/
Launch Apollo Studio

Migrating from the "engine" option


Apollo Server v2.18 deprecates the engine option to the ApolloServer constructor and provides a new way of configuring its communication with Apollo Studio. The engine option continues to work for existing functionality, but you will eventually want to update to the new API. If you don't explicitly pass engine to the ApolloServer constructor, you don't have to do anything.

Apollo Server ships with several plugins that help it integrate with Apollo Studio: the usage reporting plugin plugin, the schema reporting plugin, and the inline trace plugin. Apollo Server has some heuristics to install these plugins by default in certain circumstances (documented in the individual plugin reference pages), but otherwise they are standard Apollo Server plugins. They are configured by passing arguments to the constructor functions. Some overall graph configuration (such as your graph API key and graph variant name) is set via the apollo option to the ApolloServer constructor (or via environment variables).

Before Apollo Server v2.18, all this functionality built directly inside Apollo Server, and all of it was configured via the engine option to the ApolloServer constructor. While the engine option does continue to work for backwards compatibility, new configuration options will only be added to the plugins directly. To make it easier to undersatnd where configuration is coming from, you cannot mix and match the engine option with the three new plugins or the apollo option; if you want to start configuring the plugins directly you need to migrate all of your usage of the engine option to the plugin functions. Fortunately, this is relatively straightforward; most engine options correspond directly to an option passed to one of the plugins or on the apollo option to ApolloServer. This page lists the options on engine and explains how to migrate them to the Studio integration plugins.

If you don't explicitly pass engine to your ApolloServer constructor, you don't need to make any changes! Specifically, configuration via the environment variables APOLLO_KEY (and its legacy equivalent ENGINE_API_KEY), APOLLO_GRAPH_VARIANT (and its legacy equivalent ENGINE_SCHEMA_TAG), and APOLLO_SCHEMA_REPORTING has not changed at all.

(There is one minor change in v2.18 that happens if you don't use any of this Studio integration: upgrading to Apollo Server v2.18 will make ApolloServer start registering SIGINT and SIGTERM signal handlers which invoke ApolloServer.stop() when those signals are received. Previously these handlers were only registered when you use Studio usage or schema reporting. If this is a problem for you, disable the handlers by passing stopOnTerminationSignals: false to the ApolloServer constructor.)

Migration example

Here's a high level example showing how to migrate off of the engine option. Let's say your ApolloServer constructor looked like this, and you already set your API key using $APOLLO_KEY:

import { ApolloServer } from 'apollo-server-express';

function rewriteError(err) {
  if (err.message.matches(/hide-me/)) {
    return null;
  }
  return err;
}

function reportTiming(requestContext) {
  return requestContext.request.http?.headers?.get('no-report') !== 'true';
}

const server = new ApolloServer({
  typeDefs,
  resolvers,
  engine: {
    graphVariant: 'prod',
    handleSignals: false,
    rewriteError,
    reportTiming,
    privateVariables: ['foo', 'bar'],
    reportSchema: true,
    schemaReportingInitialDelayMaxMs: 30 * 1000,
  },
});

You can rewrite this as:

import { ApolloServer } from 'apollo-server-express';
// The plugins are always imported from apollo-server-core, no matter which
// framework you use.
import {
  ApolloServerPluginUsageReporting,
  ApolloServerPluginSchemaReporting,
} from 'apollo-server-core';

function rewriteError(err) {
  if (err.message.matches(/hide-me/)) {
    return null;
  }
  return err;
}

function includeRequest(requestContext) {
  return requestContext.request.http?.headers?.get('no-report') !== 'true';
}

const server = new ApolloServer({
  typeDefs,
  resolvers,
  apollo: {
    // You can also just set $APOLLO_GRAPH_VARIANT.
    graphVariant: 'prod',
  },
  stopOnTerminationSignals: false,
  plugins: [
    ApolloServerPluginUsageReporting({
      rewriteError,
      includeRequest,
      sendVariableValues: { exceptNames: ['foo', 'bar'] },
    }),
    ApolloServerPluginSchemaReporting({
      initialDelayMaxMs: 30 * 1000,
    }),
  ],
});

Options that move to apollo

Two engine options move to the new apollo option to the ApolloServer constructor. engine.apiKey becomes apollo.key, and engine.graphVariant becomes apollo.graphVariant. engine.schemaTag was a previous name for engine.graphVariant, so uses of that field should also change to apollo.graphVariant (apollo.schemaTag is not supported).

// This code...
const server = new ApolloServer({
  engine: {
    apiKey: 'service:xxx:yyy',
    graphVariant: 'production',
  },
});

// ... is equivalent to this code.
const server = new ApolloServer({
  apollo: {
    key: 'service:xxx:yyy',
    graphVariant: 'production',
  },
});

Note that these two options are often specified via environment variable: the API key can be specified as APOLLO_KEY (or its legacy equivalent ENGINE_API_KEY), and the graph variant can be specified as APOLLO_GRAPH_VARIANT (or its legacy equivalent ENGINE_SCHEMA_TAG). Apollo Server v2.18 does not change this behavior.

