A method for dynamically maintaining a monitoring node for embedded containers
By dynamically adjusting monitoring permissions in the embedded environment, the problems of business interruption and network overhead caused by container migration are solved. A debouncing mechanism is built to achieve load balancing and stability, ensuring the security of edge node devices and business continuity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 北京星云越动科技有限公司
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-31
AI Technical Summary
In embedded environments, traditional container management platforms suffer from frequent physical container migrations, leading to business interruptions, state loss, and high network overhead. Centralized management platforms are prone to single-point performance bottlenecks, while the dynamic load of edge node devices causes frequent system oscillations and makes them prone to crashing due to their inability to handle management tasks.
By abandoning physical container migration and adopting dynamic delegation and revocation of monitoring permissions, combined with hysteresis debouncing mechanism and network affinity algorithm, load balancing is achieved. This includes steps such as parameter initialization, global monitoring and overload judgment, affinity-based traffic distribution and permission delegation, status inspection and self-protection recycling, etc., to achieve dynamic scaling of monitoring tasks.
Break through the performance bottleneck of centralized single points, avoid business interruption and network overhead, eliminate frequent system oscillations, reduce network probing overhead, ensure the stability of edge node devices, and prevent core business crashes.
Smart Images
Figure CN122489262A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of edge computing and embedded container cluster management. Specifically, it relates to a method for dynamically maintaining monitoring nodes for embedded containers. Background Technology
[0002] With the development of edge computing, the number of containers running on node devices is constantly increasing. Traditional container management platforms typically employ centralized monitoring or load balancing solutions based on heavy container migration. However, in embedded environments, device computing power, memory, and network bandwidth are severely limited, and frequent physical container migrations can lead to business interruptions, state loss, and significant network overhead. When a centralized management platform is responsible for monitoring massive numbers of containers, it is highly susceptible to creating single-point performance bottlenecks. Furthermore, the load on edge node devices is highly dynamic; simple fixed thresholds often result in frequent system oscillations, and if edge node devices are unconditionally burdened with additional management tasks, their core business can easily crash due to resource preemption. How to achieve secure, stable, and automated dynamic scaling of monitoring computing power without physical container migration is a significant challenge currently facing container cluster management. Summary of the Invention
[0003] Technical problems to be solved This invention aims to address the problems of business interruption, state loss, and significant network overhead caused by frequent physical container migrations when existing container cluster management technologies are applied in embedded environments, as well as the single-point performance bottlenecks that centralized management platforms are prone to. Specifically, this invention provides a new dynamic maintenance method for monitoring node devices to solve the problems of frequent system oscillations caused by simple fixed thresholds on edge node devices with limited computing power and high load, and the crashes of the node device's core business due to resource preemption caused by blindly accepting management tasks. This allows for safe, stable, and automated dynamic scaling of monitoring computing power without the need for physical container migration.
[0004] Technical solution To address the aforementioned technical problems, this invention provides a method for dynamic maintenance of monitoring nodes for embedded containers, applicable to edge computing cluster environments that include a centralized management platform and multiple node devices. Please refer to... Figure 1 The core of this method lies in abandoning the traditional physical migration of containers and instead achieving load balancing through the dynamic delegation and revocation of monitoring permissions, and includes the following steps: Parameter initialization phase: Initialize the anti-shake parameters with hysteresis range based on the performance limits of the software and hardware, and set the high water level of the control platform, the single allocation target of the agent, the low water level of the agent, and the self-protection danger water level.
[0005] Global monitoring and overload determination phase: The management platform undertakes the default global monitoring task and evaluates it periodically. If the total number of containers currently directly monitored by the platform exceeds the platform's high water level, it is determined that the number of monitored containers is overloaded, triggering "monitoring fission".
[0006] Affinity-based traffic distribution and permission delegation phase: Identify node devices with idle physical resources and upgrade them to monitoring node devices. Calculate the reciprocal of the network round-trip latency of candidate node devices to generate a network affinity score. Use a greedy algorithm to accurately package monitoring tasks under target capacity constraints to complete the physical "nearest" transfer of monitoring rights. During this process, the original node device container continues to run and does not undergo migration.
