Can't migrate Apollo Gateway to Server v4

[tiagodj]

Hi,

I am having some difficulty migrating my Apollo Gateway to the new Apollo Sever v4.

The issue is that in our code we do a filtering of the Schema before creating the Server, but after creating the Gateway.

The code in v3 is:

  const gateway = new ApolloGateway(config);

  const { schema, executor } = await gateway.load();

  const publicSchema = graphqlSFilter(schema, graphqlFilterMap); <---- this filters out APIs from the public schema

  const server = new ApolloServer({
    schema: publicSchema,
    executor,
    ...
});

but in the Apollo v4 I was no able to find a configuration where this works.

How do I achieve this in v4?

I am open to a completely different approach.

The main goal is that we have 2 gateways running in parallel: one used by internal apps with all APIs, and one exposed to external public API with limited operations.

[tiagodj]

I was able to find a solution! (of course after I asked the question here)

I am converting the SDL to GraphQLSchema and back again before creating the gateway:

import { buildSchema } from "graphql";
import { printSchemaWithDirectives } from "@graphql-tools/utils";

  // ...
  const fullSchema = buildSchema(supergraphSdl);
  const filteredSchema = graphqlSFilter(fullSchema);
  const filteredSdl = printSchemaWithDirectives(filteredSchema, {
    assumeValid: true,
    commentDescriptions: false,
  });

  const gateway = new ApolloGateway({
    supergraphSdl: filteredSdl,
    ...
  });

[trevor-scheer]

Thanks for reporting back with your solution @tiagodj. You're using what we'd consider a custom executor. We do offer a migration path for your use case in the migration guide.

[nikhiltevaera]

Hey David,
I have sent you a request on Linkedin ,please let me know if you receive and then on we can continue our conversation.

[glasser]

Hi, I don't think private LinkedIn messages are the best way to provide open source support — it's best if any help I give open source users is available to help everyone. Perhaps you can minimize your issue to a small example. If you're an Apollo customer looking for private support, we have a process for that of course.

[nikhiltevaera]

Alright then let me provide some example snippets,
`
const schema = createSchema();

const executor: GatewayExecutor = async ({ operationName, queryHash, document, context, request }) => {
......
// the logic here contains caching compiled Query....
return await compiledQuery.query(schema, context, request.variables || {});
};

const server = new ApolloServer({
includeStacktraceInErrorResponses: !isProduction(),
introspection: true,
validationRules: [depthLimit(10)], // Restrict queries to a depth of 10.
plugins: [
process.env.NODE_ENV === 'production'
? ApolloServerPluginLandingPageDisabled()
: ApolloServerPluginLandingPageLocalDefault({ footer: false }),
],
gateway: {
async load() {
return { executor };
},
onSchemaLoadOrUpdate() {
return () => {};
},
async stop() {},
},
});

await server.start();
const apolloMiddleware = expressMiddleware(server, {context : ...})

//app is from express
app.use("/",apolloMiddleware);

`

So during gateway initialization , it throws an error as "Error: You must call start() before getSchemaDerivedData()",

I am building schema and schema using my custom function which returns schema of type GraphQlSchema.

I want to load schema through gateway...,Need your help on this

[glasser]

This is a little hard to follow, but you need actually implement onSchemaLoadOrUpdate — it takes a callback that you pass the schema to when you have it.

[nikhiltevaera]

let me make this simpler for you,
Earlier in apollo server 3 - Migration

i have my custom executor that i pass similar to the guide shown in here.

, i want also to load schema before the executor initializes itself...

Since i am using schema in my custom_executor , for some logic, so the server throws the issue as -" You must call start() before getSchemaDerivedData()"