ДалееСледующее руководство
Контракты данных для конвейеров машинного обучения
Технический
Техническое РУКОВОДСТВО
Цель уровня обслуживания (SLO) — это измеримая цель для индикатора обслуживания, такого как доступность или задержка в течение определенного окна.
Прослушайте это руководствоОзвучено ИИ · ~3 min
ML services can add quality and freshness indicators, but alerts should distinguish user-impacting symptoms from noisy model metrics and route each failure to an owner.
An SLO states a target for a service-level indicator (SLI) over a time window. Common SLIs include availability, latency, throughput and error rate. ML services may also monitor prediction freshness, feature availability, queue delay or quality once outcomes arrive. The SLO should reflect what users need and what the service can measure reliably. A clear denominator matters: availability might be the fraction of eligible requests that succeed, with exclusions defined. Latency objectives often focus on a percentile rather than the mean because a small slow tail can affect users even when average latency is low. Batch inference may use deadlines, completion rates and data freshness instead of request-level latency. A feature pipeline can have its own SLO if stale features make predictions less useful. Model accuracy usually cannot be measured immediately without labels, so quality SLOs may lag or use carefully validated proxies. Alerts should prompt action. Page for urgent user-impacting conditions, such as sustained errors or exhausted capacity; use tickets or dashboards for slower trends. Error budgets quantify how much unreliability is allowed under a target and can guide release pace, but they do not replace product judgment. Alerts need windows, thresholds, routing and runbooks. A noisy alert creates fatigue, while a poorly chosen indicator can stay green as user experience declines. Separate service SLOs from model-quality goals. A service can be available but serve stale or low-quality predictions. Conversely, model metrics may shift for benign population reasons while requests remain healthy. Define data sources, ownership, exclusions, aggregation and burn-rate behavior. Review incidents and user feedback to update SLOs. Monitoring should support service reliability and model oversight without turning every statistical fluctuation into an emergency page.
A prediction API defines an availability SLO over successful eligible requests and tracks the error budget consumed during a rolling period.
A latency SLO uses a percentile target, such as a specified share of requests completing within a budget, rather than averaging away slow-tail requests.
A batch scoring service monitors completion by a deadline and freshness of feature snapshots, which may matter more than per-request latency.
A model-quality alert waits for labels to mature, while a separate page fires immediately for elevated service errors; each signal has a different response owner.
ML service SLOs can become more useful when they cover user-facing reliability, feature freshness and delayed quality evidence without mixing them into one opaque score. Teams should review objectives after incidents and track error-budget use across releases. Batch workloads need completion deadlines and data-readiness checks, while online services need latency and availability views. Alerting can combine fast operational pages with slower model-quality investigations. Clear ownership and runbooks make the objective actionable and reliable as architectures, use cases and traffic patterns evolve.
Бесплатная рассылка
Три проверенных AI-истории каждое утро буднего дня, написанные простым английским. Бесплатно навсегда, без рекламы.
Одно электронное письмо каждый будний день. Отписаться в один клик. Мы никогда не продаем и не передаем ваш адрес.
Проверьте себя
Мгновенная обратная связь по каждому ответу и общий сертификат с проверяемым идентификатором после прохождения курса.
Поддержите бесплатное образование в области искусственного интеллекта. AI Understanding — это некоммерческая организация 501(c)(3). Никакой рекламы и платного доступа. Сделать пожертвование
A service-level objective (SLO) is a measurable target for a service indicator such as availability or latency over a defined window. ML services can add quality and freshness indicators, but alerts should distinguish user-impacting symptoms from noisy model metrics and route each failure to an owner.
An SLO sets a target for an SLI such as latency or availability over a time window.
Averages can hide a small but user-impacting tail of slow requests.
Batch workflows often care about timely completion and freshness rather than online request latency.
An error budget is the permitted bad-event fraction implied by the objective over its window.
Infrastructure can be healthy while prediction quality changes, and quality often has delayed labels.
Продолжайте учиться
Другие руководства, выбранные по этой теме
ДалееСледующее руководство
Контракты данных для конвейеров машинного обучения
Технический