Docs

Volcano Engine Function as a Service

Monitoring and logs

Enabling veFaaS application monitoring

This topic describes how to enable veFaaS application performance monitoring for full observability of function execution.

Background information

veFaaS is deeply integrated with APMPlus. You can enable application monitoring for functions deployed on veFaaS to implement automatic ingestion of function monitoring data into APMPlus. After you enable application monitoring, you can get capabilities such as automatic trace data collection, automatic metric generation, trace query, and log query.

Prerequisites

APMPlus has been activated, and server-side monitoring has been enabled. For detailed steps, see Enabling server-side monitoring.

A function has been created. For detailed steps, see Function type selection.

Limits

veFaaS application monitoring supports Python and Native Python runtimes. More runtimes will be supported in the future.

Procedure

Step 1: Enable application monitoring

Enable application monitoring when creating the function

When creating the function in the veFaaS console, enable application monitoring. For detailed steps, see Event functions and Web applications.

Enable application monitoring for an existing function

In the veFaaS console, select Function in the left navigation pane.

On the Function page, find the target function and click its name to go to the function details page.

On the Function settings tab, enable App Monitoring and save your configuration. Then, release the function again.

After the APMPlus monitoring service is enabled, the monitoring data related to the function will be delivered to APMPlus. APMPlus provides an integrated monitoring service covering metrics, distributed tracing, and logs, and generates out-of-the-box monitoring dashboards. It also supports monitoring pro-code applications running on Volcano Engine ModelArk, as well as AI agents such as LangChain.

Step 2: View monitoring data

In the veFaaS console, select Function in the left navigation pane.

On the Function page, find the target function and click its name to go to the function details page.

On the App monitoring tab, go to the AI Application Trace Analysis section and filter the conversation-related traces by span attributes. In this scenario, the conversation input apmplus.entity.input is used to filter traces for this request.

In the Trace Analysis section, you can view the relevant traces. When you hover over the Input/output button, the input and output details of the agent are automatically displayed.

Click TraceID to view the complete trace and topology, with three views available: flame graph, list, and topology graph.

Click the Metric tab on the right and then LLM to view the LLM metrics associated with the current trace. For example, you query metrics like token usage, LLM response latency, TPOT, and TTFT.

Select the ark_runtime.chat span in the flame graph to view the input and output details of the conversation in the Attribute section on the right. Scroll down to view more agent-related information in the Span Tag section, for example, token usage, TPOT, and TTFT.

Select a span with an alert in the flame graph to view the error stack and event details in the Event section on the right. This information facilitates trace analysis and troubleshooting.

Go back to the App monitoring tab, and in the Runtime monitoring section, you can view agent runtime information, for example, CPU time, CPU usage, memory usage, GC object count, thread count, and context switch count.

Usage example

In MaaS (Model as a Service) scenarios, use veFaaS, together with Volcano Engine ModelArk, to monitor code-intensive applications and generate LLM-related traces and metrics. For details, see Code-intensive MaaS trace monitoring.

Last updated: 2026.08.19 14:06:07