Data Storage Method and Apparatus in Distributed Storage System, and Computer Program Product

Abstract

The solution discloses a data storage method in a distributed storage system: A client generates data of a stripe and concurrently sends data of strips in the stripe to storage nodes corresponding to the strips, so that data interaction between the storage nodes is reduced, and write concurrency is improved; and the storage node receives the data of the strips sent by the client and records a sequence of data of strips corresponding to a same logical address, so that write performance of a distributed block storage system is improved and data consistency is ensured.

Description

TECHNICAL FIELD

The present invention relates to the field of information technologies, and in particular, to a data storage method and apparatus in a distributed storage system, and a computer program product.

BACKGROUND

A distributed storage system includes a storage node and a stripe, each stripe includes a plurality of strips, and a storage node corresponds to a strip in the stripe, that is, a storage node provides storage space for a strip in the stripe. Generally, as shown in FIG. 1, a plurality of storage nodes may include a primary storage node (a storage node 1). The primary storage node is configured to receive data sent by a client, and then the primary storage node selects a stripe, divides the data into data of strips, and sends data of strips stored in other storage nodes to corresponding storage nodes (a storage node 2, a storage node 3, and a storage node 4). The foregoing operations ensure data consistency in the distributed storage system, but the primary storage node may easily become a data writing bottleneck. In addition, data interaction between the storage nodes is increased, and write performance of the distributed storage system is reduced.

SUMMARY

This application provides a data storage method and apparatus in a distributed storage system, and a computer program product, without requiring a primary storage node, so that under the premise of ensuring data consistency, data interaction between storage nodes is reduced, and write performance of the distributed storage system is improved.

A first aspect of this application provides a data storage method in a distributed storage system. The distributed storage system includes M storage nodes N_j, and j is an integer ranging from 1 to M. In this method, the storage node N_j receives data of a strip SU_Nj in a stripe S_N, where the stripe S_N includes M strips SU_Nj; the storage node N_j receives data of a strip SU_Kj in a stripe S_K, where a logical address of the data of the strip SU_Kj is the same as a logical address of the data of the strip SU_Nj, N is different from K, and the stripe S_K includes M strips SU_Kj; and the storage node N_j generates a record, where the record is used to indicate that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj. In this data storage method, the storage node N_j receives the data of the strip SU_Nj and the data of the strip SU_Kj directly from a client and indicates, by using the record, that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj, so that when the storage node N_j stores a plurality of pieces of data of the same logical address, data of a latest version may be determined based on the record. In this data storage method, a primary storage node is not required, and therefore, data interaction between storage nodes is reduced, and write performance of the distributed storage system is improved.

In implementation, the record in the first aspect of this application may indicate, based on an identifier of the strip SU_Nj and an identifier of the strip SU_Kj, that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj. The data of the strip SU_Nj and the data of the strip SU_Kj may be sent by a same client or by different clients.

With reference to the first aspect of this application, in a possible implementation, the storage node N_j backs up the record to one or more other storage nodes in the M storage nodes, to improve reliability of the record. When the storage node N_j becomes faulty, the record is obtained from a storage node in which the record is backed up, and between the restored data of the strip SU_Nj and the restored data of the strip SU_Kj, the data of the strip SU_Kj is determined as latest data based on the record.

With reference to the first aspect of this application and the foregoing possible implementations, in a possible implementation, the storage node N_j receives a read request sent by a third client, where the read request includes the logical address; the storage node N_j determines, by querying the record based on the logical address, that the data of the strip SU_Kj is latest data; and the storage node N_j returns the data of SU_Kj to the third client.

With reference to the first aspect of this application and the foregoing possible implementations, in a possible implementation, the strip SUN and the strip SU_Kj are allocated from the storage node N_j by a stripe metadata server.

A second aspect of this application provides a storage node in a distributed storage system. The distributed storage system includes M storage nodes N_j, where j is an integer ranging from 1 to M. The storage node serves as the storage node N_j, and includes a plurality of units that are configured to implement the first aspect of this application and various implementations of the first aspect.

A third aspect of this application provides a storage node in a distributed storage system. The distributed storage system includes M storage nodes N_j, where j is an integer ranging from 1 to M. The storage node serves as the storage node N_j, and includes an interface and a processor that communicate with each other. The processor is configured to implement the first aspect of this application and various implementations of the first aspect.

A fourth aspect of this application provides a computer program product. The computer program product includes a computer instruction, and the computer instruction is used by a storage node in a distributed storage system. The distributed storage system includes M storage nodes N_j, where j is an integer ranging from 1 to M. The storage node serves as the storage node N_j, and runs the computer instruction to implement the first aspect of this application and various implementations of the first aspect.

