Platform For AI:Logging, monitoring, and alerting

Overview

Logs, monitoring, and alerts each play a distinct role in incident response:

• Logs: Capture detailed service output—startup messages, request handling, and error stack traces—so you can pinpoint the root cause of an issue.

• Monitoring: Surface service trends through metrics and charts (QPS, latency, error rate, resource usage) so you can detect anomalies and plan capacity.

• Alerts: Apply thresholds to monitoring metrics and notify you when something goes wrong, prompting timely action.

A typical troubleshooting path: an alert fires → monitoring narrows down the affected metric and time window → logs reveal the root cause.

Logs

EAS provides the following logging and recording capabilities. Choose based on your use case:

• Historical log queries: Query past logs on the Log tab of the service details page, with per-instance filtering, keyword search, and time-range filtering. Best for log retrieval and analysis.

• Real-time log tracking: Select an instance in the Service Instance list, then click Real-time Logs to stream the container's rolling output. Best for interactive debugging and scenarios requiring real-time visibility, such as watching token-by-token inference output.

  Note

  Real-time logs aren't available for services created before this feature launched. Recreate the service to use them.

• SLS log collection: Collect EAS service logs into SLS through three integration methods—Insights tab for quick onboarding, Cloud Monitor 2.0 Access Center for batch integration, and SLS Logtail for manual configuration—covering scenarios from single-service quick setup to advanced custom collection. For details, see Configure log collection.

• Call record persistence: Save all service request and response records to MaxCompute or SLS for auditing, call analysis, or troubleshooting. To enable, go to the Features section during service deployment and enable Save Call Records, then select a storage target:

  • MaxCompute: Select an existing MaxCompute Project (if none exists, see Create a MaxCompute project), and configure the MaxCompute Table name. The system automatically creates this table in the selected project when the service is deployed.

  • Simple Log Service: Select an existing Simple Log Service Project (if none exists, see Create a project), and configure the logstore name. The system automatically creates this Logstore in the selected project when the service is deployed.

Monitoring

EAS provides five monitoring capabilities. Choose based on monitoring scope and metric source:

• Service monitoring (per service): View basic runtime metrics such as QPS, latency, error rate, and resource usage for a single service, with Service and Instance dimension switching. Works out of the box—open the Monitoring tab on the service details page. For details, see View service monitoring metrics.

• Monitoring dashboard (region-level aggregation): Aggregate service metrics from all workspaces under the current account in the selected region onto a unified dashboard for multi-service health checks and capacity planning. Switching workspaces does not affect the data scope. For details, see Monitoring dashboard.

• Custom monitoring metrics (per service, user-reported): Report business metrics such as inference token count, batch size, and hit rate to drive auto scaling. Requires instrumentation in your service code. For details, see Custom monitoring and auto scaling metrics.

• Prometheus monitoring (account-level, external system integration): Connect EAS monitoring metrics to Prometheus through the Cloud Monitor 2.0 Access Center, query metrics with PromQL, and integrate with Grafana or other monitoring systems. Includes advanced metrics such as inference framework metrics and GPU compute metrics. Additional charges apply. For details, see Monitor EAS inference services with Prometheus.

• Tracing (per service, ARMS): ARMS-based call chain tracing for pinpointing cross-component performance bottlenecks. Some official images support one-click activation; for other images, integrate the ARMS probe into the runtime command. For details, see Enable tracing for LLM applications in EAS.

Alerts

EAS supports two alerting scenarios. Choose based on the notification trigger source:

• Metric alerts: Notify you when monitoring metrics—such as error rate, latency, or resource usage—exceed a configured threshold. Build rules from service monitoring metrics. For details, see Configure service monitoring alerts.

• CloudMonitor events: Listen for service lifecycle events—such as deployment completion, scaling, and unexpected instance exits—and send notifications. For details, see View EAS events in CloudMonitor.

FAQ