작업 이상치는 최근 데이터 패턴을 기준으로 API에서 비정상적이거나 예기치 않은 API 데이터 패턴을 식별합니다. 예를 들어 이 API 오류율 그래프에서 오류율은 오전 7시에 갑자기 증가합니다. 이 시점까지의 데이터와 비교했을 때 이러한 증가는 이상으로 분류하기에 충분할 정도로 비정상적입니다.
API 데이터의 모든 편차가 이상을 반영하는 것은 아닙니다. 대부분 무작위 변동입니다. 예를 들어 오류율에서 최대한의 경우 이상으로 여겨질 수 있는 사소한 편차가 나타나더라도 이것만으로는 이상으로 분류하기에 충분하지 않습니다.
작업 이상치에서는 API 데이터를 지속적으로 모니터링하고 통계 분석을 수행하여 실제 이상을 데이터의 무작위 변동과 구분합니다.
작업 이상치는 다음과 같은 이상 유형을 자동으로 감지합니다.
조직, 환경, 리전 수준에서 HTTP 503 오류 증가
조직, 환경, 리전 수준에서 HTTP 504 오류 증가
조직, 환경, 리전 수준에서 모든 HTTP 4xx 또는 5xx 오류 증가
조직, 환경, 리전 수준에서 90번째 백분위수(p90)에 대한 총 응답 지연 시간 증가
감지된 이상은 다음 정보를 포함합니다.
프록시 지연 시간이나 HTTP 오류 코드와 같은 이상을 일으킨 측정항목입니다.
이상의 심각도입니다. 심각도는 모델의 신뢰도 수준에 따라 적음, 보통, 심각일 수 있습니다. 신뢰도 수준이 낮으면 심각도가 적음을 나타내고 높은 신뢰도 수준은 심각함을 나타냅니다.
작업 이상치는 이전 시계열 데이터에서 API 프록시 동작의 모델을 학습하는 방식으로 작동합니다. 모델 학습을 위해 개발자가 별도로 취해야 할 조치는 없습니다. Apigee는 지난 6시간 동안의 API 데이터에서 자동으로 모델을 만들고 학습합니다.
따라서 Apigee는 이상치를 로깅하기 전에 모델을 학습하기 위해 API 프록시에 대한 최소 6시간의 데이터가 필요합니다.
학습의 목표는 모델의 정확성을 개선하는 것이며 그런 다음 과거 데이터에서 테스트할 수 있습니다. 모델의 정확성을 테스트하는 가장 간단한 방법은 오류율(거짓양성 및 거짓음성의 합계를 총 예측 이벤트 수로 나눈 값)을 계산하는 것입니다.
이상 이벤트 로깅
런타임에 작업 이상치는 API 프록시의 현재 동작을 모델에서 예측한 동작과 비교합니다. 그런 다음 작업 이상치는 특정 신뢰도 수준을 사용하여 운영 측정항목이 예측 값을 초과하는 시기를 확인할 수 있습니다. 예를 들어 5xx 오류 비율이 모델이 예측하는 비율을 초과할 때입니다.
Apigee는 이상을 감지하면 작업 이상치 대시보드에 이벤트를 자동으로 로깅합니다. 대시보드에 표시되는 이벤트 목록에는 감지된 모든 이상과 트리거된 알림이 포함됩니다.
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["이해하기 어려움","hardToUnderstand","thumb-down"],["잘못된 정보 또는 샘플 코드","incorrectInformationOrSampleCode","thumb-down"],["필요한 정보/샘플이 없음","missingTheInformationSamplesINeed","thumb-down"],["번역 문제","translationIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-08-18(UTC)"],[[["\u003cp\u003eOperations Anomalies identifies unusual API data patterns, distinguishing them from random fluctuations based on recent data.\u003c/p\u003e\n"],["\u003cp\u003eThis feature, available in the Apigee UI in Cloud console, detects increases in specific HTTP errors and latency levels.\u003c/p\u003e\n"],["\u003cp\u003eUsing Operations Anomalies requires the AAPI Ops add-on to be enabled, along with the appropriate user roles and logging access.\u003c/p\u003e\n"],["\u003cp\u003eApigee automatically trains models from the previous six hours of API data to detect anomalies, and a minimum of six hours of API data is needed to train a model.\u003c/p\u003e\n"],["\u003cp\u003eDetected anomalies are displayed on the Operations Anomalies dashboard, providing details such as the time, summary, environment, region, and severity of the anomaly.\u003c/p\u003e\n"]]],[],null,["# Operations Anomalies overview\n\n*This page\napplies to **Apigee** and **Apigee hybrid**.*\n\n\n*View [Apigee Edge](https://docs.apigee.com/api-platform/get-started/what-apigee-edge) documentation.*\n\n| **Important:** This page describes how to use Operations Anomalies, which is comparable to the [Advanced API Operations Anomaly Detection](/apigee/docs/aapi-ops) functionality in the Classic Apigee UI. Operations Anomalies is only available in the [Apigee UI in Cloud console](https://console.cloud.google.com/apigee) while Anomaly Detection is only available when using the classic [Apigee UI](https://apigee.google.com). Both are available at this time.\n\nOperations Anomalies overview\n-----------------------------\n\nOperations Anomalies identifies unusual or unexpected API data patterns on your APIs,\nbased on recent data patterns. For example,\nin this graph of API error rate, the error rate suddenly jumps up at around 7 AM. Compared\nto the data leading up to that time, this increase is unusual enough to be classified as an anomaly.\n\nNot all variations in API data represent anomalies: most\nare random fluctuations. For example, you can see some minor variations in\nerror rate leading up to the anomaly, but these are not significant enough to be categorized as an\nanomaly.\n\nOperations Anomalies continually monitors API data and performs statistical analysis to distinguish true\nanomalies from random fluctuations in the data.\n\nOperations Anomalies automatically detects these anomaly types:\n\n- Increase in HTTP 503 errors at the organization, environment, and region level\n- Increase in HTTP 504 errors at the organization, environment, and region level\n- Increase in all HTTP 4xx or 5xx errors at the organization, environment, and region level\n- Increase in the total response latency for the 90th percentile (p90) at the organization, environment, and region level\n\nA detected anomaly includes this information:\n\n- The metric that caused the anomaly, such as proxy latency or an HTTP error code.