An event-driven promotion position dynamic compensation processing method

Abstract

The present application relates to computer information processing and distributed system technical field, provide a kind of based on event-driven promotion bit dynamic compensation processing method.For solving the contradiction between real-time compensation processing and multi-party state consistency difficult to consider in traditional technology, method by encapsulating abnormal event signal as state element, according to collaborative review relationship atlas generates the collaborative party locking position map marked all associated collaborative party nodes, with state projection message multicast triggers each associated collaborative party node, each associated collaborative party node determines the optimal compensation promotion content based on the global consistent polymorphic compensation utility function, then each local commit with atomic state transaction makes the content state of each promotion bit consistent update, realizes the real-time automatic compensation of promotion bit under eliminating the intervention of distributed coordination node and the atomic consistency of multi-party state, significantly improves the real-time, consistency and system scalability of promotion bit compensation processing in distributed content management system.

Description

Technical Field

[0001] This invention relates to the fields of computer information processing and distributed systems technology, and particularly to distributed collaborative control and event-driven processing technology, specifically to an event-driven generalized bit dynamic compensation processing method. Background Technology

[0002] Promotion slot compensation processing refers to the process of reallocating and restoring the status of vacant promotion slots caused by content rejection, material changes, or push anomalies during the submission, review, and publication of information promotion content (such as advertisements, notices, and recommendation information). It is one of the core functions of a distributed content management system.

[0003] Taking a scenario of information promotion content management with multi-party collaborative review as an example: After the content provider submits promotional materials, they need to undergo compliance review by the reviewer. Once approved, the promotional content is placed on the corresponding promotional slot by the publisher. During this process, if any abnormality occurs at any stage (such as review rejection, expired materials, or failed push), the relevant promotional slot becomes vacant. It is necessary to select alternative promotional content to fill the slot, and simultaneously update the status records of each participating party for that promotional slot. Each participating party (such as the content provider, reviewer, and publisher) maintains its own data view related to that promotional slot, but the parties operate independently, and there is no unified central database.

[0004] In traditional technologies, compensation for promotion slots typically employs two methods: First, a workflow engine-based approach, where predefined compensation branches respond to anomalies, and the workflow engine sequentially calls the service interfaces of each participant to replace the content and update the status of the promotion slot. Second, an asynchronous message queue-based approach, where an anomaly triggers the delivery of a compensation message, the compensation service consumes the message, performs content selection based on merit, and writes the result to the databases of each participant. Both methods utilize an explicit centralized compensation scheduling node, which undertakes core functions such as anomaly detection, decision calculation, and operation distribution.

[0005] However, the problem with the aforementioned traditional technologies lies in the irreconcilable contradiction between the real-time nature of compensation processing and the consistency of the states of all participants. When the centralized scheduling node executes compensation decisions, it needs to query and confirm the states across multiple participants. The interaction delay introduced by this process causes the compensation operation to lag behind the occurrence of the abnormal event, and the promotion position remains in an idle state until the compensation is completed. On the other hand, simplifying the state confirmation process in pursuit of faster response speed may lead to inconsistencies in the state of the promotion position among the participants due to data asynchrony.

[0006] The problem stems from the fact that traditional technologies employ a "process-centric" architecture, treating the promotion position as an object manipulated by various participants and state correction as an operation sequence orchestrated by a centralized coordinator. Under this architecture, there is an inextricable temporal dependency between state synchronization, compensation decisions, and operation execution: to ensure consistency between the state data maintained by each participant related to the same promotion position, distributed locks or transaction coordinators must be introduced, inevitably increasing processing latency; conversely, reducing latency inevitably weakens the reliability of state synchronization.

[0007] Therefore, in traditional promotion position compensation processing, there is a technical contradiction that is difficult to reconcile with the real-time requirements of compensation processing and the consistency guarantee of multi-party states, which has become a technical problem that urgently needs to be solved in the field of distributed content management. Summary of the Invention

[0008] The technical problem to be solved by this invention is: how to achieve low-latency automatic compensation for promotion resources and atomic consistency of multi-party state data under the condition of eliminating the intervention of distributed coordination nodes.

[0009] To address the aforementioned technical problems, this invention provides an event-driven dynamic compensation method for promotion positions. By constructing an event-driven distributed autonomous state synchronization mechanism, the compensation process is transformed from centralized process orchestration into autonomous parallel solving by a group of distributed peer nodes. This differs from the operation sequence-driven mode in traditional technologies that rely on workflow engines or compensation services as centralized schedulers. It can achieve real-time automatic compensation of promotion positions without relying on centralized compensation scheduling nodes, while ensuring the consistency of state updates among the collaborating nodes. This breaks through the technical bottleneck of the incompatibility between real-time compensation and state consistency in the traditional process-centric architecture.

[0010] To address the aforementioned technical problems, this invention provides the following technical solution: an event-driven dynamic compensation method for promotion positions, comprising: responding to any collaborating node in a distributed system detecting an abnormal event signal, the collaborating node generating a state primitive containing a promotion position identifier, event type, and scope corresponding to the abnormal event signal; the collaborating node generating a collaborating locking bitmap identifying all target collaborating nodes associated with the promotion position identifier based on a preset collaborating audit relationship graph, and multicasting the state primitive and the collaborating locking bitmap as state projection messages to trigger each target collaborating node to perform local consistency solving; each target collaborating node performing local consistency solving based on a globally consistent polymorphic compensation utility function according to its local historical state and the state primitive, deterministically deriving the optimal compensation promotion content corresponding to the promotion position identifier; each target collaborating node committing the optimal compensation promotion content locally as an atomic state transaction, so that the content state of the promotion position corresponding to the promotion position identifier in each target collaborating node is consistently updated to the optimal compensation promotion content.

[0011] As a preferred embodiment of the event-driven dynamic compensation processing method for promotion positions described in this invention, a collaborative locking bitmap is generated, identifying all target collaborative nodes associated with the promotion position identifier. The state primitives and the collaborative locking bitmap are then multicast as state projection messages. This includes: obtaining a preset collaborative review relationship graph, which stores the associations between each promotion position identifier and collaborative nodes in a graph data structure. Each collaborative node is pre-assigned a unique hotkey to identify itself. The bit width of the unique hotkey is determined by the total number of collaborative nodes in the distributed system, and the position of the bit with a value of 1 in the unique hotkey is fixed to correspond to the node of that collaborative node in the distributed system. Based on the promotion position identifier, subgraph matching is performed in the preset collaborative review relationship graph to obtain collaborative nodes connected to the promotion position identifier by edges. A node set is defined; the unique hot code corresponding to each collaborating node in the collaborating node set is extracted; a logical OR operation is performed on each unique hot code, and the result is used as a collaborating party locking bitmap, wherein each bit in the collaborating party locking bitmap corresponds to a collaborating party node in the distributed system, the collaborating party node corresponding to the bit with a value of 1 in the collaborating party locking bitmap is the target collaborating party node, and the collaborating party node corresponding to the bit with a value of 0 is a collaborating party node in the distributed system that is not the target of this dynamic compensation processing; the state primitive and the collaborating party locking bitmap are encapsulated into a state projection message; the collaborating party locking bitmap is parsed into a list of network addresses of each target collaborating party node; the state projection message is sent to each target collaborating party node corresponding to the network address list in the User Datagram Protocol multicast mode, and message compression and retransmission confirmation mechanisms are enabled during transmission.

[0012] Beneficial Effects: The solution implemented in this invention encapsulates abnormal event signals into state primitives containing a generalization bit identifier, event type, and scope. This first achieves a direct mapping from abnormal events to a standardized data structure, enabling heterogeneous events to enter the compensation processing flow with a unified data structure, eliminating information parsing delays caused by differences in event formats. Based on this, a collaborative party locking bitmap identifying all target collaborative party nodes is generated according to a preset collaborative audit relationship graph, and multicast is used to trigger the process with state projection messages. This achieves precise delineation and decentralized triggering of the compensation calculation boundary, transforming the traditional serial distribution mode where a central scheduling node notifies each participant one by one into a parallel triggering mode where each target collaborative party node receives state projection messages in parallel and autonomously initiates the solution process. This eliminates the scheduling delay and single-point bottleneck of the central node. Furthermore, through… By using a globally consistent polymorphic compensation utility function, each target collaborating node performs local consistency solving based on its local historical state and state primitives, deterministically deriving the same optimal compensation promotion content. This achieves deterministic determination of the same compensation decision by each node without the participation of a central coordinating node. It transforms the traditional "master-slave" compensation mode, which relies on a central node for unified decision-making, into a "peer-to-peer" consistent solution mode where each node solves independently based on a shared function. This fundamentally eliminates the interaction delay introduced by cross-node state queries and confirmations. Finally, by having each target collaborating node submit the optimal compensation promotion content locally as an atomic state transaction, the content state of the promotion bit is updated collaboratively across nodes. This achieves atomic effectiveness and immediate consistency of the compensation result in the databases of each participating party, avoiding the inconsistency of the state window caused by step-by-step submissions in traditional technologies.

[0013] By combining the above interconnected effects, a four-in-one distributed autonomous state synchronization compensation mechanism is constructed, consisting of "event primitive mapping - bitmap locking projection - consistent solution of utility functions - individual atomic submission".

[0014] Thus, unlike the "process center" compensation mode in traditional technologies that relies on centralized scheduling nodes for process orchestration, this invention transforms abnormal events into common signals that drive the autonomous state consistency solution of each participant. It replaces the decision-making function of the central node with a shared deterministic utility function, transforming the traditional passive serial "process center" mode into an active parallel "distributed autonomous state synchronization mechanism" mode. Ultimately, it achieves low-latency automatic compensation of promotion positions and atomic consistency of multi-party states while eliminating the intervention of distributed coordination nodes, significantly improving the real-time performance, consistency, and scalability of promotion position compensation processing in distributed content management systems. Attached Figure Description

[0015] Figure 1A flowchart illustrating the event-driven dynamic compensation method for promotion positions provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the first sub-process of the event-driven dynamic compensation processing method for promotion positions provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the second sub-process of the event-driven dynamic compensation processing method for promotion positions provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to specific examples. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] This invention provides an event-driven dynamic compensation method for promotion slots. This method can be a data processing engine hosted by a distributed server cluster, applied to the automatic compensation of promotion slots and the synchronization of the status of each participant in an information promotion content management system. It can directly drive information promotion data processing applications, including but not limited to the following multi-dimensional applications: First, in the information promotion content management platform, which integrates multiple functions such as content submission, collaborative review, scheduling and release, and data statistics, when promotion slots become vacant due to reasons such as review rejection, expired materials, or failed pushes, compensation processing is required across multiple participants. By deploying the method of this invention, when any collaborative node detects an abnormal event signal, that collaborative node can directly encapsulate the abnormal event as a state primitive, generate a collaborative node locking bitmap based on the collaborative review relationship graph, and trigger each target collaborative node with state projection message multicast. Each target collaborative node performs local consistency solving based on a globally consistent polymorphic compensation utility function, deterministically derives the same optimal compensation promotion content, and then submits it locally with atomic state transactions, enabling the content state of the promotion slot to be updated collaboratively and consistently among all participants. This method eliminates the interaction delays and single-point bottlenecks caused by relying on centralized compensation scheduling nodes for process orchestration in traditional technologies, providing platform operators with real-time and automated data analysis basis for promotion slot compensation processing, supporting the formulation of precise content scheduling and resource optimization allocation strategies.

[0018] Secondly, in the ad delivery engine, by capturing and projecting abnormal events of the ad placement in real time, multiple collaborating nodes, such as advertisers, the delivery system, and the monitoring party, are triggered to determine the optimal replacement ad content in parallel and update their local status. This not only enables automatic re-delivery at the millisecond level when the ad placement is empty, avoiding traffic loss caused by the empty ad placement, but also provides an accurate exposure start time benchmark for ad delivery performance attribution analysis based on the final consistency of the status data maintained by each collaborating node.

[0019] Third, in the edge nodes of the content delivery network, by converting promotion bit abnormal events into state primitives and performing multicast projection based on the coordinator locked bitmap, each edge cache node can collaboratively and consistently determine the replacement content identifier and then update its local cache state with atomic transactions. This achieves content compensation consistency among edge nodes without the need for intervention from the central scheduling node, providing low-latency, highly available automatic compensation capability for promotion bit states for large-scale content delivery networks.