[0007] Status inspection and self-protection recovery phase: The platform traverses the surviving monitoring node devices and performs strict serial judgment. If the physical CPU load of the monitoring node device exceeds the self-protection danger level threshold, emergency self-protection is triggered, and its monitoring rights are revoked. If the physical load is healthy but the actual number of monitored containers falls below the low level, the smooth recovery of idle resources is triggered.
[0008] Beneficial effects Compared with existing technologies, this invention addresses specific problems in the background technology by transforming physical relocation into a purely logical monitoring task transfer, and by combining a hysteresis debouncing mechanism with a network affinity algorithm, achieving the following significant beneficial effects: (1) Breaking the single-point performance bottleneck of centralized systems and achieving zero-interruption scheduling: In response to the problems in the background technology that "centralized management and control platforms are prone to single-point performance bottlenecks when handling massive containers" and "frequent physical migrations lead to business interruptions and high network overhead", this invention effectively disperses the pressure of the management and control platform concurrently handling the monitoring of massive containers by comparing high-water level parameters to trigger dynamic allocation of monitoring tasks (i.e., "monitoring fission"). At the same time, the traditional "container physical migration" is transformed into a purely logical "monitoring task transfer". Devices selected as monitoring nodes do not need to migrate their original containers, eliminating single-point bottlenecks and avoiding the risks of network IO blocking and state loss caused by physical migration.
[0009] (2) Constructing a robust anti-jitter mechanism to prevent frequent system oscillations: Addressing the issue in the background technology that "highly dynamic loads on edge node devices and simple fixed thresholds easily lead to frequent system oscillations," this invention introduces industrial-grade hysteresis interval theory. By utilizing the significant capacity difference between a single allocation target calculated based on hardware performance limits and its specific proportion of low-water level in hysteresis recovery, a dynamic anti-jitter interval is constructed. This method perfectly absorbs the impact of high-frequency dynamic addition and deletion of cluster containers in edge environments, effectively preventing frequent start-stop oscillations in the monitoring topology.
[0010] (3) Precise traffic splitting based on affinity, reducing network probing overhead: This invention adopts a network affinity greedy traffic splitting algorithm, which uses the reciprocal of the network round-trip delay time calculated by concurrent probing as the affinity score for greedy packaging. This technique ensures that the transferred monitoring tasks are absolutely "nearby" in the physical network topology, greatly reducing the cross-segment probing latency and network bandwidth loss of the upgraded monitoring node devices when performing monitoring tasks.
[0011] (4) Establishing a self-protection mechanism with absolute priority for underlying load to avoid core business crashes: In response to the problem in the background technology that "node devices unconditionally undertaking management and control tasks can easily lead to the crash of their own core business", this invention adopts a strict serial status inspection mechanism. By introducing underlying physical load monitoring independent of the number of containers, the monitoring node device is given the ability to perform "emergency backoff" (i.e., reclaim the monitoring task with the highest priority) when its real-time underlying physical load exceeds the safety threshold, effectively protecting edge node devices with limited computing power from crashing and improving the overall stability of the system. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a flowchart illustrating a dynamic maintenance method for monitoring nodes of embedded containers provided in an embodiment of the present invention.
[0014] Figure 2 This is an application scenario diagram of a dynamic maintenance method for monitoring nodes of embedded containers provided by an embodiment of the present invention. Detailed Implementation
[0015] Step 1: Combining Figure 1 The "initialize parameters" step in the text is similar to... Figure 2 The "Scenario 1: Initial Stable State" refers to the state when the control platform is started.