A fifth aspect of this application provides a data storage method in a distributed storage system. The distributed storage system includes M storage nodes N_j, where j is an integer ranging from 1 to M. The method includes: A first client receives a first write request, where the first write request includes first data and a logical address; the first client divides the first data into data of one or more strips SU_Nj in a stripe S_N, where the stripe S_N includes M strips SU_Nj; the first client sends the data of the one or more strips SU_Nj to the storage node N_j; a second client receives a second write request, where the second write request includes second data and the logical address; the second client divides the third data into data of one or more strips SU_Kj of a stripe S_K, where N is different from K, and the stripe S_K includes M strips SU_Kj; and the second client sends the data of the one or more strips SU_Kj to the storage node N_j.

With reference to the fifth aspect of this application, in a possible implementation, the strip SU_Nj and the strip SU_Kj are allocated from the storage node N_j by a stripe metadata server.

With reference to the fifth aspect of this application, in a possible implementation, each piece of the data of the one or more strips SU_Nj further includes data strip status information, and the data strip status information is used to indicate whether each data strip in the stripe S_N is empty.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 is a schematic diagram of data storage in a distributed storage system in the prior art;

FIG. 2 is a schematic diagram of a distributed block storage system according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a server in a distributed block storage system according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a partition view of a distributed block storage system according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a relationship between a strip and a storage node in a distributed block storage system according to an embodiment of the present invention;

FIG. 6 is a flowchart of a method for writing data by a client in a distributed block storage system according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of determining a partition by a client in a distributed block storage system according to an embodiment of the present invention;

FIG. 8 is a flowchart of a method for storing data by a storage node in a distributed block storage system according to an embodiment of the present invention; and

FIG. 9 is a schematic structural diagram of a storage node in a distributed block storage system according to an embodiment of the present invention.

EMBODIMENTS OF THE PRESENT INVENTION

A distributed storage system in an embodiment of the present invention includes a Huawei® Fusionstorage® series and an OceanStor® 9000 series. For example, as shown in FIG. 2, the distributed storage system includes a plurality of servers such as a server 1, a server 2, a server 3, a server 4, a server 5, and a server 6. The servers communicate with each other by using InfiniBand, an Ethernet, or the like. In this embodiment of the present invention, the server in the distributed storage system is also referred to as a storage node. In actual application, a quantity of servers in the distributed storage system may be increased based on an actual requirement. This is not limited in this embodiment of the present invention.

The server in the distributed storage system includes a structure shown in FIG. 3. As shown in FIG. 3, each server in the distributed storage system includes a central processing unit (CPU) 301, a memory 302, an interface 303, a hard disk 1, a hard disk 2, and a hard disk 3. The memory 302 stores a computer instruction. The CPU 301 executes the computer instruction in the memory 302 to perform a corresponding operation. The interface 303 may be a hardware interface such as a network interface card (NIC) or a host bus adapter (HBA), or may be a program interface module or the like. The hard disk includes a solid state disk (SSD), a mechanical hard disk, or a hybrid hard disk. For example, the mechanical hard disk is an Hard Disk Drive (HDD). A hard disk interface may be a serial advanced technology attachment (SATA) interface, a serial attached small computer system interface (SAS) interface, a fiber channel (FC) interface, a peripheral component interconnect express (PCIe) interface, a non-volatile memory express (NVMe) interface, or the like. The CPU 201 may be replaced with a field programmable logic gate array (FPGA) or other hardware. Alternatively, an FPGA or other hardware and the CPU 301 jointly perform the foregoing corresponding operation. For ease of description, in this embodiment of the present invention, the CPU 301 and the memory 302 are referred to as a processor, or hardware that replaces the CPU 301 and the memory 302 is referred to as a processor, or a combination of the CPU 301, the memory 302, and other hardware is referred to as a processor.

A client in the distributed storage system writes data into the distributed storage system based on a write request from a host or reads data from the distributed storage system based on a read request from a host. The server shown in this embodiment of the present invention may serve as the client. In addition, the client may alternatively be a device independent of the server shown in FIG. 3. A form of the host in this embodiment of the present invention may be a server, a virtual machine (VM), a terminal device, or the like. This is not limited in this embodiment of the present invention. The client in the distributed storage system provides a storage resource in the distributed storage system for the host. For example, in a distributed block storage system, the client provides a block resource such as a logical unit for the host, to provide a data access operation for the host. The logical unit is also referred to as a logical unit number (LUN). In a distributed file storage system, a client provides a file resource for a host. In a distributed object storage system, a client provides an object resource for a host.

