AI agent self-diagnosis and repair system, method, and program

Abstract

In the event of failures in multiple AI agents, even if the affected AI agent is unable to start, diagnosis and repair can be performed externally, reducing investigation time and accelerating recovery. [Solution] The present invention comprises a diagnostic and repair agent that receives fault reports, acquires log information and configuration files of the target AI agent, diagnoses the problem by at least one of known pattern matching, configuration inconsistency detection, and dependency verification, repairs the problem by at least one of configuration correction, restart instruction, and rollback, and generates the results as a report.

Description

Technical Field

[0001] The present invention relates to an information processing system that hierarchically controls a plurality of agents while continuously improving the execution quality of lower-level execution agents.

Background Art

[0002] In recent years, agents that execute multi-stage processing based on instructions in natural language have become widespread, and a configuration in which a plurality of agents cooperate to solve tasks has been proposed.

[0003] Also, frameworks for evaluating the execution logs and outputs of agents, and frameworks for composing a plurality of agents to generate responses to inputs are known.

[0004] On the other hand, when a plurality of lower-level execution agents are associated with users and operated in parallel, education based on different situations, intentions, and products for each user, and operation for continuously improving the behavior of the lower-level execution agents are desired.

[0005] However, it is not easy to establish evaluation and feedback as a continuous loop, update the execution conditions of lower-level execution agents, continuously collect behavior information and product information associated with users, and operate in a form that ensures information separation and access control between users.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Summary of the Invention

[0007] In environments where multiple AI agents are in operation, if engineers individually collect logs, check settings, identify dependencies, and create and execute recovery procedures when a failure occurs, the investigation workload will increase, potentially prolonging the recovery time.

[0008] In particular, if the target AI agent is unable to start, diagnosis and recovery that rely on the AI ​​agent's internal functions cannot be performed, requiring external evidence collection and recovery operations, but these procedures tend to be highly dependent on individual expertise.

[0009] Furthermore, if multiple configuration files exist and there are inconsistencies or contradictions between those configuration files or between descriptions within those files, recovery that only fixes superficial errors is likely to lead to recurrence. In addition, mistaking dependency-related failures for configuration-related failures can lead to unnecessary changes or excessive reboots.

[0010] In view of the above problems, the present invention aims to enable external diagnosis and repair, including when the target AI agent is unable to start, thereby reducing the man-hours required for troubleshooting and accelerating recovery. [Means for solving the problem]

[0011] According to the present invention, An AI agent self-diagnosis and repair system that includes a diagnostic and repair agent in an information processing system including multiple AI agents, The aforementioned diagnostic and repair agent includes a means for receiving fault reports from users regarding the target AI agent, A means for obtaining log information and configuration files corresponding to the aforementioned target AI agent, A diagnostic means that generates a diagnostic result based on the log information and the configuration file, A repair execution means that modifies the configuration file based on the diagnostic results, An AI agent self-diagnosis and repair system is obtained, which includes a report generation means for generating a report including the diagnosis result and the processing content by the repair execution means.

Advantages of the Invention

[0012] According to the present invention, starting from a failure report, log information and configuration files are collected, and diagnosis and repair can be executed in a closed loop, so that the engineering investigation work of engineers can be reduced and the time to recovery can be shortened.

[0013] According to the present invention, even if the target AI agent cannot be started, evidence can be obtained from the outside and diagnosis and repair can be executed, so that the possibility of recovery can be increased even in case of a serious failure.

[0014] According to the present invention, by diagnosis including detection of setting errors or contradictions and verification of dependency relationships, the accuracy of cause analysis can be improved, and unnecessary changes and excessive restarts can be suppressed.

[0015] According to the present invention, by accumulating self-repair logs and using them for reports and preventive measures, knowledge for preventing recurrence of the same type of failure and improving operation can be continuously accumulated.

Brief Description of the Drawings

[0016] [Figure 1] It is a configuration diagram of the present invention. [Figure 2] It is a functional block diagram of the present invention. [Figure 3] It is a flowchart of the present invention. [Figure 4] It is a sequence diagram of the present invention.

Embodiments for Carrying Out the Invention

[0017] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description does not limit the present invention, and those skilled in the art can appropriately change or combine it.

[0018] As shown in FIG. 1, the AI agent self-diagnosis and repair system 1 includes a diagnosis and repair agent 10 and a plurality of target AI agents 20A, 20B, and 20C. The diagnosis and repair agent 10 may be referred to as the "Doctor Agent" as needed. The target AI agents 20A, 20B, and 20C may be referred to as the "Agent" as needed.

