Design method of edge distributed cluster system with survivability and decentration characteristic

Abstract

The invention discloses a design method of an edge distributed cluster system with destroy-resistant capability and decentration characteristic. The design method comprises the following steps: building a large edge-small edge-end three-layer organization distributed architecture of the edge distributed cluster system; designing a life cycle management mechanism of the edge distributed cluster system to realize cluster life cycle management; designing a cluster membership judgment service mechanism of the edge distributed cluster system so as to judge an existing cluster node state and a service state of the existing cluster node state; a data synchronization mechanism of the edge distributed cluster system is designed, and data synchronization between cluster nodes is achieved; designing a cluster destroy-resistant mechanism of the edge distributed cluster system; a distributed data storage mechanism of an edge distributed cluster system is designed, and the structural information of the whole cluster is stored in all cluster unmanned devices in the form of multiple copies. According to the method, real bee colony collaboration is realized, and the task execution efficiency of the unmanned cluster is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of cluster management, and more specifically, relates to a design method for an edge distributed cluster system having both anti-destruction capability and decentralization characteristics. Background Art

[0002] With the continuous development of science and technology, small unmanned devices have made rapid progress in payload miniaturization, battery technology, endurance, real-time communication, low cost, etc. Intelligent technologies such as cluster technology and collaborative technology have significantly improved the mission execution capabilities of small unmanned devices, allowing small unmanned devices to play a key role in daily life, and therefore have received widespread attention in the field of intelligence.

[0003] Cluster collaboration is one of the most effective means to improve the performance of unmanned device mission execution. In recent years, many studies have been conducted at home and abroad on the distributed unmanned device cluster collaboration. The decentralized swarm collaboration concept has been proven to have many advantages, which can give full play to the collective intelligence efficiency of unmanned clusters and achieve autonomous decision-making, autonomous organization, and autonomous cooperation. At the same time, it realizes decentralization, the cluster has no dominant node, and the unmanned devices perform tasks through mutual collaboration. The loss of any individual task execution capability will not affect the task completion efficiency of the entire cluster.

[0004] However, the existing unmanned device clusters have not yet achieved true swarm collaboration. The pain points and difficulties in achieving swarm collaboration lie in cluster decentralization and cluster anti-destruction. Some studies still choose a centralized distributed architecture, which is simple to implement, but if the center changes state due to its own or external reasons, the entire cluster will lose its task execution capability and lack cluster anti-destruction capability; some studies have achieved simple decentralization and anti-destruction capability, but the entire cluster is too flat, the collaborative intention cannot be efficiently converted, and the efficiency of cluster collaborative execution tasks is greatly reduced. Summary of the invention

[0005] The purpose of the present invention is to provide a method for designing an edge distributed cluster system with both anti-destruction capability and decentralization characteristics, which solves the two major pain points of unmanned clusters being difficult to decentralize and difficult to resist destruction, realizes true swarm collaboration, and effectively improves the task execution efficiency of unmanned clusters.

[0006] The technical solution to achieve the purpose of the present invention is:

[0007] A method for designing an edge distributed cluster system with both anti-destruction capability and decentralization characteristics, comprising:

[0008] Build a three-layer organizational distributed architecture of big edge-small edge-end for edge distributed cluster system;

[0009] Design the lifecycle management mechanism of edge distributed cluster systems to implement cluster lifecycle management;

[0010] Design a cluster membership determination service mechanism for edge distributed cluster systems to determine the status of existing cluster nodes and their service status;

[0011] Design a data synchronization mechanism for edge distributed cluster systems to achieve data synchronization between cluster nodes;

[0012] Design a cluster anti-destruction mechanism for edge distributed cluster systems. The cluster anti-destruction mechanism includes a fault detection mechanism and a cluster structure autonomous adjustment mechanism after a node failure. The fault detection mechanism is used to detect the status of each node in the cluster and update the list of active nodes. The cluster structure autonomous adjustment mechanism is adjusted according to the changes in the status of cluster nodes.

[0013] Design a distributed data storage mechanism for edge distributed cluster systems, where the structural information of the entire cluster is stored in multiple copies in all cluster unmanned devices.

