Distributed dynamic migration service system based on eBPF

Abstract

The invention relates to a distributed dynamic migration service system based on eBPF, which belongs to the technical field of distributed computing and operating system networks, and comprises a state sensing layer, a user layer and an intelligent decision-making layer which are in information connection in sequence, the state sensing layer is used for acquiring network indexes in real time in three dimensions in the Linux kernel layer by mounting three eBPF programs and outputting the network indexes, and the network indexes comprise global flow, delay time and retransmission rate; the user layer is used for receiving and preprocessing the global traffic, delay time and retransmission rate of the state sensing layer, and then outputting to the intelligent decision-making layer; and the intelligent decision-making layer is used for carrying out network quality comprehensive evaluation based on the preprocessed global flow, delay time and retransmission rate, generating a quality score and a quality grade, and adjusting operation parameters of a distributed consensus algorithm through an adaptive rule according to the quality score and the quality grade so as to influence and decide the switching condition of the main and standby nodes. The method has the effect of improving the switching efficiency and accuracy.

Description

Technical Field

[0001] This application relates to the fields of distributed computing and operating system network technology, and in particular to a distributed dynamic migration service system based on eBPF. Background Technology

[0002] In the era of microservices and cloud-native computing, current application networking demands increasingly higher network reliability and system stability. This has led to the development of full-duplex failover technology to enable primary / backup switching during network failures. However, traditional distributed failover systems operate at the application layer using heartbeat mechanisms, resulting in long latency (typically on the order of seconds) and long paths requiring the complete kernel protocol stack and context switching. When brief but severe network fluctuations occur on the order of microseconds or milliseconds, the application layer may be completely unaware of these fluctuations or experience significant latency, leading to excessively long fault detection times. Furthermore, primary node election strategies are often based on static configurations, such as fixed node IDs and priorities. This can result in the selection of a node with high latency as the primary node, leading to frequent failovers and system instability, ultimately impacting failover efficiency and accuracy. Summary of the Invention

[0003] To improve switching efficiency and accuracy, this application provides a distributed dynamic migration service system based on eBPF.

[0004] A distributed dynamic migration service system based on eBPF is provided.

[0005] A distributed dynamic migration service system based on eBPF includes a state awareness layer, a user layer, and an intelligent decision-making layer that are sequentially connected by information. The state-aware layer is used to collect and output network metrics in real time in three dimensions in the Linux kernel layer by mounting three eBPF programs. The network metrics include global traffic, latency, and retransmission rate. The user layer receives global traffic, latency, and retransmission rate from the state awareness layer, preprocesses them, and then outputs them to the intelligent decision-making layer. The intelligent decision layer is used to comprehensively evaluate network quality by taking the preprocessed global traffic, latency, and retransmission rate, generating a quality score and quality level. Then, based on the quality score and quality level, the operating parameters of the distributed consensus algorithm are adjusted through adaptive rules, thereby affecting the decision-making conditions for the switching of master and backup nodes.

[0006] Optionally, the state-aware layer collects network metrics in real time across three dimensions within the Linux kernel layer by mounting three eBPF programs, including; By mounting the XDP program to the specified network interface eth, packets are captured at the earliest point in the network driver layer, and atomic operations are used to update the global packet counter, thereby obtaining global traffic. By mounting the TC program to the specified network interface in the kernel layer in the outgoing and incoming directions, in-depth analysis of TCP connections and accurate measurement of RTT are achieved, thereby obtaining the latency time. The eBPF program is mounted into the kernel function using kprobe. When the kernel retransmits a data packet, the retransmission count of the corresponding connection is incremented, and the total transmission count of the corresponding connection is also incremented. The number of retransmitted data packets and the total number of transmitted data packets are obtained, and the retransmission rate is determined based on the number of retransmitted data packets and the total number of transmitted data packets.

[0007] Optionally, by mounting the XDP program to the specified network interface eth, only the Ethernet header and IP header are parsed.

[0008] Optionally, the user layer preprocesses the global traffic, including: Periodically read the data packet counter to obtain the number of packets and the period time in the current period; Calculate the current rate per packet based on the number of packets in the current cycle and the cycle time; Obtain the historical packet rate for a given time window and calculate the average packet rate; The flow deviation coefficient is determined based on the current per-packet rate and the average packet rate.