[0019] The diagnosis and repair agent 10 receives a failure report from the user 30 and performs diagnosis and repair on the target AI agents 20A, etc. The user 30 may send a failure report via a user terminal or the like.

[0020] In the configuration shown in FIG. 1, the diagnosis and repair agent 10 can access a log storage unit 60, a configuration file storage unit 70, a failure pattern DB 80, and a self-repair log DB 90. The log storage unit 60 may hold log information regarding the target AI agents 20A, etc. The configuration file storage unit 70 may hold configuration files of the target AI agents 20A, etc. The failure pattern DB 80 may hold known failure patterns. The self-repair log DB 90 may hold the history of diagnosis and repair.

[0021] The target AI agents 20A, etc. may utilize an external dependency 100. The external dependency 100 may include at least one of an external service, a data store, a model provider, an authentication infrastructure, etc.

[0022] In this embodiment, even if the target AI agents 20A, etc. are unable to start, the diagnosis and repair agent 10 can acquire log information and configuration files. Here, "acquireable even when unable to start" includes being able to acquire log information and configuration files without going through the execution process of the target AI agents 20A, etc. The acquisition path may be a path via a file system, an execution infrastructure, an administrative process, or an administrative interface, etc.

[0023] As shown in Figure 2, the diagnostic and repair agent 10 includes a symptom collection module 11, a diagnostic engine 12, a repair execution unit 13, and a report generation unit 14. These may be implemented on a single computer or distributed across multiple computers.

[0024] The symptom collection module 11 receives fault reports from users 30. Fault reports may include at least one of the following: identification information to identify the AI ​​agent in question, time of occurrence, summary of symptoms, scope of impact, or steps to reproduce the fault.

[0025] The symptom collection module 11 may generate collection requests to obtain log information and configuration files related to the target AI agent 20A, etc. The log information may include at least one of the following: execution logs, error logs, audit logs, event logs, or trace information. The configuration files may include configuration, policy, tool definitions, connection destination information, etc.

[0026] In this embodiment, the configuration file may include an agent definition file that specifies the operating conditions or operating policies of the target AI agent 20A, etc. The agent definition file may, for example, be in the format of SOUL.md, but is not limited to this.

[0027] The diagnostic engine 12 performs diagnostic processing based on log information and configuration files collected by the symptom collection module 11. The diagnostic processing may include at least one of the following: matching against known failure patterns, detecting configuration errors or inconsistencies, and verifying dependencies.

[0028] By comparing with known failure patterns, candidate failure patterns registered in the failure pattern DB80 can be identified based on the characteristics of errors included in log information, characteristics of configuration files, or symptom descriptions included in failure reports.

[0029] Detecting configuration errors or inconsistencies can reveal inconsistencies between entries within a configuration file or between multiple configuration files. These inconsistencies may manifest as missing required fields, mismatched types or ranges, conflicting values ​​for the same key, missing references, or conflicting permission policies.

[0030] Dependency verification can confirm the success or failure of communication with the external dependency 100, the success or failure of authentication, or the feasibility of requests. Dependency verification can be performed on the external dependency 100 from the diagnostic / repair agent 10 even if the target AI agent 20A, etc., is unable to start.

[0031] The repair execution unit 13 performs repairs based on the diagnostic results generated by the diagnostic engine 12. The repairs may include at least one of the following: modifying configuration files, instructing a service restart, or performing a rollback.

[0032] Modifying a configuration file may involve applying a differential patch to the configuration file stored in the configuration file storage unit 70. The repair execution unit 13 may maintain a backup of the configuration file before modifying it.

[0033] The instruction to restart the service may be an instruction to restart or redeploy the target AI agent 20A, etc. The restart or reload may be performed via the management interface provided by the execution platform of the target AI agent 20A, etc.

[0034] Executing a rollback may be a process that restores a configuration file or related resources to a predetermined past state. For example, this may include reverting to a version of the configuration file that was functioning correctly immediately before the rollback.

[0035] The report generation unit 14 generates a report that includes the diagnostic results and the processing details performed by the repair execution unit 13. The report may include at least one of the following: a candidate cause, the difference in supporting logs or settings, the repair procedure performed, and the verification results.

[0036] The report generation unit 14 can store a self-repair log in the self-repair log DB 90, which includes diagnostic results, repair procedures, success or failure of the repair, and whether or not the problem recurred. The self-repair log can be used to generate preventive measures for future failure prevention or to update the failure pattern DB 80.

