputs information to / from the distribution server 1 and other virtual nodes 2. This input / output unit includes a communication interface that transmits and receives information via a communication line, and an input / output interface that performs input / output between an input unit such as a keyboard (not shown) and an output unit such as a monitor. Is done. In the case of the virtual node 2, the input / output unit of the data management server (physical node) has the function of the input / output unit of the virtual node 2 as an entity.

  The storage unit includes storage means such as a hard disk, a flash memory, and a RAM. The above-described original data, replicated data, dummy data (original), dummy data (replicated), distributed hash table 100 (not shown), and the like. Remembered. In the case of the virtual node 2, the storage unit of the data management server (physical node) has the function of the storage unit of the virtual node 2 as an entity.

≪Missing data identification device≫
The missing data specifying device 3 generates dummy data that is paired with each of the virtual nodes 2, and stores them as dummy data (original) and dummy data (replicated) in each virtual node 2 via the distribution server 1. In addition, when the virtual node 2 disappears, the missing data specifying device 3 transmits a dummy data deletion request signal (see FIG. 7) requesting deletion of dummy data to the virtual node 2 via the distribution server 1. By receiving the response information indicating that the dummy data cannot be deleted (NG), it is detected that the data of the corresponding virtual node 2 has been lost. Further, the assigned range of the virtual node 2 paired with the dummy data is identified by referring to the (old generation) distributed hash table 100 (100b) before the change.

With reference to FIG. 4, a configuration example of the missing data specifying device 3 according to the present embodiment will be described.
As shown in FIG. 4, the missing data specifying device 3 according to the present embodiment includes a control unit 30, an input / output unit 31, and a storage unit 32.

  The input / output unit 31 inputs and outputs information to and from the distribution server 1 and the alarm monitoring device 4. The input / output unit 31 includes a communication interface that transmits and receives information via a communication line, and an input / output interface that performs input / output between an input unit such as a keyboard (not shown) and an output unit such as a monitor. Composed.

  The storage unit 32 includes storage means such as a hard disk, a flash memory, and a RAM, and the (new generation) distributed hash table 100 (100a) after the change (indicated as “distributed hash table (new)” in FIG. 4); A pre-change (old generation) distributed hash table 100 (100b) (indicated as “distributed hash table (old)” in FIG. 4) and a dummy data key table 200 (described later in detail) are stored.

  The control unit 30 controls the entire missing data identification device 3 and includes a missing data identification processing unit 300, a key generation unit 310, and an alarm notification unit 320 as shown in FIG. The control unit 30 is realized by, for example, a program stored in the storage unit 32 being developed and executed on a RAM by a CPU (not shown).

  The missing data identification processing unit 300 generates dummy data paired with each virtual node 2 and stores the dummy data in the corresponding virtual node 2, and the (new generation) distributed hash table 100 a after the change and the old (old) The increase / decrease in the virtual node 2 (data management server) is detected by comparing the distributed hash table 100b of the generation. When the missing data identification processing unit 300 detects the disappearance of the virtual node 2 (data management server), the dummy data deletion request signal (FIG. 7) with a dummy data key paired with the virtual node 2 is attached. ) And the response information indicating that the dummy data cannot be deleted is received to detect that the data of the corresponding virtual node 2 is missing. Then, the missing data identification processing unit 300 identifies the assigned range of the virtual node 2 paired with the dummy data by referring to the pre-change (old generation) distributed hash table 100b.

  The missing data identification processing unit 300 includes a node increase / decrease detection unit (increase / decrease detection unit) 301, a dummy data registration unit 302, and a dummy data deletion unit 303.

  The node increase / decrease detection unit (increase / decrease detection unit) 301 receives the distributed hash table change notification signal from the distribution server 1 via the input / output unit 31, and stores the changed (new generation) distributed hash in the storage unit 32. Store as table 100a. At that time, the node increase / decrease detection unit 301 stores the pre-change distributed hash table 100 stored in the storage unit 32 before that time as the old generation distributed hash table 100b in the storage unit 32 as it is. Keep it.

  The node increase / decrease detection unit 301 compares the new-generation distributed hash table 100a stored in the storage unit 32 with the old-generation distributed hash table 100b, and detects the difference between them to thereby detect the virtual node 2 (data management server ) Is detected. Specifically, the node increase / decrease detection unit 301 detects that the virtual node 2 has been added by inserting a new record of the virtual node 2 into the new generation distributed hash table 100a. On the other hand, the node increase / decrease detection unit 301 detects that the virtual node 2 has disappeared when the record of the existing virtual node 2 is deleted in the new generation distributed hash table 100a.

