Purpose
This article explains how to access and use the platform’s centralized logging system to inspect execution behavior, investigate failures, and troubleshoot issues across services.
Overview
The platform uses VictoriaLogs as its centralized logging solution. It provides a log query interface based on LogQL, similar in usage style to Loki, and aggregates logs across the platform so users can inspect what happened during workflow execution or service runtime.
This makes it possible to debug both offline and online behavior from a single logging interface rather than checking logs independently on each component.
When to use centralized logging
Use centralized logging when you want to:
- inspect what happened during a workflow run
- troubleshoot failing tasks or services
- investigate unexpected runtime behavior
- review logs emitted by Ray, Airflow, or other platform services
- understand execution details across multiple components
Centralized logging is one of the main entry points for debugging operational issues.
How to access the logging UI
To open the centralized logging interface:
- open the platform landing UI
- click the Logs button
This opens the VictoriaLogs interface.
How the interface works
VictoriaLogs provides a query interface that uses LogQL-style syntax. The experience is similar to querying logs in Loki: users can search, filter, and inspect log streams to understand what happened during execution.
The interface is designed for interactive investigation. It allows you to narrow results, inspect relevant log lines, and focus on the services or jobs you are interested in.
What you can inspect
The centralized logging system can be used to inspect logs from multiple parts of the platform, including:
- orchestration-related services
- Airflow-driven workflows
- Ray workloads
- deployed online services
- general platform services
Because the logs are aggregated centrally, this interface is often the fastest way to understand what happened without switching across multiple tools.
Typical workflow for log inspection
A common way to use centralized logging is:
- identify the failing or suspicious workflow, service, or time window
- open VictoriaLogs from the Logs button in the platform entry UI
- run a LogQL query or filter by the relevant component
- inspect the returned log lines
- correlate the results with what you see in Airflow, Ray, or the chat/runtime interfaces
This makes centralized logging a bridge between orchestration state and actual runtime behavior.
When centralized logging is especially useful
Investigating failed workflows
If a DAG fails in Airflow or an agentic deployment behaves unexpectedly, centralized logs can show the execution details behind the failure.
Inspecting Ray behavior
Ray logs are centralized in VictoriaLogs, so this interface can be used even when the originating execution happened inside Ray workers or Ray Serve deployments.
Understanding multi-component failures
In distributed systems, failures often span more than one service. Centralized logging makes cross-service debugging much easier.
Centralized logging versus component-local logs
Some platform components, such as Ray, also expose their own interfaces and logs. Those remain useful, but centralized logging provides an aggregated view.
A good mental model is:
- use component-local UIs when you are already focused on one subsystem
- use centralized logging when you want one place to investigate behavior across the platform
Best practices
Start with the simplest scope
Begin by narrowing the logs to the relevant time window, component, or workflow before expanding the search.
Correlate logs with the workflow stage
If you are debugging an offline flow, compare the logs with the failing DAG task or stage. If you are debugging an online system, compare them with the observed agent or tool behavior.
Use centralized logging early in debugging
It is often faster to inspect the logs first than to guess which subsystem caused the problem.
Keep the configuration nearby
Logs are easier to interpret when you can compare them to the YAML configuration that defines the relevant flow or service.
Comments
0 comments
Article is closed for comments.