[0037] Next, with reference to Figure 3, a processing procedure as an example of an AI agent self-diagnosis and repair method will be described. The procedure in Figure 3 can be implemented by a program executed by the diagnostic and repair agent 10.

[0038] The diagnostic and repair agent 10 receives fault reports from the user 30 (step S101).

[0039] Next, the diagnostic and repair agent 10 identifies the target AI agent 20A, etc., based on the identification information, etc., included in the failure report (step S102).

[0040] Next, the diagnostic and repair agent 10 acquires log information and configuration files corresponding to the target AI agent 20A, etc. (step S103). This acquisition may be performed via a route that does not involve the execution process of the target AI agent 20A, etc.

[0041] Next, the diagnostic and repair agent 10 performs diagnostic processing based on the acquired log information and configuration files and generates diagnostic results (step S104). The diagnostic processing may include at least one of the following: matching against known failure patterns, detecting configuration errors or inconsistencies, and verifying dependencies.

[0042] Next, the diagnostic and repair agent 10 generates a repair plan based on the diagnostic results (step S105). The repair plan may include the type of repair to be performed, the order of application, and the verification method.

[0043] Next, the diagnostic and repair agent 10 performs a repair according to the repair plan (step S106). The repair may include at least one of the following: modifying a configuration file, instructing a service restart, or performing a rollback.

[0044] Next, the diagnostic and repair agent 10 performs a recovery check after the repair (step S107). The recovery check may include at least one of the following: startup status, health check results, or success or failure of a predetermined operational test. If the recovery check is insufficient, a rollback may be performed to return to a safer state.

[0045] Next, the diagnostic and repair agent 10 generates a report including the diagnostic results and repair details, and accumulates a self-repair log (step S108).

[0046] Next, with reference to Figure 4, an example of the interaction from fault report reception to report generation will be explained. User 30 sends a fault report to the diagnostic and repair agent 10 (step S101).

[0047] The diagnostic and repair agent 10 identifies the target AI agent 20A, etc. (step S102) and obtains log information and configuration files from the log storage unit 60 and the configuration file storage unit 70 (step S103). Even if the target AI agent 20A, etc. is unable to start, this acquisition can be performed through a route that does not involve the execution process of the target AI agent 20A, etc.

[0048] The diagnostic and repair agent 10 may refer to the failure pattern DB 80, perform a match, and further verify dependencies on the external dependency 100 (step S104).

[0049] The diagnostic and repair agent 10 generates a repair plan (step S105), modifies or rolls back the configuration file in the configuration file storage unit 70, and instructs the target AI agent 20A, etc. to restart as necessary (step S106).

[0050] The diagnostic and repair agent 10 performs recovery verification (step S107), presents the report to the user 30, and stores the history in the self-repair log DB 90 (step S108).

[0051] The log information may include at least one of the following: time information, target identifier, event type, error details, and correlation identifier. The diagnostic / repair agent 10 may handle the log information while masking the confidential information contained therein.

[0052] The configuration file may include runtime parameters, connection information for external dependencies 100, references to authentication information, or tool definitions. The agent definition file specifies the operating policies or constraints for the target AI agent 20A, etc., and may include, for example, the range of available tools, constraints on output formats, or security policies.

[0053] The failure pattern DB80 may store associated failure pattern identifiers, symptom characteristics, log or configuration conditions to be referenced as evidence, and recommended repair procedures. The self-repair log DB90 may store the type of repair performed, applied configuration differences, whether a restart or rollback occurred, recovery confirmation results, and recurrence status.

[0054] Next, as an example, we will describe an example of how to handle startup failures caused by configuration issues. For example, we will consider a case where the target AI agent 20A fails to start due to a conflict between configuration files.

[0055] When user 30 submits a fault report (step S101), the diagnostic and repair agent 10 retrieves log information and configuration files (step S103), performs configuration error or inconsistency detection, and identifies the inconsistencies (step S104).

[0056] The diagnostic and repair agent 10 generates a repair plan (step S105), applies configuration corrections to resolve inconsistencies, and instructs the target AI agent 20A to restart if necessary (step S106). Subsequently, it performs startup status and predetermined response checks as a recovery confirmation (step S107), generates a report, and accumulates the history (step S108).

[0057] As another embodiment, an example of handling failures caused by dependencies will be described. For example, if communication or authentication to an external dependency 100 fails, the diagnostic and remediation agent 10 will present the failure as a diagnostic result based on dependency verification and may include in the report a response that includes recovery or authentication renewal on the dependency side, rather than just configuration correction.

