Use Workbench AI to solve performance issues

Workbench AI is an intelligent companion for RCA workbench. The core functionalities include:

Workbench Commentary: Utilizes GenAI to explain timelines and provide insights.
Contextual Analysis: Leverages the knowledge graph to provide context to Large Language Models (LLMs) for effective root cause analysis.
Automated Workflow: Quickly diagnose issues by automating common investigation workflows.
Natural Language Telemetry Discovery: Use natural language to effortlessly locate and query your entities, logs, and metrics.

Prerequisites

To begin using Workbench AI, first you need to enable LLM features. Refer to Enable LLM features.

Open Workbench AI

After adding entities to RCA workbench, click the blue icon in the bottom-right corner to activate Workbench AI. You can expand the tab by clicking the maximize button in the top-right corner.

To ensure better performance, you should narrow the time range to the specific incident window you want to analyze. After Workbench AI is open, allow the analysis to complete.

Natural language queries

Built-in tools empower AI to answer questions and investigate incidents using natural language. This allows you to check entity health and retrieve telemetry data like logs, which the AI then uses to enhance its incident analysis.

Many of the tool calls are anchored on a set of entities. If you don’t specify which entity, Workbench AI picks a few it thinks are relevant. If you want to be specific, use the @ command to pick specific entities when asking questions or giving instructions.

Some questions prompt Workbench AI to make multiple tool calls. For example, you can ask: “Show the request rate of all services that connect to my service”.

Use automated workflows

To streamline common troubleshooting workflows, Workbench AI offers a set of more complex tools. The most frequently used ones, such as “find relevant logs,” are accessible via the / command and highlighted in Next steps.

Find relevant amend assertions searches for amend assertions in the same namespace or cluster and examines their timestamps to see which ones are relevant to the current incident window
Find relevant logs searches for logs relevant to current incident
Find problematic connections search for first-degree connected entities
Find potential cause by common patterns search for connected entities with more complex architectural patterns, such as Outbound/Inbound dependencies, Node issue, Noisy neighbor, and so on.

Interact with timeline view

All the entities mentioned by AI output should have a context menu, so you can examine its KPIs, and add it to RCA workbench.

This helps you to compare the timeline visuals with AI-generated investigation results. Whenever Workbench AI state changes, you have a choice to restart the whole chat or just update the current chat to continue the conversation.