In an embodiment of the present invention, a distributed block storage system is used as an example. A client provides a block protocol access interface, so that the client provides a distributed block storage access point service. A host may access a storage resource in a storage resource pool in the distributed block storage system by using the client. Usually, the block protocol access interface is configured to provide a LUN for the host. When a distributed block storage system program runs on a server including a hard disk, the server serves as a storage node to store data received by the client. For example, for the server, one hard disk may serve as one storage node by default, in other words, when the server includes a plurality of hard disks, the server may serve as a plurality of storage nodes. In another implementation, when a distributed block storage system program runs on a server, the server serves as one storage node. This is not limited in this embodiment of the present invention. Therefore, for a structure of the storage node, refer to FIG. 4 and related description. During initialization of the distributed block storage system, hash space (for example, 0 to 2{circumflex over ( )}32) is divided into N equal parts. Each equal part is one partition, and the N equal parts are equally allocated based on a quantity of hard disks. For example, in the distributed block storage system, N is 3600 by default, in other words, the partitions are P1, P2, P3, . . . , and P3600. It is assumed that a current distributed block storage system includes 18 hard disks (storage nodes). In this case, each storage node carries 200 partitions. The partition P includes M storage nodes N_j. There is a correspondence between the partition and the storage node, namely, a mapping between the partition and the storage node N_j included in the partition, and the correspondence is also referred to as a partition view. As shown in FIG. 4, in an example, the partition includes four storage nodes N_j. In this case, the partition view is “P2—storage node N₁—storage node N₂—storage node N₃—storage node N₄”, where j is an integer ranging from 1 to M. During initialization of the distributed block storage system, the partition view is determined, and is subsequently adjusted based on the quantity of hard disks in the distributed block storage system. In an implementation, the client saves the partition view.

Based on a reliability requirement of the distributed block storage system, data reliability may be improved by using an erasure coding (Erasure Coding, EC) algorithm. For example, a 3+1 mode is used, three data strips and one check strip constitute a stripe. In this embodiment of the present invention, data is stored in the partition in a form of a stripe. One partition includes R stripes S_i, where i is an integer ranging from 1 to R. In this embodiment of the present invention, P2 is used as an example for description.

In the distributed block storage system, fragment management is performed on the hard disk in a unit of 8 kilobytes (KB), and allocation information of each 8 KB fragment is recorded in a metadata management area of the hard disk. The fragments of the hard disk constitute the storage resource pool. The distributed block storage system includes a stripe metadata server. A implementation may be that a stripe management program runs on one or more servers in the distributed block storage system. The stripe metadata server allocates a stripe to the partition. The partition view shown in FIG. 5 is still used as an example. As shown in FIG. 5, the stripe metadata server allocates, based on the partition view, a storage address, namely, storage space, to a strip SU_ij in a stripe S_i in the partition P2 from a storage node N_j corresponding to the partition. The stripe metadata server allocates a storage address to SU_i1 from the storage node N₁, allocates a storage address to SU₁₂ from the storage node N₂, allocates a storage address to SU_i3 from the storage node N₃, and allocates a storage address to SU_i4 from the storage node N₄. In another implementation, the stripe metadata server does not allocate a storage address to a strip SU_ij in a stripe S_i in the partition P2 from a storage node N_j corresponding to the partition. When the client writes data into the storage node, the storage node N_j allocates a storage address to the strip SU_ij in the stripe. The storage address allocated to the strip SU_ij may be a logical address, for example, a logical block address (LBA), of a hard disk in the storage node N_j. In another implementation, in an SSD that supports an open channel (Open-channel), the storage address allocated by the stripe metadata server to the strip SU_ij in the stripe from the storage node N_j may alternatively be a physical address of the SSD. In another implementation, the hard disk in the storage node N_j is a LUN, namely, a LUN connected to the storage node. Stripe metadata records a mapping relationship between a stripe identifier and a strip identifier, namely, a correspondence between S_i and a strip SU_ij in the stripe, and a strip SU_ij included in a stripe S_i can be found based on the correspondence. Further, the stripe metadata server further records a correspondence between a strip SU_ij and a stripe S_i, and based on the correspondence, a stripe S_i can be found by using a strip SU_ij, so that stripe information, for example, all strips SU_ij included in S_i, is queried.

To reduce a quantity of strip identifiers managed by the stripe metadata server, the stripe metadata server allocates a version number to an identifier of a strip in a stripe. After a stripe is released, a version number of a strip identifier of a strip in the released stripe is updated, so that the strip identifier is used as a strip identifier of a strip in a new stripe. The stripe metadata server pre-allocates a strip SU_ij to a stripe S_i, so that waiting time can be reduced when the client writes data, thereby improving write performance of the distributed block storage system. In this embodiment of the present invention, the strip SU_ij in the stripe S_i has a unique identifier in the distributed block storage system.

In this embodiment of the present invention, a logical unit allocated by the distributed block storage system is connected to the client, to perform a data access operation. The logical unit is also referred to as a logical unit number (LUN). In the distributed block storage system, one logical unit may be connected to only one client, or one logical unit may be connected to a plurality of clients, that is, a plurality of clients share one logical unit. The logical unit is provided by a storage resource pool shown in FIG. 2.

In an embodiment of the present invention, as shown in FIG. 6, this embodiment of the present invention includes the following steps:

Step 601: A first client receives a first write request.