  When the dummy data registration unit 302 sets up the cluster, that is, when the old-generation distributed hash table 100b does not exist in the storage unit 32, or detects that the virtual node 2 has been added, the dummy data registration unit 302 The key generation unit 310 is requested to generate a dummy data key corresponding to the virtual node ID. Then, the dummy data registration unit 302 generates a dummy data registration request signal (FIG. 6) and transmits it to the distribution server 1.

FIG. 6 is a diagram showing a data configuration of the dummy data registration request signal according to the present embodiment.
The dummy data registration request signal is registered with request type information indicating “registration” as a request type, and dummy data key information used as a consistent hash key in the distribution processing unit 10 of the distribution server 1. Data and value. Here, a value for specifying the virtual node 2 such as a virtual node ID is set as the value.

  Returning to FIG. 4, when the node increase / decrease detection unit 301 detects that the virtual node 2 constituting the cluster has disappeared, the dummy data deletion unit 303 detects the disappearance from the dummy data key table 200 described later. A dummy data key corresponding to the virtual node 2 is acquired, a dummy data deletion request signal (FIG. 7) is generated, and transmitted to the distribution server 1.

FIG. 7 is a diagram illustrating a data configuration of the dummy data deletion request signal according to the present embodiment.
The dummy data deletion request signal corresponds to the request type information indicating “delete” as the request type and the virtual node 2 in which the disappearance is detected in the distribution processing unit 10 of the distribution server 1 to be used as a consistent hash key. Key information of dummy data to be configured.

  Further, the dummy data deletion unit 303 receives a response signal (FIG. 8) to the dummy data deletion request signal from the distribution server 1.

FIG. 8 is a diagram illustrating a data configuration of the response signal according to the present embodiment.
The response signal stores either “OK” or “NG” as the response type.

When the dummy data deletion unit 303 receives a response signal with “OK” from the distribution server 1, the dummy data deletion unit 303 detects the virtual node 2 in which the disappearance is detected in the dummy data key table 200 (see FIG. 9). Delete the record corresponding to the node ID.
On the other hand, when receiving a response signal with “NG” from the distribution server 1, the dummy data deletion unit 303 determines that the data of the virtual node 2 corresponding to the dummy data is missing. Then, the determination result is output to the alarm notification unit 320.

  As shown in FIG. 9, the key generation unit 310 is information that associates an identifier unique to the data management server or the virtual node 2 (here, “virtual node ID”) and a key of dummy data. A dummy data key table 200 is generated.

FIG. 9 is a diagram showing a data configuration of the dummy data key table 200 according to the present embodiment.
As illustrated in FIG. 9, the dummy data key table 200 stores combinations of virtual node IDs and dummy data keys as records corresponding to the number of virtual nodes 2.

  The key generation unit 310 acquires the value of the virtual node ID registered in the dummy data key table 200, and uses this value as the dummy data input destination virtual node ID. The key generation unit 310 generates a dummy data key using the following key generation algorithm based on the dummy data input destination virtual node ID and the distributed hash table 100 (100a).

(Key generation algorithm)
FIG. 10 is a flowchart for explaining a dummy data key generation algorithm executed by the key generation unit 310 according to the present embodiment.
Here, since the hash function is a one-way function, a key cannot be generated from the hash value. Therefore, the key generation unit 310 generates a key of dummy data by finding a key included in the assigned range of the virtual node 2 by trial and error.
Note that the key generation unit 310 refers to the distributed hash table 100 (100a) when generating the dummy data key, obtains information on all virtual node IDs, and acquires each virtual node ID in the dummy data key table 200. Generate a record.

  First, the key generation unit 310 acquires the value of the virtual node ID registered in one row (acquired sequentially from the first row) of the dummy data key table 200, and sets this value as the dummy data input destination virtual node ID. Then, a random character string is generated as a key for requesting registration of dummy data to the dummy data input destination virtual node (step S1). Alternatively, the key generation unit 310 may generate a key obtained by connecting a character string randomly formed to a meaningful character string such as “dummydata”.

  Next, the key generation unit 310 calculates the hash value of the key generated in step S1 using the same hash function as that of the distribution processing unit 10 of the distribution server 1 (step S2).

