CN115344551A - Data migration method and data node - Google Patents

Abstract

The embodiment of the application discloses a data migration method, which is used for migrating a load without waiting for the completion of the migration of target data from a first data node to a second data node and reducing time delay. The method in the embodiment of the application comprises the following steps: a second data node receives a data migration request about target data sent by a first data node, wherein the data migration request comprises a memory address of the target data in the first data node; the method comprises the steps that an RDMA virtual memory space is established according to a data migration request, a memory address of the RDMA virtual memory space is mapped to a memory address of a first data node, and the RDMA virtual memory space is used for a second data node to access target data on the first data node according to a target operation request; sending an instruction for modifying the metadata to the management node, wherein the instruction for modifying the metadata is used for modifying the metadata by the management node so that a target operation request for accessing the target data is routed to the second data node; and receiving and storing the target data sent by the first data node.

Description

A data migration method and data nodes

This application is a divisional application. The application number of the original application is 201710495228.6, and the original application date is June 26, 2017. The entire content of the original application is incorporated in this application by reference.

technical field

The present application relates to the computer field, and in particular to a data migration method, a first data node and a second data node.

Background technique

With the continuous development of database technology, users' requirements for database expansion and disaster recovery capabilities continue to increase, and distributed databases have become more and more popular with users.

In existing load balancing, some data (which may be called target data or migration data) on the first data node needs to be migrated to the second data node. Before the data migration occurs, the server has some access flows (such as read/write requests) to the first data node; during the data migration process, the server’s access to the target data is still concentrated on the first data node; after the data migration is completed , the server's access to the target data will be switched to the second data node.

However, the load of the target data needs to wait until the data is copied, that is, the data is copied to the second data node before the load of the target data can be switched to the second data node, and the load migration speed is slow; the target data needs to be copied to the second data node first. The kernel buffer is sent to the network for transmission to the second data node, which consumes a lot of CPU resources and will affect the performance of the first data node to handle other loads, and the target data transmission speed is slow and the delay is large.

Contents of the invention

The embodiment of the present application provides a data migration method and a data node, which are used to migrate loads without waiting for the target data to be migrated from the first data node to the second data node, so as to increase the speed of load migration and reduce the time delay.

The first aspect of the embodiment of the present application provides a data migration method, which may include: in the load balancing solution, the second data node receives a data migration request about the target data sent by the first data node, and the data migration request includes the The memory address of the target data in the first data node; it should be understood that the target data may be hot data on the first data node, and the hot data may be data that is frequently accessed or whose access times exceed a specific threshold. The second data node establishes a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used for the second data The node accesses the target data on the first data node through RDMA according to the target operation request; the second data node sends an instruction to modify the metadata to the management node, and the instruction to modify the metadata is used by the management node to modify the metadata to A target operation request for accessing the target data is routed to the second data node; the target operation request here may be a load or an access to the target data. Metadata, also known as intermediary data and relay data, is the data describing data (data about data), mainly describing the information of data attributes (property), used to support such as indicating storage location, historical data, resource search , file recording and other functions. The second data node receives and stores the target data sent by the first data node.

In the embodiment of the present application, the management node may collect cluster information, and the cluster information may include capacity and load information of each data node and each store on the data node. The management node can trigger load balancing according to the collected cluster information. The first data node receives the trigger load balancing instruction sent by the management node, and sends a data migration request about the target data to the second data node, or the first data node sends a request to the management node Send an instruction that the load is greater than the first threshold, the second data node sends an instruction that the load is less than the second threshold to the management node, the management node can confirm that the first data node with the load greater than the first threshold is the source data node, and confirm that the load is less than the second threshold The second data node is the target data node, the management node can send an instruction to trigger load balancing to the first data node, and the first data node sends a migration request about the target data to the second data node. The second data node establishes an RDMA virtual memory space according to the data migration request about the target data, and the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, so the second data node can directly access the first data node When the target data of a data node is accessed, there is no need to wait for the target data to be replicated on the second data node before migrating the target operation request. The second data node then sends a metadata modification instruction to the management node, which can route the target operation request for accessing the target data to the second data node; after receiving the target operation request, the second data node can access the first data through RDMA Target data on the node. Further, the second data node receives the target data sent by the first data node. Using RDMA virtual memory mapping technology, the data to be migrated is mapped to the virtual memory space of the second data node, and all usage rights and loads are switched to the second data node, so that the load can be migrated without waiting for the completion of data copying, and the load can be increased speed of migration.

With reference to the first aspect of the embodiments of the present application, in the first implementation manner of the first aspect of the embodiments of the present application, the second data node receives and stores the target data sent by the first data node, and may The method includes: the second data node receives and stores the target data sent by the first data node through RDMA. In this embodiment of the application, RDMA technology is used in the process of real-time data migration of distributed database clusters, which reduces the consumption of CPU resources on the first data node by data migration, and reduces the impact of the migration process on the running of the first data node. The impact of other services, and improve the speed of data transmission.

In combination with the first aspect of the embodiment of the present application and the first implementation of the first aspect, in the second implementation of the first aspect of the embodiment of the present application, the target data is the hotspot data in the first data node Divide into one of the M data, where M is an integer greater than or equal to 2. In the embodiment of the present application, the second data node receives the target data sent by the first data node through RDMA. It may be that the first data node divides the hot data into M shares and transmits one of the M shares. In this way, Therefore, the size of the data sent from the first data node to the second data node can be reduced, shortening the unavailable time and data volume of the target data, and improving the service quality.

In combination with the first aspect of the embodiment of the present application, the first implementation of the first aspect, and the second implementation of the first aspect, in the third implementation of the first aspect of the embodiment of the present application, + the method It may also include: after the second data node sends the metadata modification instruction to the management node, the original target operation request (such as load or access) on the target data will be routed to the second data node, when the second data node receives the When the target operation requests, if the target data accessed by the target operation request is not stored on the second data node, it should be understood that after the second data node establishes an RDMA virtual memory space, the second data node can access the first data node through RDMA. During the process of receiving the target data from the data nodes, the first data node may also send the target data to the second data node, or may send the target data to the second data node through RDMA. Then the second data node accesses the target data on the first data node through RDMA according to the target operation request; if the target data accessed by the target operation request has been stored on the second data node, the second data The node performs access on the stored target data according to the target operation request. In the embodiment of the present application, if the target data accessed by the target operation request is not stored on the second data node, the memory data on the first data node can be accessed through RDMA. If the target data accessed by the target operation request is stored on the second data node, it can be directly accessed on the second data node, and there is no need to access the first data node through RDMA, otherwise, the second data node needs to send data to the first data node Nodes send target data, which wastes resources and increases access delay.