\n- The severity of the anomaly. The severity can be slight, moderate, or severe, based on its confidence level in the model. A low confidence level indicates that the severity is slight, while a high confidence level indicates that it is severe.\n\nPrerequisites for using Operations Anomalies\n--------------------------------------------\n\nTo use Operations Anomalies:\n\n- The AAPI Ops add-on must be enabled for your organization. See [Enable AAPI Ops in an organization](/apigee/docs/aapi-ops#enable).\n- Users of Operations Anomalies must have the [required roles for AAPI Ops](/apigee/docs/aapi-ops#required-roles-for-aapi-ops).\n- Users who [investigate anomalies in the dashboard](/apigee/docs/api-platform/analytics/investigate-anomalies) also need the `roles/logging.viewer` role.\n\n\u003cbr /\u003e\n\nView detected Operations Anomalies\n----------------------------------\n\nWhen Operations Anomalies detects an anomaly, it displays the anomaly details in the\nOperations Anomalies dashboard.\nYou can investigate the anomaly in the API Monitoring dashboards and\ntake appropriate action if necessary. You can also\ncreate an alert to notify you if similar events occur in future.\n\nThe Operations Anomalies dashboard in the Apigee UI is your primary source of information about\ndetected Operations Anomalies. The dashboard displays a list of recent anomalies.\n\nTo open the Operations Anomalies dashboard:\n\n1. Sign in to [Apigee UI in Cloud console](https://console.cloud.google.com/apigee).\n2. [Switch to the organization](/apigee/docs/api-platform/get-started/switch-org) that you want to monitor.\n3. In the left menu, select **Analytics \\\u003e Operations Anomalies**.\n\nThis displays the Operations Anomalies dashboard.\n\nBy default, the dashboard shows anomalies that have occurred during the previous hour.\nIf no anomalies have been detected during that time period, no rows are\ndisplayed in the dashboard. You can select a larger time range from\nthe time range menu in the top right of the dashboard.\n\nEach row in the table corresponds to a detected anomaly,\nand displays the following information:\n\n- The date and time of the anomaly.\n- A brief summary of the anomaly, including the proxy in which it occurred and the fault code that triggered it.\n- The environment in which the anomaly occurred.\n- The region where the anomaly occurred.\n- The severity of the anomaly event: slight, moderate, or severe. Severity is based on a statistical measure (p-value) of how unlikely it would be for the event to occur by chance (the more unlikely the event, the greater its severity).\n\nYou can also\n[investigate an anomaly](/apigee/docs/api-platform/analytics/investigate-anomalies)\nin the API Monitoring dashboards, which shows various graphs of recent API traffic\ndata.\n\nHow anomaly detection works\n---------------------------\n\nAnomaly detection involves the following stages:\n\n- [Train models](#train-models)\n- [Log anomaly events](#log-anomaly-events)\n\n### Train models\n\nOperations Anomalies works by training a model of the behavior of your API proxies from historical\ntime-series data. There is no action required on your part to train the model. Apigee automatically\ncreates and trains models for you from the previous six hours of API data.\nTherefore, Apigee requires a minimum of six hours of data on an API proxy to train the model before\nit can log an anomaly.\n\nThe goal of training is to improve the accuracy of the model, which can then be tested\non historical data. The simplest\nway to test a model's accuracy is to calculate its *error rate*---the\nsum of false positives and false negatives, divided by the total number of predicted events.\n\n### Log anomaly events\n\nAt runtime, Operations Anomalies compares the current behavior of your API proxies with the behavior\npredicted by the model. Operations Anomalies can then determine, with a specific confidence level,\nwhen an operational metric is exceeding the predicted value. For example, when the rate of 5xx errors\nexceeds the rate predicted by the model.\n\nWhen Apigee detects an anomaly, it automatically logs the event in the\n[Operations Anomalies\ndashboard](#view-detected-operations-anomalies). The list of events displayed in the dashboard includes all\ndetected anomalies, as well as triggered alerts."]]