[0020] Fourth, in the cross-platform information promotion data statistics system, by recording the atomic consistency of the optimal compensation promotion content and its status update timestamps submitted by each collaborating node, the system can obtain a unified exposure statistics switching benchmark among the various data statistics terminals, providing a reliable event time alignment reference for the integration of cross-platform promotion effect data.

[0021] This invention addresses the technical contradiction between the real-time requirements of compensation processing and the difficulty in ensuring consistency of multiple states, providing a reusable distributed state consistency core data processing capability for information promotion content management systems.

[0022] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] Example 1: Event-driven dynamic compensation method for promotion positions.

[0024] In this embodiment, please refer to Figure 1 , Figure 1 This is a flowchart illustrating the event-driven dynamic compensation method for promotion positions provided in an embodiment of the present invention. Figure 1 As shown, in this embodiment, the method includes the following steps S11-S14: S11: In response to any collaborating node in the distributed system detecting an abnormal event signal, the collaborating node generates a state primitive containing the generalization bit identifier, event type, and scope corresponding to the abnormal event signal.

[0025] Promotional slot: refers to a resource location or slot on the display page of promotional content (such as advertisements, notifications, recommendations, etc.) used to carry and display specific promotional content. Each promotional slot has a unique identifier. Promotional slots are the basic resource units in the information promotion content management system, and their status includes occupied (promotional content allocated), vacant (no valid promotional content), locked (being allocated), etc.

[0026] Collaborating nodes refer to independent computing instances or service processes within a distributed information promotion content management system. These nodes participate in the business processes of content submission, review, publication, and data statistics, each maintaining some status data related to its respective promotion slot. They are interconnected and collaborative due to promotion slot compensation processing. Collaborating nodes include, but are not limited to, content providers, review managers, data analysts, and publishers. These nodes communicate via a pre-defined network protocol, are logically independent, and do not have a unified central database.

[0027] Anomaly event signals refer to standardized notification messages issued by any collaborating node when it detects a change in the status of a promotion slot or an anomaly in the promotion content, used to trigger the compensation process. Sources of anomaly event signals include, but are not limited to: "Content Review Rejection" notifications from the review end, "Materials Expired" notifications from the content provider, "Push Failure" notifications from the publishing end, or "Data Anomaly" notifications from the data statistics end.

[0028] State primitives are immutable, atomic, structured data records generated by collaborating nodes that detect abnormal event signals. They are the basic computational unit driving consistent updates of promotion slot states, encapsulating heterogeneous abnormal event signals into standardized event description units, serving as common input for subsequent consistency resolution by various target collaborating nodes. State primitives may include, but are not limited to, the following fields: Promotion slot identifier, used to uniquely identify the abnormal promotion slot; Event type, used to record the specific cause of the abnormal event (e.g., "rejection," "expired material," "push failure," etc.), providing the event context information needed by each target collaborating node to determine compensation strategies during consistency resolution; Scope, used to describe the range of attributes affected by the abnormal event (e.g., the identifier of rejected promotion content, the type of expired material, etc.). State primitives may also include a generation timestamp and an event source node identifier.

[0029] The specific implementation of this step is as follows: First, each collaborating node in the distributed system deploys its own event monitoring module, which continuously listens for state change events within its jurisdiction. When any collaborating node's event monitoring module detects an abnormal event signal, this step of the processing flow is triggered.

[0030] Secondly, the event encapsulation module of the collaborating node that detects the abnormal event signal parses the detected abnormal event signal and extracts the promotion bit identifier field, event type code, and description of the affected attribute range carried in the abnormal event signal. The event type code is a predefined enumerated value, and the attribute range description specifies the specific promotion content identifier or attribute field affected. The detailed parsing and extraction process will be elaborated in subsequent Embodiment 2.

[0031] Then, the event encapsulation module of the aforementioned collaborating node encapsulates the extracted promotion bit identifier field, event type code, and description of the affected attribute range, usually along with the generation timestamp and event source node identifier, into an immutable data record as a state primitive. Here, an immutable data record means that once generated, its content cannot be modified or deleted; subsequent processing can only append to and read from it.

[0032] It should be noted that the generation method of the aforementioned state primitives is not limited to independent generation by the collaborating node that detects the abnormal event signal. In one example, the collaborating node that detects the abnormal event signal can forward the original abnormal event signal to a preset stateless event processing agent, which then encapsulates the state primitive and returns it. In another example, a multi-node redundant encapsulation mechanism can be used, where the collaborating node that detects the abnormal event signal and its neighboring nodes simultaneously generate the state primitives, and the consistency of the state primitives is ensured through comparison. Any technical means that can convert abnormal event signals into standardized structured state primitives falls within the protection scope of this invention.

[0033] For example, consider a scenario of multi-party collaborative review of promotional content management. A distributed system deploys three collaborative nodes: a content provider, a review management system, and a publishing system. One day, the content provider submits promotional content B, which is then reviewed for compliance by the review management system. After review, the review management system rejects the promotional content B. At this point, the event monitoring module of the review management system detects this "content review rejection" anomaly signal, triggering the generation process of the state primitive. The event encapsulation module of the review management system parses this anomaly signal, extracting the promotional slot identifier field Slot_01, the event type code AUDIT_REJECTED, and the description of the affected attribute range {content_id: B, status: blocked}. It can also combine this with the current generation timestamp T1 and the event source node identifier Auditor_Node, encapsulating it into an immutable data record as the state primitive P1.

[0034] In step S11, heterogeneous abnormal event signals are encapsulated into state primitives of a unified format. This transforms the abnormal events into a standardized data structure that can be jointly identified and processed by the subsequent target collaborating nodes. This provides a deterministic common input for the decentralized compensation processing flow and solves the problems of information parsing delay and processing inconsistency caused by differences in event formats.

[0035] S12: The collaborating party node generates a collaborating party locking bitmap that identifies all target collaborating party nodes associated with the promotion bit identifier based on a preset collaborating audit relationship graph, and multicasts the state primitive and the collaborating party locking bitmap as state projection messages to trigger each target collaborating party node to perform local consistency solving.

[0036] The pre-built collaborative review relationship graph refers to a pre-constructed and stored graph data structure used to describe the collaborative relationships between each promotion slot identifier and each collaborating node. In this graph data structure, nodes represent promotion slot identifiers or collaborating node, and edges represent business relationships between the promotion slot and the collaborating node that require collaboration (e.g., the collaborating node needs to participate in the content review, data statistics, or content provision of the promotion slot). The specific construction method of the pre-built collaborative review relationship graph can refer to the relevant technical means of existing graph database construction technologies in this field, and will not be elaborated here.

[0037] Target collaborating node: refers to the collaborating node in the preset collaborative review relationship graph that has an edge connection with the promotion position that has an anomaly, that is, the collaborating node that needs to participate in this dynamic compensation process.

[0038] Collaborator Locking Bitmap: This refers to a fixed-length binary bit sequence used to identify and lock all target collaborator nodes involved in the current dynamic compensation process. Its technical meaning is that by using specific bit values ​​(1 or 0) in the bitmap, the range of collaborator nodes participating in this consistency solution and atomic state transaction submission is pre-defined and immutably defined. This replaces the traditional dynamic determination method where a central scheduling node notifies each participant individually, achieving decentralized one-time locking of the computational boundary. The bit width of the Collaborator Locking Bitmap is pre-determined by the total number of collaborator nodes in the distributed system, with each bit corresponding to a specific collaborator node in the distributed system. A bit with a value of 1 indicates that the corresponding collaborator node is a target collaborator node in this dynamic compensation process, while a bit with a value of 0 indicates that the corresponding collaborator node is a non-target collaborator node in this dynamic compensation process.

[0039] State projection messages are network transmission messages that encapsulate state primitives and collaborator lock bitmaps. They are used to "project" abnormal event information and computational boundaries to all target collaborator nodes in a single, parallel manner, replacing the traditional serial notification method where a central scheduling node sends instructions to each participant individually. The "projection" action itself does not require confirmation or response from the receiver. Once the sender completes the projection, the trigger function ends, initiating each target collaborator node to autonomously and in parallel initiate a local consistency solution process based on the same state primitives and collaborator lock bitmaps carried in the message.

[0040] Local consistency solution: This refers to the process where each target collaborating node, after receiving the state projection message, independently runs a globally consistent polymorphic compensation utility function locally. It performs calculations based on its local historical state and state primitives to deterministically derive the same optimal compensation promotion content. "Local" means that each target collaborating node executes independently in its own computing environment, without needing to negotiate or interact with other target collaborating nodes.

[0041] The specific implementation of this step is as follows: First, the bitmap generation module of the collaborating node that detects the abnormal event signal obtains a preset collaborative review relationship graph. This graph stores the associations between each promotion bit identifier and the collaborating node using a graph data structure. Each collaborating node is pre-assigned a unique hotkey to identify itself. The bit width of the unique hotkey is determined by the total number of collaborating nodes in the distributed system, and the position of the bit with a value of 1 in the unique hotkey corresponds to the node's index in the distributed system. The specific process of generating the unique hotkey and constructing the bitmap will be detailed in the subsequent embodiment 4.

[0042] Secondly, the bitmap generation module performs subgraph matching in the preset collaborative review relationship graph based on the promotion bit identifier carried by the state primitive, and obtains a set of collaborative party nodes that are connected to the promotion bit identifier by an edge. This set of collaborative party nodes is the set of target collaborative party nodes for this dynamic compensation process.

[0043] Then, the bitmap generation module extracts the one-hot code corresponding to each collaborating node in the collaborating node set, performs a logical OR operation on each one-hot code, and uses the result as the collaborating node locking bitmap. Each bit in the collaborating node locking bitmap corresponds to a collaborating node in the distributed system. The collaborating node corresponding to a bit with a value of 1 is the target collaborating node, and the collaborating node corresponding to a bit with a value of 0 is a collaborating node in the distributed system that is not the target of this dynamic compensation processing.

[0044] Then, the message encapsulation module of the collaborating node encapsulates the state primitives and the collaborating party lock bitmap into a state projection message.

[0045] Then, the message sending module of the collaborating node parses the collaborating locking bitmap into a list of network addresses for each target collaborating node. It can then use User Datagram Protocol (UDP) multicast to send the state projection message to each target collaborating node corresponding to its network address, enabling message compression and retransmission acknowledgment mechanisms during transmission. The specific implementation of UDP multicast can be found in existing network communication protocols and will not be elaborated upon here. The choice of UDP multicast over TCP unicast is based on the following technical considerations: First, UDP multicast can deliver messages to multiple target nodes in a single transmission, decoupling communication overhead from the number of target nodes and significantly reducing network load in high-concurrency compensation scenarios. Second, the connectionless nature of the UDP protocol eliminates the delay in connection establishment and termination, reducing the delivery delay of the state projection message to the millisecond level. Third, supplemented by an application-layer retransmission acknowledgment mechanism, it can compensate for the inherent unreliability of the UDP protocol while maintaining low latency, achieving a reliable message delivery rate.

[0046] It should be noted that the generation method of the aforementioned collaborative party locking bitmap is not limited to logical OR operations based on one-hot codes. In one example, a preset list of node identifiers can be directly mapped to a binary bit sequence, and the bitmap can be generated by looking up a table. In another example, a compressed bitmap data structure (such as a Roaring Bitmap) can be used, which is suitable for scenarios where the number of collaborative party nodes is large and the target node is sparsely distributed. Any technical means that can compactly identify the target collaborative party node set is within the protection scope of this invention.

[0047] For example, following the previous example, the bitmap generation module of the review management end (i.e., the collaborating node that detects the abnormal event signal) obtains the preset collaborative review relationship graph. Assume there are 8 collaborating nodes in the distributed system, with node numbers N0 to N7. The promotion bit Slot_01 is associated with the content provider (N1), the review management end (N2), and the publishing end (N3). The bitmap generation module performs subgraph matching based on Slot_01 to obtain the collaborating node set {N1, N2, N3}. The unique hotkeys of each node (N1: 00000010, N2: 00000100, N3: 00001000) are extracted and logically ORed to obtain the collaborating locking bitmap 00001110. The message encapsulation module encapsulates the state primitive P1 and the collaborating locking bitmap 00001110 into a state projection message M1. The message sending module parses the collaborator lock bitmap into a list of network addresses N1, N2, and N3, and sends M1 to the three target collaborator nodes via UDP multicast, triggering each target collaborator node to perform local consistency solving.