In combination with the first aspect of the embodiments of the present application, the first implementation of the first aspect, and the second implementation of the first aspect, in the fourth implementation of the first aspect of the embodiments of the present application, the method may also be It includes: if the target operation request is a write operation request, and the write operation request indicates adding new data, then the second data node performs a write operation on the second data node according to the write operation request. Because if it is a write operation request, it is to write the target data, no matter whether the target data sent by the first data node is saved on the second data node, the second data node can write the target data on the second data node according to the write operation request Perform write operations to reduce the transmission process and delay.

In combination with the first aspect of the embodiments of the present application, the first implementation of the first aspect, and the second implementation of the first aspect, in the fifth implementation of the first aspect of the embodiments of the present application, the method may also be Including: if the target operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the target operation request has been saved on the second data node, then the second The data node performs access on the second data node according to the target operation request; if the target operation request is a read operation request, or a write operation request for modifying the target data, and the target operation request accesses the target If the data is not stored on the second data node, the second data node accesses the target data on the first data node through RDMA according to the target operation request. Regardless of whether it is a read operation request or a write operation request, if the target data sent by the first data node is stored on the second data node, the second data node can directly access the target data; if the second data node does not store the target data For the target data sent by a data node, the second data node can access the target data of the first data node through RDMA. A specific implementation method is provided, which increases the feasibility of the scheme.

The second aspect of the embodiment of the present application provides a data migration method, which may include: in a load balancing solution, the first data node sends a migration request about target data to the second data node. The data migration request includes the memory address of the target data in the first data node, and the data migration request is used for the second data node to establish a remote direct memory access RDMA virtual memory space, and the memory address of the RDMA virtual memory space is mapped to The memory address of the first data node, the RDMA virtual memory space is used for the second data node to access the target data on the first data node through RDMA according to the target operation request, and the target operation request is routed to the second data node by the management node An operation request for accessing the target data on the second data node; it should be understood that the target data may be hot data on the first data node, and hot data is data that is frequently accessed or whose access times exceed a specific threshold. The target operation request here may be a load or access to target data. Metadata, also known as intermediary data and relay data, is the data describing data (data about data), mainly describing the information of data attributes (property), used to support such as indicating storage location, historical data, resource search, File recording and other functions. The first data node sends the target data to the second data node.

In the embodiment of the present application, the management node may collect cluster information, and the cluster information may include capacity and load information of each data node and each store on the data node. The management node can trigger load balancing according to the collected cluster information. The first data node receives the trigger load balancing instruction sent by the management node, and sends a data migration request about the target data to the second data node, or the first data node sends a request to the management node Send an instruction that the load is greater than the first threshold, the second data node sends an instruction that the load is less than the second threshold to the management node, the management node can confirm that the first data node with the load greater than the first threshold is the source data node, and confirm that the load is less than the second threshold The second data node is the target data node, the management node can send an instruction to trigger load balancing to the first data node, and the first data node sends a data migration request about the target data to the second data node, and the data migration request is used for the The second data node establishes an RDMA virtual memory space, and the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, so the target data of the first data node can be directly accessed on the second data node, There is no need to wait for the target data to be replicated on the second data node before migrating the target operation request. The first data node sends the target data to the second data node. Using RDMA virtual memory mapping technology, the data to be migrated is mapped to the virtual memory space of the second data node, and all usage rights and loads are switched to the second data node, so that the load can be migrated without waiting for the completion of data copying, and the load can be increased speed of migration.

With reference to the second aspect of the embodiments of the present application, in the first implementation manner of the second aspect of the embodiments of the present application, sending the target data to the second data node by the first data node may include: the first The data node sends the target data to the second data node through RDMA. Using RDMA technology for the process of real-time data migration of distributed database clusters reduces the consumption of CPU resources on the first data node by data migration, reduces the impact of the migration process on other services running on the first data node, and improves the speed of data transmission.

In combination with the second aspect of the embodiment of the present application and the first implementation of the second aspect, in the second implementation of the second aspect of the embodiment of the present application, the first data node sends the target data to the second data node , may include: the first data node determines the hotspot data; it should be understood that the hotspot data in the first data node may be determined by the first data node, or the management node may send information about the hotspot data to the first data node, The first data node determines the hotspot data according to the information of the hotspot data. The first data node divides the hotspot data into M pieces of data, and M is an integer greater than or equal to 2; the first data node selects the target data from the M pieces of data; the first data node sends the data to the second data node Alternatively, the target data may be sent to the second data node through RDMA. In the embodiment of this application, the first data node can divide the hotspot data into M shares, and transmit one of the M shares, so that the size of the data sent from the first data node to the second data node can be reduced , which shortens the unavailable time and data volume of the target data, and improves the service quality.

With reference to the second implementation of the second aspect of the embodiment of the present application, in the third implementation of the second aspect of the embodiment of the present application, before the first data node determines the hotspot data, the method may further include: the first data node A data node receives hot data information sent by a management node; determining the hot data by the first data node may include: determining the hot data by the first data node according to the hot data information. In the embodiment of the present application, a method for determining hotspot data is provided, which increases the feasibility of the solution.

The third aspect of the embodiment of the present application provides a data migration method, which may include: the management node determines the first data node and the second data node for data migration; that is, the management node can collect cluster information, and the cluster information can include each data Capacity and load information of each store on nodes and data nodes. The management node can trigger load balancing according to the collected cluster information. The first data node receives the trigger load balancing instruction sent by the management node, and sends a data migration request about the target data to the second data node, or the first data node sends a request to the management node Send an instruction that the load is greater than the first threshold, the second data node sends an instruction that the load is less than the second threshold to the management node, the management node can confirm that the first data node with the load greater than the first threshold is the source data node, and confirm that the load is less than the second threshold The second data node is the target data node, the management node can send an instruction to trigger load balancing to the first data node, and the first data node sends a migration request about the target data to the second data node. The management node receives the metadata modification instruction sent by the second data node; the management node modifies the metadata according to the metadata modification instruction, so that the target operation request for accessing the target data on the first data node is routed to the second data node. on the data node.