  Subsequently, the key generation unit 310 refers to the distributed hash table 100 (100a) using the generated hash value of the key, and calculates the virtual node ID of the allocation destination virtual node whose assigned range is the hash value of the key. (Step S3).

  Then, the key generation unit 310 determines whether or not the calculated virtual node ID matches the dummy data input destination virtual node ID (step S4).

  If they match as a result of the determination in step S4 (step S4 → Yes), the virtual node ID (dummy data input destination virtual node ID) obtained in step S1 of the dummy data key table 200 is used as the key generated in step S1. Are registered in the dummy data key table 200 (step S5).

  On the other hand, if they do not match as a result of the determination in step S4 (step S4 → No), the process returns to step S1 to generate a random character string again as a key for dummy data.

  When the key generation unit 310 registers the dummy data key corresponding to the virtual node ID of one row of the dummy data key table 200 in step S5, the key generation unit 310 acquires the virtual node ID of the next row, and performs steps S1 to S5. repeat. The key generation unit 310 executes this for the virtual node IDs of all rows stored in the dummy data key table 200.

  In this way, the key generation unit 310 generates a dummy data key corresponding to each virtual node ID and registers it in the dummy data key table 200.

  Returning to FIG. 4, the alarm notification unit 320 corresponds to the key of the dummy data when the dummy data deletion unit 303 determines that data has been lost by receiving the response signal with “NG”. The assigned range of the virtual node 2 is specified by referring to the old generation distributed hash table 100b. Then, the alarm notification unit 320 generates a data loss detection signal with the assigned range of the virtual node 2 as the range of the hash value of the missing data key, and notifies the alarm monitoring device 4 of the data loss detection signal. Note that the alarm notification unit 320 may transmit the data loss detection signal to the alarm monitoring device 4 as an SNMP (Simple Network Management Protocol) trap, for example.

≪Alarm monitoring device≫
The alarm monitoring device 4 receives the data loss detection signal from the missing data specifying device 3, and outputs the fact that the data loss has been detected and the hash value range of the key of the missing data to the output means (not shown). .
The alarm monitoring device 4 is a device including a control unit, an input / output unit, and a storage unit (all not shown), and the alarm reception unit 40 included in the control unit receives a data loss detection signal, The administrator of the distributed database system 1000 is notified of data loss detection.

<Distributed database system missing data identification processing>
Next, the missing data specifying process executed by the distributed database system 1000 according to the present embodiment will be described.
In the distributed database system 1000, the missing data identification device 3 registers dummy data for all virtual nodes 2 when the data management server first sets up a cluster of a plurality of virtual nodes 2.
Thereafter, the missing data specifying device 3 and the distribution server 1 pair with each virtual node 2 existing in the cluster by registering and deleting dummy data in accordance with changes such as addition or disappearance of the virtual node 2. The dummy data to be maintained is always registered.
Then, when the virtual node 2 disappears, the missing data identification device 3 transmits a dummy data deletion request (dummy data deletion request signal) of the virtual node 2 that has disappeared to the distribution server 1 and deletes it. Whether or not the data of the virtual node 2 is missing data is determined.
Regarding the flow of the above missing data specifying process, (1) dummy data registration process at the time of cluster setup, (2) dummy data registration process at the time of adding a virtual node, (3) dummy data deletion process at the time of disappearance of a virtual node, (4 ) Detailed description will be given in the order of data loss detection and alarm information output processing.

≪ (1) Dummy data registration process during cluster setup≫
FIG. 11 is a sequence diagram showing dummy data registration processing executed for all virtual nodes 2 when setting up the data management server in a cluster.

First, the distribution server 1 (the distribution processing unit 10 and the alive monitoring unit 11) generates a distributed hash table 100 as shown in FIG. 5 when setting up a cluster of a plurality of virtual nodes 2 formed on the data management server. (Step S10). Then, the distributed hash table change notification unit 12 of the distribution server 1 transmits a distributed hash table change notification signal including the generated distributed hash table 100 to the missing data specifying device 3 (step S11).
The distributed hash table change notifying unit 12 transmits the distributed hash table change notification signal including the generated distributed hash table 100 to each virtual node 2, so that the same distributed hash is transmitted to each virtual node 2. The table 100 is stored. By referring to this distributed hash table 100, each virtual node 2 identifies the virtual node 2 that is the copy destination of the original data stored in itself.