The first write request includes first data and a logical address. In a distributed block storage system, the distributed block storage system provides an LBA of a LUN. The logical address is used to indicate a writing position of the first data in the LUN.

Step 602: The first client determines that the logical address is distributed in a partition P.

In this embodiment of the present invention, a partition P2 is used as an example. With reference to FIG. 4, the first client stores a partition view of the distributed block storage system. As shown in FIG. 7, the first client determines, based on the partition view, the partition in which the logical address included in the first write request is located. In an implementation, the first client generates a key Key based on the logical address, calculates a hash value of the key according to a hash algorithm, determines a partition corresponding to the hash value, and therefore determines that the logical address is distributed in the partition P2. In other words, the first data is distributed in the partition P2.

Step 603: The first client obtains a stripe S_N from R stripes, where N is an integer ranging from 1 to R.

A stripe metadata server manages a correspondence between a partition and a stripe and a relationship between a strip in a stripe and a storage node. An implementation in which the first client obtains the stripe S_N from the R stripes is as follows: The first client determines that the logical address is distributed in the partition P2, and the first client queries the stripe metadata server to obtain the stripe S_N in the R stripes included in the partition P2. The logical address is an address used for writing data by the client in the distributed block storage system. Therefore, that the logical address is distributed in the partition P represents a same meaning as that the first data is distributed in the partition P. Another implementation in which the first client obtains the stripe S_N from the R stripes may be as follows: The first client obtains the stripe S_N from a stripe in the R stripes that is allocated to the first client. A client may store a mapping relationship between a partition and a stripe. The client may cache a stripe allocated by the stripe metadata server.

Step 604: The first client divides the first data into data of one or more strips SU_Nj in the stripe S_N.

The stripe S_N includes a strip. The first client receives the first write request, caches the first data included in the first write request, and divides the cached data based on a size of the strip in the stripe. For example, the first client divides the data based on a length of the strip in the stripe to obtain data of a strip size, performs a modulo operation on a quantity M (for example, 4) of storage nodes in the partition based on the logical address of the data of the strip size, to determine a location of the data of the strip size in the stripe, namely, a corresponding strip SU_Nj, and further determines, based on the partition view, a storage node N_j corresponding to the strip SU_N. Therefore, data of strips with a same logical address is distributed in a same storage node. For example, the first data is divided into data of one or more strips SU_N. In this embodiment of the present invention, P2 is used as an example. With reference to FIG. 5, the stripe S_N includes four strips: SU_N1, SU_N2, SU_N3, and SU_N4. In an example, the first data is divided into data of two strips, namely, data of SU_N1 and data of SU_N2. Data of the strip SU_N3 may be obtained through division of data in another write request received by the first client. For details, refer to a description of a first request. Further, data of the check strip SU_N4 is generated based on the data of SU_N1, the data of SU_N2, and the data of SU_N3. The data of the check strip SU_N4 is also referred to as check data. For how to generate data of a check strip based on data of a data strip in a stripe, refer to an existing stripe implementation algorithm. Details are not described in this embodiment of the present invention.

In this embodiment of the present invention, the stripe S_N includes four strips, namely, three data strips and one check strip. When the first client caches data for more than a period of time, the first client needs to write the data into the storage node. When the data cannot fill the data stripe, for example, there is only the data of the strip SU_N1 and the data of SU_N2 that are obtained through division of the first data, the check strip is generated based on the data of SU_N1 and the data of SU_N2. Optionally, data of a valid data strip SU_Nj includes data strip status information of the stripe S_N, and the valid data strip SU_Nj is a strip that is not empty. In this embodiment of the present invention, the data of the valid data strip SU_N1 and the data of the valid data strip SU_N2 both include the data strip status information of the stripe S_N, and the data strip status information is used to indicate whether each data strip in the stripe S_N is empty. For example, 1 is used to indicate that the data strip is not empty, and 0 is used to indicate that the data strip is empty. In this case, the data strip status information included in the data of SU_N1 is 110, and the data strip status information included in the data of SU_N2 is 110, which indicates that SU_N1 is not empty, SU_N2 is not empty, and SU_N3 is empty. The data of the check strip SU_N4 generated based on the data of SU_N1 and the data of SU_N2 includes check data of the data strip status information. Because SU_N3 is empty, the first client does not need to replace the data of SU_N3 with all-0 data and write the all-0 data into the storage node N₃, thereby reducing an amount of written data. When reading the stripe S_N, the first client determines, based on the data strip status information of the stripe S_N included in the data of the data strip SU_N1 or the data of the data strip SU_N2, that SU_N3 is empty.

When SU_N3 is not empty, the data strip status information included in the data of SU_N1, the data of SU_N2, and the data of SU_N3 in this embodiment of the present invention is 111, and the data of the check strip SU_N4 generated based on the data of SU_N1, the data of SU_N2, and the data of SU_N3 includes the check data of the data strip status information.

