We've introduced a new profiling feature to help you understand the performance of your GraphQL executions. It provides detailed insights into DataFetcher calls, Dataloader usage, and execution timing. This guide will show you how to use it and interpret the results.

The Profiler is available in all versions after v25.0.beta-5. It will also be included in the forthcoming official v25.0 release.

To enable profiling for a GraphQL execution, you need to set a flag on your ExecutionInput. It's as simple as calling .profileExecution(true) when building it.

import graphql.ExecutionInput;
import graphql.GraphQL;

// ...
GraphQL graphql = GraphQL.newGraphQL(schema).build();

ExecutionInput executionInput = ExecutionInput.newExecutionInput()
        .query("query Hello { world }")
        .profileExecution(true) // Enable profiling
        .build();

graphql.execute(executionInput);

The profiling results are stored in the GraphQLContext associated with your ExecutionInput. After the execution is complete, you can retrieve the ProfilerResult object from the context.

The result object is stored under the key ProfilerResult.PROFILER_CONTEXT_KEY.

import graphql.ExecutionInput;
import graphql.ExecutionResult;
import graphql.GraphQL;
import graphql.ProfilerResult;

// ...
ExecutionInput executionInput = /* ... see above ... */
ExecutionResult executionResult = graphql.execute(executionInput);

ProfilerResult profilerResult = executionInput.getGraphQLContext().get(ProfilerResult.PROFILER_CONTEXT_KEY);

if (profilerResult != null) {
    Map<String, Object> summary = profilerResult.shortSummaryMap();
    System.out.println(summary); // or log as you prefer
}

A great way to get a quick overview about ProfilerResult is by using the shortSummaryMap() method. It returns a map with key metrics about the execution, which you can use for logging. Let's break down what each key means.

Here's a detailed explanation of the fields you'll find in the map returned by shortSummaryMap():

This is a list of maps, each detailing a DataLoader dispatch event.

This map gives you more information into the type of your DataFetchers' return values.

The keys are COMPLETABLE_FUTURE_COMPLETED, COMPLETABLE_FUTURE_NOT_COMPLETED, and MATERIALIZED. Each key maps to another map with two keys:

• count: The number of unique fields with DataFetchers that returned this result type.
• invocations: The total number of invocations across all fields that returned this result type.

Here's what the result types mean:

If you're logging the ProfilerResult from an Instrumentation, note that engine timing statistics such as startTimeNs, endTimeNs, totalRunTimeNs, engineTotalRunningTimeNs will be set to 0. This is because the timing is set after all Instrumentations have run, so it is not available from within an Instrumentation.

Apart from engine timing information, all other ProfilerResult information is still valid if accessed from within an Instrumentation.