[0009] Optionally, the user layer preprocesses the delay time, including: Calculate the average value of the delay time collected within the preset time period.

[0010] Optionally, the intelligent decision-making layer performs a comprehensive network quality assessment based on preprocessed global traffic, latency, and retransmission rate, generating a quality score and quality level, including: The global flow score is determined based on the flow deviation coefficient and the first preset lookup table. The delay time score is determined based on the average delay time and a second preset comparison table; The retransmission rate score is determined based on the retransmission rate and the third preset comparison table. The quality score is determined based on the global traffic score, latency score, retransmission rate score, preset global traffic weight value, preset latency weight value, and retransmission rate weight value. The quality level is determined based on the quality score and a preset level comparison table, which stores the correspondence between the quality score and the quality level.

[0011] Optionally, determining the quality level based on the quality score and a preset level comparison table includes: A quality score greater than 90 and less than or equal to 100 is considered excellent. A quality score greater than 75 and less than or equal to 90 is considered good. If the quality score is greater than 60 points and less than or equal to 75 points, the quality level is average. If the quality score is greater than 40 and less than or equal to 60, the quality level is poor. If the quality score is greater than 0 and less than or equal to 40, the quality level is severe.

[0012] Optionally, the operating parameters include election timeout time and heartbeat interval, and the step of adjusting the operating parameters of the distributed consensus algorithm according to quality score and quality level through adaptive rules includes: When the quality level is excellent or good, reduce the election timeout and shorten the heartbeat interval; When the quality level is average, increase the election timeout period; When the quality level is poor or severe, increase the election timeout and decrease the heartbeat interval.

[0013] Optionally, the operating parameters are set with parameter thresholds, the parameter threshold for the election timeout is 100ms-1000ms, and the parameter threshold for the heartbeat interval is less than half of the election timeout.

[0014] Optionally, the conditions affecting the switching of primary and backup nodes in decision-making include: An election is triggered when the quality level reaches the "severe" level and continues to deteriorate over a period of time. An election is triggered when the retransmission rate or latency is abnormally high and continues to deteriorate. An election is triggered when global traffic affects the abnormal operation mode of the consensus algorithm.

[0015] In summary, this application includes at least one of the following beneficial technical effects: By enabling self-awareness and self-decision-making in multi-node scenarios, and by using eBPF technology to achieve network awareness of distributed systems, we can optimize consensus algorithm parameter adjustment and election decisions, thereby improving switching efficiency and accuracy. Attached Figure Description

[0016] Figure 1 This is a logical architecture diagram of a distributed dynamic migration service system based on eBPF according to an embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating the network metric collection process of the state awareness layer in a distributed dynamic migration service system based on eBPF, according to an embodiment of this application. Detailed Implementation

[0018] The present application will be further described in detail below with reference to the accompanying drawings.

[0019] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Current methods for achieving high availability in distributed systems involve running a proxy on each node. This proxy periodically sends heartbeat packets via multicast to other nodes at the application layer for heartbeat maintenance and master / slave status maintenance. Simultaneously, a distributed consensus algorithm runs, using a fixed timeout, such as 150ms-300ms for election. When the master node fails, other nodes time out due to not receiving heartbeats, triggering a new election to elect a new master node. Election strategies are typically based on node IDs or statically configured priorities. However, this approach has certain technical limitations. Fixed timeout parameters cannot adapt to dynamically changing network environments. In good network conditions, excessively short timeouts can lead to misjudgments, while excessively long timeouts cause high failover latency during network congestion. Static election strategies cannot detect real-time network latency and topology changes between nodes, and cannot guarantee that the elected master node is the globally optimal solution. User-space monitoring proxies incur significant performance overhead, making it difficult to achieve continuous, fine-grained monitoring at the microsecond level.

[0022] Example 1

[0023] Figure 1 This is a logical architecture diagram of a distributed dynamic migration service system based on eBPF. Figure 1 The logical architecture diagram of the distributed dynamic migration service system based on eBPF includes a state awareness layer, a user layer, and an intelligent decision-making layer connected in sequence.

