CN114928615B - Load balancing method, device, equipment and readable storage medium - Google Patents

Abstract

The application discloses a load balancing method, a device, equipment and a readable storage medium, wherein when a host is newly added in a load balancing system, a load balancer receives a data packet, if a five-tuple of the data packet does not exist in a mapping table, effective RSs for load balancing are determined from RSs of all hosts contained in the load balancing system, a target RS is selected from the effective RSs, and the data packet is sent to the target RS, so that the second load balancing is completed. By adopting the scheme, after the host is newly added each time, the load balancer determines the target RS from the effective RS and sends the data packet, so that the load balancer on the newly added host can generate a mapping table aiming at the existing service in sufficient time, the mapping relation between the quintuple and the RS of the existing service is recorded in the mapping table, the load balancer is prevented from sending the data packet of the existing service to the incorrect RS, and the purpose of improving the service quality is realized.

Description

Load balancing method, device, equipment and readable storage medium

Technical field

This application relates to the field of load balancing technology, and in particular to a load balancing method, device, equipment and readable storage medium.

Background technique

With the development of cloud technology, the number of accesses faced by servers is increasing rapidly, and servers need to be able to handle large-scale concurrent service accesses. For servers with heavy loads, CPU and I/O processing capabilities quickly become bottlenecks, and the performance of a single server is always limited. Simply improving hardware performance cannot really solve the problem. Therefore, multi-server and load balancing (Load Balance, LB) technology must be used to meet the needs of large-scale concurrent service access.

Load balancing refers to using a load balancer to distribute a large amount of concurrent access or data traffic to multiple real servers, which are processed by the real servers to reduce the time users wait for responses. In cloud platforms, load balancers usually adopt the Linux Virtual Server (LVS) architecture, which is composed of a front-end load balancer (LoadBalancer, LB) and a backend real server (Real Server, RS) group. After receiving the service request from the client, LB schedules and forwards the service request to RS according to the preset scheduling forwarding policy and load balancing scheduling algorithm. RS processes the service request and returns the response result. However, LVS load balancing technology has the disadvantages of high performance overhead and LB is easily affected by load sudden increases.

In order to overcome the problems of LVS load balancing technology, the industry introduced the load balancing architecture of eXpress Data Path (XDP). However, based on the XDP architecture, when a new host is added to the cluster, the connection of the ongoing business in the cluster will be abnormally disconnected, and the abnormality is proportional to the number of new hosts, causing the ongoing business to be abnormal.

Contents of the invention

This application provides a load balancing method, device, equipment and readable storage medium. By improving the load balancer, it solves the impact on existing businesses when new hosts are added to the XDP-based architecture and achieves the purpose of improving business reliability. .

In a first aspect, embodiments of the present application provide a load balancing method, which is applied to a load balancer on a host in a load balancing system. The method includes:

Determine that there is a new host in the load balancing system, the load balancer is a load balancer on any one of the multiple hosts included in the load balancing system, and the load balancer is deployed on each of the multiple hosts. Load balancer and real server RS;

receive data packet;

The effective RS is determined from the RS of each host included in the load balancing system. The effective RS is the RS involved when the load balancer performs load balancing on the data packets of existing services before adding a host. There are services that the load balancing system is providing to users before the new host is added;

Determine a target RS from the effective RS;

Send the data packet to the target RS.

In the second aspect, embodiments of the present application provide a load balancing method, which is applied to query devices in the load balancing system. The method includes:

Receive a notification request sent by the RS on the newly added host in the load balancing system, where the notification request is used to indicate the virtual IP address VIP of the RS;

Record the VIP and the notification time point. The notification time point is the time point when the query device receives the notification request. The VIP and the notification time point are used to determine whether the RS on the newly added host is Convert to effective RS. The effective RS in the load balancing system is the RS involved when the load balancer in the load balancing system load balances the data packets of existing services. The existing services are before the new host is added. The load balancing system is providing services to users.

In a third aspect, embodiments of the present application provide a load balancing device, a load balancer integrated on a host in a load balancing system, and the device includes:

Determining module, used to determine that there is a new host in the load balancing system, the load balancer is a load balancer on any one of the multiple hosts included in the load balancing system, each of the multiple hosts Deploy load balancer and real server RS on each host;

Receiving module, used to receive data packets;

A processing module configured to determine the effective RS from the RS of each host included in the load balancing system. The effective RS is involved when the load balancer performs load balancing on data packets of existing services before adding a host. RS, the existing service is the service that the load balancing system is providing to the user before the new host is added, and the target RS is determined from the effective RS;

A sending module, configured to send the data packet to the target RS.

In the fourth aspect, embodiments of the present application provide a load balancing device integrated on a query device in a load balancing system. The device includes:

A transceiver module, configured to receive a notification request sent by the RS on the newly added host in the load balancing system, where the notification request is used to indicate the virtual IP address VIP of the RS;

A processing module configured to record the VIP and notification time point. The notification time point is the time point when the query device receives the notification request. The effective RS in the load balancing system is before the new host is added. , the RS involved when the load balancer in the load balancing system performs load balancing on the data packets of existing services. The existing services are the services that the load balancing system is providing to users before the new host is added.

In a fifth aspect, embodiments of the present application provide an electronic device, including: a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the The electronic device implements the method described in the above first aspect or various possible implementations of the first aspect.

In a sixth aspect, embodiments of the present application provide an electronic device, including: a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the The electronic device implements the method described in the above second aspect or various possible implementations of the second aspect.

In a seventh aspect, embodiments of the present application provide a computer-readable storage medium. Computer instructions are stored in the computer-readable storage medium. When executed by a processor, the computer instructions are used to implement the above first aspect or the first aspect. various possible implementations.

In an eighth aspect, embodiments of the present application provide a computer-readable storage medium. The computer-readable storage medium stores computer instructions. When executed by a processor, the computer instructions are used to implement the above second aspect or the second aspect. various possible implementations.

In a ninth aspect, embodiments of the present application provide a load balancing system, including a switch, multiple hosts, and a query device. The switch establishes a network connection with each of the multiple hosts, and the query device communicates with the multiple hosts. Each of the hosts establishes a network connection, where:

The switch is configured to perform load balancing to determine a target load balancer from the load balancers included in the plurality of hosts and send data packets;

A load balancer and a real server RS are deployed on each of the plurality of hosts, and the load balancer is used to perform the load balancing method described in the above first aspect or various possible implementations of the first aspect.