[0016] Let MAX_plat be the theoretical maximum number of concurrent monitoring containers on the management platform, and MAX_node be the theoretical maximum number of concurrent monitoring containers on a single edge node device. The system initializes the following thresholds:
[0017] 1. High water level on the platform ; 2. Agent's single allocation target ; 3. Agent low water level ; Self-protection dangerous water level (Real-time physical CPU load limit of node devices). Step 2: As Figure 1 The steps of "the management and control platform assumes default global monitoring" and Figure 2 As shown in "Scenario 1", the management platform assumes the task of monitoring the default global node devices, directly listening to all node devices (e.g., Figure 2 The system enters a periodic evaluation cycle (e.g., an evaluation is performed every 30 seconds) for node devices A, B, C, and D, as well as their internal containers.
[0018] Step 3: As Figure 1 As shown in the "Number of Containers Currently Monitored by the Statistics Platform" and the judgment branch, the management platform counts the total number of containers currently under direct monitoring. C_total .like Figure 2 As shown in the evolution process, when the number of containers in the cluster increases dynamically, if (Right now Figure 1 If the message "Platform monitoring overload" appears, proceed to step 4; otherwise, proceed to step 6.
[0019] Step 4: Enter Figure 2 The scenario shown is "Scenario 2: 'Peak Load and Dynamic Partition Monitoring'". The management platform scans the cluster to find node devices N with extremely idle physical resources (such as...). Figure 2 The previously idle node device C is given a command to start a monitoring task and upgrade it to "monitoring node device M". During this upgrade process, the original containers on node device N continue to run and no cross-node device migration operations are performed, thus partially relieving the pressure on the management platform.
[0020] Step 5: As Figure 1 As shown in the "Greedy Distribution Based on Network Affinity" and "Greedy Packaging" steps, node device N concurrently probes candidate ordinary node devices. i Round-trip delay time Calculate the devices at each node i Network affinity rating : Score based on network affinity node devices i The candidate list (candidate_list) is generated by sorting the nodes in descending order. The candidate list is then traversed, and node devices are retrieved one by one. i Network affinity rating Get the total number of containers to be allocated to the monitoring node devices. Sum With node devices i Number of containers running internally ,if Then the node device i Add to the splitter set select_list and update the total number of containers. Otherwise, the traversal stops, and the management platform transfers the monitoring rights of all node devices in the distribution set select_list to node device N, such as... Figure 2 As shown in Scenario 2, the upgraded node device C will prioritize sharing the monitoring domains with the closest physical distance (such as node device E with Score=80 and node device F with Score=50). The management platform itself will stop monitoring these devices and proceed to step 2. By combining the above techniques of comparing high-water mark parameters and delegating monitoring tasks, the concurrent pressure on the centralized management platform is effectively relieved, solving the problem of single-point performance bottlenecks easily generated by monitoring massive containers. Simultaneously, the purely logical transfer of permissions avoids business interruptions caused by physical container migration. Monitoring rights refer to assigning the monitoring tasks of some node devices and the containers running within them to the upgraded node devices for execution. This results in the management platform relinquishing its monitoring rights over these devices to the upgraded nodes. Step 6: Enter Figure 2 The scenario shown is "Scenario 3: Permission Revocation and Integration". Figure 1 The status inspection process involves the platform traversing all surviving monitoring node devices M and obtaining their real-time physical CPU load. CPU_local .if (If the physical load is too high), trigger the backup mechanism and proceed to step 8; otherwise, proceed to step 7.
[0021] Step 7: As Figure 1 Judgment 2 and Figure 2 As shown in "Scenario B: Smooth Degradation of Business Loss", obtain the monitoring node device M (e.g., Figure 2 The original monitoring node device O) is currently responsible for monitoring the number of containers. C_monitored .if (State B: Monitoring services are significantly lost, computing power is idle), skip to step 8; if not (State C: Stable operation), maintain the status quo and skip to step 6. By using the above-mentioned technical means of constructing a de-jittering interval by utilizing the single allocation target and the difference between the low water level capacity of delayed recovery, the problem of frequent system oscillation caused by dynamic changes in edge load is effectively avoided.