  When the node increase / decrease detection unit 301 of the missing data specifying device 3 receives the distributed hash table change notification signal, there is no distributed hash table 100 already registered in the storage unit 32. The distributed hash table 100 is extracted from the distributed hash table change notification signal, and the distributed hash table 100 is stored in the storage unit 32 (step S12).

  Next, the key generation unit 310 of the missing data specifying device 3 generates the dummy data key table 200 based on the virtual node ID of the distributed hash table 100 stored in the storage unit 32, and the dummy data described in FIG. The key generation process is executed (step S13). Then, the key generation unit 310 completes the dummy data key table 200 (FIG. 9) and stores it in the storage unit 32.

Subsequently, the dummy data registration unit 302 of the missing data specifying device 3 refers to the dummy data key table 200, generates a dummy data registration request signal (FIG. 6) (step S14), and transmits it to the distribution server 1.
In this dummy data registration request signal, a request type “registration”, a key of the generated dummy data, and a value are set. As the value, for example, a value specifying the virtual node 2 such as a virtual node ID is set with reference to the dummy data key table 200.

  Upon receiving the dummy data registration request signal, the distribution server 1 uses the distribution processing unit 10 to calculate the hash value by applying the dummy data key set in the dummy data registration request signal to the hash function. Then, the distribution processing unit 10 refers to the distributed hash table 100 stored in the storage unit (not shown), determines the virtual node 2 that is the distribution destination, and transfers the dummy data registration request signal (step) S15). At this time, the virtual node 2 as the transfer destination of the dummy data registration request signal is set as the dummy data input destination virtual node set by the key generation unit 310 of the missing data specifying device 3, that is, as the value of the dummy data registration request signal. Virtual node 2.

  Subsequently, the virtual node 2 that has received the dummy data registration request signal stores the dummy data in its own storage unit (step S16). For the distribution server 1 and the virtual node 2 of the data management server, the dummy data is data that is not different from normal data, and the dummy data stored here is dummy data (original). The data management unit 20 of the virtual node 2 refers to the distributed hash table 100 stored in itself according to the consistent hash algorithm, generates dummy data (duplicates) for the number of replicas set in advance, The dummy data (replica) is transmitted to and stored in another virtual node 2 in charge of data storage.

  When receiving the response information indicating that the dummy data has been registered from the predetermined virtual node 2, the distribution processing unit 10 of the distribution server 1 transmits an OK response signal to the missing data specifying device 3 (step S17).

  In this way, the dummy data registration unit 302 of the missing data specifying device 3 registers dummy data in all virtual nodes 2. As a result, preparation for missing data detection for each virtual node 2 by the missing data specifying device 3 is completed.

≪ (2) Dummy data registration process when adding virtual node≫
When a physical database management server is added after the operation of the distributed database system 1000, the virtual node 2 constituting the cluster is added. When a new virtual node 2 is added, a virtual node 2 in which the assigned range of hash values is divided among the existing virtual nodes 2 is generated. The dummy data registration process when adding a virtual node will be described below with reference to FIGS.

  FIG. 12 is a diagram for explaining division of the assigned range of the existing virtual node 2 when the virtual node 2 is added. FIG. 13 is a diagram for explaining the change of the distributed hash table 100 when the virtual node 2 is added.

When a virtual node 2 (virtual node “X” in FIG. 12) is added by adding a data management server or the like, the virtual node ID “X” of the added virtual node 2 is added as shown in FIG. A point (circle) of the hash value “x” corresponding to “” is added on the consistent hash ring. As a result, the assigned range (hash values “a” to “b”) of the existing virtual node “B” becomes the same as the assigned range (hash values “a” to “x”) of the added virtual node “X”. To the virtual node “B” in charge (hash values “x” to “b”). In the distributed hash table 100, as shown in FIG. 13, a record with a virtual node ID “X” between a record with a virtual node ID “A” and a record with a virtual node ID “B”. Will be inserted.
As a result, it is necessary to newly perform dummy data registration processing for the added virtual node “X”. Furthermore, for the virtual node “B”, the assigned range of the hash value is divided, so that the already registered dummy data belongs to the assigned range of the virtual node “X” or the assigned range of the virtual node “B”. Since it does not know whether it belongs, after deleting the existing dummy data, the dummy data is re-registered in the assigned range of the virtual node “B” after the division.

