Explore traces

Learn how to observe how requests propagate through your distributed systems, understand the interactions between microservices, and trace the life of the request through your app’s architecture.

Distributed tracing in Axiom allows you to observe how requests propagate through your distributed systems. This could involve a user request going through several microservices, and resources until the requested information is retrieved and returned. By tracing these requests, you’re able to understand the interactions between these microservices, pinpoint issues, understand latency, and trace the life of the request through your app’s architecture.

Traces and spans

A trace is a representation of a single operation or transaction as it moves through a system. A trace is made up of multiple spans.

A span represents a logical unit of work in the system with a start and end time. For example, an HTTP request handling process might be a span. Each span includes metadata like unique identifiers (trace_id and span_id), start and end times, parent-child relationships with other spans, and optional events, logs, or other details to help describe the span’s operation.

Trace schema overview

Field	Type	Description
trace_id	String	Unique identifier for a trace
span_id	String	Unique identifier for a span within a trace
parent_span_id	String	Identifier of the parent span
name	String	Name of the span for example, the operation
kind	String	Type of the span (for example, client, server, producer)
duration	Timespan	Duration of the span
error	Boolean	Whether this span contains an error
status.code	String	Status of the span (for example, null, OK, error)
status.message	String	Status message of the span
attributes	Object	Key-value pairs providing additional metadata
events	Array	Timestamped events associated with the span
links	Array	Links to related spans or external resources
resource	Object	Information about the source of the span

This guide explains how you can use Axiom to analyze and interrogate your trace data from simple overviews to complex queries.

Browse traces with the OpenTelemetry app

The Axiom OpenTelemetry app automatically detects any OpenTelemetry trace data flowing into your datasets and publishes an OpenTelemetry Traces dashboard to help you browse your trace data.

The following fields are expected to display the OpenTelemetry Traces dashboard: duration, kind, name, parent_span_id, service.name, span_id, and trace_id.

Navigate the app

• Use the Filter Bar at the top of the app to narrow the charts to a specific service or operation.
• Use the Search Input to find a trace ID in the selected time period.
• Use the Slowest Operations chart to identify performance issues across services and traces.
• Use the Top Errors list to quickly identify the worst-offending causes of errors.
• Use the Results table to get an overview and navigate between services, operations, and traces.

View a trace

Click a trace ID in the results table to show the waterfall view. This view allows you to see that span in the context of the entire trace from start to finish.

Customize the app

To customize the app, use the fork button to create an editable duplicate for you and your team.

Query traces

In Axiom, trace events are just like any other events inside datasets. This means they’re directly queryable in the UI. While this is can be a powerful experience, it’s important to note some important details to consider before querying:

• Directly aggregating upon the duration field produces aggregate values across every span in the dataset. This is usually not the desired outcome when you want to inspect a service’s performance or robustness.

• For request, rate, and duration aggregations, it’s best to only include the root span using isnull(parent_span_id).

Waterfall view of traces

To see how spans in a trace are related to each other, explore the trace in a waterfall view. In this view, each span in the trace is correlated with its parent and child spans.

To explore spans within a trace using the OpenTelemetry Traces app, follow these steps:

1. Click the Dashboards tab.
2. Click OpenTelemetry Traces.
3. In the Slowest Operations chart, click the service that contains the trace.
4. In the list of trace IDs, click the trace you want to explore.
5. Explore how spans within the trace are related to each other in the waterfall view. To reveal additional options such as collapsing and expanding child spans, right-click a span.

To try out this example, go to the Axiom Playground.

Run in Playground

To access the waterfall view from the Query tab, follow these steps:

1. Ensure the dataset you work with has trace data.

2. Click the Query tab.

3. Run a query that returns the _time and trace_id fields. For example, the following query returns the number of spans in each trace:

   ['otel-demo-traces']
   | summarize count() by trace_id
   

4. In the list of trace IDs, click the trace you want to explore. To reveal additional options such as copying the trace ID, right-click a trace.

5. Explore how spans within the trace are related to each other in the waterfall view. To reveal additional options such as collapsing and expanding child spans, right-click a span. Event names are displayed on the timeline for each span.

To try out this example, go to the Axiom Playground.