[0048] In step S12, a collaborative party locking bitmap is generated based on a preset collaborative audit relationship graph, and multicast is triggered by a state projection message. This transforms the serial distribution mode in traditional technology, where the central scheduling node notifies each participant one by one, into a parallel triggering mode where each target collaborative party node receives the projection message in parallel and starts solving autonomously. This provides accurate computational boundaries and synchronous start-up opportunities for the subsequent local consistency solution of each target collaborative party node, and solves the problem of delayed compensation processing caused by the scheduling delay of the central node.

[0049] S13: Each of the target collaborating nodes, based on its local historical state and the state primitive, performs a local consistency solution based on a globally consistent polymorphic compensation utility function to deterministically determine the optimal compensation promotion content corresponding to the promotion bit identifier.

[0050] Local historical status: This refers to the collection of historical status data related to the ad placement maintained by each target collaborating node in its local storage. Since each target collaborating node plays a different business role in the distributed system and maintains status data of different dimensions, the local historical status of each target collaborating node is usually different in terms of data content, data dimensions, and data structure. For example, the content provider maintains the sequence of ad content that has been placed on this ad placement and the material status of each ad content (e.g., valid, expired, etc.); the review management maintains the historical review records of each ad content on this ad placement (e.g., approval time, rejection reason classification, etc.); and the data statistics maintains historical exposure, click-through rate, conversion rate, and other statistical indicators for each ad content on this ad placement.

[0051] The globally consistent polymorphic compensation utility function is a composite weighted evaluation function deployed with the same version and parameter configuration on all target collaborating nodes. It is used to comprehensively score candidate promotional content across multiple dimensions to deterministically select the globally optimal compensation promotional content. The polymorphic compensation utility function is a composite weighted function, and its calculation factors may include content quality factors, constraint satisfaction factors, and state continuity factors. The specific calculation methods for each factor will be detailed in subsequent embodiments 5, 6, and 7. The global consistency of the polymorphic compensation utility function is maintained through version synchronization by a pre-defined distributed configuration center; the specific synchronization mechanism will be detailed in subsequent embodiment 8.

[0052] The deterministic conclusion is that each target collaborating node, based on the same globally consistent polymorphic compensation utility function, inputs the same state primitives and candidate promotion content set. Although their local historical states may be different, the polymorphic compensation utility function is designed as a weighted structure in which the final score ranking is dominated by the common input and the local historical state only plays a minor role. This ensures that each target collaborating node will inevitably obtain the same optimal solution after independent calculation, without the need for inter-node negotiation or consensus protocol.

[0053] The optimal compensation promotion content refers to the candidate promotion content with the highest total utility score that is deterministically selected from the set of candidate promotion content by each target collaborating node after running the polymorphic compensation utility function. This optimal compensation promotion content will be used to replace the promotion content that has malfunctioned, thus completing the compensation process.

[0054] The specific implementation of this step is as follows: First, after receiving the state projection message, each target coordinating node starts a consensus solving process in its local compensation solver.

[0055] Secondly, each target collaborating node obtains a set of candidate promotional content. The methods for obtaining the set of candidate promotional content include, but are not limited to: reading all currently valid promotional content that meets the basic attribute requirements of the promotional slot (such as the size and position of the promotional slot) from the promotional content pool maintained by the content provider, and using it as the set of candidate promotional content.

[0056] Then, for each candidate promotional content in the candidate promotional content set, the compensation solver of each target collaborating node calculates the content quality factor, constraint satisfaction factor, and state connectivity factor. The specific factor calculation process will be detailed in subsequent embodiments 5, 6, and 7.

[0057] Then, the compensation solver of each target collaborating node performs a weighted sum of the content quality factor, constraint satisfaction factor, and state connection factor to obtain the total utility score of the candidate promotion content. The weight coefficients of each factor are preset weight vectors, with the constraint satisfaction factor having a larger weight in the total utility score, while the state connection factor only plays a minor adjustment role.

[0058] Finally, the compensation solver of each target collaborating node determines the candidate promotion content with the highest total utility score as the optimal compensation promotion content corresponding to the promotion bit identifier. Since the compensation solver of each target collaborating node runs a globally consistent polymorphic compensation utility function, and the inputs are the same state primitives and the same set of candidate promotion content, each target collaborating node will inevitably obtain the same total utility score ranking and the same optimal compensation promotion content after independent calculation.

[0059] It should be noted that the method of obtaining the aforementioned candidate promotion content set is not limited to reading it directly from the content provider. In one example, the candidate promotion content set can be obtained and encapsulated into the state projection message when the collaborating node that detects the abnormal event signal generates the state projection message, and each target collaborating node can directly parse it from the state projection message. In another example, each target collaborating node can obtain the candidate promotion content set from its own locally cached copy of the promotion content. Any technical means that enables each target collaborating node to obtain the same candidate promotion content set is within the protection scope of this invention.

[0060] For example, following the previous example, after receiving the state projection message M1, the content provider (N1), the review management (N2), and the publishing (N3) each start their local compensation solvers. Each compensation solver obtains a set of candidate promotional content (assuming it includes content C, content D, and content E). For content C, the content quality factor (e.g., Q(C) = 0.85), constraint satisfaction factor (e.g., S(C) = 0.90), and state connection factor (e.g., A(C) = 0.75) are calculated respectively, and the weighted sum is used to obtain a total utility score of 0.82. For content D, the total utility score is 0.76; for content E, the total utility score is 0.70. Content C has the highest total utility score, and the compensation solvers of each target collaborating node deterministically determine that the optimal compensation promotional content for promotion slot Slot_01 is content C.

[0061] In step S13, each target collaborating node performs a local consistency solution based on a globally consistent polymorphic compensation utility function and deterministically derives the same optimal compensation promotion content. This transforms the traditional "master-slave" compensation mode, which relies on a central node for unified decision-making, into a "peer-to-peer" consistent solution mode where each target collaborating node solves independently based on a shared function. This solves the interaction delay problem introduced by cross-node state query and confirmation, and achieves deterministic autonomous consistency in the compensation decisions of each target collaborating node without the participation of a central coordinating node.

[0062] S14: Each of the target collaborating nodes submits the optimal compensation promotion content locally as an atomic state transaction, so that the content status of the promotion bit corresponding to the promotion bit identifier in the local area of ​​each target collaborating node is consistently updated to the optimal compensation promotion content.

[0063] Atomic state transactions refer to a sequence of database operations in a Database Management System (DBMS) that possess the ACID properties of atomicity, consistency, isolation, and durability. In this invention, atomic state transactions ensure that when each target collaborating node updates the status data related to promotion slots, the update operations involving multiple fields such as promotion content fields, review status fields, and exposure count fields are executed as an indivisible whole. Either all operations are successfully committed, or all are rolled back, thereby guaranteeing the consistency of the local state of each target collaborating node.

[0064] The specific implementation of this step is as follows: First, after each target collaborating node completes the local consistency solution and obtains the optimal compensation promotion content, each node starts an atomic state transaction in its local database management system.

[0065] Secondly, within this atomic state transaction, each target collaborating node updates the value of the current promotion content field corresponding to the promotion bit identifier to the identifier of the optimal compensation promotion content. Furthermore, each target collaborating node, based on its own role, synchronously updates other state fields related to the promotion bit. The specific multi-field atomic update operation will be detailed in subsequent embodiment 9.

[0066] Then, each target collaborating node submits the atomic state transaction. If the atomic state transaction is successfully submitted, the state of the promotion position in the local database of each target collaborating node is synchronously updated to the optimal compensation promotion content. The state data related to the promotion position maintained by each target collaborating node all point to the same optimal compensation promotion content, thus achieving consistent state updates of the promotion position at the distributed level, and the compensation process is completed.

[0067] Explanation regarding the consistent local updates of promotional content status across all target collaborating nodes: In this embodiment, the distributed system does not have a unified central database. The content status of the promotion slots is not stored as an independent data entity in a central node, but is logically distributed in the local databases of each target collaborating node. Each target collaborating node maintains different dimensions of status data related to the promotion slots according to its role in the distributed system. For example, the content provider maintains the occupancy relationship between the promotion slots and the promotional content, the review management maintains the review status associated with the promotion slots, and the data statistics maintains the exposure statistics benchmark associated with the promotion slots.

[0068] When each target collaborating node commits the same optimal compensation promotion content as an atomic state transaction locally, the state data related to the promotion slot in the local database of each target collaborating node all point to the optimal compensation promotion content. At this point, the content state of the promotion slot at the distributed layer is consistently updated to the optimal compensation promotion content. In other words, the consistent update of the content state of the promotion slot itself is naturally achieved through the consistent update of the local state data of each target collaborating node. Both describe different perspectives of the same technical fact—the former describes it from the overall perspective of the promotion slot, while the latter describes it from the individual perspective of each target collaborating node.

[0069] It should be noted that the above-mentioned atomic state transaction submission method is not limited to independent submission by each target collaborating node. In one example, a two-phase commit protocol can be used to coordinate the atomic state transaction submission of each target collaborating node, but this is an optional implementation and not a necessary technical feature of this invention. Any technical means that enables each target collaborating node to complete atomic updates locally is within the protection scope of this invention.

[0070] For example, following the previous example, the content provider (N1) initiates an atomic state transaction locally, updating the current promotion content field of the promotion slot Slot_01 to the identifier of content C, and simultaneously updating the usage status of content C to "occupied"; the review management end (N2), in its local atomic state transaction, updates the review status field corresponding to the promotion slot Slot_01 to "pending review"; the publishing end (N3), in its local atomic state transaction, resets the exposure count field of the promotion slot Slot_01 to zero and updates the statistics start timestamp to the current system time T2. After the three target collaborating nodes each submit their local atomic state transactions, the content status of the promotion slot Slot_01 is collaboratively updated to content C, and the compensation process is completed.

[0071] In step S14, each target collaborating node submits the optimal compensation promotion content locally in an atomic state transaction, enabling the content status of the promotion position to be updated collaboratively across all target collaborating nodes. This achieves atomic effectiveness and immediate consistency of the compensation results in the databases of all participating parties, solving the problem of inconsistent state windows caused by step-by-step submissions in traditional technologies.

[0072] In this embodiment, through sequential processing in steps S11-S14, the abnormal event signal is encapsulated into a state primitive; each target coordinating node is triggered by the coordinating party locking bitmap and state projection message; each target coordinating node performs local consistency solution based on a globally consistent polymorphic compensation utility function and deterministically obtains the same optimal compensation promotion content; and each node submits its atomic state transaction locally to coordinate the update of the promotion position's content state. This fully realizes the event-driven dynamic compensation processing method for promotion positions. Its core lies in: transforming abnormal events into a common signal that drives the autonomous state consistency of each participating party; replacing the decision-making function of the central node with a shared deterministic utility function; and transforming the traditional passive serial "process center" compensation mode into an active parallel "distributed autonomous state synchronization mechanism" compensation mode. This provides a core technical solution for dynamic compensation processing of promotion positions in a distributed content management system.

[0073] Example 2: Specific implementation of structured encapsulation and immutable writing of state primitives.

[0074] In this embodiment, please refer to Figure 2 , Figure 2 This is a schematic diagram of the first sub-process of the event-driven dynamic compensation method for promotion positions provided in an embodiment of the present invention. Figure 2 As shown, in this embodiment, a state primitive is generated that includes the generalization bit identifier, event type, and scope corresponding to the abnormal event signal, including: S21: Analyze the abnormal event signal and extract the promotion bit identifier field, event type code, and description of the affected attribute range carried in the abnormal event signal; S22: Encapsulate the promotion bit identifier field, the event type code, and the description of the affected attribute range into an immutable data record as a state primitive, wherein the immutable data record further includes a generation timestamp and an event source node identifier; S23: Calculate the hash value of each field in the state primitive, and embed the hash value as the data fingerprint of the state primitive into the state primitive; S24: The embedded state primitive is appended to an append-only state log, and each record in the append-only state log is linked together by the hash value of the previous record.

[0075] Data fingerprint: In this invention, a data fingerprint refers to a fixed-length digital digest generated by hashing the content of each field in a state primitive using a preset hash function. This digest uniquely identifies the current version of the state primitive. Any modification to the content of the state primitive will result in a change to the data fingerprint; therefore, the data fingerprint can be used to verify the integrity and immutability of the state primitive. The specific implementation of the hash operation can refer to existing cryptographic hash algorithms in the field, and will not be elaborated here. In this invention, SHA-256 can be selected as the preset hash function, based on the following technical considerations: First, SHA-256 has strong collision resistance; the probability of different state primitive contents generating the same hash value is computationally infeasible, ensuring the uniqueness of the data fingerprint. Second, SHA-256 outputs a fixed 256-bit digest, facilitating storage and comparison. Third, SHA-256 is widely standardized and its computational overhead is controllable; the real-time performance of the state primitive encapsulation will not be significantly reduced due to hash operations.