The query device is configured to perform the load balancing method described in the above second aspect or various possible implementations of the second aspect.

In the load balancing method provided by the embodiment of this application, when the load balancer on each host in the load balancing system detects a new host in the load balancing system, once a load balancer receives a data packet, if the load balancer does not exist in the mapping table For the five-tuple of the data packet, the effective RS for load balancing is determined from the RS of each host included in the load balancing system, the target RS is selected from the effective RS and the data packet is sent to the target RS, thereby completing the second step. Secondary load balancing. Using this solution, each time a new host is added, the load balancer determines the target RS from the effective RS and sends a data packet to ensure that the load balancer on the new host has sufficient time to generate mappings for existing services. Table, the mapping table records the mapping relationship between the quintuple of the existing business and the RS to prevent the load balancer from sending the data packets of the existing business to the incorrect RS. That is to say, every time a new host is added, the load balancer on the host takes effect but the RS does not immediately participate in load balancing. Instead, the mapping table of the load balancer on the newly added host saves all five existing services. After the mapping relationship between tuples and RSs is established, the RS on the newly added host will be used for load balancing to ensure that the data packets of existing services after the new host are added are sent to the correct RS, thereby improving service quality without any changes. switch to reduce the impact and operational complexity.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a schematic architectural diagram of a load balancing system to which the load balancing method provided by the embodiment of the present application is applicable;

Figure 2 is a schematic diagram of the working level of the XDP-based load balancing architecture on the host provided by the embodiment of this application;

Figure 3 is a flow chart of the load balancing method provided by the embodiment of the present application;

Figure 4 is another flow chart of the load balancing method provided by the embodiment of the present application;

Figure 5 is a schematic diagram of the application scenario of the load balancing method provided by the embodiment of the present application;

Figure 6 is a schematic diagram of a load balancing device provided by an embodiment of the present application;

Figure 7 is a schematic diagram of another load balancing device provided by an embodiment of the present application;

Figure 8 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

With the development of cluster technology, load balancing, as the core of cluster technology, has also received widespread attention and research. While load balancing ensures that the load of the back-end servers is equal, it also needs to ensure that data packets belonging to the same connection are sent to the same back-end server. Packets belonging to the same connection have the same quintuple.

Load balancing aims to provide high-availability, high-performance cluster technology. It provides a cheap, effective and transparent method to expand the bandwidth of the cluster, increase the throughput of the cluster, strengthen the cluster's ability to process network data, and ensure high availability of services.

The existing four-layer load balancing technology uses the LVS architecture, which is a load balancing technology based on TCP/IP. It is implemented through the IP Virtual Server (IPVS) module and works in the kernel. The LVS architecture has the following disadvantages:

First, the performance overhead is serious.

Based on the LVS architecture, data packet forwarding needs to go through the network card and kernel system to achieve forwarding. From a design perspective, in order to ensure security and isolation, the operating system must go through the following processes when data packets pass through the kernel: allocating socket buffer (SKB), socket locking, TCP (or UDP)... ...Protocol stack, kernel mode and user mode switching. Causes serious performance overhead. At the same time, the LVS architecture needs to ensure synchronization between load balancers in the cluster, resulting in an abnormal increase in traffic and high load balancer performance overhead.

Second, the load balancer is susceptible to sudden load increases.

Based on the LVS architecture, due to high performance overhead, large data traffic will increase the possibility of equipment failure. Therefore, in order to reduce the machine failure rate, by placing the load balancer and the RS that the backend actually provides services on different hosts, the number of load balancers is less than the number of RS, which makes the load balancer vulnerable to load bursts. Influence. This problem is exacerbated by the fact that packets go through the regular Linux network stack before being processed by the load balancer.

In order to overcome the problems of LVS load balancing technology, the industry introduced a load balancing architecture based on XDP. In this architecture, the data packets are filtered or processed directly before entering the user mode. Compared with the LVS-based load balancing architecture, the XDP-based load balancing architecture has better performance. The load balancing architecture based on XDP has the following advantages:

a. Timely and efficient data packet processing.

Under the XDP-based load balancing architecture, after the load balancer is enabled in driver mode, the load balancer processes the data packet after the network card receives the data packet and before the kernel intercepts the data packet. During the processing, for each inbound data packet, the load balancer always calls the Berkeley Packet Filter (BPF) program. If the network card has multiple queues, the load balancer calls the BPF program in parallel for each queue. Furthermore, the BPF routines used to process packets are lock-free and use BPF mappings unique to each CPU. Based on this parallel mechanism, the performance of load balancing scales linearly with the number of receive queues of the network card.

b. There is no need to ensure synchronization between load balancers in the cluster.

The XDP-based load balancing architecture uses a consistent hash algorithm instead of the traditional load balancing algorithm, which avoids the need for LVS-based load balancing architecture to ensure synchronization between load balancers in the cluster, thus avoiding traffic anomalies and load balancing. The problem of high server performance overhead.

c. Provide a fast path

The XDP-based load balancing architecture provides a fast path for high-speed processing of data packets. The load balancer and RS are deployed on the same host and run together without any performance degradation. At the same time, it enhances its ability to adapt to load surges, host failures, and maintenance.

The similarity between the XDP-based load balancing architecture and the LVS-based load balancing architecture is that the load balancer needs to establish a routing protocol with the switch in the computer room, such as the Open Shortest Path First (OSPF) protocol, in order to provide The virtual IP address (vertical IP, VIP) of the load balancer is advertised to the outside world. After that, the client can send the data packet to the corresponding load balancer in the cluster. Finally, the load balancer selects an RS from multiple RSs based on the load balancing algorithm and sends the data packet to the RS.

Normally, when the cluster provides services to the outside world, the number of hosts in the cluster increases and decreases unavoidably and frequently. Actual usage found that when a new host was added to the cluster, the data packet originally sent to RS A of the cluster was sent to RS B, causing an abnormality in the ongoing business within the cluster. Moreover, the proportion of anomalies increases with the number of new hosts.

In order to add new machines to the cluster, the industry considers improving the switches. But in reality, a switch is connected to many hosts, and each host has traffic. If you change the switch, it means: when you need to add machines or replace machines in the cluster due to planning changes or machine downtime, etc., if you change the switch, you need to back up all the traffic of the entire cluster, otherwise it will cause problems within the cluster. It affects ongoing business, consumes manpower and material resources, leads to poor business quality, and causes customer complaints. Among them, ongoing services are called old services, existing services, etc., and connections used to transmit data packets of ongoing services are called old connections.

