Learn about MetricSets in APM π
MetricSets are key performance indicators, like request rate, error rate, and request duration, that are calculated from traces and spans in Splunk APM. There are 2 categories of MetricSets: Troubleshooting MetricSets (TMS), used for high-cardinality troubleshooting, and Monitoring MetricSets (MMS), used for real-time monitoring. MetricSets are similar to the metric time series (MTS) used in Splunk Infrastructure Monitoring to populate charts and generate alerts. See Metric time series to learn more. MetricSets are MTS that are specific to Splunk APM.
Monitoring MetricSets π
Monitoring MetricSets (MMS) are metric time series (MTS) that power the monitoring capabilities in Splunk APM, including charts and dashboards. MMS power the APM landing page and the dashboard view. MMS are also the metrics that detectors monitor to generate alerts.
MMS are available for a specific endpoint or for the aggregate of all endpoints in a service.
Endpoint-level MMS reflect the activity of a single endpoint in a service, while service-level MMS aggregate the activity of all of the endpoints in the service. MMS are created for spans where the span.kind
has a value of SERVER
or CONSUMER
.
Spans might lack a kind
value, or have a different kind
value, in the following situations:
The span originates in self-initiating operations or inferred services
An error in instrumentation occurs.
MMS retention period π
Splunk Observability Cloud stores MMS for 13 months by default.
Available default MMS metrics and dimensions π
MMS are available for the APM components listed in the following table. Each MMS also has a set of dimensions you can use to monitor and alert on service performance. In addition to the following default MMS, you can create custom MMS to deep dive on your MMS. See Create a Monitoring MetricSet with a custom dimension.
Metric name |
Dimensions |
Custom dimension available? (Yes/No) |
---|---|---|
|
|
Yes |
|
|
No |
|
|
Yes |
|
|
No |
|
|
No |
Monitoring MetricSets in APM are generated as histogram metrics. Histogram metrics represent a distribution of measurements or metrics, with complete percentile data available. Data is distributed into equally sized intervals, allowing you to compute percentiles across multiple services and aggregate datapoints from multiple metric time series. Histogram metrics provide an advantage over other metric types when calculating percentiles, such as the p90 percentile for a single MTS. See more in Metric types. For histogram MMS, there is a single metric for each component.
Previously, MMS were classified as either a counter or gauge metric type. The previous MMS included 6 metrics for each component.
Description |
Histogram MMS |
MMS (deprecated) |
---|---|---|
Request count |
|
|
Minimum request duration |
|
|
Maximum request duration |
|
|
Median request duration |
|
|
Percentile request duration |
|
|
Percentile request duration |
|
|
Example metrics in APM π
A histogram MTS uses the following syntax using SignalFlow:
histogram(metric=<metric_name>[,filter=<filter_dict>][,resolution=<resolution>)
The following table displays example SignalFlow functions:
Description |
Histogram MMS |
Previous MMS (deprecated) |
---|---|---|
Aggregate count of all MTS |
|
|
P90 percentile for single MTS |
|
|
Combined p90 for multiple services |
|
|
Note
Because an aggregation is applied on histogram(), to display all of the metric sets separately, each dimension needs to be applied as a groupby.
Use MMS within Splunk APM π
Use MMS for alerting and real-time monitoring in Splunk APM. You can create charts, dashboards, and alerts based on Monitoring MetricSets.
Task |
Documentation |
---|---|
Create charts |
|
Create dashboards |
|
Create an alert |
|
Monitor services in APM dashboards |
Troubleshooting MetricSets π
Troubleshooting MetricSets (TMS) are metric time series (MTS) you can use for troubleshooting high-cardinality identities in APM. You can also use TMS to make historical comparisons across spans and workflows.
Splunk APM indexes and creates Troubleshooting MetricSets for several span tags by default. For more details about each of these tags, see Default indexed span tags in APM. You canβt modify or stop APM from indexing these span tags.
You can also create custom TMS by indexing additional span tags and processes. To learn how to index span tags and processes to create new Troubleshooting MetricSets, see Index span tags to create Troubleshooting MetricSets.
Available TMS metrics π
Every TMS creates the following metrics, known as request, error, and duration (RED) metrics. RED metrics appear when you select a service in the service map. See Scenario: Kai investigates the root cause of an error with the Splunk APM service map to learn more about using RED metrics in the service map.
Request rate
Error rate
Root cause error rate
p50, p90, and p99 latency
The measurement precision of Troubleshooting MetricSets is 10 seconds. Splunk APM reports quantiles from a distribution of metrics for each 10-second reporting window.
Use TMS within Splunk APM π
TMS appear on the service map and in Tag Spotlight. Use TMS to filter the service map and create breakdowns across the values of a given indexed span tag or process.
See View dependencies among your services in the service map and Analyze service performance with Tag Spotlight.
TMS retention period π
Splunk Observability Cloud retains TMS for the same amount of time as raw traces. By default, the retention period is 8 days.
For more details about Troubleshooting MetricSets, see Reference and best practices for span tags and Troubleshooting MetricSets.
Comparing Monitoring MetricSets and Troubleshooting MetricSets π
Because endpoint-level and service-level MMS include a subset of the TMS metrics, you might notice that metric values for a service are different depending on the context in Splunk APM. This is because MMS are the basis of the dashboard view and MMS can only have a kind
of SERVER
or CONSUMER
. In contrast, TMS are the basis of the troubleshooting and Tag Spotlight views and TMS arenβt restricted to specific metrics.
For example, values for checkout
service metrics displayed in the host dashboard might be different from the metrics displayed in the service map because there are multiple span kind
values associated with this service that the MMS that power the dashboard donβt monitor.
To compare MMS and TMS directly, restrict your TMS to endpoint-only data by filtering to a specific endpoint. You can also break down the service map by endpoint.