[0076] Append-only state logs refer to a data storage structure that only supports appending new records to the end of the log and does not support modifying or deleting existing records. Each record in an append-only state log is linked together through the hash value of the previous record, forming an integrity verification chain that traces back from the latest record to the beginning of the log. Specific implementation methods for append-only state logs can be found in existing log structure storage technologies (such as log structure merge trees), and will not be elaborated upon here.

[0077] The specific implementation of this embodiment is as follows: First, the event encapsulation module of the collaborating node that detects the abnormal event signal parses the abnormal event signal, extracting the promotion bit identifier field, event type code, and description of the affected attribute range carried in the abnormal event signal. The parsing method can be based on field matching using a preset abnormal event signal template.

[0078] Secondly, the event encapsulation module encapsulates the extracted promotion bit identifier field, event type code, and description of the affected attribute range, along with the generated timestamp and event source node identifier, into an immutable data record, which serves as a state primitive.

[0079] Then, the event encapsulation module calculates the hash value of each field in the state primitive. Specifically, the fields of the state primitive are serialized into a byte stream in a preset order, the serialized byte stream is input into the SHA-256 hash function to obtain a 256-bit hash value, and the hash value is used as the data fingerprint of the state primitive and embedded in the data fingerprint field of the state primitive.

[0080] Finally, the event encapsulation module appends the state primitives with embedded data fingerprints to an append-only state log. Specifically, the state log maintenance module obtains the hash value of the last record in the append-only state log, serializes the state primitives, and writes them together with the hash value of the last record as a new record to the end of the log. The hash value of the new record is then used for chaining association with the next record.

[0081] It should be noted that the encapsulation method for the aforementioned immutable data records is not limited to embedding SHA-256 hash values. In one example, a digital signature method can be used, where a collaborating node signs the state primitive using its private key, and the signature value serves as an immutable credential. In another example, a Merkle tree structure can be used, organizing multiple state primitives into leaf nodes of a Merkle tree, with the Merkle root serving as proof of aggregate integrity. Any technical means that can guarantee the immutability and verifiability of state primitives falls within the protection scope of this invention.

[0082] For example, following the previous example, the event encapsulation module of the audit management end parses the abnormal event signal, extracts the promotion slot identifier field Slot_01, the event type code AUDIT_REJECTED, and the description of the affected attribute range {content_id: B, status: blocked}, and combines them with the generated timestamp T1 and the event source node identifier Auditor_Node to encapsulate it into a state primitive P1. The SHA-256 hash value of each field is calculated to obtain the data fingerprint "a3b2c1d4e5f6...". The data fingerprint is embedded into the corresponding field of P1, and the embedded P1 is appended to the append-only state log. The hash value of the record preceding the write position of P1 is "H_prev", and the write of P1 includes the content of P1 and "H_prev", with the hash value of P1 being "H_P1".

[0083] In this embodiment, by encapsulating state primitives as immutable data records and embedding data fingerprints, and appending them to an append-only state log in a hash chain-like manner, an immutable verification chain and a complete traceability path for state primitives are constructed. Thus, based on the mapping of events to state primitives in Embodiment 1 above, the anti-counterfeiting and integrity protection of state primitives are further realized, solving the problem that the basis for compensation processing is unreliable due to the possibility of tampering or loss of state primitive content, and significantly improving the data credibility and auditability of compensation processing.

[0084] Example 3: Specific implementation of event preprocessing and intelligent filtering.

[0085] In this embodiment, before generating a state primitive containing the generalization bit identifier, event type, and scope corresponding to the abnormal event signal, the method further includes: Obtain the abnormal event signal; Extract the text description information and structured fields from the abnormal event signals; The text description information is converted into a word vector sequence through a preset word embedding matrix, and then concatenated with the encoded vector of the structured field to obtain the event feature vector; The event feature vector is input into a preset binary classification fully connected network to obtain the confidence level that the abnormal event signal belongs to a preset target event category that needs to trigger dynamic compensation processing. When the confidence level exceeds a preset confidence level threshold, the event category of the abnormal event signal is determined to be the target event category, and the step of generating the state primitive is executed.

[0086] A pre-trained word embedding matrix refers to a parameter matrix that maps words in natural language text to dense vectors of a fixed dimension. In this invention, the pre-trained word embedding matrix is ​​used to convert textual description information in abnormal event signals into numerical vectors for subsequent processing by a fully connected binary classification network. The pre-trained word embedding matrix can be obtained by extracting textual description information as training data from large-scale information dissemination corpora, such as historical abnormal event logs accumulated in a distributed system, and pre-training it using word embedding training methods known in the art (such as Word2Vec, GloVe, or FastText). In this invention, word embedding matrices are chosen over one-hot encoding or term frequency-inverse document frequency (TF-IDF) as the text vectorization method based on the following technical considerations: First, word embedding matrices can capture the semantic similarity between words, making descriptions of semantically similar abnormal events closer in the vector space, which helps the binary classification network learn more generalizable event category boundaries; Second, word embedding matrices output dense vectors of fixed dimensions, typically set to 100 to 300 dimensions, preferably 200 dimensions. Compared to the high-dimensional sparse vectors of one-hot encoding, this significantly reduces the input dimension of the subsequent fully connected binary classification network, improving the training efficiency and inference speed of the model. The specific word embedding training process can be found in existing natural language processing techniques in this field, and will not be elaborated upon here.

[0087] Event feature vector: This refers to a fixed-length numerical vector formed by concatenating a sequence of word vectors obtained by transforming the textual description information in the abnormal event signal through a word embedding matrix with the encoded vector of the structured field. The event feature vector is used as input to a pre-defined binary classification fully connected network for event classification.

[0088] The pre-defined binary classification fully connected network refers to a pre-trained feedforward neural network used to map event feature vectors to binary classification confidence scores. The structure of this feedforward neural network may include an input layer, at least one hidden layer, and an output layer. The output layer may employ a sigmoid activation function, outputting a confidence value between 0 and 1. The training samples for the pre-defined binary classification fully connected network are a pre-labeled set of historical abnormal event signal samples, with each sample labeled as either "belonging to the target event category" or "not belonging to the target event category." Specific network structure design and training processes can refer to existing neural network model building techniques in this field, and will not be elaborated upon here.

[0089] Target event category: This refers to a predefined set of abnormal event categories that require dynamic compensation processing. An abnormal event belonging to the target event category is any event whose occurrence will result in an empty promotion slot, requiring compensation processing. Once such an event occurs, it will enter the generation of state primitives and subsequent compensation processing flow. Abnormal event signals that do not belong to the target event category (such as non-critical log records, automatically recovered transient fluctuation events, etc.) will be filtered out and will not trigger subsequent compensation processing flow.

[0090] The specific implementation of this embodiment is as follows: First, the event filtering module of the collaborating node that detects the abnormal event signal obtains the abnormal event signal.

[0091] Secondly, the event filtering module extracts text description information and structured fields from the abnormal event signals. The text description information consists of the text content in natural language describing the event details (such as the reason for rejection); the structured fields are the event attributes in key-value pairs (such as the event source module and associated promotional slot type).

[0092] Then, the event filtering module converts the text description information into a word vector sequence using a preset word embedding matrix. Specifically, after segmenting the text description information into words, it queries the preset word embedding matrix to obtain the word vector corresponding to each word, arranges the word vectors in chronological order to obtain a word vector sequence, and performs average pooling on the word vector sequence to obtain the text feature vector.

[0093] Then, the event filtering module encodes the structured fields. Specifically, for the value of each structured field, if it is a categorical field, it can be converted into a binary vector using one-hot encoding; if it is a numerical field, it can be mapped to the 0-1 range using a preset normalization function. The encoded vectors of each structured field are concatenated to obtain the structured feature vector.

[0094] Then, the event filtering module concatenates the text feature vector and the structural feature vector to obtain the event feature vector.

[0095] Then, the event filtering module inputs the event feature vector into a preset binary classification fully connected network. The preset binary classification fully connected network consists of an input layer, a hidden layer (e.g., 128 neurons with ReLU activation function), and an output layer (e.g., 1 neuron with Sigmoid activation function) connected in sequence. The input dimension is the same as the event feature vector dimension, and the output is a confidence value between 0 and 1.

[0096] Then, the event filtering module compares the confidence level with a preset confidence threshold (typically 0.75, 0.80, or 0.85, preferably 0.80). When the confidence level exceeds the preset confidence threshold, the event category of the abnormal event signal is determined to be the target event category, and the step of generating state primitives is initiated; otherwise, the abnormal event signal is filtered out, and the subsequent compensation processing flow is not triggered.

[0097] It should be noted that the selection of the above event classification model is not limited to the architecture of word embedding plus fully connected network. In one example, a pre-trained language model based on the Transformer architecture (such as a lightweight variant of BERT) can be used, which is suitable for scenarios where the semantics of abnormal event signals are complex and require deep semantic understanding; in another example, a gradient boosting decision tree (such as XGBoost) can be used, which is suitable for scenarios where abnormal event signals have rich structured features. Any technical means that can classify abnormal event signals to determine whether compensation processing needs to be triggered is within the scope of protection of this invention.

[0098] Furthermore, in this embodiment, "event category" and "event type" contained in the aforementioned state primitive are two concepts at different levels, serving different functions in the technical process. Specifically, the event category refers to the output result of the classification model performing binary classification on the abnormal event signal, used to determine whether the abnormal event belongs to a preset target event category that needs to trigger dynamic compensation processing. Its semantics focus on the triggering decision of "whether compensation needs to be triggered." On the other hand, the event type refers to an information field encapsulated in the state primitive, used to record the specific cause of the abnormal event, such as "rejection of review," "expired material," "failed push," etc. Its semantics focus on the information record of "what specifically caused the abnormal event," providing the event context information required by each target collaborating node to determine the compensation strategy during consistency solving. When the classification model determines that the event category of the abnormal event signal is the target event category, it enters the step of generating the state primitive. During the generation of the state primitive, the specific event cause information carried by the abnormal event signal is extracted and written into the event type field of the state primitive.

[0099] For example, following the previous example, the event filtering module of the review management end obtains the abnormal event signal, extracts the text description information "Promotional content B was rejected due to containing illegal advertising language" and the structured field {Event Source: Review Module, Associated Promotional Position Type: Information Flow Promotional Position}. The text description information is transformed through a preset word embedding matrix (200 dimensions) and average pooled to obtain the text feature vector V_text (200 dimensions). The structured field is encoded to obtain the structure feature vector V_struct (e.g., 10 dimensions). V_text and V_struct are concatenated to obtain the event feature vector V_event (210 dimensions). V_event is input into a preset binary classification fully connected network, and the output confidence score is 0.92. Since the confidence score of 0.92 exceeds the preset confidence threshold of 0.80, the event category is determined to be the target event category, and the process proceeds to the step of generating state primitives.

[0100] In this embodiment, by introducing an intelligent event filtering step based on word embedding and binary classification network before generating state primitives, the triggering decision of abnormal event signals is changed from unconditional full triggering to intelligent classification triggering based on semantic understanding. Thus, based on the event-driven compensation achieved in Embodiment 1 above, it further realizes the active filtering of non-compensation events, solves the problem of system resource waste and invalid compensation operations caused by triggering compensation processing for all abnormal events, and significantly improves the resource utilization efficiency and response accuracy of the compensation processing system.

[0101] Example 4: Specific implementation of cooperative locking bitmap generation and reliable multicast projection.