In this embodiment of the present invention, the data of the data strip SU_N further includes metadata, such as an identifier of the data strip SU_N and a logical address of the data of the data strip SU_N.

Step 605: The first client sends the data of the one or more strips SU_N to the storage node N_j.

In this embodiment of the present invention, the first client sends the data of SU_N1 obtained through division of the first data to a storage node N₁, and sends the data of SU_N2 obtained through division of the first data to a storage node N₂. The first client may concurrently send the data of the strip SU_Nj in the stripe S_N to the storage node N_j, without requiring a primary storage node, so that data interaction between storage nodes is reduced, write concurrency is improved, and write performance of the distributed block storage system is improved.

Further, when only the first client writes data into the LUN provided by the distributed block storage system, the first client receives a second write request. The second write request includes second data and the logical address described in FIG. 6. The first client determines, according to the algorithm described in the process in FIG. 6, that the logical address is distributed in the partition P2. The first client obtains a stripe S_K from the R stripes and divides the second data into data of one or more strips SU_Kj in the stripe S_K, such as data of SU_K1 and data of SU_K2. The first client sends the data of the one or more strips SU_Kj to the storage node N_j. That is, the data of SU_K1 is sent to the storage node N₁, and the data of SU_K2 is sent to the storage node N₂, where K is an integer ranging from 1 to R, and N is different from K. In this embodiment of the present invention, that the logical address is distributed in the partition P represents a same meaning as that the second data is distributed in the partition P. Further, data of the valid data strip SU_Kj includes data strip status information of a stripe KY. Further, the data of the data strip SU_Kj further includes metadata of the data of the data strip SU_Kj, such as an identifier of the strip SU_Kj and a logical address of the data of the strip SU_Kj. For a further description, refer to the description of the first client in FIG. 6, and details are not repeatedly described herein. For a meaning of that the first client obtains the stripe S_K from the R stripes, refer to the meaning of that the first client obtains the stripe S_N from the R stripes, and details are not repeatedly described herein.

In another implementation, when a plurality of clients such as the first client and a second client write data into the LUN provided by the distributed block storage system, the second client may perform the foregoing operation of the second write request performed by the first client.

Further, when one logical unit is connected to a plurality of clients such as the first client and the second client, the second client receives a third write request, and the third write request includes third data and the logical address described in FIG. 6. The second client determines, according to the algorithm described in the process in FIG. 6, that the logical address is distributed in the partition P2. The second client obtains a stripe S_K from the R stripes and divides the third data into data of one or more strips SU_Kj in the stripe S_K, such as data of SU_K1 and data of SU_K2. The second client sends the data of the one or more strips SU_Kj to the storage node N_j. That is, the data of SU_K1 is sent to the storage node N₁, and the data of SU_K2 is sent to the storage node N₂, where K is an integer ranging from 1 to R, and N is different from K. That the logical address is distributed in the partition P represents a same meaning as that the third data is distributed in the partition P. For a meaning of that the second client obtains the stripe S_K from the R stripes, refer to the meaning of that the first client obtains the stripe S_N from the R stripes, and details are not repeatedly described herein. Further, data of the valid data strip SU_Kj includes data strip status information of the stripe S_K. Further, the data of the data strip SU_Kj further includes at least one of an identifier of the second client and a timestamp TPK of the stripe S_K obtained by the second client. Further, the data of the data strip SU_Kj further includes metadata, such as an identifier of the data strip SU_Kj and a logical address of the data of the data strip SU_Kj. For a further description of the second client, refer to the description of the first client in FIG. 6, and details are not repeatedly described herein.

In the embodiment shown in FIG. 6, the client divides the data into the data of the strips, and sends the data of the strips to the corresponding storage nodes, without requiring the primary storage node, so that pressure on the primary storage node is reduced, and data interaction between the storage nodes is reduced. In addition, the data of the strips in the stripe is concurrently written into the corresponding storage nodes, so that write performance of the distributed block storage system is improved.

Corresponding to the embodiment of the first client shown in FIG. 6, as shown in FIG. 8, a storage node N_j performs the following steps.

Step 801: The storage node N_j stores data of a strip SU_Nj in a stripe S_N.

With reference to the embodiment shown in FIG. 6, the storage node N₁ receives the data of SU_N1 sent by the first client, the storage node N₂ receives the data of SU_N2 sent by the first client, the storage node N₃ receives the data of SU_N3 sent by the first client, and a storage node N₄ receives the data of SU_N4 sent by the first client. The stripe S_N includes M strips SU_Nj, namely, the strip SU_N1 to a strip SU_NM.

Step 802: The storage node N_j stores data of a strip SU_Kj in a stripe S_K.