Obviously, the existing XDP-based load balancing architecture cannot add new machines to the cluster, which greatly limits the scalability and availability of the cluster.

Based on this, embodiments of the present application provide a load balancing method, device, equipment and readable storage medium. By improving the load balancer, the load balancer has the function of querying the mounting duration of each RS to solve the problem of XDP-based The architecture reduces the impact on existing services when new servers are added, while achieving the purpose of improving service reliability without changing the switches.

Figure 1 is a schematic architectural diagram of a load balancing system to which the load balancing method provided by the embodiment of the present application is applicable. Please refer to Figure 1. The load balancing system includes a host 11, a query device 12, a switch 13 and a terminal device 14. A network connection is established between the query device 12 and each host 11 . A network connection is established between each host 11 and the switch 13 , and a network connection is established between the terminal device 14 and the switch 13 .

Host 11 is also called a machine, server, etc., and XDP-based load balancer 111 and RS112 are deployed on each host 11. The load balancer 111 has the function of querying the VIP of each RS112, and can obtain the VIP of the RS112 on each host 11 and the VIP notification time point through the query device 12. It should be noted that since the load balancer 111 and RS12 are deployed on each host 11, the VIP of the host 11, the load balancer 111 on the host 11, and the VIP of RS112 on the host 11 are the same.

The query device 12 is also called the IP watcher component. It can be a machine independent of the host 11, or it can be a host 11. The load balancer 111 and RS112 are deployed on the host 11 and at the same time, each host in the load balancing system is saved. 11 VIP and VIP notification time. Each time a new host 11 is added in the load balancing system, RS112 of the new host 11 notifies the VIP of the new host 11 to the query device 12, and the query device 12 records the VIP of RS112 and the notification time point. It can be seen that the query device 12 is used to collect and record the VIP and notification time points of each RS 112 in the load balancing system. The notification time point is the time point when host 11 is added.

The switch 13 is used to connect the terminal device 14 and the host 11 for the first load balancing. After receiving the data packet from the terminal device 14, the switch 13 selects one host 11 from the plurality of hosts 11 and sends the data packet. After the load balancer 111 of the selected host 11 receives the data packet, it executes the load balancing method provided by the embodiment of this application to perform the second load balancing, selects the target RS from multiple RSs 112, and sends the data packet to the target RS.

The terminal device 14 may be hardware or software. When the terminal device 14 is hardware, the terminal device 14 is, for example, a mobile phone, a tablet computer, a personal computer, an e-book reader installed with an Android operating system, a Microsoft operating system, a Symbian operating system, a Linux operating system or an Apple iOS operating system. Laptops, desktop computers, etc. When the terminal device 14 is software, it can be installed in the hardware device listed above. In this case, the terminal device 14 can be, for example, multiple software modules or a single software module, which is not limited by the embodiments of the present application.

Please refer to Figure 1. The load balancer 111 uses Fast Reroute (FRR), quagga, etc. to perform peer-to-peer interconnection with the switch 13, and notifies the switch 13 of the VIP of the load balancer 111. Then, the switch 13 uses the Equal-Cost Multipath Routing (ECMP) mechanism to select a target load balancer from the load balancer 111 that has advertised the VIP and sends the data packet, thus completing the first load balancing. When the load balancer 111 advertises the VIP to the switch 13, the Open Shortest Path First (OSPF) routing protocol is usually used.

According to the above, it can be seen that there are two kinds of VIP notifications in the embodiment of the present application. One is that the load balancer 111 uses the OSPF routing protocol to notify the VIP of the load balancer 111 to the outside world, that is, to the switch 13 and so on. One is that RS112 notifies the query device 12 of the VIP of RS112.

After the first load balancing, after the data packet reaches the load balancer 111 through the switch 13, the load balancer 111 will select the target RS from the RS112 on the host 11 included in the load balancing system, and send the data packet to the target RS. This completes the second load balancing. In order to ensure that data packets of the same service are sent to the same target RS, the load balancer 111 will use a consistent hash algorithm after receiving the first data packet of a service. The consistent hash algorithm determines the target RS based on the five-tuple of the data packet, the VIP of each RS in the load balancing system, the number of RS, etc. After determining the target RS for the first data packet of a service, the load balancer 111 records the mapping relationship between the five-tuple of the first data packet and the target RS in the local mapping table. In this way, each time subsequent load balancer 111 receives a data packet of the same service, the target RS can be determined by querying the mapping table according to the five-tuple of the data packet.

That is to say, in the embodiment of this application, each time the load balancer 111 receives a data packet, it first queries the mapping table according to the five-tuple of the data packet. If there is a five-tuple of the data packet in the mapping table, it directly adds the five-tuple to the data packet. The RS corresponding to the tuple is used as the target RS. If the five-tuple of the data packet does not exist in the mapping table, the effective RS is determined, and the target RS is determined from the effective RS.

Using this solution, the load balancer first checks the mapping table to determine the target RS each time it performs the second load balancing. Only when the target RS cannot be determined by querying the mapping table, the target RS is determined from the effective RS. , to avoid the performance consumption caused by repeatedly executing consistent hashing.

Optionally, in the above embodiment, after the load balancer determines the target RS from the effective RS, it also adds a mapping relationship between the five-tuple of the data packet and the target RS in the mapping table. In this way, every time the mapping relationship between a new quintuple and RS is determined, a new record is added to the mapping table. Since the quintuple of data packets of the same service is the same, after subsequently receiving the data packet of the same service, the target RS can be determined by directly querying the mapping table without performing consistent hashing operations, which is fast and performance-consuming. Low.

Using this solution, by storing the mapping relationship between the quintuple and the target RS in the mapping table, the performance consumption caused by subsequent repeated execution of consistent hashing can be avoided.

Optionally, in the above embodiment, when the load balancer determines the target RS from the effective RS, it performs a consistent hash operation based on the VIP of the effective RS, the quintuple of the data packet, the number of effective RS, etc., thereby determining the target RS. TargetRS.

In addition, since all data packets of the same service have the same five-tuple, when the number of RSs in the load balancing system remains unchanged, no matter which load balancer 111 receives the data packet, it can all The data packet is sent to the same destination RS. Therefore, synchronization of individual load balancers 111 within the load balancing system is not required.