In this embodiment of the application, the management node can determine the first data node and the second data node of the migration target data, and the management node receives the instruction to modify the metadata sent by the second data node; the management node according to the instruction to modify the metadata The metadata is modified so that a target operation request for accessing target data on the first data node is routed to the second data node. The second data node can access the target data on the first data node through RDMA, and the target operation request can be switched to the second data node without waiting for the target data on the first data node to be copied to the second data node.

With reference to the third aspect of the embodiment of the present application, in the first possible implementation manner of the third aspect of the embodiment of the present application, the management node determining the first data node and the second data node for data migration may include: if the When the management node detects that a new data node joins the database cluster, the management node selects a data node with the largest load from the database cluster as the first data node, and the new data node as the second data node. An implementation manner of how to determine the first data node and the second data node for data migration is provided, which increases the feasibility of the solution.

With reference to the third aspect of the embodiment of the present application, in the second possible implementation manner of the third aspect of the embodiment of the present application, the management node determining the first data node and the second data node for data migration may include: if the If the management node does not detect that there is a new data node joining the database cluster, the management node selects a data node with the largest load from the database cluster as the first data node, and selects a data node with the least load as the second data node. Another implementation manner of how to determine the first data node and the second data node for data migration is provided, which increases the feasibility of the solution.

The fourth aspect of the embodiment of the present application provides a data node, which has the function of migrating load without waiting for the target data to be migrated from the first data node to the second data node, increasing the speed of load migration, and reducing the time delay. This function may be implemented by hardware, or may be implemented by executing corresponding software on the hardware. The hardware or software includes one or more modules corresponding to the above functions.

The fifth aspect of the embodiment of the present application provides a data node, which has the function of migrating load without waiting for the target data to be migrated from the first data node to the second data node, increasing the speed of load migration, and reducing the delay. This function may be implemented by hardware, or may be implemented by executing corresponding software on the hardware. The hardware or software includes one or more modules corresponding to the above functions.

The sixth aspect of the embodiment of the present application provides a management node, which can migrate loads without waiting for the target data to be migrated from the first data node to the second data node, so as to increase the speed of load migration and reduce the delay. This function may be implemented by hardware, or may be implemented by executing corresponding software on the hardware. The hardware or software includes one or more modules corresponding to the above functions.

The seventh aspect of the embodiment of the present application provides a data node, which may include:

a transceiver, a processor, a memory and a bus, the transceiver, the processor and the memory connected by the bus;

The memory is used to store operation instructions;

The transceiver is configured to receive a data migration request about the target data sent by the first data node, where the data migration request includes the memory address of the target data in the first data node; send an instruction to modify metadata to the management node, The instruction for modifying metadata is used by the management node to modify the metadata, so that the target operation request for accessing the target data is routed to the second data node; and storing the target data sent by the first data node;

The processor is configured to establish a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used for the second The data node accesses the target data on the first data node through RDMA according to the target operation request.

The eighth aspect of the embodiment of the present application provides a data node, which may include:

a transceiver, a memory, and a bus, the transceiver and the memory connected via the bus;

The memory is used to store operation instructions;

The transceiver is configured to send a data migration request about the target data to the second data node, the data migration request includes the memory address of the target data in the first data node, and the data migration request is used for the second data node Establishing a remote direct memory access RDMA virtual memory space, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used by the second data node to access the first data node through RDMA according to the target operation request For the target data on a data node, the target operation request is an operation request routed by the management node to the second data node to access the target data; and the target data is sent to the second data node.

A ninth aspect of the embodiment of the present application provides a management node, which may include:

a transceiver, a processor, a memory and a bus, the transceiver, the processor and the memory connected by the bus;

The memory is used to store operation instructions;

The processor is configured to determine a first data node and a second data node for data migration; modify the metadata according to the instruction for modifying the metadata, so that a target operation request for accessing target data on the first data node is routed to the second data node On the second data node;

The transceiver is configured to receive an instruction for modifying metadata sent by the second data node.

The tenth aspect of the embodiment of the present application provides a distributed database system. The distributed database system includes a first data node and a second data node. The first data node described in the optional implementation manner; the second data node is the second data node described in the second aspect of the present application or any optional implementation manner of the second aspect.

The eleventh aspect of the embodiments of the present invention provides a storage medium. It should be noted that the technical solution of the present invention essentially or the part that contributes to the existing technology or all or part of the technical solution can be produced by software. It is embodied in the form that the computer software product is stored in a storage medium for storing computer software instructions used by the above-mentioned equipment, which includes data nodes or management The program designed by the node.

The storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and other media capable of storing program codes.

The twelfth aspect of the embodiments of the present invention provides a computer program product containing instructions, which, when run on a computer, causes the computer to execute the first aspect of the present application or any optional implementation of the first aspect, or, the present application The second aspect or any optional implementation manner of the second aspect, or the method described in the third aspect of the present application or any optional implementation manner of the third aspect.

It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

In this embodiment of the application, the second data node establishes the RDMA virtual memory space according to the data migration request about the target data sent by the first data node, and the second data node sends an instruction to modify the metadata to the management node, and the management node can according to The instruction for modifying the metadata modifies the metadata so that the target operation request for accessing the target data is routed to the second data node; thus, the target operation request does not need to be migrated until the target data is copied on the second data node, Reduce the delay; the first data node sends the target data to the second data node. Using RDMA virtual memory technology, the target data is mapped to the RDMA virtual memory space of the second data node, and all usage rights and loads are switched to the second data node, so that the load can be migrated without waiting for the copy of the target data to be completed, improving the The speed of load migration; further, RDMA technology can be used in the process of real-time data migration of distributed database clusters, which reduces the consumption of CPU resources on the first data node by data migration and reduces the impact of the migration process on the first data node. The impact of other running businesses and increased data transmission speed. The target data is migrated in small pieces, thereby shortening the unavailable time and data volume of the target data, and improving the service quality. Save time and reduce delay.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings that are required in the description of the embodiments and prior art. Obviously, the accompanying drawings in the following description are only some implementations of the present application For example, other drawings can also be obtained from these drawings.