[0058] In this embodiment, when the repair execution unit 13 makes changes to the configuration file, it retains a backup of the file before the change and performs a rollback if the recovery confirmation is insufficient, thereby ensuring a safe recovery. Furthermore, the scope of automatic repair execution can be controlled, and under predetermined conditions, manual approval may be required.

[0059] The present invention can be implemented in a cloud environment, an on-premises environment, or a combination thereof. The execution mode of the target AI agent 20A, etc., may be a container, a virtual machine, or a physical server, etc.

[0060] The present invention may be provided as a program that causes a computer to execute each function of the diagnostic and repair agent 10. The program may be distributed via a communication line or provided by being recorded on a recording medium.

[0061] Next, we will describe some modifications. The following modifications may be combined as appropriate without departing from the spirit of the present invention.

[0062] <Modification 1> A configuration may be used that uses a management process or management agent as a path to acquire log information and configuration files without going through the execution process of the target AI agent 20A, etc. In this case, the diagnostic / repair agent 10 can acquire evidence via the management process or management agent, and the possibility of diagnosing when the target AI agent 20A, etc. is unable to start can be increased.

[0063] According to this modified example 1, an AI agent self-diagnosis and repair system can be configured that includes a path that enables the acquisition of evidence regardless of the operating status of the target AI agent 20A, etc.

[0064] <Modification 2> Matching with known failure patterns may be implemented using similarity search or a trained classifier in addition to the rule-based approach. In this case, the history stored in the self-repair log DB90 may be used as features and reflected in the ranking of candidate diagnostic results or the determination of whether automatic repair is possible.

[0065] According to this modified example 2, an AI agent self-diagnosis and repair system can be configured that improves the accuracy of diagnostic candidates or the validity of repair selection based on past repair history.

[0066] <Modification 3> Instead of centrally deploying the diagnostic and repair agents 10, they may be distributed across multiple operational domains, with a hierarchical structure where a higher-level management node performs aggregation and reporting. This can improve the scalability of data collection and repair in large-scale environments containing a large number of target AI agents.

[0067] According to this modified example 3, an AI agent self-diagnosis and repair system can be configured that operates a large number of target AI agents using distributed or hierarchical diagnostic and repair agents.

[0068] <Modification 4> When the repair execution unit 13 performs the repair, it may include impact assessment, phased restart, or application of a canary-type approach. For example, if multiple target AI agents share the same configuration template, the configuration modification may be applied to only some of the targets first, and then expanded to the remaining targets after recovery confirmation.

[0069] According to this modified example 4, an AI agent self-diagnosis and repair system can be configured that reduces operational risk by including the application and verification of repair steps.

[0070] <Modification 5> The reports generated by the report generation unit 14 may be linked with an external system for ticket management or operational management and managed as incident records, approvals, and audit trails. In this case, in addition to presenting remediation procedures, approval history and execution history may be stored together.

[0071] According to this modified example 5, an AI agent self-diagnosis and repair system can be configured that integrates reporting and history management in conjunction with the operational management system.

[0072] The embodiments and variations described above are illustrative examples, and the present invention is not limited thereto. [Explanation of symbols]

[0073] 1. AI Agent Self-Diagnosis and Repair System 10 Diagnostic and Repair Agents 11. Symptom Collection Module 12 Diagnostic Engines 13 Repair Execution Unit 14 Report generation section 20A Target AI Agents 20B Target AI Agents 20C Target AI Agents 30 users 60 Log Storage Section 70 Configuration File Storage Section 80 Failure Patterns DB 90 Self-healing log DB 100 External dependencies

Claims

1. An information processing system that includes multiple AI agents is equipped with a diagnostic and repair agent. It is an AI agent self-diagnosis and repair system, The aforementioned diagnostic and repair agent includes a means for receiving fault reports from users regarding the target AI agent, Acquisition means for acquiring log information corresponding to the target AI agent and a configuration file including an agent definition file that defines the operating conditions or operating policy of the target AI agent, even when the target AI agent is unable to start, via a route that does not involve the execution process of the target AI agent. A diagnostic means that generates a diagnostic result by detecting inconsistencies between descriptions in the configuration file or between multiple configuration files based on the log information and the configuration file, and by confirming the success or failure of communication or authentication with external dependencies used by the target AI agent. A repair execution means that modifies the configuration file based on the diagnostic results, A report generation means that generates a report including the diagnostic results and the processing details by the repair execution means, An AI agent self-diagnosis and repair system characterized by having [a certain feature].

