Docs » Drill down into service performance with indexed tags and MetricSets in SignalFx µAPM » Which span tags to index in SignalFx µAPM

Which span tags to index in SignalFx µAPM 🔗

Important

The original µAPM product, released in 2019, is now called µAPM Previous Generation (µAPM PG). In the documentation, µAPM now refers to the product released on March 31, 2020.

If you are using µAPM Previous Generation (µAPM PG), see µAPM PG Traces, Spans, Metrics, and Metadata.

Index only span tags you want drill down into to gain insights about the performance of your infrastructure, or to address a specific incident. Some span tags provide a level of cardinality that just isn’t useful. For example, indexing query_id would generate MetricSets for every unique query, and in most cases there’s no reason for this level of cardinality.

Also avoid indexing span tags that represent ephemeral resources like container_id.

Consider which span tags are worth creating MetricSets for. Here are some questions you can ask about your environment:

  • Are there any attributes I look at when an incident occurs?

    If you’re running Kubernetes, you could index k8s.pod.name to view the performance of services by specific Kubernetes pods.

  • Do I run multiple versions or builds of code at the same time?

    You could index tags for version or build_id to break down your infrastructure according to specific versions or builds of your applications.

  • Do I deploy services in multiple regions or fault domains?

    It could be useful to view metrics for services by specific region span tags to identify issues with resources in specific regions or zones.

  • Do I monitor multiple products?

    To get a better idea about how services for a particular product are running, you could use a span tag like product_category to view metrics for traces for a single product type rather than viewing traces for multiple product types at the same time.

How indexing span tags provides insights into service performance 🔗

Indexing span tags generates Troubleshooting MetricSets for services in your environment. A Troubleshooting MetricSet is a set of unique span tag key-value pairs. Each Troubleshooting MetricSet generates metrics that you can use to drill down into service performance.

The measurement precision of Troubleshooting MetricSets is 10 seconds. SignalFx reports quantiles from a distribution of metrics for each 10-second reporting window.

These metrics are what Troubleshooting MetricSets generate:

  • Request rate
  • Error rate
  • Root-cause error rate
  • Latency

Because endpoint is an indexed tag, you can break down the performance of each service by each endpoint you collect. You can even break down service-to-service calls, known as edges. To do so, mouse over or click edges from the Troubleshooting tab on the µAPM page like this:

../../_images/apm-metricets-troubleshooting-mouseover.png

You can also see these metrics in service panels when you select a service from the Troubleshooting tab on the µAPM page:

../../_images/apm-metricsets-metrics.png

Types of span tags you can index 🔗

There are three types of span tags you can index:

Type Example Description
Global Tenant class, application ID A tag with only a single value across a trace. This associates the span tag value with an entire trace.
All services Version, host, status code A tag that exists for most or all services. This associates the span tag value with services in a trace. The value could change across services in a trace.
Specific service(s) Job ID, database name A tag that exists for only a single service in a trace. This associates the span tag value with a single service in a trace.

Span tags µAPM automatically indexes 🔗

µAPM automatically indexes and creates Troubleshooting MetricSets for these span tags:

Tag Example Type Description
environment production, lab Global The environment where the service associated with the span runs.
Endpoint /auth/valid, /checkout/{cartId} All services If the root span contains the span.kind = server tag, the operation name for the root span of a service.
Operation /auth/valid, /checkout/{cartId} All services The operation name for the root span of a service.
HTTP Method GET, POST, PUT All services The HTTP method of the span, if any. The actual span tag field name can be http.method or requestMethod, but HTTP Method is the tag you use to filter and break down services.
Kind server, client, producer All services The type of service associated with the span. The actual span tag field name is span.kind, but Kind is the tag you use to filter and break down services.
Service The name of a service you specify when configuring instrumentation. All services The service you instrumented and are exporting trace data from.

Limits on indexing span tags 🔗

The number of the span tags you can index depends on the number of Troubleshooting MetricSets µAPM generates from indexed span tags. µAPM generates a Troubleshooting MetricSet for every unique set of indexed span tag values. This is also known as the cardinality contribution of indexed tags. The number of Troubleshooting MetricSets you can generate is limited. Keep this limit in mind when indexing span tags. For more information, see SignalFx Microservices APM Pricing.

To determine the total number of possible Troubleshooting MetricSets, multiply the number of indexed tag values for each indexed tag associated with each unique set of service, endpoint, operation, and environment values.

For example, there are two services, frontend and checkoutservice. Each service exists in two regions. The frontend service has five endpoints, and the checkoutservice has two endpoints. region and endpoint are indexed tags for each service. There are no other indexed tags in this scenario.

The maximum number of Troubleshooting MetricsSets that can exist for these two services is 14 ((2 * 5) + (2 * 2)). This doesn’t always mean there will be 14 Troubleshooting MetricSets. If you collect traces in a certain region or with a certain endpoint, the Troubleshooting MetricSet exists only for that region or endpoint.

Here is a table that illustrates the number of possible Troubleshooting MetricSets given two services and two indexed tags with multiple values:

# Service Region Endpoint
1 frontend west /currency
2 frontend west /cart
3 frontend west /checkout
4 frontend west /shipping
5 frontend west /product
6 frontend east /currency
7 frontend east /cart
8 frontend east /checkout
9 frontend east /shipping
10 frontend east /product
11 checkoutservice west /placeholder
12 checkoutservice west /queueplaceholder
13 checkoutservice east /placeholder
14 checkoutservice east /queueplaceholder