  For the virtual node “B” in which the assigned range is divided, the dummy data is not deleted and re-registered, and the existing dummy data is assigned to the assigned range of the new virtual node “X” based on the dummy data key table 200. It may be determined whether the data belongs to or belongs to the range in charge of the original virtual node “B”, and dummy data registration processing may be performed on the virtual node 2 on which no dummy data is registered.

  Next, the flow of dummy data registration processing will be described. 14 and 15 are sequence diagrams for explaining the dummy data registration process when the virtual node 2 is added.

  As shown in FIG. 14, first, when the alive monitoring unit 11 of the distribution server 1 detects the addition of a new virtual node 2 (data management server) (step S20), the distributed virtual machine table 100 is referred to and an existing virtual table 2 is referred to. Based on the size of the assigned range of the node 2, for example, the virtual node 2 is selected in order of increasing assigned range, and the virtual node to be divided is selected from the existing virtual nodes 2.

  The alive monitoring unit 11 adds a point (circle) of the hash value of the virtual node ID of the virtual node 2 to be newly added to the assigned range of the selected virtual node 2 as a consistent hash ring. To divide (see FIG. 12). This is realized by the life and death monitoring unit 11 adding the record of the newly added virtual node 2 in the distributed hash table 100 immediately before the record of the virtual node to be divided (see FIG. 13). As described above, the alive monitoring unit 11 adds a new virtual node 2 and changes the distributed hash table 100 (step S21).

  Subsequently, when the distributed hash table 100 is changed as described above, the distributed hash table change notification unit 12 of the distribution server 1 generates a distributed hash table change notification signal including the changed distributed hash table 100, and the missing data. It transmits to the specific device 3 (step S22). Similarly, the distributed hash table change notification unit 12 transmits a distributed hash table change notification signal including the changed distributed hash table 100 to each virtual node 2 (including the added virtual node 2). deep. In this way, processing can be executed based on the same distributed hash table 100 in the distributed database system 1000.

  When the node increase / decrease detection unit 301 of the missing data specifying device 3 receives the distributed hash table change notification signal, since the previous distributed hash table 100 exists in the storage unit 32, it is the second and subsequent registrations of the distributed hash table. Recognize. Then, the node increase / decrease detection unit 301 sets the distributed hash table 100 already existing in the storage unit 32 as the old generation distributed hash table 100 (100b), and uses the newly received distributed hash table 100 as the new generation distributed hash table 100. (100a) is stored in the storage unit 32 (step S23).

  The node increase / decrease detection unit 301 compares the new generation distributed hash table 100a with the old generation distributed hash table 100b (step S24). Then, as a result of the comparison, the node increase / decrease detection unit 301 detects that a new virtual node 2 has been added to the new generation distributed hash table 100a (step S25).

  When the node increase / decrease detection unit 301 detects that the virtual node 2 has been added, the dummy data registration unit 302 executes dummy data registration processing for the added virtual node 2 (hereinafter, “procedure A”). Sometimes called). On the other hand, in the distributed hash table 100, the assigned range is divided and changed for the virtual node 2 of the record next to the added virtual node 2 record, so the dummy data deletion unit 303 deletes the existing dummy data, Thereafter, the dummy data registration unit 302 executes a registration process again (hereinafter, referred to as “procedure B”, see FIG. 15). This will be specifically described below.

(Procedure A: Dummy data registration process to the added virtual node)
First, the key generation unit 310 of the missing data specifying device 3 adds a record of the virtual node ID of the virtual node 2 added to the dummy data key table 200 (FIG. 9) stored in the storage unit 32.
Then, the key generation unit 310 executes the dummy data key generation processing described with reference to FIG. 10 based on the added virtual node ID of the virtual node 2 and the new generation distributed hash table 100a. A dummy data key corresponding to the virtual node ID of the virtual node 2 is generated (step S26). Then, the key generation unit 310 registers the generated dummy data key in the dummy data key table 200.

  Subsequently, the dummy data registration unit 302 of the missing data specifying device 3 refers to the dummy data key table 200 and generates a dummy data registration request signal (FIG. 6) (step S27). Here, the dummy data registration unit 302 sets the request type to “registration”, and generates a dummy data registration request signal in which a key and a value of the generated dummy data are set. Here, in the value, a value that identifies the added virtual node 2 such as the virtual node ID of the added virtual node 2 is set. Then, the dummy data registration unit 302 transmits the generated dummy data registration request signal to the distribution server 1.