FIG. 1 is a schematic diagram of a scene applied in the prior art;

FIG. 2 is a schematic diagram of load migration in the prior art;

FIG. 3 is a system architecture diagram applied in the prior art;

Fig. 4 is a schematic flow chart of data migration in the prior art;

FIG. 5 is an overall architecture diagram of a distributed cluster applied in the embodiment of the present application;

FIG. 6 is a schematic diagram of an embodiment of a method for data migration in the embodiment of the present application;

FIG. 7 is a schematic flow diagram of triggering load balancing in an embodiment of the present application;

FIG. 8 is a schematic diagram of load migration in the embodiment of the present application;

FIG. 9 is a schematic diagram of data block fragmentation transmission in the embodiment of the present application;

FIG. 10 is a schematic flow diagram of the fragmentation transmission algorithm in the embodiment of the present application;

FIG. 11 is a schematic diagram of an embodiment of a data node in the embodiment of the present application;

FIG. 12A is a schematic diagram of another embodiment of a data node in the embodiment of the present application;

FIG. 12B is a schematic diagram of another embodiment of a data node in the embodiment of the present application;

FIG. 13 is a schematic diagram of another embodiment of a data node in the embodiment of the present application.

Detailed ways

The embodiment of the present application provides a data migration method and a data node, which are used to migrate loads without waiting for the target data to be migrated from the first data node to the second data node, which improves the speed of load migration and reduces the time delay.

In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiment of the application will be described below in conjunction with the drawings in the embodiment of the application. Obviously, the described embodiment is only a part of the application Examples, but not all examples. Based on the embodiments in this application, all should belong to the protection scope of this application.

With the continuous development of database technology, users' requirements for database expansion and disaster recovery capabilities continue to increase, and distributed databases have become more and more popular with users. A very important advantage of distributed databases over traditional databases is that they can migrate data from high-load data nodes to low-load data nodes; or migrate data from old data nodes to newly added data nodes, thereby Implement backup and load balancing. And this will inevitably cause a lot of network overhead and performance loss. Real-time data migration between servers has become one of the biggest bottlenecks in distributed databases.

Therefore, the distributed database adopts the remote direct memory access (Remote Direct Memory Access, RDMA) technology to allow the network card to directly read the application layer data, bypassing the traditional Transmission Control Protocol (Transmission Control Protocol, TCP) / Internet Protocol (Internet Protocol, IP ), thereby greatly improving the speed of network transmission and reducing the consumption of Central Processing Unit (CPU) resources by data migration, thereby bringing new ideas to solve the bottleneck of data migration.

Among them, RDMA means that one computer can access the memory data of another remote computer, and can read/write the memory data of the remote computer just like reading/writing the local memory. Its realization is mainly through the RDMA zero-copy network technology, which enables the network adapter (Network Interface Card, NIC) to directly transfer data to and from the application memory, thus eliminating the need to copy data between the application memory and the kernel memory , which greatly saves CPU resources and improves the speed of data transmission.

As shown in Figure 1, it is a schematic diagram of a scene applied by the prior art, and a client of a mail system is installed on a terminal (including a mobile phone, a personal computer (PC) and a tablet, etc.), referred to as a mail client; The mail client on the network is connected to the server of the mail system through the network; the server will send read/write requests to the underlying data nodes. In order to achieve load balancing, the target data on the first data node will be migrated to the second data node. Before the data migration occurs, the server has some access flows (such as read/write requests) to the first data node; during the data migration process, the server’s access to the target data is still concentrated on the first data node; after the data migration is completed , the server's access to the target data will be switched to the second data node.

As shown in FIG. 2 , it is a schematic diagram of load migration in the prior art. When the data migration starts, first mark the location at the start of the data migration in the operation log (Log) of the first data node, and then retrieve the data items to be migrated on the first data node, referred to as target data or migration data, and retrieve After the completion, copy all retrieved data items that need to be migrated to the second data node through the network; after the copy is completed, update all the operation requests for the target data starting from the marked position in the operation log to the second data node ;Finally, switch all the accesses and loads to the target data to the second data node; so far, the data migration process ends.

However, the load of the target data needs to wait until the data on the second data node is copied before switching to the second data node, and the load migration speed is slow; the target data needs to be copied to the kernel buffer first, and then sent to the network , the CPU resource consumption of the first data node is very large, which will affect the performance of the first data node for processing other loads, and the data transmission speed is relatively slow and the time delay is relatively large.

As shown in FIG. 3 , it is a system architecture diagram applied in the prior art. The functions of each component are listed as follows:

Virtual Machine (VM): virtual machine instance;

Virtual machine monitor (Hypervisor): abstract the hardware layer device into a virtual device and provide it to the virtual machine instance;

Host Migration Agent (Host Migration Agent): monitors whether memory has been modified, adds the modified memory block to the RDMA queue, then asynchronously selects a suitable memory block from the RDMA queue and initializes the RDMA transfer process of the selected memory block;

RDMA communication manager (RDMA CM): responsible for registering the memory block to be transferred to the RDMA network adapter;

RDMA Network Adapter (RDMANetwork Interface Card, RNIC): A network card with RDMA function.

That is, the host migration agent on the first data node will monitor the memory situation in real time, and once it finds that a memory block has been modified, it will immediately add the memory block to a queue. At the same time, the host migration agent also asynchronously selects a memory block from the queue, registers it on the RDMA network card, and transmits it to the second data node through RDMA. Among them, the selection of memory blocks from the queue must follow certain rules: first, the size of the selected memory block and the memory block already being transferred cannot exceed a threshold; second, in the memory block that loads the first rule, You can select the earliest/latest modified memory block, or the most frequently modified memory block.

As shown in FIG. 4 , which is a schematic flow chart of data migration in the prior art, each modified memory block has to go through the following state change processes. On the first data node, after the memory block is modified, it is first in the queue; after being selected, it will be locked; then registered to the RDMA network card; and then transferred to the second data node through RDMA; after the transfer is completed, the memory block is transferred from RDMA network card logout; finally unlock the memory block. Here, data migration is performed in units of data blocks. During the migration, a large amount of data may be locked and unavailable on the selected memory block for a long time, thereby affecting the quality of service.

As shown in FIG. 5 , it is an overall architecture diagram of a distributed cluster applied in the embodiment of the present application. In the embodiment of this application, the migration data of the first data node is mapped to the RDMA virtual memory space of the second data node by using the RDMA virtual memory mapping technology, and all usage rights and loads are switched to the second data node, thereby The load can be migrated without waiting for the completion of data replication, which improves the speed of load migration and reduces the delay; the use of RDMA technology for the process of real-time data migration of distributed database clusters can also reduce the consumption of CPU resources of the first data node by data transmission. Reduce the impact of the migration process on other running businesses, and increase the speed of data transmission; further, the first data node migrates the data blocks into blocks, which can shorten the unavailable time and data volume of the target data, and improve service quality.