2. An AI agent self-diagnosis and repair system according to claim 1, wherein the agent definition file includes at least one of the range of tools available to the target AI agent, constraints on the output format, or a safety policy.

3. An AI agent self-diagnosis and repair system according to claim 1 or 2, wherein the diagnostic means further performs a comparison with known failure patterns based on the error characteristics contained in the log information, the characteristics of the configuration file, or the symptom description contained in the failure report.

4. An AI agent self-diagnosis and repair system according to claim 1, characterized in that the detection of inconsistencies by the diagnostic means includes detecting at least one of the following: missing essential items, mismatch in type or range, contradictory values ​​for the same key, absence of a referenced object, or conflict in authorization policies.

5. An AI agent self-diagnosis and repair system according to claim 1, wherein the external dependency includes at least one of an external service, a data store, a model provision unit, or an authentication infrastructure, and the system is characterized by confirming the success or failure of communication or authentication with the external dependency.

6. An AI agent self-diagnosis and repair system according to claim 1, wherein the repair execution means, prior to modifying the configuration file, maintains the configuration file as a backup and modifies the configuration file by applying a differential patch to the configuration file.

7. An AI agent self-diagnosis and repair system according to claim 1, wherein the repair execution means provides a service restart instruction to prompt the restart or redeployment of the target AI agent.

8. An AI agent self-diagnosis and repair system according to claim 1, wherein the repair execution means performs a rollback to restore the configuration file or related resources to a predetermined past state.

9. An AI agent self-diagnosis and repair system according to claim 1, wherein the report generation means generates a report that includes at least one of a candidate cause, a difference in the underlying log or settings, the repair procedure performed, and the confirmation result.

10. An AI agent self-diagnosis and repair system according to claim 1, wherein the diagnosis and repair agent further comprises storage means for accumulating a self-repair log that includes at least one of the diagnosis result, repair procedure, success or failure of repair, and whether or not the problem recurs.

11. An AI agent self-diagnosis and repair system according to claim 10, wherein the diagnosis and repair agent updates the failure pattern DB based on the self-repair log stored in the storage means.

12. An AI agent self-diagnosis and repair system according to claim 1, wherein the path includes a path through a file system, an execution platform, a management process, a management agent, or a management interface.

13. An AI agent self-diagnosis and repair system according to claim 1, wherein the receiving means identifies the target AI agent among the plurality of AI agents based on the identification information contained in the failure report.

14. An AI agent self-diagnosis and repair system according to claim 3, characterized in that the comparison with known failure patterns by the diagnostic means includes identifying the known failure pattern based on at least one of the log information and the configuration file.

15. An AI agent self-diagnosis and repair method executed in an information processing system including multiple AI agents, The process of receiving bug reports from users regarding the target AI agent, A step of obtaining log information corresponding to the target AI agent and a configuration file including an agent definition file that defines the operating conditions or operating policy of the target AI agent, even if the target AI agent is unable to start, by a route that does not involve the execution process of the target AI agent. The process involves generating a diagnostic result by detecting inconsistencies between descriptions in the configuration file or between multiple configuration files based on the log information and the configuration file, and by confirming the success or failure of communication or authentication with external dependencies used by the target AI agent. A step of modifying the configuration file based on the diagnostic results, A step of generating a report that includes the diagnostic results and the processing details resulting from the modification of the configuration file, An AI agent self-diagnosis and repair method characterized by including [a specific element].

16. An AI agent self-diagnosis and repair method according to claim 15, characterized in that the path includes a path through a file system, an execution platform, a management process, a management agent, or a management interface.

17. An AI agent self-diagnosis and repair method according to claim 15 or 16, wherein after the step of modifying the configuration file, The process of instructing the AI ​​agent to restart its service, If the recovery confirmation result after the service restart is insufficient, the process includes performing a rollback to restore the configuration file or related resources to a predetermined past state, An AI agent self-diagnosis and repair method characterized by further including the following.

18. An AI agent self-diagnosis and repair method according to claim 15, wherein the step of generating the report is characterized in that the report includes at least one of a candidate cause, a difference in the underlying log or settings, the repair procedure performed, and the confirmation result.

19. A program that causes a computer to execute the AI ​​agent self-diagnosis and repair method described in claim 15.

20. A computer-readable recording medium on which the program described in claim 19 is recorded.

Images