[0102] In this embodiment, please refer to Figure 3 , Figure 3 This is a schematic diagram of the second sub-process of the event-driven dynamic compensation method for promotion positions provided in an embodiment of the present invention. Figure 3 As shown, in this embodiment, a collaborator locking bitmap is generated, identifying all target collaborator nodes associated with the promotion bit identifier. The state primitive and the collaborator locking bitmap are then multicast as state projection messages, including: S31: Obtain a preset collaborative review relationship graph. The preset collaborative review relationship graph stores the association between each promotion bit identifier and the collaborative node in a graph data structure. Each collaborative node is pre-assigned a unique hot code to identify itself. The bit width of the unique hot code is determined by the total number of collaborative nodes in the distributed system. The position of the bit with a value of 1 in the unique hot code is fixed to the node of the collaborative node in the distributed system. S32: Based on the promotion position identifier, perform subgraph matching in the preset collaborative review relationship graph to obtain a set of collaborative party nodes that are connected to the promotion position identifier by an edge; S33: Extract the unique hot code corresponding to each collaborative node in the collaborative node set; S34: Perform a logical OR operation on each of the one-hot codes and use the result as a collaborator locking bitmap. Each bit in the collaborator locking bitmap corresponds to a collaborator node in the distributed system. The collaborator node corresponding to the bit with a value of 1 in the collaborator locking bitmap is the target collaborator node, and the collaborator node corresponding to the bit with a value of 0 is a collaborator node in the distributed system that is not the target of this dynamic compensation processing. S35: Encapsulate the state primitive and the cooperative locking bitmap into a state projection message; S36: Parse the cooperative party locking bitmap into a list of network addresses for each of the target cooperative party nodes; S37: The state projection message is sent to each of the target cooperating nodes corresponding to the network address list via User Datagram Protocol multicast, and message compression and retransmission confirmation mechanisms are enabled during transmission.

[0103] One-hot code: A one-hot code is a set of binary codewords, where each codeword has exactly one 1 bit and all other bits are 0. In this invention, one-hot codes are used to assign a unique digital identity to each collaborating node in a distributed system. This digital identity occupies a fixed position in the full bitmap. The bit width of the one-hot code is determined by the total number N of collaborating nodes in the distributed system, meaning each one-hot code is an N-bit binary number. The one-hot code is assigned as follows: the N collaborating nodes are numbered sequentially from 0 to N-1. The one-hot code for the collaborating node numbered i is an N-bit binary number where the i-th bit is 1 and all other bits are 0.

[0104] The specific implementation of this embodiment is as follows: First, the bitmap generation module of the collaborating node that detects the abnormal event signal obtains the preset collaborative review relationship graph. The preset collaborative review relationship graph is stored in a graph database or in-memory graph structure as a graph data structure. The node types in the graph include promotion position nodes and collaborating node nodes, and the edge type is "association", indicating that the collaborating node participates in the business process of the promotion position.

[0105] Secondly, the bitmap generation module uses the promotion bit identifier as the query condition to perform subgraph matching in the preset collaborative review relationship graph. The subgraph matching process is as follows: starting from the promotion bit node corresponding to the promotion bit identifier, it expands outward one hop along the "association" edge, collects all neighboring nodes of type collaborative node, and obtains the collaborative node set.

[0106] Then, the bitmap generation module extracts the one-hot code field value from the metadata of each collaborating node in the collaborating node set to obtain each one-hot code.

[0107] Then, the bitmap generation module performs a bitwise logical OR operation on each one-hot code. The rule for the logical OR operation is: when two binary bits are logically ORed, the result is 1 if at least one of them is 1; the result is 0 if both are 0. Since only one bit in each one-hot code is 1 and the position of the 1 bit in each code is different, the result of the logical OR operation is a sequence of binary bits in which the corresponding bits of all target collaborating nodes are 1 and the rest are 0, i.e., the collaborating party locking bitmap.

[0108] Then, the bitmap generation module encapsulates the state primitives and the collaborator lock bitmap into a state projection message. The message body structure of the state projection message includes a message header (containing a message type identifier and a message sequence number) and a message body (containing state primitive fields and collaborator lock bitmap fields).

[0109] Then, the message sending module parses the collaborator locking bitmap into a list of network addresses for each target collaborator node. The parsing process is as follows: iterate through each bit of the collaborator locking bitmap. If the bit is 1, look up the collaborator node identifier corresponding to that bit through a preset mapping table between bit positions and node identifiers. Then, look up the corresponding network address through a preset mapping table between node identifiers and network addresses, and add the network address to the address list.

[0110] Finally, the message sending module sends the status projection message to each target collaborator node corresponding to the network address list via UDP multicast. During sending, message compression (using a preset compression algorithm, such as LZ4 or Snappy, to compress the message body to reduce the amount of data transmitted over the network) and a retransmission acknowledgment mechanism are enabled (a timer is started after sending; if no acknowledgment response is received from the target collaborator node within a preset timeout period, automatic retransmission is performed).

[0111] It should be noted that the reliable delivery mechanism for UDP multicast described above is not limited to application-layer retransmission acknowledgments. In one example, forward error correction (FEC) coding can be used, where the state projection message is redundantly encoded at the sending end, and the receiving end recovers the lost data packets by decoding, avoiding retransmission delays. In another example, a retransmission method based on negative acknowledgments can be used, where the receiving end only sends a retransmission request when a data packet loss is detected. Any technical means that can provide reliable delivery guarantees based on UDP multicast falls within the protection scope of this invention.

[0112] For example, following the previous example, the bitmap generation module of the review management end obtains a preset collaborative review relationship graph. This graph contains 8 collaborative nodes with an 8-bit unique hotkey width. The unique hotkey for node N1 (content provider) is 00000010, for N2 (review management end) it is 00000100, and for N3 (publisher) it is 00001000. The bitmap generation module performs subgraph matching using Slot_01 to obtain the collaborative node set {N1, N2, N3}. After extracting the unique hotkeys, it performs a logical OR operation: 00000010 OR 00000100 OR 00001000 = 00001110, which is the collaborative locking bitmap. The message encapsulation module encapsulates the state primitive P1 and the collaborative locking bitmap 00001110 into a state projection message M1. The message sending module resolves 00001110 into a list of network addresses N1, N2, and N3, and sends it to M1 via UDP multicast, using LZ4 compression and a maximum of 3 retransmission confirmations.

[0113] In this embodiment, a cooperative locking bitmap is generated by subgraph matching based on graph data structure and one-hot code logical OR operation. Reliable delivery of state projection messages is achieved by using UDP multicast supplemented by compression and retransmission confirmation mechanisms. A complete decentralized projection triggering mechanism is constructed. Based on the boundary determination and multicast triggering achieved in Embodiment 1 above, the accurate and efficient encoding of computational boundaries and reliable, low-latency delivery of projection messages are further realized. This solves the problem of omission or erroneous triggering of compensation processing caused by ambiguous computational boundaries and unreliable message delivery, and significantly improves the reliability of the compensation processing system and the efficiency of network communication.

[0114] Example 5: Specific implementation of the solution framework for the polymorphic compensation utility function.

[0115] In this embodiment, the scope of the state primitive includes constraints; based on a globally consistent polymorphic compensation utility function, a local consistency solution is performed to deterministically derive the optimal compensation promotion content corresponding to the promotion bit identifier, including: Obtain a collection of candidate promotional content; For each candidate promotion content in the candidate promotion content set, calculate the content quality factor of the candidate promotion content, the constraint satisfaction factor between the candidate promotion content and the constraint conditions carried by the state primitive, and the state connection factor between the candidate promotion content and the local historical state. The total utility score of the candidate promotion content is obtained by weighted summing of the content quality factor, the constraint satisfaction factor and the state connection factor. The candidate promotion content with the highest total utility score is determined as the optimal compensation promotion content corresponding to the promotion position identifier.

[0116] Content Quality Factor: This refers to a numerical indicator used to quantitatively evaluate the quality level of candidate promotional content. The content quality factor can be calculated using existing content quality assessment models in this field. Its input consists of the material attributes of the candidate promotional content (such as material resolution, copy length, historical click-through rate, historical conversion rate, etc.), and its output is a normalized quality score between 0 and 1. The specific calculation method for the content quality factor can refer to relevant technical means of existing content quality assessment technologies in this field and be applied to the quality assessment scenario of promotional content; it will not be elaborated further here.

[0117] Constraint satisfaction factor: refers to a numerical index used to quantitatively evaluate the degree of matching between candidate promotion content and the constraints carried by the state primitive. The specific calculation method of constraint satisfaction factor will be explained in detail in the subsequent Example 6.

[0118] State continuity factor: refers to a numerical indicator used to quantitatively evaluate the consistency between candidate promotion content and local historical state. It is used to measure the continuity and smoothness of the promotion strategy after replacing invalid content with the candidate promotion content. The specific calculation method of the state continuity factor will be explained in detail in the subsequent embodiment 7.

[0119] The specific implementation of this embodiment is as follows: First, the compensation solver of each target collaborating node obtains a set of candidate promotional content, which can be derived from the promotional content pool maintained by the content provider.

[0120] Secondly, for each candidate promotion content in the candidate promotion content set, the compensation solver calls the content quality assessment module, constraint satisfaction assessment module, and state connection assessment module respectively to calculate the content quality factor, constraint satisfaction factor, and state connection factor. The calculation results of each factor can be normalized values ​​between 0 and 1.

[0121] Then, the compensation solver performs a weighted summation of the content quality factor, constraint satisfaction factor, and state connectivity factor. The formula for the weighted summation is: Total utility score = w1 × Content quality factor + w2 × Constraint satisfaction factor + w3 × State connectivity factor. Here, w1, w2, and w3 are preset weight coefficients, and their sum is 1.

[0122] The weighting coefficients can be configured as follows: w1 ranges from 0.15 to 0.25, preferably 0.20; w2 ranges from 0.50 to 0.70, preferably 0.60; and w3 ranges from 0.10 to 0.25, preferably 0.20. The range and preferred values ​​of these weighting coefficients can be determined using hyperparameter tuning methods such as grid search or Bayesian optimization on historical compensation data. The goal is to ensure that each target collaborating node deterministically arrives at the same optimal compensation promotion content after local consistency solving. The largest value for w2 ensures that the constraint satisfaction factor dominates the total utility score, resulting in a high degree of consistency among target collaborating nodes in their judgment of whether the promotion content satisfies the current constraints, reflecting the decisive influence of the common input on the ranking result. A smaller value for w3 ensures that the state connection factor only plays a minor role, and the differences in calculation results caused by different local historical states among the target collaborating nodes are insufficient to change the ranking result determined by the common input. If w2 is below 0.50, the dominance of constraint satisfaction may be weakened, and in extreme cases, the nodes of each objective cooperating party may produce different ranking results due to differences in their local historical states; if w3 is above 0.25, the fine-tuning effect of the state connection factor may be excessively amplified, which may also affect the determinism of the solution results.

[0123] Finally, the compensation solver identifies the candidate promotion content with the highest total utility score as the optimal compensation promotion content corresponding to the promotion position identifier.

[0124] It should be noted that the calculation method for the total utility score is not limited to a simple weighted summation of the three factors. In one example, a multiplicative synthesis method can be used: Total Utility Score = Content Quality Factor × Constraint Satisfaction Factor × State Connection Factor, which is suitable for scenarios where there is a strong multiplicative relationship between the factors. In another example, the Top-Ideal-Solution (TOPSIS) method can be used, which is suitable for scenarios with a large number of candidate promotional content and requiring comprehensive consideration. Any technical means that can comprehensively evaluate the quality of candidate promotional content based on multiple factors falls within the protection scope of this invention.

[0125] For example, continuing from the previous example, the compensation solver of the content provider (N1) obtains a set of candidate promotional content {content C, content D, content E}. For content C, the content quality factor is calculated to be 0.85, the constraint satisfaction factor to be 0.90, and the state connection factor to be 0.75; for content D, the three factors are 0.80, 0.70, and 0.85 respectively; for content E, the three factors are 0.75, 0.60, and 0.90 respectively. A weighted summation is performed using weight configurations w1=0.20, w2=0.60, and w3=0.20: The total utility score for content C is 0.20 × 0.85 + 0.60 × 0.90 + 0.20 × 0.75 = 0.17 + 0.54 + 0.15 = 0.86. The total utility score for content D = 0.20 × 0.80 + 0.60 × 0.70 + 0.20 × 0.85 = 0.16 + 0.42 + 0.17 = 0.75; The total utility score for content E is 0.20 × 0.75 + 0.60 × 0.60 + 0.20 × 0.90 = 0.15 + 0.36 + 0.18 = 0.69.

[0126] Among them, content C had the highest total utility score and was determined to be the optimal compensation promotion content.

[0127] The following explanation addresses the differences in local historical states of each target collaborating node and the consistency of the solution results: In this embodiment, each target collaborating node plays a different role in the distributed system, maintaining different dimensions of state data related to the promotion slot. For example, the content provider maintains the sequence of promotional content already deployed on the promotion slot and the material status of each promotional content; the review management maintains the historical review records of each promotional content on the promotion slot; the data statistics maintains the historical exposure statistics of each promotional content on the promotion slot, and so on. Therefore, the local historical states of each target collaborating node are typically different in terms of data content, data dimensions, and data structure.