[0014] Furthermore, the three-layer organizational distributed architecture adopts a tree structure, in which the large edge is the leader of the entire cluster, responsible for receiving all tasks of the cluster and assigning tasks to small edges in the cluster; the small edge plays a connecting role, receiving instructions from the large edge to obtain the assigned task information, and needs to manage the end nodes in their respective groups downward, further decompose the tasks obtained from the large edge, and then assign them to the end nodes; excluding the large edge and the small edge, the remaining nodes in the tree structure are all ends, and the ends are responsible for executing the tasks after the large edge and the small edge are distributed and processed.

[0015] Furthermore, the lifecycle management mechanism performs cluster lifecycle management through the cluster big edge interface, group interface and node interface. The cluster big edge interface creates, deletes and queries the entire cluster structure through the big edge node name and node ID; the group interface is used to adjust the group structure, including the creation, query and deletion of a certain group structure; the node interface is used for each surviving node in the cluster to perform operations such as node list query, node status query, node deletion and node exit from the current group, and can also create nodes that are not in the cluster structure and add nodes to specified groups.

[0016] Furthermore, the cluster membership determination service mechanism operates in two service modes, request / response and publish / subscribe, during cluster startup and cluster operation.

[0017] Furthermore, the data synchronization mechanism adopts polling mechanism communication and Gossip protocol communication. When the number of cluster nodes is lower than the set value or the network signal is higher than the set value, the polling mechanism is used to complete the data synchronization between nodes; otherwise, the Gossip protocol is used for information synchronization, and the Gossip protocol uses a random method to propagate information to the entire network.

[0018] Furthermore, the Gossip protocol uses direct mail, anti-entropy propagation and rumor propagation to achieve data synchronization and update. For scenarios where the number of nodes is lower than the set value, direct mail is used for data synchronization; for scenarios where full data synchronization is required, such as new nodes joining the cluster or dead nodes rejoining the cluster, anti-entropy propagation is used; for incremental data synchronization between nodes, rumor propagation is used for event propagation.

[0019] Furthermore, the Gossip protocol uses a Push&Pull method for data interaction.

[0020] Furthermore, the fault detection mechanism adopts a point-to-point random detection mechanism to separate the fault detector and the membership update propagation component of the problem node, specifically including:

[0021] During the entire detection cycle T, node A randomly selects a node B from the cluster node list, sends a detection message to it, and then waits for node B to return a response message; if a response message from node B is received within the preset response time, it means that node B is alive; if no response message is received within the preset response time, node A starts the indirect detection process;

[0022] Within the detection period T, if node A does not receive any response from node B directly or through the indirect detection process, the status of node B in the member list is marked as dead, and the node B status information is sent to the remaining nodes through the propagation component.

[0023] Furthermore, the indirect detection process refers to node A randomly selecting K active member nodes and sending request response messages to them. After receiving the request response messages, the K active nodes will send a detection message to node B; if node B sends a response message to the K active nodes within the response time, the response message will be forwarded to node A.

[0024] Furthermore, the cluster structure autonomous adjustment mechanism is distinguished according to the different roles of the failed nodes, namely, end node failure, small edge node failure and large edge node failure. When an end node failure occurs, the structure of the large edge node and the small edge node will not change, and only the state of the failed node will change; when a small edge node failure occurs, the structure of the large edge node will not change, but the structure of the group where the small edge node is located will change, and the group will reselect one of the surviving end nodes under the small edge node as the new small edge node; when a large edge node failure occurs, both the large edge node structure and the small edge node structure will change, and first the surviving small edge node under the large edge node will be selected as the new large edge node, and then the end node will be selected from the original group of the small edge node that has become the large edge node to become the new small edge node.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: starting from the perspective of multi-agent cluster collaborative control, the present invention comprehensively considers the two major pain points of clusters being difficult to decentralize and difficult to resist destruction, and proposes an edge cluster organizational architecture consisting of large edge-small edge-end. The cluster member nodes are effectively managed through a cluster life management mechanism and a cluster membership determination service, and the cluster data synchronization mechanism is used to ensure the effectiveness of the services of each node in the cluster. A cluster resistance mechanism is designed to ensure the reliability and availability of the cluster, and to enhance the cluster's survivability in the face of failures, thereby ensuring the success rate of the tasks performed by the final distributed cluster. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the organizational structure of the large edge-small edge-end edge cluster of the present invention;

[0027] Figure 2 It is a schematic diagram of the point-to-point fault detection mechanism of the present invention;