In Figure 5, the upper part is the physical structure of the database cluster, and the lower part is the logical structure of the data. As can be seen from Figure 5, there are 5 data nodes in the cluster (generally referring to 5 machines, not limited to 5), and each data node has 3 storage stores (not limited to 3). As can be seen from the lower part of Figure 5, each store stores multiple paragraphs range (range means that all the data stored in the database is divided into multiple paragraphs, each paragraph is called a range, for example, according to the ASCII initial character , you can divide all the data into multiple ranges such as 0-9, a-c, c-e, etc.). Each store stores several ranges. For example, store1 of the data node node1 stores 6 ranges including 0-9, a-c, e-g, g-i, i-k, and k-m.

For the needs of disaster recovery, each range can be copied into several copies and placed on different stores of different data nodes, so that when one of the data nodes fails, other data nodes still save the data node copy. As can be seen from Figure 5, multiple ranges are stored on each store, different patterns represent different ranges, and the same pattern represents the same range. For example, store1 of node1 and store1 of node2 in the figure both store copies of Range a-c.

During the operation of the database, it is inevitable to encounter a situation that some ranges are accessed more and some ranges are accessed less. This will cause some data nodes to be accessed more, and some data nodes to be accessed less, that is, the load is unbalanced. A range that is visited more often can be called a hotspot range, and a data node that is visited more can be called a hotspot data node. When the access rate of a hot data node reaches a certain limit, the performance of the data node will decrease accordingly, and the delay of range access on the data node will also increase, and even cause interruption and crash.

In order to avoid serious consequences caused by unbalanced load, database clusters need to perform load migration. Generally, some ranges on hot data nodes can be migrated to other data nodes. But the process of load migration often has a lot of overhead. For example, the copying of range data consumes a lot of network bandwidth. Moreover, traditional network transmission requires the participation of the CPU. The data to be transmitted needs to be copied from the user space to the kernel space, and then processed by the TCP/IP protocol stack. Sending out, the receiving process is equally complicated. Therefore, the load migration process will put a greater burden on the hot data nodes that were originally overburdened, and the consequence may be that the data nodes hang directly.

The embodiment of the present application adopts the RDMA virtual memory mapping technology to directly map the range to be migrated to the second data node, so that there is no need to wait for the completion of data copying, and the load can be quickly migrated and the delay can be reduced. By introducing RDMA technology, RDMA technology is used in the data replication process of load migration, which greatly improves the speed of data transmission. More importantly, the whole process of RDMA transmission does not require the CPU of the first data node to participate, so compared with the traditional The network transmission method, especially the burden on the first data node is greatly reduced.

The method of data migration in the technical solution of the present application will be further described below in the form of an embodiment. As shown in FIG. 6, it is a schematic diagram of an embodiment of the method of data migration in the embodiment of the present application, including:

601. The management node determines a first data node and a second data node for data migration.

In the embodiment of this application, in a possible implementation manner, the management node periodically collects relevant information of each data node in the cluster; that is, the management node collects cluster information at regular intervals, and the cluster information may include Each data node and the capacity and load information of each store on the data node. The first data node may also be called a source data node, and the second data node may also be called a target data node.

In another possible implementation, the first data node may send an instruction to the management node that the load exceeds the first threshold, and the instruction that the load exceeds the first threshold may indicate that the first data node is the source data node, and the second data node An instruction that the load is lower than the second threshold is sent to the management node, and the instruction that the load is lower than the second threshold may indicate that the second data node is the target data node. Or, the management node receives the instruction sent by the first data node that the load exceeds the first threshold, determines that the first data node is the source data node, and receives the instruction sent by the second data node that the load is lower than the second threshold, and determines that the second data node node is the target data node.

Wherein, the management node determines the first data node and the second data node for data migration according to the cluster information, which may include but not limited to the following implementations:

(1) If the management node detects that a new data node joins the database cluster, the management node selects a data node with the largest load from the database cluster as the first data node, and uses the new data node as the second data node.

(2) If the management node does not detect that there is a new data node joining the database cluster, the management node selects a data node with the largest load from the database cluster as the first data node, and selects a data node with the smallest load as the second data node.

Exemplarily, as shown in FIG. 7 , it is a schematic flow diagram of triggering load balancing. If the management node finds that a new data node joins the database cluster, it immediately selects a store with the largest load from the database cluster as the source store, and picks a hotspot range with the largest load from the store as the migration data, and transfers the newly added data node to the database cluster. Any store is used as the target store to trigger load migration; if no new data node is added, it depends on whether the load of any store exceeds the threshold. Pick a hotspot range with the largest load as the migration data. Then find a store with the smallest load in the database cluster as the target store to trigger load balancing; if neither new data nodes are added nor the store load exceeds the threshold, load balancing will not be triggered.

It should be noted that the management node may be an independent server, or a module or unit integrated on the server, such as a management module or a routing module, which is not specifically limited.

602. The first data node sends a data migration request about the target data to the second data node.

In this embodiment of the application, the data migration request may include the memory address of the target data in the first data node, the data migration request is used for the second data node to establish an RDMA virtual memory space, and the memory address of the RDMA virtual memory space is mapped to the first data node The memory address of a data node, the RDMA virtual memory space is used for the second data node to access the target data on the first data node through RDMA according to the target operation request, and the target operation request is routed by the management node to access the target data on the second data node Action request. Usually, the hotspot data is the target data here, and can also be understood as the migration data on the first data node. It should be understood that hot data refers to data that is frequently accessed or whose access times exceed a specific threshold.

In a possible implementation, before the first data node sends a data migration request about the target data to the second data node, the first data node and the second data node may receive an instruction to trigger load balancing from the management node , the first data node may send a data migration request about the target data to the second data node according to the instruction triggering load balancing, and the second data node is ready to receive the target data according to the instruction triggering load balancing.