With reference to the embodiment shown in FIG. 6, the storage node N₁ receives the data of SU_K1 sent by the first client, the storage node N₂ receives the data of SU_K2 sent by the first client, the storage node N₃ receives data of SU_K3 sent by the first client, and the storage node N₄ receives data of SU_K4 sent by the first client. The stripe S_K includes M strips SU_Kj, namely, the strip SU_K1 to a strip SU_KM. A logical address of the data of the strip SU_Kj is the same as a logical address of the data of the strip SU_Nj. In another embodiment of the present invention, the data of the strip SU_Kj in the stripe S_K is sent by a second client.

Step 803: The storage node N_j generates a record, where the record is used to indicate that the data of the strip SU_N reaches the storage node N_j before the data of the strip SU_Kj.

In an implementation of this embodiment of the present invention, the record indicates, based on an identifier of the strip SU_N and an identifier of the strip SU_Kj, that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj, so that data consistency is ensured based on the record. That is, the identifier of the strip SU_Nj and the identifier of the strip SU_Kj are used as an entry. In the entry, a sequence of the data of the two strips in reaching the storage node N_j may be represented by a sequence of the identifiers of the two strips. In another implementation, in the entry, a sequence of the data of the two strips in reaching the storage node N_j may alternatively be represented by a combination of the identifiers of the two strips and other symbols.

In this embodiment of the present invention, the storage node N_j backs up the record to one or more other storage nodes in M storage nodes. For example, using the storage node N₁ as an example, the record may be backed up to one or more of the storage node N₂, the storage node N₃, and the storage node N₄, so that reliability of the record can be improved, and the record is prevented from being lost when the storage node N₁ becomes faulty. The storage node N_j stores an identifier of the storage node in which the record is backed up. In another implementation, because the stored identifier of the storage node in which the record is backed up is lost when the storage node N_j becomes faulty, the data of the strip SU_Kj between the restored data of the strip SU_Nj and the restored data of the strip SU_Kj cannot be determined as latest data. To prevent this, a backup relationship of the record between the M storage nodes may be alternatively determined by a stripe metadata server.

In this embodiment of the present invention, the storage node N_j further records a mapping between the logical address of the data of the strip SU_N and the identifier of the strip SU_Nj, and a mapping between the logical address of the data of the strip SU_Kj and the identifier of the strip SU_Kj. For ease of querying the record stored by the storage node N_j, the logical address of the data of the strip may be used as an index to organize the record. The storage node N_j stores the data of the strip SU_Nj and records a mapping between the identifier of the strip SU_Nj and an address for storing the data of the strip SU_Nj. The storage node N_j stores the data of the strip SU_Kj and records a mapping between the identifier of the strip SU_Kj and an address for storing the data of the strip SU_Kj. In this embodiment of the present invention, for a strip storing check data, namely, a check strip, the storage node needs to record a mapping between an identifier of the check strip and an address for storing the data of the check strip, and does not need to record a mapping between a logical address of the data of the check strip and the identifier of the check strip, because the data of the check strip has no logical address.

In this embodiment of the present invention, the storage node N_j stores a mapping between a logical address of data of a strip and an identifier of the strip in a cache according to a sequence of receiving the data of the strip. Therefore, the storage node N_j generates the record based on the mapping. In this embodiment of the present invention, a record is not required to be generated immediately after the data of the strip is received, and generating the record may be performed as a background task, so that performance of processing the data of the strip by the storage node N_j is improved. When the storage node is configured to store the data of the check strip, namely, the check data, the storage node N_j stores, in the cache according to a sequence of receiving the data of the check strip, the mapping between the identifier of the strip and the logical address of the data of the check strip stored in the storage node N_j, and the storage node N_j generates the record based on the mapping. In another implementation, the storage node N_j stores the data of the check strip in the cache according to the sequence of receiving the data of the check strip, and the storage node N_j generates the record based on a sequence of data in the cache.

In this embodiment of the present invention, the storage node N_j receives a read request sent by the client, and the read request includes a logical address. Because the logical address of the data of the strip SU_Nj is the same as the logical address of the data of the strip SU_Kj, the storage node N_j further records a mapping between the logical address of the data of the strip SU_Nj and the identifier of the strip SU_Nj, and a mapping between the logical address of the data of the strip SU_Kj and the identifier of the strip SU_Kj. Therefore, latest data, namely, data of a latest version, between the data of the strip SU_Nj and the data of the strip SU_Kj needs to be determined, to ensure data consistency based on the record. The storage node N_j determines, by querying the stored record based on the logical address, that the data of the strip SU_Kj is the latest data. The client may be the first client or the second client, or may be a client other than the first client and the second client.