  Then, the distribution server 1 that has received the dummy data registration request signal determines the added virtual node 2 as the distribution destination based on the dummy data key set in the dummy data registration request signal, and outputs the dummy data registration request signal. Transfer (step S28).

  Next, the virtual node 2 that has received the dummy data registration request signal stores the dummy data (original) in its storage unit (step S29). The virtual node 2 generates dummy data (replica), and transmits and stores the replicated data to the other virtual node 2 in charge according to the consistent hash algorithm.

  When the dummy data is correctly registered in the predetermined virtual node 2, the distribution processing unit 10 of the distribution server 1 transmits an OK response signal to the missing data specifying device 3 (step S30).

(Procedure B: Deletion and re-registration processing of dummy data to an existing virtual node)
Next, a description will be given of deletion and re-registration processing of dummy data to an existing virtual node 2 in which the assigned range is divided and changed by adding a new virtual node 2.

  As shown in FIG. 15, first, the dummy data deletion unit 303 of the missing data specifying device 3 refers to the dummy data key table 200 (FIG. 9) and determines the dummy data key of the virtual node 2 in which the assigned range is divided. Obtaining and setting the request type as “delete”, a dummy data deletion request signal is generated (step S31). Then, the dummy data deletion unit 303 transmits the generated dummy data deletion request signal to the distribution server 1.

  Then, the distribution server 1 that has received the dummy data deletion request signal determines the corresponding virtual node 2 as the distribution destination according to the consistent hash algorithm based on the dummy data key set in the dummy data deletion request signal, The dummy data deletion request signal is transferred (step S32).

  The virtual node 2 that has received the dummy data deletion request signal deletes the dummy data (original) (step S33). In addition, the virtual node 2 transmits a dummy data deletion request signal to the other virtual node 2 that stores the dummy data (replication) to delete the dummy data (replication).

  When the dummy data is deleted, the distribution server 1 transmits an OK response signal to the missing data specifying device 3 (step S34).

  The dummy data deletion unit 303 of the missing data specifying apparatus 3 that has received the OK response signal to the dummy data deletion request signal, from the dummy data key table 200 in the storage unit 32, the virtual node of the virtual node 2 to which the assigned range is divided The ID record is deleted (step S35).

  Subsequently, the dummy data registration unit 302 executes dummy data registration processing (re-registration) of the virtual node 2 in which the assigned range is divided (step S36). The dummy data registration process (re-registration) is the same as steps S26 to S30 (see FIG. 14) of the procedure A described above.

  In this way, even when a new virtual node 2 is added, it is possible to maintain a state in which dummy data associated with each virtual node 2 is stored in each virtual node 2.

≪ (3) Dummy data deletion process when virtual node disappears≫
Next, dummy data deletion processing when the virtual node 2 disappears due to a failure of the data management server or the like will be described with reference to FIG. Here, a case where dummy data can be deleted in the dummy data deletion processing will be described.
FIG. 16 is a sequence diagram for explaining dummy data deletion processing when the virtual node 2 disappears. When the virtual node 2 (data management server) disappears, the missing data specifying device 3 determines whether or not data loss (both original data and duplicated data disappeared) depends on whether or not dummy data can be deleted. . When a certain virtual node 2 (data management server) disappears, the distribution server 1 transfers the dummy data deletion request to the other virtual node 2 storing the duplicate data of the corresponding data. Here, the fact that the corresponding data can be deleted means that the data does not disappear in the other virtual node 2, and the fact that the corresponding data cannot be deleted means that the duplicate data is stored. This is because it means that the data has disappeared also in the other virtual node 2 that has been stored.

  First, when the alive monitoring unit 11 of the distribution server 1 detects the occurrence of a failure in the virtual node 2 (step S40), the record of the virtual node 2 in which the failure has occurred is deleted with reference to the distributed hash table 100 (FIG. 5). Then, the distributed hash table 100 is changed (step S41).

  Subsequently, when the distributed hash table 100 is changed as described above, the distributed hash table change notification unit 12 of the distribution server 1 generates a distributed hash table change notification signal including the changed distributed hash table 100, and the missing data. It transmits to the specific device 3 (step S42). Similarly, the distributed hash table change notification unit 12 transmits a distributed hash table change notification signal including the changed distributed hash table 100 to each virtual node 2.