In another possible implementation, before the first data node sends a data migration request about the target data to the second data node, the first data node sends an instruction to the management node that the load exceeds the first threshold, and the second data node Send an instruction that the load is lower than the second threshold to the management node, the management node can determine that the first data node is the source data node, and the second data node is the target data node, and the management node can send an instruction to trigger load balancing to the first data node. The first data node may send a data migration request about the target data to the second data node according to the instruction triggering load balancing, and the second data node is ready to receive the target data according to the instruction triggering load balancing.

Optionally, the management node can also determine the target data to be migrated from the first data node to the second data node, that is, the hot data, and then send information about the hot data to the first data node, and the first data node can determine according to the information about the hot data hot data.

603. The second data node establishes an RDMA virtual memory space according to the data migration request.

In this embodiment of the application, the second data node establishes an RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used by the second data node according to the target The operation requests to access target data on the first data node through RDMA.

604. The second data node sends an instruction to modify the metadata to the management node.

In this embodiment of the present application, after the virtual memory space of the second data node is established, the second data node may send an instruction to modify the metadata to the management node, and the management node receives the instruction to modify the metadata sent by the second data node. It should be noted that metadata (Metadata), also known as intermediary data and relay data, is data describing data (dataabout data), mainly describing information about data attributes (property), used to support such as indicating storage Data, resource search, file recording and other functions.

605. The management node modifies the metadata according to the metadata modification instruction, so that the target operation request for accessing the target data on the first data node is routed to the second data node.

In this embodiment of the application, after the management node receives the metadata modification instruction sent by the second data node, the management node can modify the metadata according to the metadata modification instruction, so that the target data originally routed to the first data node All target operation requests are routed to the second data node.

Exemplarily, as shown in FIG. 8 , it is a schematic diagram of load migration in the embodiment of the present application. By using RDMA memory mapping technology, the migration data block P on the first data node S1 can be directly mapped to the data block P1 on the second data node S2, so that S2 can access the data on P as if it were accessing its own memory . Then, by modifying the metadata of the database cluster, the management node can route the operation requests to P that should have been routed to S1 to P1 of S2.

In this way, the second data node can access the data block P on S1 in the same way as accessing the data block P1 on S2, and S2 will handle all the load on P (although the actual data block P on S1 still needs to be accessed, But the CPU is the CPU of S2), and the mapping process from P to P1 and the network transmission process are transparent to users, thus realizing the instantaneous migration of load on P. Moreover, since the RDMA network transmission process does not require CPU participation, the load migration process will not consume the CPU resources of S1, so it will not affect other services running on S1. In addition, through memory mapping, the process of migrating the data block P on S1 to S2 is converted into the internal data copy process of S2 (P1 → P2), which is managed by S2 in a unified manner, which can improve the migration efficiency. It should be noted that, in this embodiment of the application, the target operation request can be understood as a load.

606. The second data node accesses the received target operation request.

In this embodiment of the application, when the second data node receives the target operation request, if the target data accessed by the target operation request is not stored on the second data node, the second data node accesses the first data through RDMA according to the target operation request The target data on the node; if the target data requested by the target operation has been stored on the second data node, the second data node will access the stored target data according to the target operation request.

In several other implementations, as follows:

(1) If the target operation request is a write operation request, and the write operation request indicates adding new data, the second data node performs a write operation on the second data node according to the write operation request.

(2) If the target operation request is a read operation request, and the target data accessed by the target operation request has been stored on the second data node, then the second data node performs access on the second data node according to the target operation request.

(3) If the target operation request is a read operation request, and the target data accessed by the target operation request is not stored on the second data node, the second data node accesses the target data on the first data node through RDMA according to the target operation request.

(4) If the target operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the target operation request has been stored on the second data node, Then the second data node performs access on the second data node according to the target operation request.

(5) If the target operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the target operation request is not stored on the second data node, Then the second data node accesses the target data on the first data node through RDMA according to the target operation request.

607. The first data node sends the target data to the second data node.

In the embodiment of this application, after the second data node takes over the load of the migrated data on the first data node, it can start to copy the migrated data (generally hotspot range) on the first data node to the second data node; After the migration data is copied, the load migration ends. The second data node receives and stores the target data sent by the first data node. It should be noted that hot data can be understood as data whose number of visits or visit rate exceeds a threshold. Optionally, the first data node may also send the target data to the second data node through RDMA, then the second data node may receive the target data sent by the first data node through RDMA.

The first data node sends the target data to the second data node, which may include: the first data node determines the hot data; the first data node divides the hot data into M data, and M is an integer greater than or equal to 2; The target data is selected from the M data; the first data node sends the target data to the second data node, or sends the target data to the second data node through RDMA.

Wherein, the hotspot data determined by the first data node may be determined by the first data node according to its own capacity, load information, access rate and other information, or it may be the information of the hotspot data of the first data node determined by the management node Send it to the first data node, and the first data node determines the hotspot data according to the information of the hotspot data.

When the first data node sends target data to the second data node through RDMA, a fragmented transmission method may be used. Exemplarily, as shown in FIG. 9 , it is a schematic diagram of fragmented data block transmission in the embodiment of the present application. Because the data may be operated by the first data node and the second data node at the same time during the migration process, the data is locked and unavailable during the migration process. If the locked data is too large, the migration time will be correspondingly longer, and a large amount of data will be unavailable for a long time, which will inevitably lead to a decline in service quality and even service interruption. Therefore, in the embodiment of this application, large data blocks are divided into multiple small blocks (generally divided into blocks by page) for migration, thereby reducing the amount of unavailable data and shortening the time of data unavailable, thereby improving service quality and avoiding interruption.

In the schematic diagram shown in Figure 9:

1) Divide the migration data block P1 into multiple small data blocks (for example, one page per small data block), and migrate the small data blocks to P2 one by one;

2) Lock the small data block being migrated (read/write are not accessible);

3) During the replication process, all write requests for the migrated data are routed to P2 for execution;

4) For read or update operation requests, a migration status table can be saved, which records the migration status of each small data block (copy completed, copying, not copied). For the data blocks that have been copied, directly perform read or update operations on P2; for data blocks that are not copied or are being copied, perform read or update operations on P1.

It should be noted that the data block P1 is actually a mirror image mapped from the migrated data block P of the first data node to the target server, so if the above-mentioned block migration algorithm is directly used instead of the RDMA technology, each small block When the data is transferred to the second data node, the metadata information of the block of data must be modified, and the migration speed will inevitably be greatly reduced.