When a new host 11 is added to the load balancing system, it is equivalent to adding load balancer 111 and RS112 at the same time. Switch 13 uses the OSPF routing protocol to perform the first load balancing to determine the target load balancer. It is very likely that the data packet will be sent to the load balancer 111 on the newly added host 11, and the mapping table of the load balancer 111 is empty.

The addition of host 11 causes the number of RS112 to change, which in turn causes the result of the consistent hashing algorithm to change. Therefore, in the traditional XDP-based load balancing architecture, when the load balancer 111 receives a data packet, and the data packet is a data packet of an existing service, the load balancer 111 determines the target based on the consistent hash algorithm. The RS may not be the correct RS. For example, the load balancer 111 sends the data packet that should have been sent to RS A to RS B. After RS B receives the data packet, since RS B does not have information such as the quintuple of the existing service, it directly discards the data packet or feeds back a reset (RST), causing the existing service to be abnormal. Obviously, the first load balancing of switch 13 and the second load balancing of load balancer 111 caused the proportion of existing business exceptions to soar when host 11 was added to the load balancing system.

It can be understood that the existing business means that before the new host 11 is added, there is at least one host 11 in the load balancing system, and the quintuple of this business and the target RS exist in the mapping table of the load balancer 111 on the host 11. Correspondence. For example, for a video service, before adding host 11, there were three hosts 11 in the load balancing system. Correspondingly, there were three load balancers in the load balancing system. If the mapping relationship between the quintuple of the video service and the target RS is recorded in the mapping table of all or part of the load balancer, it means that the video service is an existing service.

In the embodiment of this application, each time a new host 11 is added, RS112 on the newly added host 11 reports the VIP to the query device 12, and the query device 12 records the VIP and notification time point of the RS112 on the newly added host 11. Moreover, every time a new host 11 is added, the load balancer 111 function on the new host 11 is enabled, and the RS112 function on the newly added host 11 is not enabled for the time being. Instead, the new host 11 is enabled after a preset period of time. function on RS112. After the RS112 function on the new host 11 is enabled, the load balancer 111 will consider the RS112 when performing consistent hashing operations.

After receiving the data packet, the load balancer 111 determines the RS112 whose mounting time exceeds the preset time from multiple RS112s as the effective RS112. Afterwards, the target RS is determined from the effective RS112 and the data packet is sent. That is to say, every time a new host 11 is added, RS112 on the newly added host 11 does not immediately participate in the consistent hashing algorithm, but only participates in the consistent hashing algorithm after a preset period of time. Within the preset time period, initially, the mapping table of the load balancer 111 on the newly added host 11 is empty. After that, every time a data packet is received, if the five-tuple of the data packet does not exist in the mapping table, consistent hashing will be performed based on the effective RS112, the five-tuple of the data packet, etc. to determine the target RS and send the data packet. At the same time, the load balancer 111 on the newly added host 11 records the mapping relationship between the quintuple and the target RS in the mapping table.

In this way, after a preset period of time has elapsed for the new host 11, the mapping table of the load balancer 111 on the new host 11 will record the mapping relationship between the five-tuple packets of all existing services and the target RS. After the preset time, RS112 on the newly added host 11 takes effect. At this time, if the consistent hash algorithm is used to determine the target RS of the data packet of the existing service, an error is likely to occur. However, for data packets of existing services, the load balancer 111 first queries the mapping table, so it can determine the correct target RS. For the data packet of the new service, the load balancer 111 performs a consistent hash algorithm to determine the target RS based on the currently effective RS112, the five-tuple of the new service data packet, etc. and adds a new record to the mapping table.

For the load balancer 111 on the original host 11 in the load balancing system, if there is a mapping relationship between the five-tuple of the data packet of the existing service and the target RS in the mapping table of the load balancer 111, then through query mapping table can determine the correct target RS. If there is no mapping relationship between the five-tuple of the data packet of the existing service and the target RS in the mapping table, the target RS can also be determined and the data packet is sent by performing consistent hashing based on the effective RS112, the five-tuple of the data packet, etc. .

It should be understood that the numbers of the host 11, the query device 12, the switch 13 and the terminal device 14 in Figure 1 are only illustrative. In actual implementation, any number of hosts 11, query devices 12, switches 13 and terminal devices 14 can be deployed according to actual needs.

Figure 2 is a schematic diagram of the working level of the XDP-based load balancing architecture on the host provided by the embodiment of this application. Please refer to Figure 2. The working layers on the host include: network card driver, load balancer, kernel user process, etc. Among them, the load balancer is an XDP-based load balancer, and the user process is a related process in the user mode. When the RS on the host is an edge node in the Content Delivery Network (CDN), the user state has Squid cache and shark (scheduling) components.

Please refer to Figure 2. The host is enabled in driver mode. Data packet processing is performed after the network card receives the data packet and before the kernel intercepts it. This avoids the overhead caused by switching between user mode and kernel mode. Therefore, compared with LVS-based Load balancing architecture. The XDP-based load balancing architecture provided by the embodiments of this application can reduce performance overhead.

Next, based on the load balancing system shown in Figure 1, the load balancing method described in the embodiment of the present application will be described in detail. For example, please refer to Figure 3, which is a flow chart of a load balancing method provided by an embodiment of the present application. This embodiment is explained from the perspective of a load balancer. This embodiment includes:

301. The load balancer determines that there is a new host in the load balancing system.

Wherein, the load balancer is a load balancer on any one of the multiple hosts included in the load balancing system, and the load balancer and the real server RS are deployed on each of the multiple hosts. Since the load balancer and RS are deployed on each host, every time a new host is added, it means that one load balancer and one RS are added respectively.

In the embodiment of this application, two load balancing are included, which are hereinafter referred to as the first load balancing and the second load balancing respectively. The first load balancing refers to switching to select a target load balancer from the load balancers of multiple hosts. , and sends the packet to the target load balancer. The second load balancing means that the target load balancer selects the target RS from the RS of multiple hosts and sends the data packet to the target RS. The target load balancer and target RS may be on the same host or on different hosts.

In the embodiment of this application, the load balancer is a load balancer on each host in the load balancing system after a new host is added. That is to say, each load balancer in the load balancing system, including the load balancer on the original host in the load balancing system, and the load balancer on the newly added host, each load balancer detects the new load balancer in the load balancing system. After adding a host, the load balancing method described in the embodiment of this application is executed.

After the load balancer determines that a new host has been added to the load balancing system, once the load balancer receives a data packet from the switch, it indicates that the load balancer is the target load balancer.