[0128] Although the local historical states of each target collaborating node are different, the polymorphic compensation utility function in this embodiment is designed such that, given a promotion bit identifier and state primitive, each target collaborating node, after independently running the polymorphic compensation utility function, can deterministically obtain the same optimal compensation promotion content. The implementation mechanism is as follows: the polymorphic compensation utility function is constructed as a weighted summation structure containing factors related to common inputs and factors related to private inputs. Through the weighting ratio between each factor, the final total utility score ranking result is dominated by the common input (i.e., the state primitive and the candidate promotion content set), while the impact of the differences in the local historical states of each target collaborating node on the ranking result is limited to a range that does not change the final optimal solution.

[0129] Specifically, the constraint satisfaction factor is entirely determined by the attribute matching relationship between the state primitives and the candidate promotion content. All target collaborating nodes calculate the same result for this factor, which has a significant weight in the total utility score (e.g., the preferred value for w2 can be 0.60). This ensures that all target collaborating nodes have a consistent judgment on whether the promotion content satisfies the current constraint conditions, reflecting the decisive influence of the common input on the ranking result. The state connection factor is calculated using local historical states. Different target collaborating nodes may obtain different connection factor values ​​due to different local historical states. However, this connection factor is configured to play only a minor adjustment role in the total utility score (e.g., the preferred value for w3 can be 0.20). In most cases, the difference in its score is insufficient to change the ranking result determined by the common input.

[0130] Regarding the technical function of the state connection factor in boundary cases, it should be noted that in actual engineering scenarios, there may be two or more candidate promotional contents in the candidate promotional content set that score exactly the same on the content quality factor and constraint satisfaction factor. In this case, the state connection factor becomes the decision-making basis to break the tie—each target collaborating node, based on its local historical state, uses the same cosine similarity weighted average algorithm to calculate the connection degree between each equally excellent candidate promotional content and the historical compensation content. Since the historical compensation content is public information commonly known to all target collaborating nodes, the ranking direction of the connection factor values ​​calculated by each target collaborating node is consistent—candidate promotional contents more similar to the historical compensation content will receive higher connection factor scores. Therefore, even in boundary cases, each target collaborating node can deterministically derive the same optimal compensation promotional content, ensuring global consistency in compensation decisions.

[0131] Through the above mechanism, each target coordinating node can complete the consensus solution locally without inter-node communication negotiation and determine the same optimal compensation promotion content, thus achieving consistent convergence of compensation decisions under the condition of eliminating the intervention of distributed coordination nodes.

[0132] In this embodiment, by constructing a utility function solution framework that uses a weighted summation of three factors—content quality, constraint satisfaction, and state connection—the selection of compensation content is transformed from a simple replacement in a single dimension to a comprehensive optimal solution in multiple dimensions. Thus, based on the consistent solution achieved in Embodiment 1 above, it further realizes multi-dimensional quantification and global optimization of compensation decisions, solving the problem that the quality of the selected replacement content is poor or that it is out of touch with historical promotion strategies due to a single compensation strategy. This significantly improves the rationality of compensation content selection and the continuity of promotion effects.

[0133] Example 6: Algorithm-level implementation of constraint satisfaction factors.

[0134] In this embodiment, calculating the constraint satisfaction factor between the candidate promotion content and the constraint conditions carried by the state primitive includes: Parse the scope of the state primitive to obtain at least one constraint, which includes an audit status constraint and a promotional content format constraint. The attributes of the candidate promotion content are matched and verified one by one with the constraints to obtain the satisfaction score of each constraint. The constraint satisfaction factor is obtained by weighted summation of the satisfaction scores.

[0135] Constraints: These refer to the restrictive requirements on the compensation promotion content recorded in the scope of the state primitive. Constraints include, but are not limited to, review status constraints (such as requiring the replacement content to be in a review-approved or pending status, and not in a review-rejected status) and promotion content format constraints (such as requiring the size, file format, and text length of the replacement content to match the display specifications of the promotion slot).

[0136] The specific implementation of this embodiment is as follows: First, the constraints of each target collaborating node satisfy the scope of the parsing state primitive in the evaluation module, and extract the list of constraints contained therein. Each constraint can contain a constraint type identifier and constraint parameters.

[0137] Secondly, for each extracted constraint, the constraint satisfaction evaluation module matches and verifies the corresponding attributes of the candidate promotional content against that constraint. The specific matching and verification rules are as follows: For review status constraints, it checks whether the current review status of the candidate promotional content meets the constraint requirements. If it does, the satisfaction score is 1.0; if it does not, the satisfaction score is 0.0. For promotional content format constraints, it checks whether the corresponding format attributes of the candidate promotional content (such as size, file type, text length, etc.) are within the allowed range of the constraint parameters. If they match completely, the satisfaction score is 1.0; if they match partially, a score between 0.0 and 1.0 can be given based on the degree of matching; if they do not match at all, the satisfaction score is 0.0.

[0138] Then, the constraint satisfaction evaluation module performs a weighted summation of the satisfaction scores of each constraint. The formula for the weighted summation is: Constraint Satisfaction Factor = Σ(λ_i × s_i), where λ_i is the weight coefficient of the i-th constraint, s_i is the satisfaction score of the i-th constraint, and Σλ_i = 1.

[0139] The typical configuration of the weighting coefficients can be as follows: the weight range for the review status constraint can be 0.50 to 0.70, with a preferred value of 0.60; the weight range for the promotion content format constraint can be 0.30 to 0.50, with a preferred value of 0.40. The range and preferred values ​​of these weighting coefficients can be determined using hyperparameter tuning methods on historical compensation data, with the goal of ensuring that each target collaborating node deterministically arrives at the same optimal compensation promotion content after local consistency solving. The weighting of review status constraints over promotional content format constraints is based on the following technical considerations: Review status directly determines whether promotional content is compliant, constituting a hard constraint. If a candidate promotional content's review status does not meet the constraint requirements, even if its format perfectly matches, it should not be selected as compensation content. Therefore, review status constraints need to occupy a higher weight to ensure that candidate promotional content that does not meet review status constraints is at a significant disadvantage in the overall utility score. Promotional content formats (such as size, file type, etc.) can be adapted and adjusted to a certain extent through technical means, constituting an adaptable constraint. Its weighting needs to reflect that the degree of format matching is an important reference for selecting compensation content, but should not exceed the priority of review status compliance. If the weight of review status constraints is lower than 0.50, the dominance of rigid constraints may be weakened, and in extreme cases, there may be a risk that candidate promotional content with a matching format but non-compliant review may be selected as the optimal compensation promotional content. If the weight of promotional content format constraints is lower than 0.30, the reference value of format matching may be excessively weakened, making it difficult to select the most format-matching content from multiple compliant candidate promotional content.

[0140] It should be noted that the calculation method for the constraint satisfaction factor is not limited to weighted summation. In one example, a minimum value synthesis method can be used, where the constraint satisfaction factor is the minimum value of the satisfaction scores of all constraints, which is suitable for scenarios where all constraints are hard constraints. In another example, a penalty function method can be used, applying an exponentially decaying penalty score to unsatisfied constraints. Any technical means that can evaluate the degree of compliance of candidate promotional content based on multiple constraints falls within the protection scope of this invention.

[0141] For example, following the previous example, the constraints of the review management terminal (N2) satisfy the scope of the state primitive P1 parsed by the evaluation module, extracting two constraints: review status constraint (requiring the review status of the promotional content to be "passed" or "pending review") and promotional content format constraint (requiring the size of the promotional content to be "300 pixels wide × 250 pixels high"). Based on this: For candidate content C, its review status is "pending review," satisfying the review status constraint, s1=1.0; its size is "300 pixels wide × 250 pixels high," a perfect match, s2=1.0. Using weight configurations λ1=0.60 and λ2=0.40, the constraint satisfies factor=0.60×1.0+0.40×1.0=1.0.

[0142] For candidate content D, its review status is "rejected", s1=0.0; size matching, s2=1.0; constraint satisfaction factor=0.60×0.0+0.40×1.0=0.40.

[0143] In this embodiment, by refining the calculation of constraint satisfaction factors into a specific algorithm process of "analyzing constraints → matching and verifying one by one → weighted summation", the abstract "constraint satisfaction" is concretized into numerical indicators that can be calculated and verified item by item. Thus, based on the utility function solution framework implemented in Embodiment 5 above, the precise quantification and interpretability of constraint satisfaction assessment are further realized. This solves the problem of unreasonable selection of compensation content or non-compliance with business rules due to ambiguity in constraint assessment logic, and significantly improves the accuracy and credibility of constraint satisfaction assessment.

[0144] Example 7: Specific implementation of the algorithm for the state transition factor.

[0145] In this embodiment, calculating the state transition factor between the candidate promotion content and the local historical state includes: The promotional content associated with the promotional slot identifier and used in the most recent successful compensation is obtained from the local historical state and used as historical compensation content. Extract the first content category tag and the first audience targeting tag of the candidate promotion content, and extract the second content category tag and the second audience targeting tag of the historical compensation content; Calculate the first cosine similarity between the first content category tag and the second content category tag, and calculate the second cosine similarity between the first audience targeting tag and the second audience targeting tag; The first cosine similarity and the second cosine similarity are weighted and averaged, and the weighted average is used as the state connection factor.

[0146] Content category tags: These are tag vectors used to describe the content category to which the promoted content belongs, such as multi-tag encoded vectors for categories like "Technology," "Finance," "Sports," and "Entertainment." Content category tags are pre-labeled by the content provider when submitting promoted content, or automatically generated by the content classification model.

[0147] Audience targeting tags: These are tag vectors used to describe the characteristics of the target audience for promoted content, such as encoded vectors for dimensions like age range, gender, region, and interests. Audience targeting tags are pre-set by the content provider when submitting promoted content.

[0148] The specific implementation of this embodiment is as follows: First, the state connection assessment module of each target collaborating node retrieves the promotion content identifier used when the promotion bit was last successfully compensated (i.e., the atomic state transaction was successfully committed) from the local historical state storage, using the promotion bit identifier as an index, and obtains the detailed information of the corresponding promotion content as the historical compensation content.

[0149] Secondly, the status transition assessment module extracts the first content category tag and the first audience targeting tag of the candidate promotion content, as well as the second content category tag and the second audience targeting tag of the historical compensation content.

[0150] Then, the state connection evaluation module calculates the cosine similarity between the first content category tag and the second content category tag to obtain the first cosine similarity. The formula for calculating the cosine similarity is: cos_sim(A, B) = (A·B) / (||A||× ||B||), where A·B is the dot product of the two vectors, and ||A|| and ||B|| are the magnitudes of the two vectors, respectively.

[0151] Then, the state connection evaluation module calculates the cosine similarity between the first audience targeting label and the second audience targeting label to obtain the second cosine similarity.

[0152] Finally, the state connection evaluation module performs a weighted average of the first cosine similarity and the second cosine similarity. The formula for the weighted average is: State connection factor = α × First cosine similarity + (1-α) × Second cosine similarity, where α is a preset weight coefficient, ranging from 0.40 to 0.60, preferably 0.50.

[0153] Furthermore, the range and optimal values ​​of the aforementioned weighting coefficients can be determined using hyperparameter tuning methods on historical compensation data. This aims to optimize the selected optimal compensation promotion content to maintain consistency with historical campaign strategies in terms of promotional effectiveness. Setting the optimal value of α to 0.50, i.e., weighting the cosine similarity of content category tags and the cosine similarity of audience targeting tags equally, is based on the following technical considerations: content category and audience targeting are two independent and complementary dimensions for evaluating the continuity of promotional content. The former reflects the thematic continuity of the promotional content, while the latter reflects the consistency of the target audience. Both are equally important in measuring the smooth connection between compensation content and historical promotion strategies. If α is higher than 0.60, the weight of the audience targeting dimension is excessively compressed, potentially leading to selected compensation content that is thematically consistent but deviates from the historical strategy in terms of audience coverage. If α is lower than 0.40, the weight of the content category dimension is excessively compressed, potentially leading to selected compensation content that matches the audience but jumps in theme, both of which are detrimental to the stable continuation of promotional effectiveness.

[0154] It should be noted that the calculation method for the aforementioned state connection factor is not limited to cosine similarity weighted average. In one example, Jaccard similarity can be used to calculate the degree of overlap between tags, which is suitable for scenarios where tags are binary vectors; in another example, the normalized value of Euclidean distance can be used as a similarity measure, and a time decay factor can be added to give more recent historical compensation content a higher reference weight. Any technical means that can quantify the continuity and smoothness between candidate promotion content and historical compensation content are within the protection scope of this invention.