As shown in FIG. 10 , it is a schematic flowchart of a fragment transmission algorithm. After the copying of the data block P on the first data node starts, the migrated data will be fragmented (generally by page), and then copied piece by piece; then the small data block being copied is locked to make it inaccessible for reading/writing; when there is an operation When requesting access to migrated data, check whether the operation request is a write operation request. If it is a write operation request, it will be routed directly to the second data node; if it is not a write operation (such as a read operation request), it will be judged whether the small data block is The copying has been completed. If the copying has been completed, the read operation request is routed to the second data node for execution; if the copying is not completed, the read operation request is routed to the first data node for execution.

In this embodiment of the application, the management node determines the first data node and the second data node for data migration; the first data node sends a data migration request about the target data to the second data node; the second data node receives the first data node The data migration request sent about the target data, the data migration request includes the memory address of the target data in the first data node; the second data node establishes an RDMA virtual memory space according to the data migration request, and the memory address of the RDMA virtual memory space is mapped to the first data node The memory address of a data node, the RDMA virtual memory space is used for the second data node to access the target data on the first data node through RDMA according to the target operation request; the second data node sends an instruction to modify metadata to the management node; the management node receives The instruction to modify the metadata sent by the second data node; the management node modifies the metadata according to the instruction to modify the metadata, so that the target operation request for accessing the target data on the first data node is routed to the second data node; the first data node passes The RDMA sends the target data to the second data node; the second data node receives and stores the target data sent by the first data node through RDMA.

That is, the second data node establishes an RDMA virtual memory space according to the data migration request about the target data sent by the first data node, and the second data node sends an instruction to modify the metadata to the management node, and the management node can modify the metadata according to the instruction to modify the metadata Metadata, so that the target operation request for accessing the target data is routed to the second data node, reducing the delay; thus, there is no need to wait until the target data is replicated on the second data node before migrating the target operation request; the first data node uses RDMA Send the target data to the second data node. Using RDMA virtual memory technology, the target data is mapped to the RDMA virtual memory space of the second data node, and all usage rights and loads are switched to the second data node, so that the load can be migrated without waiting for the copy of the target data to be completed, improving the The speed of load migration; that is, RDMA technology is used in the process of real-time data migration of distributed database clusters, which reduces the consumption of CPU resources on the first data node by data migration, and reduces the impact of the migration process on other businesses running on the first data node , and increase the speed of data transmission. Furthermore, the target data is migrated in small pieces, thereby shortening the unavailable time and data volume of the target data, and improving the service quality.

The method of data migration in the embodiment of this application has been described above, and the data nodes in the embodiment of this application will be described below, as shown in Figure 11, which is a schematic diagram of an embodiment of a data node in the embodiment of this application, including:

The receiving module 1101 is configured to receive a data migration request about the target data sent by the first data node, the data migration request includes the memory address of the target data in the first data node; receive and store the target data sent by the first data node;

The processing module 1102 is configured to establish a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used for the second data node to operate according to the target Request to access target data on the first data node through RDMA;

The sending module 1103 is configured to send an instruction for modifying metadata to the management node, where the instruction for modifying metadata is used for the management node to modify the metadata, so that the target operation request for accessing the target data is routed to the second data node.

Optionally, in some embodiments of the embodiments of the present application,

The receiving module 1101 is specifically configured to receive and store target data sent by the first data node through RDMA.

Optionally, in some embodiments of the present application, the target data is one of the hotspot data divided into M data in the first data node, where M is an integer greater than or equal to 2.

Optionally, in some embodiments of the present application,

The processing module is further configured to: if the target operation request is a write operation request, and the write operation request indicates adding new data, the processing module performs a write operation on the second data node according to the write operation request.

Optionally, in some embodiments of the present application,

The processing module 1102 is further configured to: if the target operation request is a read operation request, or a write operation request to modify the target data, and the target data accessed by the target operation request has been saved on the second data node, the processing module Requesting access on the second data node; if the target operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the target operation request is not stored on the second data node, the processing module The target data on the first data node is accessed through RDMA according to the target operation request.

As shown in Figure 12A, it is a schematic diagram of another embodiment of the data node in the embodiment of the present application, including:

The sending module 1201 is configured to send a data migration request about the target data to the second data node, the data migration request includes the memory address of the target data in the first data node, and the data migration request is used for the second data node to establish remote direct memory access RDMA virtual memory space, the memory address of the RDMA virtual memory space is mapped to the memory address of the first data node, and the RDMA virtual memory space is used for the second data node to access the target data on the first data node through RDMA according to the target operation request, and the target operation The request is an operation request routed by the management node to the second data node to access the target data; and the target data is sent to the second data node.

Optionally, in some embodiments of the present application,

The sending module 1201 is specifically configured to send target data to the second data node through RDMA.

Optionally, in some embodiments of the present application,

The sending module 1201 is specifically used to determine hot data; divide the hot data into M pieces of data, where M is an integer greater than or equal to 2; select target data from the M pieces of data; and send the target data to the second data node through RDMA.

Optionally, in some embodiments of the present application, on the basis of what is shown in FIG. 12A , as shown in FIG. 12B , it is a schematic diagram of another embodiment of a data node. The data node also includes:

A receiving module 1202, configured to receive hotspot data information sent by the management node;

The sending module 1201 is specifically configured to determine hotspot data according to hotspot data information.

As shown in FIG. 13 , it is a schematic diagram of another embodiment of the data node in the embodiment of the present application.

The data nodes may have relatively large differences due to different configurations or performances, and may include one or more central processing units (central processing units, CPU) 1322 (for example, one or more processors) and memory 1332, one or one The storage medium 1330 (such as one or more mass storage devices) for storing the application program 1342 or the data 1344 above. Wherein, the memory 1332 and the storage medium 1330 may be temporary storage or persistent storage. The program stored in the storage medium 1330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on data nodes. Furthermore, the central processing unit 1322 may be configured to communicate with the storage medium 1330, and execute a series of instruction operations in the storage medium 1330 on the data node.

The data node can also include one or more power supplies 1326, one or more wired or wireless network interfaces 1350, one or more input and output interfaces 1358, and/or, one or more operating systems 1341, such as Windows Server™, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

The steps performed by the first data node and the second data node in the above embodiment can be based on the data node structure shown in FIG. 13 , and will not be repeated here.