Options that move to the ApolloServer constructor

Prior to Apollo Server v2.18, the usage reporting functionality registered one-shot handlers for the SIGINT and SIGTERM signals, which it used to send one final usage report before re-sending the signal to itself to continue shutdown. These signals handlers were installed by default if you enabled usage or schema reporting, and could be disabled by passing engine.handleSignals: false.

In Apollo Server v2.18, termination signal handling is the responsibility of Apollo Server as a whole rather than something specific to usage reporting. Apollo Server itself now registers these one-shot signal handlers, which trigger ApolloServer.stop(). This allows any plugin that implements the new serverWillStop callback to hook into shutdown logic, not just the usage reporting code.

Similarly to before, these signal handlers are registered by default but can be disabled by via an option. We've changed the option name to stopOnTerminationSignals: false as it is more explicit about the behavior.

// This code...
const server = new ApolloServer({
  engine: {
    handleSignals: false,
  },
});

// ... is equivalent to this code.
const server = new ApolloServer({
  stopOnTerminationSignals: false,
});

Options for ApolloServerPluginUsageReporting

The majority of engine options configure how usage reporting works. Many of them can be moved directly to a call to ApolloServerPluginUsageReporting in the plugins array.

These engine options work exactly like the ApolloServerPluginUsageReporting options of the same name:

  • calculateSignature
  • debugPrintReports
  • endpointUrl
  • generateClientInfo
  • maxAttempts
  • maxUncompressedReportSize
  • minimumRetryDelayMs
  • reportErrorFunction
  • reportIntervalMs
  • requestAgent
  • rewriteError
  • sendHeaders
  • sendReportsImmediately
  • sendVariableValues

The engine option tracesEndpointUrl was another name for endpointUrl and also becomes the ApolloServerPluginUsageReporting option endpointUrl.

The engine option reportTiming can be either a function or a boolean. If you passed a function here, you can pass the same function as the includeRequest option to ApolloServerPluginUsageReporting. If you passed true here, you don't need to do anything special: this just means to enable usage reporting, which is the default if an API key is provided. If you passed false here, that means you didn't want usage reporting even though you've configured ApolloServer with an API key (perhaps you only want schema reporting); in that case, you should use the ApolloServerPluginUsageReportingDisabled plugin (see example below).

The engine option maskErrorDetails was deprecated in Apollo Server 2.5 and replaced by rewriteError. While engine.maskErrorDetails still works for backwards compatibility, there is no maskErrorDetails option to ApolloServerPluginUsageReporting. If you previously passed engine.maskErrorDetails, you can instead pass rewriteError: () => new GraphQLError('<masked>') to ApolloServerPluginUsageReporting.

The privateVariables and privateHeaders engine options were deprecated in Apollo Server 2.7 and replaced by sendVariableValues and sendHeaders. While they still work on engine for backwards compatibility, you need to use the newer version if calling ApolloServerPluginUsageReporting directly.

  • engine.privateVariables: true is equivalent to sendVariableValues: { none: true }
  • engine.privateVariables: false is equivalent to sendVariableValues: { all: true }
  • engine.privateVariables: ['x', 'y'] is equivalent to sendVariableValues: { exceptNames: ['x', 'y'] }

engine.privateHeaders values can be translated to sendHeaders values in the same way.

import { ApolloServer } from 'apollo-server-express';
import {
  ApolloServerPluginUsageReporting,
  ApolloServerPluginUsageReportingDisabled,
} from 'apollo-server-core';

// This code...
const server = new ApolloServer({
  engine: {
    reportTiming: (requestContext) => ...,
    sendReportsImmediately: true,
    maskErrorDetails: true,
    privateHeaders: ['my-api-key'],
  },
});

// ... is equivalent to this code.
const server = new ApolloServer({
  plugins: [ApolloServerPluginUsageReporting({
    includeRequest: (requestContext) => ...,
    sendReportsImmediately: true,
    rewriteError: () => new GraphQLError('<masked>'),
    sendHeaders: { exceptNames: ['my-api-key'] },
  })],
});


// This code...
const server = new ApolloServer({
  engine: {
    reportTiming: false,
  },
});

// ... is equivalent to this code.
const server = new ApolloServer({
  plugins: [ApolloServerPluginUsageReportingDisabled()],
});

Options for ApolloServerPluginSchemaReporting

A few engine options configure how schema reporting works.

The engine.reportSchema option (or its legacy equivalent engine.experimental_schemaReporting) is replaced by the choice of whether or not to include the schema reporting plugin at all. If you passed true for either of these options, you'll want to add a call to ApolloServerPluginSchemaReporting() to your plugins array. Alternatively, you can set the APOLLO_SCHEMA_REPORTING environment variable to true, which will have the saem effect. If you passed false for either of these options, just don't call ApolloServerPluginSchemaReporting at all (and don't set the environment variable.