In this embodiment of the present invention, when the storage node N_j becomes faulty and the data of the strip SU_Nj and the data of the strip SU_Kj are lost, the stripe metadata server restores the data of the strip SU_Nj based on data of other strips of the stripe S_N on M−1 storage nodes, and restores the data of the strip SU_Kj based on data of other strips of the stripe S_K on the M−1 storage nodes. For example, when the storage node N₁ becomes faulty and the data of the strip SU_N1 and the data of the strip SU_K1 are lost, the data of the strip SU_N1 and the data of the strip SU_K1 are restored in the foregoing manner. In another implementation, restoring the data of the strip may be implemented by the client. The storage node N_j obtains the record from another storage node in which the record is backed up, and determines, between the restored data of the strip SU_Nj and the restored data of the strip SU_Kj based on the record, that the data of the strip SU_Kj is latest data, to ensure data consistency based on the record. The storage node N_j queries the backup relationship of the record between the M storage nodes from the stripe metadata server, determines an identifier of a storage node that stores a record of the storage node N_j, and obtains the record of the storage node N_j from the storage node based on the identifier of the storage node.

With reference to various implementations of the embodiments of the present invention, an embodiment of the present invention provides a storage node in a distributed storage system as a storage node N_j in the distributed storage system. As shown in FIG. 9, the storage node N_j includes a receiving unit 901 and a generation unit 902. The receiving unit 901 is configured to receive data of a strip SU_Nj in a stripe S_N and data of a strip SU_Kj in a stripe S_K, where the stripe S_N includes M strips SU_Nj and the stripe S_K includes M strips SU_Kj, a logical address of the data of the strip SU_Kj is the same as a logical address of the data of the strip SU_Nj, and N is different from K. The generation unit 902 is configured to generate a record, where the record is used to indicate that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj. The record indicates, based on an identifier of the strip SU_N and an identifier of the strip SU_Kj, that the data of the strip SU_Nj reaches the storage node N_j before the data of the strip SU_Kj. In an implementation, the data of the strip SU_Nj and the data of the strip SU_Kj are sent by a first client. In another implementation, the data of the strip SUN is sent by a first client, and the data of the strip SU_Kj is sent by a second client.

Further, the storage node N_j shown in FIG. 9 further includes a backup unit that is configured to back up the record to one or more other storage nodes in the M storage nodes.

Further, the storage node N_j shown in FIG. 9 further includes a query determining unit and a returning unit. The receiving unit is further configured to receive a read request sent by a third client, where the read request includes the logical address. The query determining unit is configured to determine, by querying the record based on the logical address, that the data of the strip SU_Kj is latest data. The returning unit is configured to return the data of SU_Kj to the third client.

Further, the storage node N_j shown in FIG. 9 further includes an obtaining and determining unit that is configured to: when the storage node N_j becomes faulty, obtain the record from a storage node in which the record is backed up, and determine, between the restored data of the strip SU_Nj and the restored data of the strip SU_Kj based on the record, that the data of the strip SU_Kj is latest data.

For an implementation of the storage node shown in FIG. 9 in this embodiment of the present invention, refer to the storage node such as the storage node N_j in the embodiments of the present invention. In another implementation, a unit in the storage node shown in FIG. 9 may be a software module that may run on a server to make the storage node complete various implementations described in the embodiments of the present invention.

An embodiment of the present invention further provides a client, including a plurality of units that are configured to implement an operation of the client in this embodiment of the present invention. For an implementation of the client, refer to the storage node such as the storage node N_j in the embodiments of the present invention. In another implementation, a unit implemented by the client may be a software module that may run on a server to make the storage node complete various implementations described in the embodiments of the present invention.

Correspondingly, the embodiments of the present invention further provide a computer readable storage medium and a computer program product. The computer readable storage medium and the computer program product include computer instructions used to implement various solutions described in the embodiments of the present invention.

Identifiers used to describe the stripe, the data strip, the check strip, and the storage node in the embodiments of the present invention are merely intended to describe the embodiments of the present invention more clearly, and a similar identifier is not required in actual product implementation. Therefore, the identifiers used to describe the stripe, the data strip, the check strip, and the storage node in the embodiments of the present invention are not intended to limit the present invention.

In another implementation of this embodiment of the present invention, a corresponding strip and storage node may not be queried based on a partition.

In the several embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the unit division in the described apparatus embodiment is merely logical function division and may be another division in actual implementation. For example, a plurality of units or components may be combined or may be integrated into another system, or some features may be ignored or not be performed. In addition, the displayed or discussed mutual couplings or direct couplings or communications connections may be implemented by using some interfaces. The indirect couplings or communications connections between the apparatuses or units may be implemented in electrical, mechanical, or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units may be selected based on actual requirements to achieve the objectives of the solutions in the embodiments.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each of the units may exist alone physically, or two or more units are integrated into one unit.

Claims

1. A data storage method in a distributed storage system, wherein the distributed storage system comprises M storage nodes Nj, j is an integer ranging from 1 to M, and the method comprises: receiving, by the storage node Nj, data of a strip SUNj in a stripe SN, wherein the stripe SN comprises M strips SUNj;receiving, by the storage node Nj, data of a strip SUKj in a stripe SK, wherein a logical address of the data of the strip SUKj is the same as a logical address of the data of the strip SUNj, N is different from K, and the stripe SK comprises M strips SUKj; andgenerating, by the storage node Nj, a record, wherein the record is used to indicate that the data of the strip SUNj reaches the storage node Nj before the data of the strip SUKj.