302. The load balancer receives the data packet.

For example, the switch performs the first load balancing, selects the target load balancer from the load balancers of multiple hosts and sends the data packet to the target load balancer. Accordingly, the target load balancer receives the packet. The data packet may be a service request initiated by the user's terminal device. After receiving the data packet, the target RS responds to the service request according to the data packet.

From the perspective of the switch, like the traditional XDP-based load balancing architecture, the OSPF routing protocol is used to perform the first load balancing to determine the target load balancer, and it is very likely that the data packet will be sent to the load balancer on the newly added host. , the mapping table of the load balancer on the newly added host is empty. Of course, the switch is also likely to send the packet to the load balancer on the original host.

303. The load balancer determines the effective RS from the RS of each host included in the load balancing system. The effective RS is the RS involved when the load balancer performs load balancing on data packets of existing services before adding a host. The existing services are services that the load balancing system is providing to users before the new host is added.

In the embodiment of this application, each time the load balancer receives a data packet, it queries the local mapping table. If the five-tuple of the data packet does not exist in the mapping table, it performs a consistent hash operation to determine the target RS. In the traditional XDP-based load balancing architecture, when a new host is added, the load balancer performs consistent hashing based on the quintuple of the data packet, the VIPs of all RSs in the load balancing system, and the number of RSs, resulting in the determination of The target RS is incorrect.

In the embodiment of this application, every time a host is added, the RS on the host does not take effect immediately, that is, it does not immediately participate in load balancing. Instead, all existing RSs are saved in the mapping table of the load balancer on the newly added host. After determining the mapping relationship between the business quintuple and the RS, the RS on the newly added host will be used for load balancing.

Therefore, when the load balancer cannot determine the target RS by looking up the mapping table and performs consistent hashing, it determines the effective RS from all RSs. The effective RS may provide at least one RS with existing services, or may not provide any existing services. However, regardless of whether existing services are provided or not, the effective RS is the RS involved when the load balancer performs consistent hashing on the data packets of existing services before adding a host.

For example, there are two hosts in the load balancing system, namely host 1 and host 2. Load balancer 1a and RS1b are deployed on host 1, and load balancer 2a and RS2b are deployed on host 2. Add host 3, and deploy load balancer 3a and RS3b on host 3. A mapping table is stored on load balancer 1a, which stores the corresponding relationship between quintuple A and RS2b. The mapping tables on load balancer 2a and load balancer 3a are empty, that is, RS2b has provided services to the outside world, and RS1b No business was provided.

That is to say, before the new host is added, there is an existing service, and the quintuple of the data packet of this service is quintuple A. Before adding a new host, after the load balancer (load balancer 1a or load balancer 2a) receives the data packet whose quintuple is quintuple A for the first time, it will use the VIP, RS2b, and RS of quintuple A and RS1b. Consistent hashing is performed on the quantity, etc., and the target RS is determined to be RS2b. Obviously, before adding host 3, the RS involved when the load balancer performs consistent hashing on the data packets of existing services includes RS1b and RS2b. Therefore, the effective RS includes RS1b and RS2b.

304. The load balancer determines the target RS from the effective RS.

For example, the load balancer does not directly select the target RS from all RSs in the load balancing system, but first determines the effective RS from all the RSs, and then selects the target RS from the effective RSs.

For example, there are two hosts in the load balancing system, namely host 1 and host 2. Load balancer 1a and RS1b are deployed on host 1, and load balancer 2a and RS2b are deployed on host 2. Add host 3, and deploy load balancer 3a and RS3b on host 3. A mapping table is stored on load balancer 1a, which stores the corresponding relationship between quintuple A and RS2b. The mapping table on load balancer 2a stores the corresponding relationship between quintuple B and RS1b. The mapping table on load balancer 3a stores The mapping table is empty. That is to say, before the new host is added, there are two existing services, and the quintuples of the data packets of the two services are quintuple A and quintuple B respectively.

Assume that the switch performs load balancing for the first time and selects load balancer 1a. After receiving the data packet, load balancer 1a determines that the quintuple of the data packet is quintuple A. Therefore, the load balancer 1a determines that the target RS is RS2b according to the five-tuple query mapping table.

Assume that the switch performs load balancing for the first time and selects load balancer 2a. After receiving the data packet, load balancer 2a determines that the quintuple of the data packet is quintuple A. However, since the mapping table is empty, load balancer 2a determines the effective RS, which includes RS1b and RS2b. Afterwards, load balancer 2a executes a consistent hash algorithm to determine the target RS as RS2b.

Assume that the switch performs load balancing for the first time and selects load balancer 3a. After receiving the data packet, load balancer 3a determines that the quintuple of the data packet is quintuple A. However, since the mapping table is empty, load balancer 3a determines the effective RS, which includes RS1b and RS2b. Afterwards, load balancer 3a executes a consistent hash algorithm to determine the target RS to be RS2b.

For another example, there were originally 100 hosts in the load balancing system. When new hosts were added, a batch of hosts were added. This batch of hosts totaled 10, resulting in a total of 110 hosts in the load balancing system. The RS on the 100 hosts Numbered from 100 to 110. When a batch of hosts is added, the load balancer on each of the 110 hosts detects the new host. When the load balancer on any host receives a data packet, if the target RS cannot be determined by querying the mapping table, the load balancer determines the effective RS. Assume that the RS numbers 1-90 are before the new host is added. When the load balancer performs load balancing on data packets of existing services, the load balancer determines the target RS based on the quintuple of the data packet, the number of VIPs and RSs of each RS in the 90 RSs.

305. The load balancer sends the data packet to the target RS.

In the embodiment of this application, the load balancer that sends the data packet, that is, the target load balancer and the target RS are located on the same host or different hosts.

In the load balancing method provided by the embodiment of this application, when the load balancer on each host in the load balancing system detects a new host in the load balancing system, once a load balancer receives a data packet, if the load balancer does not exist in the mapping table For the five-tuple of the data packet, the effective RS for load balancing is determined from the RS of each host included in the load balancing system, the target RS is selected from the effective RS and the data packet is sent to the target RS, thereby completing the second step. Secondary load balancing. Using this solution, each time a new host is added, the load balancer determines the target RS from the effective RS and sends a data packet to ensure that the load balancer on the new host has sufficient time to generate mappings for existing services. Table, the mapping table records the mapping relationship between the quintuple of the existing business and the RS to prevent the load balancer from sending the data packets of the existing business to the incorrect RS. That is to say, every time a new host is added, the load balancer on the host takes effect but the RS does not immediately participate in load balancing. Instead, the mapping table of the load balancer on the newly added host saves all five existing services. After the mapping relationship between tuples and RSs is established, the RS on the newly added host will be used for load balancing to ensure that the data packets of existing services after the new host are added are sent to the correct RS, thereby improving service quality without any changes. switch to reduce the impact and operational complexity.