[0028] Figure 3 This is the cluster organization adjustment status after a cluster node failure in the present invention. DETAILED DESCRIPTION

[0029] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0030] The present invention proposes a method for designing an edge distributed cluster system with both anti-destruction capability and decentralization characteristics, which specifically includes:

[0031] 1. Three-tier organizational structure design of big edge-small edge-end

[0032] The edge cluster management system adopts a distributed architecture, consisting of multiple nodes with the same hardware and software conditions and the same service capabilities. In order to improve the efficiency of task execution and task completion, each node in the cluster is given a different role. The big edge is the leader of the entire cluster architecture. There is only one big edge in a cluster. The big edge manages all the small edges in the cluster and splits the cluster tasks and distributes them to all small edges. There can be multiple small edges, which play a connecting role in the cluster. Each small edge needs to manage several ends and obtain task information from the big edge, and further disassemble and distribute it to the subordinate ends. The end is the foundation of the entire architecture and the actual subject of task execution.

[0033] like Figure 1 As shown, the cluster structure adopts a tree structure, which consists of three layers of organizational structure, namely large edge, small edge and end. Figure 1 In the example, node 1 is positioned as a large edge. Only one large edge is needed in the entire cluster, which is the leader of the entire cluster. It is responsible for receiving all tasks in the cluster and reasonably allocating tasks to the small edges in the cluster. Figure 1 In the diagram, nodes 2, 7, and 11 are positioned as small edges. The entire cluster can determine the number of small edges based on the needs of real-world scenarios. In a cluster, small edges receive instructions from large edges to obtain assigned task information, and need to manage the end nodes in their respective groups, further decompose the tasks obtained from large edges, and then reasonably assign them to the end nodes. The remaining nodes in the cluster structure diagram are positioned as ends, which are the most basic role positioning in the cluster structure and are responsible for the specific execution of tasks after large and small edge distribution processing.

[0034] 2. Cluster lifecycle management design

[0035] The interface is designed according to the cluster lifecycle management requirements in real scenarios, and is encapsulated as a cluster lifecycle management service, which includes functions such as cluster creation and deletion, node joining and exiting in the cluster, etc. Its application scenarios mainly include the adjustment of cluster structure during system initialization and operation.

[0036] During the initialization and operation of the cluster, the state of the cluster nodes will change due to active request instructions and the influence of the operating environment. The following interfaces are designed to implement cluster lifecycle management. The lifecycle management interface can be divided into cluster big edge interface, group interface and node interface. The cluster big edge interface includes the creation, deletion and query of the entire cluster structure through the big edge node name and node id. The group interface acts on the adjustment of the group structure, including the creation, query and deletion of a certain group structure. The node interface can act not only on each surviving node in the cluster, but also on the nodes that were not originally in the cluster structure. The interface functions that act on the surviving nodes in the cluster include node list query, node status query, node deletion and node exit from the current group. The interface functions that act on the nodes that were not originally in the cluster structure include node creation and node joining the specified group.

[0037] 3. Cluster membership determination service mechanism design

[0038] A cluster is a collection of networked nodes. Membership determination in a cluster refers to checking the status of existing cluster nodes and their service status, which is the key to cluster structure management and its anti-destruction capabilities. The cluster membership service can be run during cluster startup and cluster operation, and can be set to request / response and publish / subscribe service modes.

[0039] When the cluster is started, the corresponding cluster structure is determined on the big edge node according to the task requirements. At this time, the big edge node will create a member list according to the node status in the existing cluster structure, and then synchronize the cluster structure built by this node to the remaining nodes in the cluster through cluster data synchronization. After the cluster nodes complete the cluster structure data synchronization, each node will store a copy of the member list. This list copy will rejoin the cluster after the node loses connection to provide structural change information, so that each node can know the latest node membership information. During the operation of the cluster, node joining and node leaving will cause changes in the cluster structure, which will also cause changes in the membership in the cluster. Changes in cluster membership will occur in the following two situations: one is that the cluster administrator actively initiates node joining and exiting commands, and the other is the node stop, exit or crash event under the influence of external factors.