[0024] The state-aware layer is used to collect and output network metrics in real time from three dimensions in the Linux kernel layer by mounting three eBPF programs; The user layer is used to receive network metrics output by the state awareness layer, preprocess them, and then output them to the intelligent decision-making layer. The intelligent decision-making layer is used to conduct a comprehensive evaluation of network quality based on preprocessed network indicators, generate quality scores and quality levels, and then dynamically adjust the operating parameters of the distributed consensus algorithm according to the quality scores and quality levels, thereby affecting the decision-making conditions for the switching of master and backup nodes. Ultimately, it realizes a highly efficient autonomous intelligent switching system with millisecond-level detection of network anomalies to ensure system stability and greatly improve system performance.

[0025] It's important to understand that a distributed system connects multiple computer devices within a network, with each device acting as a node, enabling state synchronization and transitions. The eBPF (Extended Berkeley Packet Filter) upon which distributed systems are based is a technology that allows user-provided programs to be executed securely and efficiently within the operating system kernel without modifying kernel source code or loading kernel modules. A distributed consensus algorithm is an algorithm in a distributed system that enables all nodes to reach a consensus on a certain value (such as who the master node is).

[0026] Figure 2 This is a flowchart illustrating the network metric collection process for the state-aware layer of an eBPF-based distributed dynamic migration service system, such as... Figure 2 As shown, the state-aware layer collects three dimensions of network metrics in real time within the Linux kernel layer by mounting three eBPF programs: By mounting the XDP program to the specified network interface eth, packets are captured at the earliest point in the network driver layer. Only the Ethernet header and IP header are parsed, and the global packet counter is updated using atomic operations to obtain global traffic. This avoids lock contention. In the global statistics mapping process, global traffic is stored with 0 as the key and global traffic as the value.

[0027] By mounting the TC program to the specified network interface in the kernel layer in both the egress and ingress directions, deep analysis of TCP connections and accurate RTT measurement are achieved. Specifically, in the egress direction, when a TCP packet (such as a SYN packet or a data packet) leaves the protocol stack, the program extracts its connection identifier, namely the 5-tuple (source IP, destination IP, source port, destination port, protocol) and sequence number. The packet timestamp mapping uses the 5-tuple and sequence number as keys and the current high-precision timestamp, i.e., the sending timestamp T_send (obtained via bpf_ktime_get_ns()), as the value, and stores it in the eBPF hash map. In the ingress direction, the TC program captures the TCP ACK packet returned by the peer and queries the corresponding T_send in the mapping based on its ACK number and the reverse 5-tuple. If a match is found, a single RTT value is calculated: RTT = T current time - T_send. The system collects RTT samples from multiple TCP interactions over a period of time. RTT sample refers to the total time it takes for a data packet to be sent from the local machine to the peer machine and for the peer machine to return an acknowledgment packet (ACK).

[0028] The eBPF program is mounted into kernel functions (such as the `tcp_retransmit_skb` function) via `kprobe`. When the kernel retransmits a data packet, the retransmission count for the corresponding connection is incremented. This is also mounted into the kernel function (e.g., `tcp_transmit_skb` function) to increment the total transmission count for the corresponding connection, supplementing the TC retransmission detection. This process yields both the number of retransmitted data packets and the total number of transmitted data packets. Connection metrics are stored using an LRU hash map for connection mapping. The key is the data packet 5-tuple, and the value is the connection metric structure, i.e., the number of retransmitted data packets (corresponding to...). Figure 2 The retransmission count and the total number of data packets sent (corresponding to) Figure 2 The "send count" is used to further determine the retransmission rate based on the number of retransmitted data packets and the total number of sent data packets.

[0029] By mounting the above three eBPF programs into the Linux kernel layer, we can obtain global traffic, latency, and retransmission rate from three dimensions, which are the three network metrics.

[0030] After the user layer receives the three network metrics as input values, the user layer performs preprocessing on them.

[0031] In this process, the user layer preprocesses the global traffic, periodically (e.g., every second) reads the packet counter, calculates the current packet rate (number of packets in the current period / period time), maintains the historical packet rate for a time window (e.g., 60 seconds), calculates the average packet rate as the traffic baseline, and then determines the traffic deviation coefficient K based on the current packet rate and the average packet rate. The formula for calculating the traffic deviation coefficient K is the current packet rate / average packet rate.

[0032] The user layer preprocesses the delay time and calculates the average delay time of multiple RTT samples collected over a period of time, avg_rtt_ms.