  When the node increase / decrease detection unit 301 of the missing data specifying device 3 receives the distributed hash table change notification signal, since the previous distributed hash table 100 exists in the storage unit 32, it is the second and subsequent registrations of the distributed hash table. Recognize. Then, the node increase / decrease detection unit 301 sets the distributed hash table 100 already existing in the storage unit 32 as the old generation distributed hash table 100 (100b), and uses the newly received distributed hash table 100 as the new generation distributed hash table 100. (100a) is stored in the storage unit 32 (step S43).

  The node increase / decrease detection unit 301 compares the new generation distributed hash table 100a with the old generation distributed hash table 100b (step S44). Then, as a result of the comparison, the node increase / decrease detection unit 301 detects that the virtual node 2 existing in the old generation distributed hash table 100b has been deleted in the new generation distributed hash table 100a (step S45). ).

  When the node increase / decrease detection unit 301 detects that the virtual node 2 has been deleted, the dummy data deletion unit 303 of the missing data identification device 3 refers to the dummy data key table 200 (FIG. 9), and the deleted virtual The dummy data key of the node 2 is acquired, the request type is set to “delete”, and a dummy data deletion request signal is generated (step S46). Then, the dummy data deletion unit 303 transmits the generated dummy data deletion request signal to the distribution server 1. Subsequent processes (steps S47 to S51) executed by the missing data specifying device 3 are the same as the processes of the above-described steps S32 to S36 (see FIG. 15), and thus description thereof is omitted.

Here, when the existing virtual node 2 disappears due to a failure of the data management server or the like, the assigned range of the virtual node 2 is merged with another virtual node 2 according to the consistent hash algorithm. The merge-destination virtual node 2 is the virtual node 2 of the next record of the virtual node 2 that disappeared in the old-generation distributed hash table 100b. If the virtual node 2 has also disappeared, the virtual node 2 of the next record becomes the merge destination, and if the virtual node 2 has also disappeared, the virtual node 2 of the next record becomes the merge destination, Hereinafter, the record of the virtual node 2 that has not disappeared becomes the merge destination.
The missing data specifying device 3 (dummy data deletion unit 303 and dummy data registration unit 302) executes dummy data deletion processing and re-registration processing on the merged virtual node 2.
By doing so, even when the virtual node 2 disappears due to a failure of the data management server or the like, the state in which the dummy data associated with each virtual node 2 is stored in each virtual node 2 is maintained. be able to.

≪ (4) Data loss detection and alarm information output processing≫
Next, data loss detection and alarm information output processing by the missing data specifying device 3 will be described.
In the distributed database system 1000 according to the present embodiment, when a plurality of data management devices (physical nodes) exceeding the number of data replications fail at a time, data loss occurs in units of the assigned range of the virtual node 2, Dummy data included in the assigned range is also missing. At this time, since the virtual node 2 that has managed the missing data also disappears, the above-described processing of steps S40 to S47 (see FIG. 16) is attempted as processing when the virtual node 2 disappears. However, the dummy data cannot be deleted because the dummy data is also missing along with the missing data. Therefore, the missing data specifying device 3 receives the response information “NG” from the distribution server 1. This process and after will be described in detail with reference to a sequence diagram.

  FIG. 17 is a sequence diagram for explaining data loss detection and alarm information output processing when the virtual node 2 disappears.

  The virtual node 2 that has received the dummy data deletion request signal tries to delete the dummy data. However, since all the dummy data (original) and dummy data (duplicate) have disappeared, the dummy data does not exist and is deleted. I can't. Therefore, the distribution server 1 transmits an NG response signal to the missing data identification device 3 (step S60).

  When receiving the NG response signal, the dummy data deleting unit 303 of the missing data specifying device 3 determines that the data of the virtual node 2 corresponding to the dummy data is missing (detects data loss: step S61).

  When data loss is detected, the alarm notification unit 320 uses the virtual node ID of the virtual node 2 in which the deletion of dummy data is NG and the hash value of the key of the missing data based on the distributed hash table 100b of the previous generation. Is specified (step S62). The range of the hash value of the key of the missing data is a range in charge of the virtual node 2 in which the deletion of the dummy data is NG.

  The alarm notification unit 320 generates a data loss detection signal including information on the hash value range of the lost data (step S63), and notifies the alarm reception unit 40 of the alarm monitoring device 4 of the data loss detection signal.

  As described above, in the distributed database system 1000 according to the present embodiment, when a data management server that constitutes a cluster fails, data loss is detected by the dummy data deletion process of the missing data specifying device 3, The range of the hash value of the key of the missing data can be specified. Therefore, it is possible to provide a distributed database system that can be quickly recovered even if the data management server fails.