[0040] 4. Cluster data synchronization mechanism design

[0041] In a distributed cluster system, data synchronization among nodes in the cluster is the basis of cluster structure management. First of all, data synchronization needs to ensure that data is replicated and stored on multiple nodes, increase data redundancy, and ensure the continuity and stability of cluster node services. In addition, each node in a distributed cluster must maintain data consistency. When any node in the cluster updates data, other nodes will also be notified and update their own data in a timely manner. Therefore, the cluster needs to have a data synchronization mechanism with data consistency and high availability to maintain the existing structure of the cluster and cluster changes after node damage. This method provides two synchronization methods, namely polling mechanism communication and Gossip protocol communication.

[0042] When the number of cluster nodes is small or the network status is good, this method uses polling mechanism communication to complete data synchronization between nodes. When the cluster is initialized, the big edge is responsible for establishing the cluster structure and storing a copy of the member list. Finally, it polls the node members in the member list in order to synchronize data, so that each node in the cluster knows the existing cluster structure. Polling communication is a simple implementation mechanism and provides direct control over the process. However, due to the characteristic of polling to send requests periodically, it will generate unnecessary network traffic and delays, which will cause additional burden on network resources.

[0043] Since the polling mechanism imposes an additional burden on network resources, this method uses the Gossip protocol for information synchronization in environments with a large number of cluster nodes or weak network. The Gossip protocol uses a random method to propagate information to the entire network. Any node can propagate messages to the entire network when the network is connected, and the cluster node load does not increase much as the number of cluster nodes increases. The Gossip protocol mainly uses three mechanisms: direct mail, anti-entropy, and rumor mongering to achieve data updates. Different event propagations can be selected for different applicable scenarios.

[0044] For small-scale scenarios, this method uses direct mailing to synchronize data. When any node has data, it starts to traverse the node member list and sends messages to other nodes to inform the node of the data update status.

[0045] For scenarios where new nodes or reconnected nodes in the cluster need full data synchronization, this method uses anti-entropy propagation. Anti-entropy propagation propagates all data at a fixed frequency, which can avoid the problem of inconsistent data among cluster nodes due to packet loss in UDP transmission or the addition of new nodes. Anti-entropy propagation is applied to two scenarios, namely periodic data synchronization and data synchronization of newly added nodes. Periodic data synchronization refers to any node in the cluster randomly selecting any node according to the set anti-entropy period and cluster node list, and eliminating the differences between the two by exchanging data with each other to achieve final consistency of data. Data synchronization of newly added nodes means that the cluster leader will always pay attention to the number and status of nodes in the cluster. When a new node appears in the cluster or the node status changes from dead to alive, the newly added node or the node with status change will randomly select a node in the member list for data synchronization.

[0046] Some data between cluster nodes do not need to be updated by global synchronization. Incremental synchronization is more effective. For incremental synchronization of data between nodes, this method uses rumor propagation to propagate events. When a node generates new data, the node becomes active and periodically sends new data to the rest of the cluster until all nodes have completed the update of the new data. The new data to be sent by the active node is named rumor message, and each rumor data contains a timestamp. The active node will only send rumors with the latest timestamp, and the rumor message will be marked as invalid after the set time.

[0047] In the process of global synchronization and incremental synchronization of cluster nodes, the data interaction mode between nodes is involved. Under the Gossip protocol, the interaction modes of cluster nodes can be divided into three types: Push, Pull and Push&Pull. The Push mode is that the node with new information is used as the initiating node, and any node in the cluster is randomly contacted. After the contacted node receives the information, it is updated to the latest data. The Pull mode is that the node without new information is used as the initiating node, and a node is randomly selected from the cluster and information is obtained from the node. The Pull&Push mode refers to the node that initiates information exchange randomly selecting a node in the cluster to send information, and at the same time obtaining data from the other party to update local data. Compared with the other two interaction modes, the Push&Pull mode has the best data synchronization effect and the fastest convergence speed. Therefore, the present invention adopts the Push&Pull mode for data interaction.

[0048] 5. Cluster anti-destruction mechanism design

[0049] Distributed clusters need to have a cluster anti-destruction mechanism to ensure the reliability and availability of the cluster, and improve the survivability of the cluster in the face of threats and attacks, so as to ensure the success rate of the tasks performed by the distributed cluster. This method uses a tree structure to simulate the management hierarchy, adopts the edge cluster organizational architecture of large edge-small edge-end, and sets the attributes of each tree node in the tree structure to different roles. In the cluster anti-destruction process, the core lies in the cluster node fault detection mechanism and the autonomous adjustment of the cluster structure after the cluster node is damaged. The fault detection mechanism is used to check the status of each node in the cluster and update the list of active nodes. The autonomous adjustment of the cluster structure is to make corresponding adjustments according to the changes in the status of the cluster nodes.