[0155] For example, following the previous example, the state connection evaluation module of the content provider (N1) queries the local historical state and finds that the most recently successfully compensated content used for the promotion slot Slot_01 is content A (historical compensation content). The content category tags of candidate content C are [Technology: 0.8, Finance: 0.1, Sports: 0.1], and the audience targeting tags are [Age 25-35: 0.7, Male: 0.6, First-tier cities: 0.8]. The content category tags of historical compensation content A are [Technology: 0.9, Finance: 0.05, Sports: 0.05], and the audience targeting tags are [Age 25-35: 0.8, Male: 0.5, First-tier cities: 0.7]. The first cosine similarity is calculated to be approximately 0.97, and the second cosine similarity is approximately 0.99. Using α=0.50, the state connection factor is 0.50×0.97+0.50×0.99=0.98.

[0156] In this embodiment, by introducing cosine similarity calculation based on two dimensions—content category and audience targeting—the abstract concept of "state connection" is quantified into a calculable numerical indicator. Thus, based on the utility function solution framework implemented in Embodiment 5 above, it further realizes the accurate measurement of the smoothness and continuity between compensation content and historical promotion strategies. This solves the problem of fluctuations in promotion effects caused by ignoring the continuity of historical promotion strategies when selecting compensation content, and significantly improves the consistency of compensation decisions and the stability of user experience.

[0157] Example 8: Specific implementation of version synchronization of polymorphic compensation utility function.

[0158] In this embodiment, the polymorphic compensation utility function is synchronized on each of the target collaborating nodes through a preset distributed configuration center, which maintains the version number and function parameter configuration of the polymorphic compensation utility function. Before each of the aforementioned target collaborating nodes runs the polymorphic compensation utility function, the following is also included: Each target collaborating node pulls the latest version number of the polymorphic compensation utility function from the preset distributed configuration center and compares it with the current version number on its local machine. If there is a discrepancy, the function parameter configuration of the polymorphic compensation utility function in the local system will be updated synchronously from the preset distributed configuration center.

[0159] The pre-defined distributed configuration center refers to an independently deployed coordination service component used for centralized management and distribution of configuration information of nodes in a distributed system. Its implementation includes, but is not limited to, configuration centers built on distributed consistent key-value storage systems such as etcd, Consul, or ZooKeeper. The pre-defined distributed configuration center maintains the version number and function parameter configuration of the polymorphic compensation utility function, including the weight coefficients of each factor, the weight configuration of constraints, and the calculation parameters of state transition factors.

[0160] The specific implementation of this embodiment is as follows: First, before each execution of the local consistency solution, each target coordinating node reads the version number of the currently stored polymorphic compensation utility function from its local function version record.

[0161] Secondly, each target collaborating node sends a version query request to the preset distributed configuration center to retrieve the latest version number of the current polymorphic compensation utility function.

[0162] Then, each target collaborating node compares its local version number with the latest version number. If they match, the local polymorphic compensation utility function is used directly for solving the problem; if they do not match, the complete function parameter configuration corresponding to the latest version number is retrieved from the preset distributed configuration center, the function parameter configuration and version number record in the local storage are updated, and then the updated polymorphic compensation utility function is used for solving the problem.

[0163] It should be noted that the above-mentioned version synchronization triggering method is not limited to fetching before each solution. In one example, a listener-push mechanism can be used, where each target collaborating node registers a listener for function parameter configuration on a preset distributed configuration center. When the configuration changes, the preset distributed configuration center actively pushes an update notification to each target collaborating node. In another example, a combination of periodic fetching and event triggering can be used, adding an active push channel for urgent changes on top of periodic fetching. Any technical means that can ensure that each target collaborating node uses the same version of the polymorphic compensation utility function is within the protection scope of this invention.

[0164] In this embodiment, by introducing a preset distributed configuration center to synchronize the version of the polymorphic compensation utility function, the function version management is transformed from manual independent maintenance by each target collaborating node to automatic synchronization maintenance based on the central configuration. Thus, based on the utility function solution achieved in Embodiment 1 above, it further ensures that each target collaborating node is running the same version of the function at any time, solving the problem of discrepancies in the solution results caused by inconsistent function versions of each target collaborating node, and significantly improving the reliability of distributed consistency solution.

[0165] Example 9: Specific implementation of atomic transaction commit and compensation completion event publication.

[0166] In this embodiment, each of the target collaborating nodes submits the optimal compensation promotion content locally as an atomic state transaction, including: Each of the target collaborating node performs the following operations within a local database transaction: Update the value of the current promotion content field corresponding to the promotion position identifier to the identifier of the optimal compensation promotion content; Update the value of the review status field corresponding to the promotion position identifier to the pending review status; Reset the count value of the exposure count field corresponding to the promotion position identifier to zero, and update the statistics start timestamp to the current system time; After the local database transaction is successfully committed, the identifier of the optimal compensation promotion content and the promotion position identifier are sent as a compensation completion event to a preset distributed state monitoring bus to notify each external service node that has subscribed to the state change event of the promotion position identifier.

[0167] The pre-defined distributed state monitoring bus refers to an independently deployed communication infrastructure used to implement event-driven message publishing and subscription in a distributed system. Its implementation includes, but is not limited to, event buses built on message middleware such as Apache Kafka, RabbitMQ, or Redis Streams. The pre-defined distributed state monitoring bus allows external service nodes to subscribe to state change events identified by specific promotion bits. When the event publication is complete, each external service node that has subscribed to the event can asynchronously receive the notification and perform subsequent processing.

[0168] The specific implementation of this embodiment is as follows: First, each target collaborating node initiates an atomic state transaction in its local database management system.

[0169] Secondly, in this atomic state transaction, each target collaborating node performs the following update operations: updates the value of the current promotion content field corresponding to the promotion position identifier to the identifier of the optimal compensation promotion content; updates the value of the review status field corresponding to the promotion position identifier to the pending review status (indicating that the new compensation content needs to go through the review process again); resets the count value of the exposure count field corresponding to the promotion position identifier to zero, and updates the statistical start timestamp to the current system time (establishing an independent statistical benchmark for the new compensation content).

[0170] Then, commit the atomic state transaction. If the commit is successful, all the above update operations will take effect simultaneously; if the commit fails (e.g., due to transaction rollback caused by concurrency conflicts), retry will be performed according to the preset retry strategy.

[0171] Finally, after the atomic state transaction is successfully committed, each target collaborating node encapsulates the identifier of the optimal compensation promotion content and the promotion position identifier into a compensation completion event message and sends it to the preset distributed state listening bus. External service nodes (such as cache refresh service, front-end push service, etc.) that have subscribed to the state change event of this promotion position identifier can asynchronously execute the corresponding subsequent processing upon receiving the compensation completion event.

[0172] It should be noted that the method of publishing the compensation completion event is not limited to individual publication by each target collaborating node. In one example, an event aggregation method can be used, where each target collaborating node sends the compensation completion event to a preset event aggregator, which then publishes the aggregated compensation completion event after confirming that all target collaborating nodes have completed their submissions. In another example, the submission status information of each target collaborating node can be attached to the compensation completion event, allowing subscribers to determine whether all target collaborating nodes have successfully submitted. Any technical means that can notify downstream services after compensation is completed falls within the protection scope of this invention.

[0173] For example, following the previous example, the content provider (N1) initiates an atomic state transaction, updates the current promotional content field of promotion slot Slot_01 to content C, updates the review status to "pending review," resets the exposure count field to zero, updates the statistics start timestamp to T2, and commits the transaction successfully. Content C and Slot_01 are encapsulated as a compensation completion event and sent to a preset distributed state listening bus. Upon receiving this event, the cache refresh service refreshes the content cache of promotion slot Slot_01; upon receiving this event, the front-end push service pushes a compensation completion notification to the publishing end.

[0174] In this embodiment, by refining the atomic transaction commit operation into multi-field atomic updates and publishing a compensation completion event after successful commit, a complete "commit-notification" closed loop is constructed. Based on the atomic commit implemented in Embodiment 1 above, this embodiment further realizes the complete implementation of the compensation result and loosely coupled linkage with downstream services. It solves the problem of cache not being refreshed or front-end display delay caused by downstream services not being aware of the compensation completion in a timely manner, and significantly improves the closed-loop integrity of the compensation processing and the system's collaborative efficiency.

[0175] Example 10: Specific implementation of decentralized consistency verification and inconsistency state directional recovery.

[0176] In this embodiment, after each of the target collaborating nodes commits the optimal compensation promotion content locally as an atomic state transaction, the process further includes: After a preset verification waiting time window, the collaborating node that detects the abnormal event signal sends a status verification request for the promotion bit identifier to each of the target collaborating nodes. Each of the target collaborating nodes responds to the status verification request by returning the currently promoted content identifier corresponding to the promoted position identifier stored locally to the collaborating node that detected the abnormal event signal; The collaborating node that detects the abnormal event signal compares whether the current promotion content identifier returned by each target collaborating node is consistent with the identifier of the optimal compensation promotion content. If there are inconsistent target collaborating nodes, the collaborating node that detects the abnormal event signal will resend the state projection message to the inconsistent target collaborating node according to the preset retry strategy, so as to trigger the inconsistent target collaborating node to re-perform local consistency solution and atomic state transaction commit.

[0177] The specific implementation of this embodiment is as follows: First, after the collaborating node that detects the abnormal event signal completes the multicast transmission of the state projection message, it starts a timer with a preset verification waiting time window. The value of the preset verification waiting time window should be greater than the estimated maximum time taken for each target collaborating node to complete the local consistency solution and atomic state transaction submission. Its value range can be from 500 milliseconds to 2000 milliseconds, preferably 1000 milliseconds.

[0178] The estimated maximum time can be determined by simulating abnormal event signals to trigger the compensation process in the test or pre-release environment of the distributed system. End-to-end time data from receiving the state projection message to completing the atomic state transaction commit for each target collaborating node is collected, and the maximum value of multiple samples or a preset percentile (such as the 95th or 99th percentile) is used as the estimated maximum time. If the verification waiting time window is set too short, normal processing fluctuations or network latency may cause some target collaborating nodes to be misjudged as inconsistent before completing the commit, triggering unnecessary retries. If it is set too long, it will prolong the overall closed-loop time of the compensation process and reduce the real-time performance of the compensation process.

[0179] Secondly, when the timer reaches the preset verification waiting time window, the verification module of the collaborating node that detected the abnormal event signal sends a status verification request for the promotion position identifier to each target collaborating node. The status verification request carries the promotion position identifier and the identifier of the optimal compensation promotion content.

[0180] Then, each target collaborating node responds to the status verification request, queries the value of the current promotion content field corresponding to the promotion position identifier in its local database, and returns the current promotion content identifier to the collaborating node that detected the abnormal event signal.

[0181] Then, the verification module of the collaborating node that detected the abnormal event signal compares whether the current promotion content identifier returned by each target collaborating node is consistent with the identifier of the optimal compensation promotion content.

[0182] Finally, if all returned current promotion content identifiers are consistent, the verification passes, and the compensation process completes the closed loop. If at least one target collaborating node returns a current promotion content identifier that is inconsistent with the identifier of the optimal compensation promotion content, the collaborating node that detected the abnormal event signal resends the original state projection message to the inconsistent target collaborating node according to a preset retry strategy. This triggers the inconsistent target collaborating node to re-execute the local consistency solution and atomic state transaction commit process. The preset retry strategy may include an upper limit on the number of retries and a retry interval. The typical value for the upper limit on the number of retries can be 2, 3, or 4 times, preferably 3 times; the typical value for the retry interval can be 200 milliseconds, 300 milliseconds, or 500 milliseconds, preferably 300 milliseconds. The pre-defined retry strategy is based on the following technical considerations: In real-world distributed systems, the inconsistency in the state of target collaborating nodes is usually caused by transient failures, such as network jitter leading to packet loss of state projection messages, database deadlocks or transaction conflicts occurring when the target collaborating node is processing a commit. These transient failures can resolve themselves within a very short time. After resending the state projection message, the target collaborating node re-performs local consistency resolution and atomic state transaction commit, which usually successfully completes the state update, achieving directional recovery of inconsistent states and restoring consistency to the generalized state of all target collaborating nodes. By setting an upper limit on the number of retries, the inconsistency problem caused by transient failures can be effectively resolved within a limited number of retries, while avoiding the system falling into an infinite retry loop due to permanent failures (such as target collaborating node crashes or data corruption).