2. The method according to claim 1, wherein the record indicates, based on an identifier of the strip SUNj and an identifier of the strip SUKj, that the data of the strip SUNj reaches the storage node Nj before the data of the strip SUKj.

3. The method according to claim 1, wherein the method further comprises: backing up, by the storage node Nj, the record to one or more other storage nodes in the M storage nodes.

4. The method according to claim 1, wherein the data of the strip SUNj and the data of the strip SUKj are sent by a first client.

5. The method according to claim 1, wherein the data of the strip SUNj is sent by a first client, and the data of the strip SUKj is sent by a second client.

6. The method according to claim 1, wherein the strip SUNj and the strip SUKj are allocated from the storage node Nj by a stripe metadata server.

7. The method according to claim 1, wherein the method further comprises: receiving, by the storage node Nj, a read request sent by a third client, wherein the read request comprises the logical address;determining, by the storage node Nj by querying the record based on the logical address, that the data of the strip SUKj is latest data; andreturning, by the storage node Nj, the data of SUKj to the third client.

8. The method according to claim 3, wherein the method further comprises: when the storage node Nj becomes faulty, obtaining the record from a storage node in which the record is backed up, and determining, between the restored data of the strip SUNj and the restored data of the strip SUKj based on the record, that the data of the strip SUKj is latest data.

9. A storage node in a distributed storage system, wherein the distributed storage system comprises M storage nodes Nj, j is an integer ranging from 1 to M, and the storage node serves as the storage node Nj, and comprises an interface and a processor that communicate with each other, wherein the processor is configured to: receive data of a strip SUNj in a stripe SN, wherein the stripe SN comprises M strips SUNj;receive data of a strip SUKj in a stripe SK, wherein a logical address of the data of the strip SUKj is the same as a logical address of the data of the strip SUNj, N is different from K, and the stripe SK comprises M strips SUKj; andgenerate a record, wherein the record is used to indicate that the data of the strip SUNj reaches the storage node NJ before the data of the strip SUKj.

10. The storage node according to claim 9, wherein the record indicates, based on an identifier of the strip SUNj and an identifier of the strip SUKj, that the data of the strip SUNj reaches the storage node Nj before the data of the strip SUKj.

11. The storage node according to claim 9, wherein the processor is further configured to back up the record to one or more other storage nodes in the M storage nodes.

12. The storage node according to claim 9, wherein the data of the strip SUNj and the data of the strip SUKj are sent by a first client.

13. The storage node according to claim 9, wherein the data of the strip SUNj is sent by a first client, and the data of the strip SUKj is sent by a second client.

14. The storage node according to claim 9, wherein the strip SUNj and the strip SUKj are allocated from the storage node Nj by a stripe metadata server.

15. The storage node according to claim 9, wherein the processor is further configured to: receive a read request sent by a third client, wherein the read request comprises the logical address;determine, by querying the record based on the logical address, that the data of the strip SUKj is latest data; andreturn the data of SUKj to the third client.

16. The storage node according to claim 11, wherein the processor is further configured to: when the storage node Nj becomes faulty, obtain the record from a storage node in which the record is backed up, and determine, between the restored data of the strip SUNj and the restored data of the strip SUKj based on the record, that the data of the strip SUKj is latest data.

17. A computer program product, wherein the computer program product comprises a computer instruction, and the computer instruction is used by a storage node in a distributed storage system, wherein the distributed storage system comprises M storage nodes Nj, j is an integer ranging from 1 to M, and the storage node serves as the storage node Nj, and runs the computer instruction to: receive data of a strip SUNj in a stripe SN, wherein the stripe SN comprises M strips SUNj;receive data of a strip SUKj in a stripe SK, wherein a logical address of the data of the strip SUKj is the same as a logical address of the data of the strip SUNj, N is different from K, and the stripe SK comprises M strips SUKj; andgenerate a record, wherein the record is used to indicate that the data of the strip SUNj reaches the storage node Nj before the data of the strip SUKj.

18. The computer program product according to claim 17, wherein the record indicates, based on an identifier of the strip SUNj and an identifier of the strip SUKj, that the data of the strip SUNj reaches the storage node Nj before the data of the strip SUKj.

19. The computer program product according to claim 17, wherein the storage node serves as the storage node Nj, and runs the computer instruction to further: back up the record to one or more other storage nodes in the M storage nodes.

20. The computer program product according to claim 17, wherein the storage node serves as the storage node Nj, and runs the computer instruction to further: receive a read request sent by a third client, wherein the read request comprises the logical address;determine, by querying the record based on the logical address, that the data of the strip SUKj is latest data; andreturn the data of SUKj to the third client.