Optionally, in the above embodiment, in the process of determining the effective RS from the RS of each host included in the load balancing system, the load balancer determines the mounting time of the RS of each host, and sets the mounting time of the RS that exceeds the preset time. RS as the effective RS. The mounting time refers to the time between the notification time point and the reception time point. The notification time point is the time point when the host's RS notifies the VIP to the query device, and the reception time point is the time point when the load balancer receives the data packet. By comparing the mounting time and the preset time, the purpose of accurately determining the effective RS is achieved.

In one method, each load balancer records the time point when the host joins the load balancing system, and uses this time point as the notification time point. Each load balancer in the load balancing system can communicate with each other. Therefore, the load balancer can determine the time point when other hosts join the load balancing system, that is, the notification time point. The notification time point of the host is the notification time point of the RS on the host.

When the load balancer receives a data packet and needs to perform consistent hashing, it can determine the mounting time of the RS based on the notification time point of each host's RS and the reception time point of the received data packet. The RS whose mounting time exceeds the preset time will be regarded as the effective RS. The preset time duration is, for example, 10 minutes, 8 minutes, 20 minutes, etc. The embodiments of this application are not limiting. The preset time period can at least satisfy the requirement that the load balancer on the newly added host performs the second load balancing, so that the mapping table of the load balancer stores the mapping relationship between the 5-tuples of all existing services and the RS.

In another way, as shown in Figure 1, a query device is set up. When the load balancer detects a new host and needs to perform consistent hashing on the received data packet, it sends a query request to the query device to request the notification time point of the RS of each host included in the load balancing system. After receiving the query request, the query device sends a query response to the load balancer, where the query response carries the notification time point of the RS of each host included in the load balancing system. Afterwards, the load balancer determines the mounting duration of the RS of each host based on the receiving time point and the notification time point of the RS of each host.

For example, the query_VIP function is added to the load balancer to obtain the notification time point of the RS of each host included in the load balancing system by querying the device. In this way, each time the load balancer performs consistent hashing, it does not need to interact with other load balancers in the load balancing system to obtain the notification time point of the RS. Instead, it only needs to interact with the query device once to determine the time point. The load balancing system contains the notification time point of each host's RS, and then determines the mounting time of each host's RS.

Using this solution, by setting up a query device in the load balancing system and adding a VIP query function on the load balancer, the purpose of quickly determining the mounting time of each host's RS can be achieved.

Optionally, in the above embodiment, the load balancer can flexibly determine whether there is a new host in the load balancing system.

In one approach, the load balancer detects whether the number of hosts in the load balancing system has changed. When the number of hosts in the load balancing system changes, it is determined that there are new hosts in the load balancing system. When the number of hosts in the load balancing system does not change, the load balancer considers that there are no new hosts in the load balancing system.

For example, each load balancer detects the number of RSs in the load balancing system in real time. Once the number of RS changes, it means that the number of hosts changes. Therefore, the load balancer can quickly determine whether to add or reduce hosts in the load balancing system by detecting the number of RSs.

In another method, when the load balancer determines whether there is a new host in the load balancing system, it determines whether there is an RS in the RS of each host included in the load balancing system whose mounting time does not exceed the preset time. This is because when the mounting time of an RS does not exceed the preset time and the RS is not effective, it cannot participate in consistent hashing operations.

When the RS of each host included in the load balancing system contains an RS whose mounting time does not exceed the preset time, it is determined that there is a new host in the load balancing system, and the new host is the one whose mounting time does not exceed the preset time. RS. At this time, once the load balancer receives the data packet and needs to perform consistent hashing, it determines the effective RS from all RSs included in the load balancing system, and determines the target RS from the effective RS.

When the mounting time in the RS of each host included in the load balancing system exceeds the preset time, the load balancer considers that there are no new hosts in the load balancing system. At this time, once the load balancer receives the data packet and needs to perform consistent hashing, it directly determines the target RS from all RSs included in the load balancing system.

With this solution, it is considered that RS cannot be used for consistent hashing operations when it is not effective. Therefore, hosts with RSs that have been mounted for less than the preset time are regarded as new hosts to avoid subsequent consistent hashing operations. Errors are made to achieve the purpose of improving the service quality of existing services.

Figure 4 is another flow chart of the load balancing method provided by the embodiment of the present application. This embodiment is explained from the perspective of the query device. This embodiment includes:

401. The query device receives a notification request sent by the RS on the newly added host in the load balancing system, where the notification request is used to indicate the virtual IP address VIP of the RS.

402. Record the VIP and notification time point. The notification time point is the time point when the query device receives the notification request. The effective RS in the load balancing system is before the new host is added. The RS involved when the load balancer in the load balancing system performs load balancing on the data packets of existing services. The existing services are services that the load balancing system is providing to users before the new host is added.

For example, each time a new host is added, the RS on the new host sends a notification request to the query device to pass the VIP. After the query device receives the notification request, it records the VIP and notification time of the new host. It can be understood that the time point when a new host is added and the time point when the RS on the new host is initially mounted is the notification time point, that is, the time point when the query device receives the notification request.

Using this solution, by setting up a query device to record the VIP and notification time points of each newly added host, it is convenient for the load balancer to quickly obtain the mounting time of each RS from the query device, and then quickly determine it from all RSs in the load balancing system. Effective upon birth RS.

Below, an example is used to explain the above load balancing method in detail. For example, please refer to Figure 5 , which is a schematic diagram of an application scenario of the load balancing method provided by the embodiment of the present application.

Please refer to Figure 5. Before the new host is added, there are three hosts in the load balancing system, namely host 1, host 2 and host 3. Load balancer 1a and RS1b are deployed on host 1, and load balancer 2a and RS1b are deployed on host 2. RS2b, load balancer 3a and RS3b are deployed on host 3. At 18:00 sharp, host 4 is added, and load balancer 4a and RS4b are deployed on host 4. A mapping table is stored on load balancer 1a, which stores the corresponding relationship between quintuple A and RS2b. The mapping table on load balancer 2a stores the corresponding relationship between quintuple B and RS1b. Load balancer 3a and load The mapping table on equalizer 4a is empty.