[0050] The fault detection mechanism of the present invention adopts a point-to-point random detection mechanism and separates the fault detector from the membership update propagation component of the problem node, which can solve the problem that the traditional heartbeat detection is not scalable and the message load grows exponentially with the number of cluster nodes. The fault detector avoids using heartbeat detection and instead uses random point detection of the process to achieve scalability.

[0051] The specific fault detection process is as follows: Figure 2 As shown. During the entire detection cycle T, node A randomly selects a node B from the cluster node list, sends a detection message to it, and then waits for node B to return a response message. If a response message from node B is received within the preset response time, it means that node B is alive. If no response message is received within the preset response time, node A needs to start the indirect detection process.

[0052] The indirect detection process means that node A will randomly select K active member nodes and send request response messages to them. After receiving the request response messages, the K active nodes will send detection messages to node B. If node B sends a response message to the K active nodes within the response time, the response message needs to be forwarded to node A. If node A does not receive any response directly from node B or indirectly from the remaining selected K active nodes within the detection cycle T, the status of node B in the member list will be marked as dead, and the node B status information will be sent to the remaining nodes through the propagation component.

[0053] The cluster organization adjustment after the cluster node failure can be distinguished according to the different roles of the failed node, namely end node failure, small edge node failure and large edge node failure. The specific adjustment process is as follows: Figure 3As shown. When an end node fails, the structure of the large edge node and the small edge node will not change, and only the state of the failed node will change in the end. When a small edge node fails, the structure of the large edge node will not change, but the structure of the group where the small edge node is located will change. The group will reselect one of the surviving end nodes under the small edge node as the new small edge node. When a large edge node fails, the structure of both the large edge node and the small edge node will change. First, select the surviving small edge node under the large edge node as the new large edge node, and then select the end node in the original group of the small edge node that became the large edge node to become the new small edge node.

[0054] 6. Design of distributed data storage mechanism

[0055] The distributed data storage mechanism is used to achieve cluster anti-destruction and anti-interference at the data level. The structural information of the entire cluster is not stored separately in the big edge, but is stored in the form of multiple copies in all cluster unmanned devices.

[0056] Starting from the perspective of multi-agent cluster collaborative control, the present invention comprehensively considers the two major pain points of clusters being difficult to achieve decentralization and difficult to have anti-destruction capabilities, and proposes an edge cluster organizational architecture consisting of large edge-small edge-end. The cluster member nodes are effectively managed through a cluster life management mechanism and a cluster membership determination service, and the cluster data synchronization mechanism is used to ensure the effectiveness of the services of each node in the cluster. A cluster anti-destruction mechanism is designed to ensure the reliability and availability of the cluster, and to enhance the cluster's survivability in the face of failures, thereby ensuring the success rate of the tasks executed by the final distributed cluster.

[0057] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope.

[0058] In summary, the content of this specification should not be understood as a limitation on the present invention, and any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A method for designing an edge distributed cluster system with both anti-destruction capability and decentralization characteristics, characterized in that: include: Build a three-layer organizational distributed architecture of big edge-small edge-end for edge distributed cluster system; Design the lifecycle management mechanism of edge distributed cluster systems to implement cluster lifecycle management; Design a cluster membership determination service mechanism for edge distributed cluster systems to determine the status of existing cluster nodes and their service status; Design a data synchronization mechanism for edge distributed cluster systems to achieve data synchronization between cluster nodes; Design a cluster anti-destruction mechanism for edge distributed cluster systems. The cluster anti-destruction mechanism includes a fault detection mechanism and a cluster structure autonomous adjustment mechanism after a node failure. The fault detection mechanism is used to detect the status of each node in the cluster and update the list of active nodes. The cluster structure autonomous adjustment mechanism is adjusted according to the changes in the status of cluster nodes. Design a distributed data storage mechanism for edge distributed cluster systems, where the structural information of the entire cluster is stored in multiple copies in all cluster unmanned devices.