[0022] Step 8: As Figure 1 The steps of "centralizing monitoring authority" and "downgrading node devices", and Figure 2Scenario 3 illustrates "reclaiming control of all node devices" and "downgrading to a regular business node device." The management platform reclaims the monitoring rights from monitoring node device M, allowing direct monitoring from the platform again. Then, the monitoring process of monitoring node device M is completely shut down, downgrading it back to a regular node device, and proceeding to step 2. By prioritizing the verification of the underlying physical load and triggering emergency backoff, the problem of edge nodes blindly undertaking additional management tasks leading to the collapse of their core business is solved, ensuring the absolute security of edge-side services.
Claims
1. A method for dynamic maintenance of monitoring nodes for embedded containers, applied to an edge computing cluster environment comprising a centralized management and control platform and multiple node devices, wherein, The node device runs containers, and the centralized management platform is used to remotely monitor the node device and the containers running therein via a network, and to perform control operations. The platform is characterized by the following steps: a. Initialization parameters: Based on the theoretical maximum monitoring capacity of the centralized management and control platform, initialize the high water level parameters of the centralized management and control platform; and based on the hardware performance limits of the node devices, initialize the anti-shake parameters of the node devices with hysteresis range, including single allocation target and hysteresis recovery low water level and underlying physical load threshold. b. Status Inspection: The centralized management and control platform periodically obtains the total number of containers currently directly monitored by it based on the set timed tasks; c. Dynamic task allocation: Compare the total number of containers with the high water level parameter of the centralized management and control platform to determine whether to trigger the upgrade of idle node devices, and allocate the monitoring tasks of some node devices and the containers running in them to the upgraded node devices for execution; d. Dynamic maintenance: Serially evaluate the physical CPU load of surviving monitoring node devices and the actual number of monitored containers to determine whether to trigger self-protection backoff or resource idle reclamation, thereby dynamically maintaining the overall monitoring topology of the system.
2. The method according to claim 1, characterized in that, The entire process of dynamic scaling and load balancing of node devices does not involve any physical migration of containers across node devices. Devices selected as monitoring nodes only need to start independent monitoring tasks, and the original containers on them continue to run continuously.
3. The method according to claim 1, characterized in that, The anti-jitter parameters initialized in step 'a' with a hysteresis range include a single allocation target and a hysteresis recovery low water level. The single allocation target is determined based on the product of the theoretical maximum concurrent monitoring container number of the node device and a first preset ratio. The hysteresis recovery low water level is determined based on the product of the single allocation target and a second preset ratio. The capacity difference between the single allocation target and the hysteresis recovery low water level constitutes the anti-jitter range to prevent frequent start-stop oscillations of the node device performing the monitoring task.
4. The method according to claim 1, characterized in that, The dynamic traffic splitting in step c specifically employs a network affinity greedy traffic splitting algorithm. The specific steps of this algorithm include: obtaining the network round-trip latency of candidate idle node devices through concurrent probing, calculating a network affinity score based on the reciprocal of the network round-trip latency, and generating a candidate list by sorting the scores in descending order; subsequently, using a greedy algorithm to traverse the candidate list, and under the constraint that the cumulative number of containers does not exceed the single allocation target, assigning the monitoring tasks of the node device with the highest score and the containers running in it to the upgraded node device for execution.
5. The method according to claim 1, characterized in that, The dynamic maintenance described in section d employs a serial decision mechanism for the status inspection of node devices performing monitoring tasks. This serial decision mechanism specifically includes: firstly verifying the real-time underlying physical load of the node device performing the monitoring task; if its load exceeds the underlying physical load threshold, then triggering the emergency recovery operation of the highest priority monitoring task; if the underlying physical load does not exceed the underlying physical load threshold, then continuing to verify whether the number of containers it is currently responsible for monitoring is lower than the hysteresis recovery low water level, and if it is lower than the hysteresis recovery low water level, then executing the resource idle recovery operation of the monitoring task.