In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present invention will be generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website, computer, server or data center Transmission to another website site, computer, server, or data center by wired (eg, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center integrated with one or more available media. The available medium may be a magnetic medium (for example, a floppy disk, a hard disk, or a magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, a Solid State Disk (SSD)).

Optionally, in some embodiments of the present application, a computer-readable storage medium is provided, including instructions, which, when run on a computer, cause the computer to execute the first data node or the second data node in FIG. 6 above. the method described.

Optionally, in some embodiments of the present application, a computer program product containing instructions is provided, and when it runs on a computer, the computer executes the above-mentioned first data node or the second data node in FIG. 6. Methods.

Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and other media that can store program codes.

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still understand the foregoing The technical solutions described in each embodiment are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the application.

Claims (22)

1. A method for data migration, comprising: The second data node receives a data migration request about the target data from the first data node, where the data migration request includes a memory address of the target data in the first data node; The second data node establishes a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space is used for The second data node accesses data on the first data node; The second data node sends an instruction to modify metadata to the management node, and the instruction to modify metadata is used by the management node to modify the metadata, so that an operation request for accessing the target data is routed to the first Two data nodes; The second data node receives the target data from the first data node. 2. The method according to claim 1, wherein the second data node receives and stores the target data sent by the first data node, comprising: The second data node receives and stores the target data sent by the first data node through RDMA. 3. The method according to claim 1 or 2, wherein the target data is a part of hotspot data of the first data node. 4. The method according to any one of claims 1-3, wherein the method further comprises: If the operation request is a write operation request, and the write operation request indicates adding new data, the second data node performs a write operation on the second data node according to the write operation request. 5. The method according to any one of claims 1-3, wherein the method further comprises: If the operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the target operation request has been stored on the second data node, then the second The second data node performs access on the second data node according to the operation request; If the operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the operation request is not stored on the second data node, the second The data node accesses the target data on the first data node through RDMA according to the operation request. 6. A method for data migration, comprising: The first data node sends a data migration request about the target data to the second data node, the data migration request includes the memory address of the target data in the first data node, and the data migration request is used for the second data node Two data nodes establish a remote direct memory access RDMA virtual memory space, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space is used for the second data node to access the data on the first data node; The first data node sends the target data to the second data node. 7. The method according to claim 6, wherein the sending of the target data by the first data node to the second data node comprises: The first data node sends the target data to the second data node through RDMA. 8. The method according to claim 6 or 7, wherein the sending of the target data by the first data node to the second data node comprises: The first data node determines hotspot data; The first data node divides the hotspot data into M pieces of data, where M is an integer greater than or equal to 2; The first data node selects the target data from the M shares of data; The first data node sends the target data to the second data node. 9. The method according to claim 8, wherein before the first data node determines the hotspot data, the method further comprises: The first data node receives the hotspot data information sent by the management node; The first data node determines hotspot data, including: The first data node determines the hotspot data according to the hotspot data information. 10. A data node, characterized in that, comprising: A receiving module, configured to receive a data migration request about target data from a first data node, where the data migration request includes a memory address of the target data in the first data node; receive and store the first data The target data sent by the node; A processing module, configured to establish a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space is used for all The second data node accesses data on the first data node; A sending module, configured to send an instruction to modify metadata to a management node, where the instruction to modify metadata is used by the management node to modify the metadata so that an operation request for accessing the target data is routed to the second data node. 11. The data node according to claim 10, characterized in that, The receiving module is specifically configured to receive and store the target data sent by the first data node through RDMA. 12. The data node according to claim 10 or 11, wherein the target data is hotspot data of the first data node. 13. The data node according to any one of claims 10-12, characterized in that, The processing module is further configured to: if the operation request is a write operation request, and the write operation request indicates adding new data, the processing module writes on the second data node according to the write operation request operate. 14. The data node according to any one of claims 10-12, characterized in that, The processing module is further configured to: if the operation request is a read operation request, or a write operation request for modifying the target data, and the target data accessed by the operation request has been stored in the second data node, the processing module performs access on the second data node according to the operation request; It is also used if the operation request is a read operation request, or a write operation request for modifying the data, and the target data accessed by the operation request is not stored on the second data node, then the The processing module accesses the target data on the first data node through RDMA according to the operation request. 15. A data node, characterized in that, comprising: A sending module, configured to send a data migration request about target data to a second data node, where the data migration request includes a memory address of the target data in the first data node, and the data migration request is used for the The second data node establishes a remote direct memory access RDMA virtual memory space, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space is used for the second data node to access all Describe the data on the first data node. 16. The data node according to claim 15, characterized in that, The sending module is specifically configured to send the target data to the second data node through RDMA. 17. The data node according to claim 15 or 16, characterized in that, The sending module is specifically used to determine hotspot data; divide the hotspot data into M pieces of data, where M is an integer greater than or equal to 2; select the target data from the M pieces of data; The node sends the target data. 18. The data node according to claim 15, wherein the data node further comprises: The receiving module is used to receive the hotspot data information sent by the management node; The sending module is specifically configured to determine the hotspot data according to the hotspot data information. 19. A data node, comprising: a transceiver, a processor, a memory, and a bus, the transceiver, the processor, and the memory being connected through the bus; The memory is used to store operation instructions; The transceiver is configured to receive a data migration request about target data from a first data node, where the data migration request includes a memory address of the target data in the first data node; and send modification metadata to a management node An instruction for modifying metadata is used for the management node to modify the metadata so that a target operation request for accessing the target data is routed to the second data node; receiving and storing the first data The target data sent by the node; The processor is configured to establish a remote direct memory access RDMA virtual memory space according to the data migration request, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space uses Access target data on the first data node at the second data node. 20. A data node, characterized in that, comprising: a transceiver, a memory and a bus, the transceiver and the memory are connected through the bus; The memory is used to store operation instructions; The transceiver is configured to send a data migration request about target data to a second data node, where the data migration request includes a memory address of the target data in the first data node, and the data migration request is used for The second data node establishes a remote direct memory access RDMA virtual memory space, the memory address of the RDMA virtual memory space corresponds to the memory address of the first data node, and the RDMA virtual memory space is used for the second data node Access data on the first data node. 21. A computer-readable storage medium, characterized by comprising instructions, which, when run on a computer, cause the computer to execute the method according to any one of claims 1-9. 22. A computer program product containing instructions, characterized in that when it is run on a computer, it causes the computer to execute the method according to any one of claims 1-9.

Images