2. The method for designing an edge distributed cluster system with both invulnerability and decentralization characteristics according to claim 1, characterized in that: The three-layer organizational distributed architecture adopts a tree structure, in which the big edge is the leader of the entire cluster, responsible for receiving all tasks of the cluster and assigning tasks to the small edges in the cluster; the small edge plays a connecting role, receiving instructions from the big edge to obtain the assigned task information, and needs to manage the end nodes in their respective groups, further decompose the tasks obtained from the big edge, and then assign them to the end nodes; except for the big edge and the small edge, the remaining nodes in the tree structure are all ends, and the ends are responsible for executing the tasks after the big edge and the small edge are distributed and processed.

3. The method for designing an edge distributed cluster system with both invulnerability and decentralization according to claim 1, characterized in that: The lifecycle management mechanism performs cluster lifecycle management through the cluster big edge interface, group interface and node interface. The cluster big edge interface creates, deletes and queries the entire cluster structure through the big edge node name and node id; the group interface is used to adjust the group structure, including the creation, query and deletion of a certain group structure; the node interface is used for each surviving node in the cluster to perform node list query, node status query, node deletion and node exit from the current group, and at the same time, create nodes for nodes that are not originally in the cluster structure and add nodes to the specified group.

4. The method for designing an edge distributed cluster system with both invulnerability and decentralization according to claim 1, characterized in that: The cluster membership determination service mechanism operates in two service modes, request / response and publish / subscribe, during cluster startup and cluster operation.

5. The method for designing an edge distributed cluster system with both invulnerability and decentralization characteristics according to claim 1, characterized in that: The data synchronization mechanism adopts polling mechanism communication and Gossip protocol communication. When the number of cluster nodes is lower than the set value or the network signal is higher than the set value, the polling mechanism is used to complete the data synchronization between nodes; otherwise, the Gossip protocol is used for information synchronization. The Gossip protocol uses a random method to propagate information to the entire network.

6. The method for designing an edge distributed cluster system with both invulnerability and decentralization characteristics according to claim 5, characterized in that: The Gossip protocol uses direct mail, anti-entropy propagation and rumor propagation to achieve data synchronization and update. For scenarios where the number of nodes is lower than the set value, direct mail is used for data synchronization; for scenarios where new nodes are added to the cluster or dead nodes rejoin the cluster and full data synchronization is required, anti-entropy propagation is used; for incremental data synchronization between nodes, rumor propagation is used for event propagation.

7. The method for designing an edge distributed cluster system with both invulnerability and decentralization according to claim 6, characterized in that: The Gossip protocol uses the Push & Pull method for data interaction.

8. The method for designing an edge distributed cluster system with both invulnerability and decentralization according to claim 1, characterized in that: The fault detection mechanism adopts a point-to-point random detection mechanism to separate the fault detector and the membership update propagation components of the problem node, specifically including: During the entire detection cycle T, node A randomly selects a node B from the cluster node list, sends a detection message to it, and then waits for node B to return a response message; if a response message from node B is received within the preset response time, it means that node B is alive; if no response message is received within the preset response time, node A starts the indirect detection process; Within the detection period T, if node A does not receive any response from node B directly or through the indirect detection process, the status of node B in the member list is marked as dead, and the node B status information is sent to the remaining nodes through the propagation component.

9. The method for designing an edge distributed cluster system with both invulnerability and decentralization characteristics according to claim 8, characterized in that: The indirect detection process refers to that node A randomly selects K active member nodes and sends a request response message to them. After receiving the request response message, the K active nodes will send a detection message to node B; If node B sends a response message to K active nodes within the response time, the response message is forwarded to node A.

10. The method for designing an edge distributed cluster system with both invulnerability and decentralization characteristics according to claim 1, characterized in that: The cluster structure autonomous adjustment mechanism is distinguished according to the different roles of the failed nodes, namely, end node failure, small edge node failure and large edge node failure. When an end node failure occurs, the structure of the large edge node and the small edge node will not change, and only the state of the failed node will change; when a small edge node failure occurs, the structure of the large edge node will not change, but the structure of the group where the small edge node is located will change, and the group will reselect one of the surviving end nodes under the small edge node as the new small edge node; when a large edge node failure occurs, both the large edge node structure and the small edge node structure will change, and firstly, the surviving small edge node under the large edge node will be selected as the new large edge node, and then the end node will be selected from the original group of the small edge node that has become the large edge node to become the new small edge node.

Images