[0183] It should be noted that the triggering method for the above-mentioned verification waiting time window is not limited to a fixed time window. In one example, an adaptive time window can be used, dynamically adjusting the window length based on the average processing time of historical compensation processes; in another example, an event-driven confirmation mechanism can be used, where each target collaborating node actively sends a confirmation message to the collaborating node that detected the abnormal event signal after the atomic state transaction is successfully committed. When the collaborating node that detected the abnormal event signal receives all confirmation messages within a preset timeout period, the verification is considered successful. Any technical means that can verify the consistency of the states of each target collaborating node under a decentralized architecture falls within the protection scope of this invention.

[0184] This embodiment constructs a decentralized consistency verification closed loop by adding delayed verification, comparison, and verification and inconsistency state-oriented recovery steps based on a preset retry strategy after atomic commit. The preset retry strategy effectively solves the state inconsistency problem caused by transient failures such as network jitter, database deadlock, or transaction conflicts through a limited number of retries. Simultaneously, the upper limit on the number of retries avoids infinite retry loops caused by permanent failures. Therefore, based on the distributed compensation processing implemented in Embodiment 1 above, it further realizes automatic verification of the state consistency of each target collaborating node and targeted repair of inconsistencies in transient failure scenarios. It solves the state inconsistency problem caused by the failure of some target collaborating nodes to commit or network partitions, significantly improving the final consistency of the compensation processing and the system's ability to recover inconsistencies, thus enhancing the system's robustness in a distributed environment.

[0185] Example 11: An integrated example of combining multiple technical features.

[0186] The event-driven dynamic compensation processing method for promotion positions described in the above embodiments can be recombined with the technical features included in different embodiments as needed to obtain a combined implementation scheme, but all are within the protection scope claimed by this invention.

[0187] For example, combining the basic scheme of Example 1 with the hash chain state log scheme of Example 2 and the graph matching and one-hot code bitmap scheme of Example 4 can constitute an enhanced compensation processing system that has both state primitive integrity protection and accurate computational boundary encoding capabilities. In this combined scheme, the state primitive is embedded with a hash data fingerprint after encapsulation and appended to an append-only state log. The cooperating party locks the bitmap efficiently through graph data structure subgraph matching and one-hot code logical OR operation, which significantly improves the data reliability and computational boundary accuracy of the compensation processing.

[0188] For example, combining the basic scheme of Example 1 with the three-factor utility function solution framework of Example 5, the constraint satisfaction factor algorithm of Example 6, and the state connection factor algorithm of Example 7 can constitute a refined solution compensation decision-making system with multi-dimensional comprehensive evaluation capabilities. In this combined scheme, the selection of compensation content comprehensively considers three dimensions: content quality, constraint satisfaction, and historical connection. Each dimension is supported by specific algorithmic processes, and the decision-making process is quantifiable and interpretable.

[0189] For example, combining the intelligent event filtering scheme of Example 3 with the function version synchronization scheme of Example 8 can constitute a precise compensation processing system with entry filtering and global consistency solution guarantee. In this combined scheme, abnormal event signals are filtered by a semantic classification model before entering the compensation processing flow, ensuring that processing is triggered only for events that require compensation; at the same time, the polymorphic compensation utility functions of each target collaborating node are kept consistent through a preset distributed configuration center, ensuring the consistency of the solution results.

[0190] The above combinations are merely illustrative and do not constitute a limitation on the scope of protection of this invention. Those skilled in the art can flexibly select and combine technical features from different embodiments according to the needs of actual application scenarios to obtain the optimal technical solution adapted to a specific scenario.

[0191] In one embodiment, a computer device is provided, which may be the device corresponding to a collaborating node, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of an event-driven generalized bit dynamic compensation processing method.

[0192] Those skilled in the art will understand that the methods provided in the embodiments of the present invention can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. The methods can also be implemented as a computer program product stored in one or more computer-readable storage media, including but not limited to: disks, optical disks, read-only memory (ROM), random access memory (RAM), flash memory, etc. When the computer program product is executed by one or more data processing devices (such as computers), the devices perform the steps as described in any of the foregoing method embodiments.

[0193] The software tools, components, or models not belonging to our company that appear in the embodiments of this invention are merely illustrative examples and do not represent actual use.

[0194] The data collection in this embodiment of the invention complies with the requirements of relevant laws and regulations, such as the Data Security Law of the People's Republic of China, the Personal Information Protection Law of the People's Republic of China, GDPR (General Data Protection Regulation of the European Union), or information security standards of other countries and regions.

[0195] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamic compensation of promotion positions based on event-driven mechanisms, characterized in that, include: In response to any collaborating node in the distributed system detecting an abnormal event signal, the collaborating node generates a state primitive containing the generalization bit identifier, event type, and scope corresponding to the abnormal event signal; The collaborating party nodes generate a collaborating party locking bitmap that identifies all target collaborating party nodes associated with the promotion bit identifier based on a preset collaborating review relationship graph. The state primitive and the collaborating party locking bitmap are then multicast as state projection messages to trigger each target collaborating party node to perform local consistency solving. Each of the target collaborating nodes, based on its local historical state and the state primitive, performs a local consistency solution based on a globally consistent polymorphic compensation utility function to deterministically determine the optimal compensation promotion content corresponding to the promotion bit identifier. Each of the target collaborating nodes submits the optimal compensation promotion content locally as an atomic transaction, so that the content status of the promotion bit corresponding to the promotion bit identifier on each of the target collaborating nodes is consistently updated to the optimal compensation promotion content.

2. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, Generate a state primitive containing the generalization bit identifier, event type, and scope corresponding to the abnormal event signal, including: The abnormal event signal is parsed, and the promotion bit identifier field, event type code, and description of the affected attribute range carried in the abnormal event signal are extracted. The promotion bit identifier field, the event type code, and the description of the affected attribute range are encapsulated into an immutable data record as a state primitive. The immutable data record also includes a generation timestamp and an event source node identifier. Calculate the hash value of each field in the state primitive, and embed the hash value as the data fingerprint of the state primitive into the state primitive; The embedded state primitives are appended to an append-only state log, and each record in the append-only state log is linked together by the hash value of the previous record.

3. The event-driven dynamic compensation method for promotion positions as described in claim 1 or 2, characterized in that, Before generating a state primitive containing the generalization bit identifier, event type, and scope corresponding to the abnormal event signal, the process also includes: Obtain the abnormal event signal; Extract the text description information and structured fields from the abnormal event signals; The text description information is converted into a word vector sequence through a preset word embedding matrix, and then concatenated with the encoded vector of the structured field to obtain the event feature vector; The event feature vector is input into a preset binary classification fully connected network to obtain the confidence level that the abnormal event signal belongs to a preset target event category that needs to trigger dynamic compensation processing. When the confidence level exceeds a preset confidence level threshold, the event category of the abnormal event signal is determined to be the target event category, and the step of generating the state primitive is executed.

4. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, Generate a collaborator locking bitmap that identifies all target collaborator nodes associated with the promotion bit identifier, and multicast the state primitives and the collaborator locking bitmap as state projection messages, including: Obtain a preset collaborative review relationship graph. The preset collaborative review relationship graph stores the association between each promotion bit identifier and the collaborative node in a graph data structure. Each collaborative node is pre-assigned a unique hot code to identify itself. The bit width of the unique hot code is determined by the total number of collaborative nodes in the distributed system. The position of the bit with a value of 1 in the unique hot code is fixed to the node of the collaborative node in the distributed system. Based on the promotion position identifier, subgraph matching is performed in the preset collaborative review relationship graph to obtain a set of collaborative party nodes that are connected to the promotion position identifier by an edge; Extract the unique hot code corresponding to each collaborative node in the collaborative node set; A logical OR operation is performed on each of the one-hot codes, and the result is used as a collaborator locking bitmap. Each bit in the collaborator locking bitmap corresponds to a collaborator node in the distributed system. The collaborator node corresponding to the bit with a value of 1 in the collaborator locking bitmap is the target collaborator node, and the collaborator node corresponding to the bit with a value of 0 is a collaborator node in the distributed system that is not the target of this dynamic compensation processing. The state primitives and the cooperative locking bitmap are encapsulated into a state projection message; The coordinate party locking bitmap is parsed into a list of network addresses for each of the target coordinate party nodes; The state projection message is sent to each of the target collaborating nodes corresponding to the network address list using User Datagram Protocol multicast, and message compression and retransmission confirmation mechanisms are enabled during transmission.

5. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, The scope of the state primitive includes constraints; based on a globally consistent polymorphic compensation utility function, a local consistency solution is performed to deterministically derive the optimal compensation promotion content corresponding to the promotion bit identifier, including: Obtain a collection of candidate promotional content; For each candidate promotion content in the candidate promotion content set, calculate the content quality factor of the candidate promotion content, the constraint satisfaction factor between the candidate promotion content and the constraint conditions carried by the state primitive, and the state connection factor between the candidate promotion content and the local historical state. The total utility score of the candidate promotion content is obtained by weighted summing of the content quality factor, the constraint satisfaction factor and the state connection factor. The candidate promotion content with the highest total utility score is determined as the optimal compensation promotion content corresponding to the promotion position identifier.

6. The event-driven dynamic compensation method for promotion positions as described in claim 5, characterized in that, Calculating the constraint satisfaction factor between the candidate promotion content and the constraints carried by the state primitive includes: Parse the scope of the state primitive to obtain at least one constraint, which includes an audit status constraint and a promotional content format constraint. The attributes of the candidate promotion content are matched and verified one by one with the constraints to obtain the satisfaction score of each constraint. The constraint satisfaction factor is obtained by weighted summation of the satisfaction scores.

7. The event-driven dynamic compensation method for promotion positions as described in claim 5, characterized in that, Calculating the state transition factor between the candidate promotion content and the local historical state includes: The promotional content associated with the promotional slot identifier and used in the most recent successful compensation is obtained from the local historical state and used as historical compensation content. Extract the first content category tag and the first audience targeting tag of the candidate promotion content, and extract the second content category tag and the second audience targeting tag of the historical compensation content; Calculate the first cosine similarity between the first content category tag and the second content category tag, and calculate the second cosine similarity between the first audience targeting tag and the second audience targeting tag; The first cosine similarity and the second cosine similarity are weighted and averaged, and the weighted average is used as the state connection factor.

8. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, The polymorphic compensation utility function is synchronized on each of the target collaborating nodes through a preset distributed configuration center, which maintains the version number and function parameter configuration of the polymorphic compensation utility function. Before each of the aforementioned target collaborating nodes runs the polymorphic compensation utility function, the following is also included: Each target collaborating node pulls the latest version number of the polymorphic compensation utility function from the preset distributed configuration center and compares it with the current version number on its local machine. If there is a discrepancy, the function parameter configuration of the polymorphic compensation utility function in the local system will be updated synchronously from the preset distributed configuration center.

9. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, Each of the target collaborating nodes shall submit the optimal compensation promotion content locally as an atomic state transaction, including: Each of the target collaborating node performs the following operations within a local database transaction: Update the value of the current promotion content field corresponding to the promotion position identifier to the identifier of the optimal compensation promotion content; Update the value of the review status field corresponding to the promotion position identifier to the pending review status; Reset the count value of the exposure count field corresponding to the promotion position identifier to zero, and update the statistics start timestamp to the current system time; After the local database transaction is successfully committed, the identifier of the optimal compensation promotion content and the promotion position identifier are sent as a compensation completion event to a preset distributed state monitoring bus to notify each external service node that has subscribed to the state change event of the promotion position identifier.

10. The event-driven dynamic compensation method for promotion positions as described in claim 1, characterized in that, After each of the target collaborating nodes commits the optimal compensation promotion content locally as an atomic state transaction, the process further includes: After a preset verification waiting time window, the collaborating node that detects the abnormal event signal sends a status verification request for the promotion bit identifier to each of the target collaborating nodes. Each of the target collaborating nodes responds to the status verification request by returning the currently promoted content identifier corresponding to the promoted position identifier stored locally to the collaborating node that detected the abnormal event signal; The collaborating node that detects the abnormal event signal compares whether the current promotion content identifier returned by each target collaborating node is consistent with the identifier of the optimal compensation promotion content. If there are inconsistent target collaborating nodes, the collaborating node that detects the abnormal event signal will resend the state projection message to the inconsistent target collaborating node according to the preset retry strategy, so as to trigger the inconsistent target collaborating node to re-perform local consistency solution and atomic state transaction commit.

Images