Axiom helps you leverage the power of timestamped event data. A common use case of event data is observability (o11y) in the field of software engineering. Observability is the ability to explain what’s happening inside a software system by observing it from the outside. It allows you to understand the behavior of systems based on their outputs such as telemetry data, which is a type of event data. Software engineers most often work with timestamped event data in the form of logs or metrics. However, Axiom believes that event data reflects a much broader range of interactions, crossing boundaries from engineering to product management, security, and beyond. For a more general explanation of event data in Axiom, see Events.

​
Types of event data in observability

Traditionally, observability has been associated with three pillars, each effectively a specialized view of event data:
  • Logs: Logs record discrete events, such as error messages or access requests, typically associated with engineering or security.
  • Traces: Traces track the path of requests through a system, capturing each step’s duration. By linking related spans within a trace, developers can identify bottlenecks and dependencies.
  • Metrics: Metrics quantify state over time, recording data like CPU usage or user count at intervals. Product or engineering teams can then monitor and aggregate these values for performance insights.
In Axiom, these observability elements are stored as event data, allowing for fine-grained, efficient tracking across all three pillars.

​
Logs and traces support

Axiom excels at collecting, storing, and analyzing timestamped event data. For logs and traces, Axiom offers unparalleled efficiency and query performance. You can send logs and traces to Axiom from a wide range of popular sources. For information, see Send data to Axiom .

​
Metrics support

For metrics data, Axiom is well-suited for event-level metrics that behave like logs, with each data point representing a discrete event. For example, you have the following timestamped data in Axiom: 
{
  "job_id": "train_123",
  "user_name": "acme",
  "timestamp": "2024-10-08T15:30:00Z",
  "node_host": "worker-01",
  "metric_name": "gpu_utilization",
  "metric_value": 87.5,
  "training_type": "image_classification"
}
You can easily query and analyze this type of metrics data in Axiom. The query below computes the average GPU utilization across nodes: 
dataset
| summarize avg(metric_value) by node_host, bin_auto(_time)
Axiom’s support for metrics data currently comes with the following limitations:
  • Axiom doesn’t support pre-aggregated metrics such as scrape samples.
  • Axiom isn’t optimized for high-dimensional metric time series with a very large number of metric/label combinations.
Support for these types of metrics data is coming soon in the first half of 2025.