The preset time period is preset on the query device 12 to 10 minutes, that is, starting from 18:00, the function of the load balancer 4a is enabled, but the function of RS4b is not enabled and does not participate in consistent hashing operations.

During the period from 18:00 to 18:10, switch 13 receives the data packet from terminal device 14 and performs the first load balancing. Among them, the quintuple of the data packet is quintuple A, that is, the data packet is a data packet of an existing service. If host 4 is not added, switch 13 sends the data packet to load balancer 1a. However, since load balancer 4a is turned on and switch 13 performs the first load balancing, it is very likely that the data packet that should be sent to load balancer 1a will be sent to load balancer 2a, load balancer 3a or load balancer 4a. .

No matter which load balancer it is sent to, the load balancer determines that the valid RS includes RS1b, RS2b and RS3b. Therefore, by performing a consistent hash operation, the target RS is determined to be RS2b, and then the data packet is sent to RS2b.

Similarly, when the quintuple of the data packet is quintuple B and the first load balancing should be sent to load balancer 2a, the switch is likely to send the data packet to load balancer 1a and load balancer 3a. Or load balancer 4a, the processing method is the same as above, and will not be described again here.

During the period from 18:00 to 18:10, load balancer 4a can establish a mapping table that stores the mapping relationship between the five-tuple and RS of all existing services.

After 18:10, after the switch receives the data packet of the existing service, there is no new load balancer for the switch (because the new load balancer was turned on at 18:00), therefore, the switch will not Splitting the packets wrongly. After the data packet arrives at the load balancer, the result of the consistent hash operation changes due to the presence of a new RS (the new RS is enabled after 18:10). However, since after 18:10, each load balancer, including the new load balancer at 18:00, has stored the mapping table of the 5-tuple and RS mapping relationships of all existing services, therefore, the mapping is queried first surface. If there is a five-tuple of the data packet in the mapping table, it means that the data packet is an existing service, and the RS corresponding to the five-tuple is directly used as the target RS. If the five-tuple of the data packet does not exist in the mapping table, it means that the data packet is a data packet of a new service, and the load balancer (load balancer 1a, load balancer 2a, load balancer 3a or load balancer 4a) determines The valid RS output includes RS1b, RS2b, RS3b and RS4b. A consistent hash operation is performed based on the five-tuple of the new service data packet and the VIP and quantity of the valid RS to determine the target RS. The target RS may be RS1b, RS2b, RS3b or RS4b. Afterwards, the load balancer adds the mapping relationship between the five-tuple of the new service's data packet and the target RS in the local mapping table.

The following are device embodiments of the present application, which can be used to execute method embodiments of the present application. For details not disclosed in the device embodiments of this application, please refer to the method embodiments of this application.

Figure 6 is a schematic diagram of a load balancing device provided by an embodiment of the present application. The load balancing device 600 includes: a determining module 61 , a receiving module 62 , a processing module 63 and a sending module 64 .

Determination module 61 is used to determine that there is a new host in the load balancing system. The load balancer is a load balancer on any one of the multiple hosts included in the load balancing system. Deploy load balancer and real server RS on each host;

Receiving module 62, used to receive data packets;

The processing module 63 is used to determine the effective RS from the RS of each host included in the load balancing system. The effective RS is involved when the load balancer performs load balancing on the data packets of existing services before adding a host. RS, the existing service is the service that the load balancing system is providing to the user before the new host is added, and the target RS is determined from the effective RS;

The sending module 64 is used to send the data packet to the target RS.

In a feasible implementation manner, when the processing module 63 determines the effective RS from the RS of each host included in the load balancing system, it is used to determine the mounting duration of the RS of each host, and the mounting duration is The RS that exceeds the preset duration is regarded as the effective RS. The mounting duration refers to the duration between the notification time point and the reception time point. The notification time point is the time point when the host's RS notifies the virtual IP address VIP to the query device. , the receiving time point is the time point when the load balancer receives the data packet.

In a feasible implementation, the sending module 64 is also used to send a query request to the query device to request to obtain the notification time point of the RS of each host included in the load balancing system;

The receiving module 62 is also configured to receive a query response from the query device, where the query response carries the notification time point of the RS of each host included in the load balancing system;

The processing module 63 is configured to determine the mounting duration of the RS of each host based on the receiving time point and the notification time point of the RS of each host.

In a feasible implementation, the determination module 61 is used to determine whether there is an RS among the RSs of each host included in the load balancing system whose mounting duration does not exceed the preset duration; when each RS included in the load balancing system When there is an RS in the host's RS whose mounting duration does not exceed the preset duration, it is determined that there is a new host in the load balancing system.

In a feasible implementation, the determination module 61 is used to detect whether the number of RSs in the load balancing system changes; when the number of RSs in the load balancing system changes, determine whether the load balancing system There are new hosts in .

In a feasible implementation manner, when the processing module 63 determines the effective RS from the RS of each host included in the load balancing system, it is used to query the mapping table according to the five-tuple of the data packet. The corresponding relationship between the five-tuple and the RS is stored in the table; when the five-tuple of the data packet does not exist in the mapping table, the effective RS is determined from the RS of each host included in the load balancing system.

In a feasible implementation manner, after the processing module 63 determines the target RS from the effective RS, it is also used to add a five-tuple of the data packet and the target RS to the mapping table. Mapping relations.

In a feasible implementation manner, when the processing module 63 determines the target RS from the effective RS, it is used to determine the target RS based on the five-tuple of the data packet, the VIP of each effective RS, and the value of the effective RS. A consistent hash is performed on the number to determine the target RS.

The load balancing device provided by the embodiment of the present application can perform the actions of the load balancer in the above embodiment. Its implementation principles and technical effects are similar and will not be described again here.

Figure 7 is a schematic diagram of another load balancing device provided by an embodiment of the present application. The load balancing device 700 includes: a transceiver module 71 and a processing module 72 .

The transceiver module 71 is configured to receive a notification request sent by the RS on the newly added host in the load balancing system, where the notification request is used to indicate the virtual IP address VIP of the RS;

The processing module 72 is configured to record the VIP and the notification time point. The notification time point is the time point when the query device receives the notification request. The VIP and the notification time point are used to determine the new notification time point. Whether the RS on the increased host is converted into an effective RS. The effective RS in the load balancing system is the RS involved when the load balancer in the load balancing system performs load balancing on the data packets of existing services. The existing RS The service is the service that the load balancing system is providing to the user before adding the host.