[0033] The user layer inputs the preprocessed network metrics into the intelligent decision layer to conduct a comprehensive evaluation of network quality, generating a quality score and quality level.

[0034] The global flow score is determined based on the flow deviation coefficient K and a preset first lookup table, namely: When K is greater than or equal to 0.8 and less than or equal to 1.2, the global traffic score is 100. When K is greater than 1.2 and less than or equal to 2.0, the global traffic score is 100 - (K - 1.2). 25; When K is greater than 2.0 and less than or equal to 5.0, the global traffic score is 80 - (K - 2.0). 20; When K is greater than 5.0, the global traffic score is 20.

[0035] The delay time score is determined based on the average value avg_rtt_ms and a second preset lookup table, i.e.: When avg_rtt_ms is less than or equal to 10ms, the delay time score is 100. When avg_rtt_ms is greater than 10ms and less than or equal to 30ms, the delay score is 100-20. (avg_rtt_ms-10) / 20; When avg_rtt_ms is greater than 30mA and less than or equal to 100ms, the latency score is 80-20. (avg_rtt_ms-30) / 70; When avg_rtt_ms is greater than 100ms and less than or equal to 300ms, the delay time score is 60-30. (avg_rtt_ms-100) / 200; When avg_rtt_ms is greater than 300ms, the latency score decays exponentially to max(0,30-(avg_rtt_ms-300) / 10).

[0036] The retransmission rate (retrans_rate) is calculated by dividing the number of retransmitted data packets by the total number of sent data packets. The retransmission rate score is determined based on the retransmission rate (retrans_rate) and a third preset lookup table. When retrans_rate is less than or equal to 0.01, the retransmission rate score is 100. When retrans_rate is greater than 0.01 and less than or equal to 0.05, the retransmission rate score is 100-20. (retrans_rate-0.01) / 0.04; When retrans_rate is greater than 0.05 and less than or equal to 0.1, the retransmission rate score is 80-30. (retrans_rate-0.05) / 0.05; When retrans_rate is greater than 0.1, the retransmission rate score exhibits exponential decay as max(0, 50-50) = 0.1. (retrans_rate-0.1) / 0.9).

[0037] In summary, global traffic score, latency score, and retransmission rate score are obtained. Weights are assigned to global traffic, latency, and retransmission rate using a preset weight table. A quality score is calculated based on these scores and their corresponding weights. In this invention, since the retransmission rate reflects network reliability and has the greatest impact on the stability of the distributed system, its weight is 40%. Latency reflects system response speed, so its weight is 35%. Global traffic reflects network health and its impact on system predictability, so its weight is 25%. Therefore, the quality score equals the retransmission rate score. 40%+ delay time score 35%+ Global Traffic Score 25%, and then the quality level is determined based on the quality score and a preset level comparison table. The preset level comparison table stores the correspondence between the quality score and the quality level, that is: A quality score greater than 90 and less than or equal to 100 is considered excellent. A quality score greater than 75 and less than or equal to 90 is considered good. If the quality score is greater than 60 points and less than or equal to 75 points, the quality level is average. If the quality score is greater than 40 and less than or equal to 60, the quality level is poor. If the quality score is greater than 0 and less than or equal to 40, the quality level is severe.

[0038] After determining the quality level, the quality score is statistically observed over a period (set according to the actual situation). If the quality score continuously declines, an intelligent election strategy is triggered. The system dynamically and adaptively adjusts the operating parameters of the distributed consensus algorithm based on the quality level to optimize performance and adapt to network conditions. The operating parameters include election timeout, heartbeat interval, and voting timeout. The adaptive rules are as follows: When the quality level is excellent or good, the election timeout time is reduced and the heartbeat interval is shortened, and a more aggressive operating strategy is adopted. When the quality level is average, the election timeout period should be appropriately increased. When the quality level is poor or severe, the election timeout time is significantly increased and the heartbeat interval is reduced, adopting a conservative operating strategy.

[0039] Meanwhile, the present invention also includes parameter boundary protection, wherein the election timeout is 100ms-1000ms and the heartbeat interval must be less than half of the election timeout.