  Note that the distributed database system 1000 according to the present embodiment is not limited to the above-described embodiment. For example, in this distributed database system 1000, the distribution destination of the distribution server 1 is described as a plurality of virtual nodes 2. However, the distribution destination is not necessarily the virtual node 2, and a plurality of data managements in which the distribution destination is a physical node. Even if it is a server, the process which should be performed by the missing data specifying device 3 and the features of the present invention are not changed.

1 Distribution server 2 Virtual node (data management server)
3 Missing data identification device 4 Alarm monitoring device 5 Client device 10 Distribution processing unit 11 Alive monitoring unit 12 Distributed hash table change notification unit (change notification unit)
20 Data management unit 30 Control unit 31 Input / output unit 32 Storage unit 40 Alarm reception unit 100 Distributed hash table (distributed hash information)
100a Distributed hash table (distributed hash information) (new)
100b Distributed hash table (distributed hash information) (old)
200 Dummy data key table 300 Missing data identification processing unit 301 Node increase / decrease detection unit (change increase / decrease detection unit)
302 Dummy data registration unit 303 Dummy data deletion unit 310 Key generation unit 320 Alarm notification unit 1000 Distributed database system

Claims (2)

A distribution server that distributes the request signal received from the client device to a plurality of data management servers, a plurality of the data management servers that execute processing on data stored by itself based on the request signal received from the distribution server, A distributed database system comprising a missing data identification device that detects missing data stored in a data management server,
The distribution server arranges hash values calculated by applying an identifier unique to the data management server to a predetermined hash function on the consistent hash ring in ascending order, and is unique to the data attached to the received request signal. Adopting a distribution method to determine the data management server to which the received request signal is distributed by arranging the hash value calculated by applying the key to the predetermined hash function,
Each of the data management servers stores data corresponding to the request signal distributed to itself as original data, and with respect to another data management server having a predetermined number of replicas adjacent on the consistent hash ring, Adopting a redundancy method of storing the duplicate data in another data management server by generating and sending duplicate data of the original data,
The distribution server is
A storage unit for storing distributed hash information for storing a range of hash values of the key of the data, indicating a responsible range to which each of the data management servers is assigned on the consistent hash ring;
A distribution processing unit that calculates a hash value of the key attached to the received request signal and transfers the request signal to the data management server of the distribution destination determined based on the distributed hash information;
A life and death monitoring unit that monitors the life and death of the plurality of data management servers, and changes the assigned range of the distributed hash information when there is a change in the life and death state,
A change notification unit that transmits a distributed hash information change notification signal including the changed distributed hash information to the missing data specifying device, and
The missing data specifying device includes:
A storage unit that stores the distributed hash information included in the received distributed hash information change notification signal as a new generation distributed hash information, and stores the distributed hash information received before the reception as an old generation distributed hash information; ,
An increase / decrease detection unit that compares the new generation distributed hash information with the old generation distributed hash information and determines an increase / decrease in the data management server;
As a result of the determination, if the number of data management servers has increased, or if the old generation distributed hash information is not stored, dummy data is generated in association with each of the data management servers, and the generation A dummy data registration unit that registers the dummy data by transmitting a dummy data registration request signal including the dummy data to the data management server that is a pair of the dummy data via the distribution server;
If the data management server has decreased as a result of the determination, a dummy data deletion request signal for requesting deletion of dummy data associated with the decreased data management server is sent via the distribution server to the decrease. A dummy data deletion unit that detects the loss of data by not being able to delete the dummy data associated with the reduced data management server,
The range of the hash value of the key of the data indicated by the assigned range on the consistent hash ring of the reduced data management server is extracted based on the distributed hash information of the old generation, and the original data and the predetermined data And a warning notification unit that identifies the key range of missing data in which duplicated data of the number of duplicates is lost. The missing data specifying device includes:
A range that is assigned to a data management server that is generated by randomly generating a key and having the hash value calculated by applying the key to the predetermined hash function extracted from the data management server indicated in the new-generation distributed hash information Until the key is included and the calculation of the hash value of the key is repeated, and when the key is included in the assigned range to be the distribution destination of the extracted data management server, the key is extracted. A key generation unit that determines the key of the dummy data associated with the data management server,
The distributed database system according to claim 1, further comprising:

Images