In a feasible implementation, the transceiver module 71 is also used to receive a query request from a load balancer. The load balancer receives the data packet from the load balancer of each host included in the load balancing system. A load balancer; sends a query response to the load balancer, where the query response carries the notification time point of the RS of each host included in the load balancing system.

The load balancing device provided by the embodiment of the present application can perform the action of querying the device in the above embodiment. Its implementation principles and technical effects are similar and will not be described again here.

FIG. 8 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 8, the electronic device 800 is, for example, the above-mentioned query device or host. The electronic device 800 includes:

Processor 81 and memory 82;

The memory 82 stores computer instructions;

The processor 81 executes the computer instructions stored in the memory 82, so that the processor 81 executes the load balancing method implemented by the load balancer in the host; or, causes the processor 81 to execute the load balancing method implemented by the query device as above. Equilibrium approach.

The specific implementation process of the processor 81 can be found in the above method embodiments. The implementation principles and technical effects are similar and will not be described again in this embodiment.

Optionally, the electronic device 800 further includes a communication component 83 . Among them, the processor 81, the memory 82 and the communication component 83 can be connected through the bus 84.

Embodiments of the present application also provide a computer-readable storage medium. Computer instructions are stored in the computer-readable storage medium. When the computer instructions are executed by a processor, they are used to implement the load balancing implemented by the load balancer in the host as above. method; or, implement the load balancing method implemented by the query device as above.

Embodiments of the present application also provide a computer program product. The computer program product includes a computer program. When the computer program is executed by a processor, it implements the load balancing method implemented by the load balancer in the host as above; or implements the load balancing method implemented by the query device as above. Equilibrium approach.

Other embodiments of the present application will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary technical means in the technical field that are not disclosed in this application. . It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

It is to be understood that the present application is not limited to the precise structures described above and illustrated in the accompanying drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (12)

1. A load balancing method, characterized in that it is applied to a load balancer on a host in a load balancing system, and the method includes: Determine that there is a new host in the load balancing system, the load balancer is a load balancer on any one of the multiple hosts included in the load balancing system, and the load balancer is deployed on each of the multiple hosts. Load balancer and real server RS; receive data packet; Query the mapping table according to the five-tuple of the data packet. When the five-tuple of the data packet does not exist in the mapping table, the effective RS is determined from the RS of each host included in the load balancing system, The mapping table stores the corresponding relationship between five-tuples and RSs. The effective RS is the RS involved when the load balancer performs load balancing on the data packets of existing services before adding a host. The existing services are The services the load balancing system is providing to users before adding new hosts; Determine a target RS from the effective RS; Send the data packet to the target RS. 2. The method according to claim 1, wherein determining the effective RS from the RS of each host included in the load balancing system includes: Determine the mounting duration of the RS of each host, and use the RS whose mounting duration exceeds the preset duration as the effective RS. The mounting duration refers to the duration between the notification time point and the reception time point, and the notification time point is The time point at which the RS of the host notifies the query device of the virtual IP address VIP, and the receiving time point is the time point at which the load balancer receives the data packet. 3. The method according to claim 2, wherein determining the mounting duration of the RS of each host includes: Send a query request to the query device to request to obtain the notification time point of the RS of each host included in the load balancing system; Receive a query response from the query device, where the query response carries the notification time point of the RS of each host included in the load balancing system; The mounting duration of the RS of each host is determined based on the receiving time point and the notification time point of the RS of each host. 4. The method according to any one of claims 1-3, characterized in that determining that a new host exists in the load balancing system includes: Determine whether among the RSs of each host included in the load balancing system, there is an RS whose mounting duration does not exceed the preset duration; When the RS of each host included in the load balancing system contains an RS whose mounting duration does not exceed the preset duration, it is determined that there is a new host in the load balancing system. 5. The method according to any one of claims 1-3, characterized in that determining that a new host exists in the load balancing system includes: Detect whether the number of RSs in the load balancing system changes; When the number of RSs in the load balancing system changes, it is determined that a new host exists in the load balancing system. 6. The method according to claim 1, characterized in that after determining the target RS from the effective RS, it further includes: Add a mapping relationship between the five-tuple of the data packet and the target RS in the mapping table. 7. The method according to any one of claims 1-3, characterized in that determining the target RS from the effective RS includes: Consistent hashing is performed according to the five-tuple of the data packet, the VIP of each of the effective RSs, and the number of the effective RSs to determine the target RS. 8. A load balancing method, characterized in that it is applied to a query device in a load balancing system, and the method includes: Receive a notification request sent by the RS on the newly added host in the load balancing system, where the notification request is used to indicate the virtual IP address VIP of the RS; Record the VIP and the notification time point. The notification time point is the time point when the query device receives the notification request. The VIP and the notification time point are used to determine whether the RS on the newly added host is Convert to effective RS. The effective RS in the load balancing system is the RS involved when the load balancer in the load balancing system load balances the data packets of existing services. The existing services are before the new host is added. The load balancing system is providing services to users. The load balancer queries the mapping table according to the five-tuple of the received data packet. The mapping table stores the corresponding relationship between the five-tuple and the RS; when the mapping When the five-tuple of the data packet does not exist in the table, the effective RS is determined from the RS of each host included in the load balancing system. 9. The method of claim 8, further comprising: Receive a query request from a load balancer, which is the load balancer that receives the data packet among the load balancers of each host included in the load balancing system; Send a query response to the load balancer, where the query response carries the notification time point of the RS of each host included in the load balancing system. 10. A load balancing system, characterized in that it includes: a switch, a plurality of hosts and a query device, the switch establishes a network connection with each of the plurality of hosts, and the query device establishes a network connection with the plurality of hosts. Each host in establishes a network connection, where: The switch is configured to perform load balancing to determine a target load balancer from the load balancers included in the plurality of hosts and send data packets; A load balancer and a real server RS are deployed on each of the plurality of hosts, and the load balancer is used to perform the method according to any one of claims 1-7; The query device is used to perform the method as claimed in claim 8 or 9. 11. An electronic device, comprising a processor, a memory, and a computer program stored on the memory and executable on the processor, characterized in that when the processor executes the computer program, the electronic device The device implements the method according to any one of claims 1 to 9. 12. A computer-readable storage medium with a computer program stored thereon, characterized in that when the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.