[0040] In this system, a node refers to an independent server instance running a complete system service within a distributed cluster. Each node has a unique identity and can communicate with other nodes via the network. Election is a core process in the distributed consensus algorithm. Its main function is to ensure that when the system needs to determine or change the master node, all nodes participate in voting to jointly elect a new master node according to predetermined rules. After a new master node is established, the system switches to the new node as the master node, aiming to achieve high availability of the service.

[0041] This system proposes to intelligently trigger elections based on network quality assessment results, in addition to conventional election triggering conditions (such as leader heartbeat timeout). Elections are triggered when the quality level reaches the "severe" level and continues to deteriorate over a period of time, attempting to switch to a node with better network quality; elections are triggered when the retransmission rate or RTT latency is abnormally high and continues to deteriorate; and elections are triggered when abnormal patterns in global traffic may affect the consensus algorithm's operation. Based on raw network metrics collected from the state-aware layer, the system quantifies complex network states into intuitive quality levels through weighted quality scoring, providing data support for subsequent intelligent decision-making.

[0042] Example 2

[0043] The objective of this invention is to verify whether, under simulated network failure conditions, the system can correctly trigger the decision-making mechanism and complete host switching based on network metrics collected by eBPF, with a switching time of less than 20ms. To verify the effectiveness of the technical solution of this invention, a specific implementation example and verification method are given below in conjunction with the logical architecture diagram shown in Figure 1.

[0044] First, the environment was deployed using three servers equipped with domestically produced S2500 processors and the Galaxy Kylin V10 operating system. These servers were connected via a gigabit Ethernet switch to form a distributed cluster, ensuring interconnectivity on the backbone network and time synchronization across all servers. Next, the eBPF program was deployed. The eBPF programs (XDP, TC, kprobe) described in the technical solution were compiled and loaded on each node, and their loading status was verified using tools such as bpftool. Then, the user-layer service was deployed, and a user-layer monitoring and decision-making service was installed and configured. This service periodically reads network metrics from the eBPF mapping. Finally, the consensus algorithm was initialized by deploying a distributed consensus algorithm (such as Raft) on each node and configuring initial parameters such as an election timeout of 300ms and a heartbeat interval of 150ms. Monitoring programs were also installed on each node. The entire process involved each user-layer node independently evaluating its own network quality based on the network metric output collected by the eBPF program, adjusting its local consensus algorithm parameters according to the evaluation results, and triggering an election when pre-set conditions were met. The specific verification process is as follows: Initially, node A is assumed to be the initial master node, and nodes B and C are backup nodes, with all nodes having an "Excellent" quality rating. A fault injection is performed on the current master node, and node A's network connection is manually disconnected to simulate a network failure. Nodes B and C, through their eBPF programs, detect in real-time communication anomalies with node A. The Real-Time To-Time (RTT) rapidly increases to the timeout threshold, and the retransmission rate sharply rises to nearly 100%. Global traffic towards node A drops sharply. Based on these metrics, the user-layer service calculates a quality score, and node A's quality rating rapidly deteriorates from "Excellent" to "Severe." Intelligent decision-making is then triggered. According to the triggering conditions in the technical solution, nodes B and C detect that the master node's quality rating has reached "Severe" and immediately trigger a high-priority election. During the election, each node participates in the election based on its real-time network quality score, and the node with the higher quality score (let's say node B) is elected as the new master node. After the election, the system performs a fast state switch, synchronizing the service connection state to node B. Verification results were obtained by monitoring the application logs of nodes B and C (using `tail -f [application log file]`), which showed a complete log sequence of fault detection → quality assessment → election triggering → state switching. Actual testing showed that the entire switchover time, from the occurrence of the fault (network disconnection) to the new master node (node ​​B) starting to provide service, was ≤20ms. After the switchover was complete, the cluster state was node B (master node) and node C (standby node).

[0045] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A distributed dynamic migration service system based on eBPF, characterized in that, It includes a state perception layer, a user layer, and an intelligent decision-making layer that are connected in sequence; The state-aware layer is used to collect and output network metrics in real time in three dimensions in the Linux kernel layer by mounting three eBPF programs. The network metrics include global traffic, latency, and retransmission rate. The user layer receives global traffic, latency, and retransmission rate from the state awareness layer, preprocesses them, and then outputs them to the intelligent decision-making layer. The intelligent decision layer is used to comprehensively evaluate network quality by taking the preprocessed global traffic, latency, and retransmission rate, generating a quality score and quality level. Then, based on the quality score and quality level, the operating parameters of the distributed consensus algorithm are adjusted through adaptive rules, thereby affecting the decision-making conditions for the switching of master and backup nodes.

