What does request rate measure?

Request rate is the average number of requests per minute for the selected time range. It reflects throughput to your federated graph router.

What error types are included in error rate?

Error rate includes both 4xx and 5xx errors. It is expressed as an average error percentage for the selected time range.

Can I filter metrics by client?

Yes. Filter by client name and client version. Client filtering requires that your client applications send GraphQL-Client-Name and GraphQL-Client-Version headers (or the Apollo-compatible equivalents) with each request.

What does the error rate over time chart show?

The chart shows the number of GraphQL requests and errors over the selected time period. It helps identify when an error spike started and how long it lasted.

Does the metrics view require additional configuration?

No additional configuration beyond the Cosmo Router with OTEL instrumentation enabled. For client-level filtering, configure client identification headers in your client applications.

Metrics Analytics for Federated GraphQL | Cosmo by WunderGraph

Q: How is P95 latency calculated?

P95 latency is the average p95 response time for the selected time range. The upper bucket boundary is 10 seconds — requests taking longer than 10 seconds are counted as 10 seconds in the calculation.

The problem

API health is hard to assess without the right numbers

Teams need a fast, accurate read on system status. Without aggregated metrics, that read requires manual work — and usually comes too late.

Performance assessment takes too long

Without aggregated metrics, understanding system health means manually pulling data from multiple places. By the time you have an answer, a slowdown may have already affected users.

No single number for error rate

Error data sits across different services and log sinks. Calculating total error percentage across a federated graph is a manual exercise that teams rarely do until an incident forces it.

Latency issues are hard to attribute

Without P95 latency filtered by operation or client, you cannot tell whether a slowdown is systemic or confined to a specific query or application.

Our solution

Three metrics. One clear view.

Metrics Analytics presents the three most important performance indicators for your federated graph in a single visual interface. Time-series charts show how each metric evolves. Filters let you zoom in on specific operations or clients without switching tools.

What you see

Metrics Analytics aggregates telemetry from the Cosmo Router collected via OpenTelemetry.
Request rate shows average requests per minute for the selected time range.
P95 latency shows the average p95 response time for the selected range. The upper bucket boundary is 10 seconds.
Error rate shows the average error percentage including both 4xx and 5xx errors.
Time-series charts show how requests and errors change over the selected period.
Filters narrow metrics by operation name, client name, and client version.

System health assessed in seconds, not minutes.

Metrics Analytics

Before & After

Before Cosmo	With Cosmo
Multiple dashboards for different metrics	Single unified metrics view
Delayed awareness of performance issues	At-a-glance visibility into request rate and latency
Aggregating error data from multiple services	Consolidated error rate across the entire federated graph
Manual calculation of percentiles	Automatic P95 latency calculation

The three metrics

Request rate, P95, error rate

Request rate: Average requests per minute. Shows throughput and reveals traffic spikes or drops.
P95 latency: The average p95 response time for the selected range. Upper bucket boundary is 10 seconds.
Error rate: Average error percentage for the selected range. Includes 4xx and 5xx errors.
Error rate over time: Time-series chart showing GraphQL requests and errors together.

How Metrics Analytics works

01

Avg requests per minute.

Request rate

Average requests per minute for the selected time range. Spikes and drops in request rate are visible at a glance, making traffic anomalies easy to spot.

02

P95. Upper bound: 10s.

P95 latency

The average p95 response time for the selected period. The upper bucket boundary is 10 seconds — requests taking longer than 10 seconds are counted as 10 seconds.

03

4xx and 5xx combined.

Error rate

Average error percentage for the selected time range. Includes both 4xx and 5xx errors. The error rate over time chart shows GraphQL requests and errors together.

04

Operation, client name, version.

Filtering

Narrow any metric by operation name, client name, or client version. Client filtering requires client identification headers on requests.

What is included

Built into every Cosmo plan

Available on Free, Pro, and Enterprise.

Instant health assessment

Request rate, P95 latency, and error rate visible the moment you open the view. No configuration needed beyond the router.

Trend visualization

Time-series charts show how request volume and error count change over the selected period. Spot when a problem started and whether it is getting better or worse.

Granular filtering

Filter metrics by operation name, client name, and client version. Isolate whether a performance issue is systemic or confined to a specific query or application.

Time range flexibility

Analyze metrics over custom date ranges or predefined periods. Available ranges respect your plan's data retention limits.

Monitor your federated graph performance

Deploy the Cosmo Router and metrics start flowing into Cosmo Studio automatically.

Start Free Read the Docs

Request rate, latency, and error rate at a glance