Run in Playground

Span duration histogram

In the waterfall view of traces, Axiom warns you about slow and fast spans. These spans are outliers because they’re at least a standard deviation over or under the average duration of spans that have the same span name and service name. Hold the pointer over the SLOW or FAST label to see additional information about the span type such as average and maximum duration. In addition, Axiom displays a histogram about the durations of spans that have the same span name and service name as the span you selected. By default, the histogram shows a one-hour window around the selected span.

The span duration histogram can be useful in the following cases, among others:

• You look at a span and you’re not familiar with the typical behavior of the service that created it. You want to know if you look at something normal in terms of duration or an outlier. The histogram helps you determine if you look at an outlier and might drill down further.
• You’ve found an outlier. You want to investigate and look at other outliers. The histogram shows you what the baseline is and what’s not normal in terms of duration. You want to filter for the outliers and see what they have in common.
• You want to see if there was a recent change in the typical duration for the selected span type.

To narrow the time range of the histogram, click and select an area in the histogram.

Example queries

Below are a collection of queries that can help get you started with traces inside Axiom. Queries are all executable on the Axiom Play sandbox.

Number of requests, average response

['otel-demo-traces']
| where isnull(parent_span_id)
| summarize count(),
            avg(duration),
            percentiles_array(duration, 95, 99, 99.9)
  by bin_auto(_time)

Top five slowest services by operation

['otel-demo-traces']
| summarize count(), avg(duration) by name
| sort by avg_duration desc
| limit 5

Top five errors per service and operation

['otel-demo-traces']
| summarize topk(['status.message'], 5) by ['service.name'], name
| limit 5

Semantic Conventions

OpenTelemetry defines Semantic Conventions which specify standard attribute names and values for different kinds of operations and data. Attributes that follow semantic conventions will be available as nested fields under the attributes field, such as attributes.http.method, attributes.db.system, etc.

For example, if a span represents an HTTP request, it may include the following attributes:

• attributes.http.method: The HTTP request method. For example, GET, POST, etc.
• attributes.http.url: The full HTTP request URL.
• attributes.http.status_code: The HTTP response status code.

Similarly, resource attributes that follow semantic conventions are available under the resource field, such as resource.host.name, resource.host.id, resource.host.os, etc.

Custom attributes that don’t match any semantic conventions are nested under the attributes.custom map field.

Querying custom attributes

Trace spans often include many custom attributes under the attributes.custom field. These custom attributes are stored as nested key-value pairs.

To access nested custom attributes, you can use Axiom Processing Language (APL) for example:

['otel-demo-traces'] 
| where ['attributes.custom']['app.synthetic_request'] == true

If you frequently need to query the same nested attribute, consider creating a virtual field for it:

1. Go to “Datasets” and click the f(x) button
2. Define the new virtual field, for example:

['otel-demo-traces']
| extend useragent = ['attributes.custom']['User-Agent']

3. You can then query the virtual field like any other field in the UI or APL.

To create a typed virtual field, you can specify the type, e.g.:

| extend deployment_id = tostring(['attributes.custom']['deployment_id'])

Span links

Span links allow you to associate one span with one or more other spans, establishing a relationship between them that indicates the operation of one span depends on the other. Span links can connect spans within the same trace or across different traces.

Span links are useful for representing asynchronous operations or batch-processing scenarios. For example, an initial operation triggers a subsequent operation, but the subsequent operation may start at some unknown later time or even in a different trace. By linking the spans, you can capture and preserve the relationship between these operations, even if they’re not directly connected in the same trace.

How it works

Span links in Axiom are based on the OpenTelemetry specification. When instrumenting your code, you create span links using the OpenTelemetry API by passing the SpanContext (containing trace_id and span_id) of the span to which to link. Links are specified when starting a new span by providing them in the span configuration. The OpenTelemetry SDK includes the link information when exporting spans to Axiom. Links are recorded at span creation time so that sampling decisions can consider them.

View span links

1. Run the following APL query to find traces with span links, for example:

['dataset'] 
| where isnotempty(links)

Run in Playground

2. Click on a trace in the results and select the trace_id.
3. In the trace details view, find the links section. This displays the trace_id and span_id associated with each linked span, as well as other attributes of the link.
4. Click View span to navigate to a linked span, either in the same trace or a different trace.