2. The distributed dynamic migration service system based on eBPF according to claim 1, characterized in that, The state awareness layer collects network metrics in real time across three dimensions at the Linux kernel layer by mounting three eBPF programs, including: By mounting the XDP program to the specified network interface eth, data packets are captured at the earliest point in the network driver layer, and atomic operations are used to update the global packet counter, thereby obtaining the global traffic. By mounting the TC program to the specified network interface in the kernel layer in the outgoing and incoming directions, in-depth analysis of TCP connections and accurate measurement of RTT are achieved, thereby obtaining the latency time. The eBPF program is mounted into the kernel function using kprobe. When the kernel retransmits a data packet, the retransmission count of the corresponding connection is incremented, and the total transmission count of the corresponding connection is also incremented. The number of retransmitted data packets and the total number of transmitted data packets are obtained, and the retransmission rate is determined based on the number of retransmitted data packets and the total number of transmitted data packets.

3. The distributed dynamic migration service system based on eBPF according to claim 2, characterized in that, The method involves mounting the XDP program to the specified network interface eth, which parses only the Ethernet header and IP header.

4. A distributed dynamic migration service system based on eBPF according to claim 2, characterized in that, The user layer preprocesses global traffic, including: Periodically read the data packet counter to obtain the number of packets and the period time in the current period; Calculate the current rate per packet based on the number of packets in the current cycle and the cycle time; Obtain the historical packet rate for a given time window and calculate the average packet rate; The flow deviation coefficient is determined based on the current per-packet rate and the average packet rate.

5. A distributed dynamic migration service system based on eBPF according to claim 4, characterized in that, The user layer preprocesses the delay time, including: Calculate the average value of the delay time collected within the preset time period.

6. A distributed dynamic migration service system based on eBPF according to claim 5, characterized in that, The intelligent decision-making layer performs a comprehensive network quality assessment based on preprocessed global traffic, latency, and retransmission rate, generating a quality score and quality level, including: The global flow score is determined based on the flow deviation coefficient and the first preset lookup table. The delay time score is determined based on the average delay time and a second preset comparison table; The retransmission rate score is determined based on the retransmission rate and the third preset comparison table. The quality score is determined based on the global traffic score, latency score, retransmission rate score, preset global traffic weight value, preset latency weight value, and retransmission rate weight value. The quality level is determined based on the quality score and a preset level comparison table, which stores the correspondence between the quality score and the quality level.

7. A distributed dynamic migration service system based on eBPF according to claim 6, characterized in that, The process of determining the quality level based on the quality score and a preset level comparison table includes: A quality score greater than 90 and less than or equal to 100 is considered excellent. A quality score greater than 75 and less than or equal to 90 is considered good. If the quality score is greater than 60 points and less than or equal to 75 points, the quality level is average. If the quality score is greater than 40 and less than or equal to 60, the quality level is poor. If the quality score is greater than 0 and less than or equal to 40, the quality level is severe.

8. A distributed dynamic migration service system based on eBPF according to claim 7, characterized in that, The operating parameters include election timeout time and heartbeat interval. Adjusting the operating parameters of the distributed consensus algorithm according to quality scores and quality levels using adaptive rules includes: When the quality level is excellent or good, reduce the election timeout and shorten the heartbeat interval; When the quality level is average, increase the election timeout period; When the quality level is poor or severe, increase the election timeout and decrease the heartbeat interval.

9. A distributed dynamic migration service system based on eBPF according to claim 8, characterized in that, The operating parameters are set with parameter thresholds. The parameter threshold for the election timeout is 100ms-1000ms, and the parameter threshold for the heartbeat interval is less than half of the election timeout.

10. A distributed dynamic migration service system based on eBPF according to claim 7, characterized in that, The conditions affecting the switching of primary and backup nodes in decision-making include: An election is triggered when the quality level reaches the "severe" level and continues to deteriorate over a period of time. An election is triggered when the retransmission rate or latency is abnormally high and continues to deteriorate. An election is triggered when global traffic affects the abnormal operation mode of the consensus algorithm.

Images