The engine.overrideReportedSchema option (and its legacy equivalent engine.experimental_overrideReportedSchema) works exactly like the overrideReportedSchema option to ApolloServerPluginSchemaReporting. However, if you want to configure this option, you should also pass the same value to ApolloServerPluginUsageReporting's option of the same name, so that schema IDs match up between your schema and usage reporting. This may require adding a call to ApolloServerPluginUsageReporting if you were otherwise depending on its default behavior.

The engine.schemaReportingInitialDelayMaxMs option (and its legacy equivalent engine.experimental_schemaReportingInitialDelayMaxMs) works exactly like the initialDelayMaxMs option to ApolloServerPluginSchemaReporting.

The engine.schemaReportingUrl option works exactly like the endpointUrl option to ApolloServerPluginSchemaReporting.

import { ApolloServer } from 'apollo-server-express';
import {
  ApolloServerPluginSchemaReporting,
  ApolloServerPluginUsageReporting,
} from 'apollo-server-core';

// This code...
const server = new ApolloServer({
  engine: {
    reportSchema: true,
  },
});

// ... is equivalent to this code. (Or you can just set
// $APOLLO_SCHEMA_REPORTING to 'true'.)
const server = new ApolloServer({
  plugins: [ApolloServerPluginSchemaReporting()],
});


// This code...
const server = new ApolloServer({
  engine: {
    reportSchema: true,
    overrideReportedSchema: SCHEMA_TEXT,
  },
});

// ... is equivalent to this code.
const server = new ApolloServer({
  plugins: [
    ApolloServerPluginUsageReporting({
      overrideReportedSchema: SCHEMA_TEXT,
    }),
    ApolloServerPluginSchemaReporting({
      overrideReportedSchema: SCHEMA_TEXT,
    }),
  ],
});

Options for ApolloServerPluginInlineTrace

By default, Apollo Server enables inline tracing on federated implementing services. One engine option can be used to configure how inline tracing works. If you specified the engine.rewriteError function and your service is an implementing service that is combined with others via federation (but not if this server is the Apollo gateway which combines multiple implementing services), then you should pass that rewriteError function to the ApolloServerPluginInlineTrace plugin function. Note that engine.rewriteError is also (more commonly) used with graphs that report directly to Apollo's servers with via the usage reporting plugin; you shouldn't add the inline trace plugin unless you're sure you're in a federated implementing service!

import { ApolloServer } from 'apollo-server-express';
import { ApolloServerPluginInlineTrace, } from 'apollo-server-core';

// This code...
const server = new ApolloServer({
  engine: {
    rewriteError: (err) => ...,
  },
});

// ... is equivalent to this code, IF THIS IS A
// FEDERATED IMPLEMENTING SERVICE.
const server = new ApolloServer({
  plugins: [ApolloServerPluginInlineTrace({
    rewriteError: (err) => ...
  })],
});

Note that the default behavior of whether inline tracing is enabled changed in v2.18. In v2.18, inline tracing is enabled for any federated implementing service, unless explicitly disabled with ApolloServerPluginInlineTraceDisabled. In previous versions, inline tracing was enabled for any federated implementing service unless an Apollo API key was provided; this special case is removed in v2.18, which means that the same graph can both report usage to Apollo's servers and include inline traces in responses. These functionalities also both now log at startup, so it should be easy to see if they are unintentionally simultaneously enabled.

Migrating from engine: false

If you passed engine: false to the ApolloServer constructor, the main effect was to disable inline tracing even if the current graph is a federated implementing service.

Prior to Apollo Server 2.13, this also had the effect of disabling usage reporting even if you configured an API key. The change in Apollo Server 2.13 appears to have been unintentional and the backwards-compatibility handling of engine: false in Apollo Server TODO(no-engine) restores the old behavior of disabling usage reporting.

If your code had engine: false because it is a federated implementing service but you do not want to enable inline tracing, you should use the new ApolloServerPluginInlineTraceDisabled plugin instead:

import { ApolloServer } from 'apollo-server-express';
import { ApolloServerPluginInlineTraceDisabled } from 'apollo-server-core';

// In a federated implementing service, this code...
const server = new ApolloServer({
  engine: false,
});

// ... is equivalent to this code.
const server = new ApolloServer({
  plugins: [ApolloServerPluginInlineTraceDisabled()],
});

If your code had engine: false because you wanted to disable usage reporting even though you are passing in an Apollo graph API key (via $APOLLO_KEY or $ENGINE_API_KEY), you should use the new ApolloServerPluginUsageReportingDisabled plugin instead. (However, be aware that usage reporting was not actually disabled by engine: false starting with Apollo Server 2.13.)

import { ApolloServer } from 'apollo-server-express';
import { ApolloServerPluginUsageReportingDisabled } from 'apollo-server-core';

// When $APOLLO_KEY or $ENGINE_API_KEY is set, this code...
const server = new ApolloServer({
  engine: false,
});

// ... is equivalent to this code.
const server = new ApolloServer({
  plugins: [ApolloServerPluginUsageReportingDisabled()],
});

Passing engine: true to the ApolloServer constructor was allowed but did not have a different effect from not passing engine at all.

Edit on GitHub