Neo4j-graphql running out of connections since update to v4.4.4

Drivers & Stacks GraphQL & GRANDstack
, ,
1

Hi All,

I'm running a Node.js Apollo GraphQL API using neo4j-graphql. This is connected to an AuraDB instance of Neo4j (v4).

I've recently updated @neo4j/graphql from 3.5.1 to 4.4.4 (along with updating neo4j-driver from 4.4.1 to 5.8.0 as a peer dependency), but fairly quickly ran into an issue where the connection pool was maxed out at 100 connections and new connection acquisition was timing out with the following error:

error in apollo [GraphQLError: Connection acquisition timed out in 60000 ms. Pool status: Active conn count = 100, Idle conn count = 0.]

I have not changed the schema in any way apart from to accommodate the breaking changes between v3 and v4, so I wasn't expecting to see an issue of this nature after updating.

Does anyone have similar experience after updating neo4j/graphql (or neo4j-driver)? I'm wondering if the latest version isn't closing connections properly and they're piling up.

Any help would be appreciated!

Thanks,

Simon.

2

Hi Simon!

That's not something we've seen reported by anyone else for now, so it might be something particular about your setup.

Could you possibly provide a sample of how you're using the ApolloServer and Neo4jGraphQL? It could be a quick fix.

Best regards,

Michael

3

Hi Michael,

Thanks for the quick response.

I don't think we're doing anything especially unusual, we create a neo4j-driver and schema in the normal way, then start the Apollo Server as follows:

const neoSchema = new Neo4jGraphQL({ typeDefs, driver }); // typeDefs is our raw graphQL schema, driver is the neo4j driver, initialised with a bolt URI and basic auth.

neoSchema
  .getSchema()
  .then((schema) => {
    const server = new ApolloServer({
      schema,
      introspection: true,
      plugins: [ApolloServerPluginLandingPageGraphQLPlayground()],
      context: createContext,
      formatError: (err) => {
        console.error("error in apollo", err);
        return err;
      }
    });

    server.listen(process.env.PORT || 3001).then(({ url }) => {
      console.log(`Server ready at ${url}`);
    });
  })
  .catch((e) => {
    console.error(e);
  });

The createContext function is just taking data from the HTTP request and sticking it into a jwt object for use in authentication and authorization directives

We're running the API on an AWS Elastic Beanstalk instance.

Please let me know if there's any other info I can supply.

Thanks,

Simon.

4

As you say, I don't see anything particularly unusual. What's the version of ApolloServer you're using?

5

apollo-server is at 3.6.1

6

Could you possibly try the latest version of ApolloServer (@apollo/server) to see if it makes any difference?

7

yep, will give that a go

8

Hi Michael,

Apologies for the delay; Christmas got in the way.

I've updated to @apollo/server @ v4.9.5 but I'm still getting the same issue.

Testing on my local, where I'm the only client connecting, I can fire off queries that get quick responses (~30-50ms) again and again until I hit 100 queries. Any further queries above 100 timeout after 60s. So it seems that the 100 connections in the driver's connection pool are not being released/re-used. In this test scenario I wouldn't expect the number of connections in the pool to go above 1 as each query is completed before I make the next request.

After updating Apollo, my startup code looks like this, which is pretty much the same as the "Getting Started" documentation:

const startServer = async () => {
  try {
    const server = new ApolloServer({
      schema: await neoSchema.getSchema()
    });

    const { url } = await startStandaloneServer(server, {
      context: createContext,
      listen: { port: parseInt(process.env.PORT) || 3001 }
    });

    console.log(`Server ready at ${url}`);
  } catch (error) {
    console.error(error);
  }
};

startServer();

Any other suggestions would be welcome; I'm stumped with this one.

Thanks,

Simon.

9

Hey Simon, I hope you had a nice time off!

Is there any chance we could see your driver configuration (without the credentials)?

Best regards,

Michael

10

Sure,

I'm initialising the driver object as follows:

import { driver as neo4JDriver, auth } from "neo4j-driver";
import { neo4jConnectionUri, neo4jUsername, neo4jPassword } from "./env";

export const driver = neo4JDriver(neo4jConnectionUri, auth.basic(neo4jUsername, neo4jPassword));
11

Hi Michael,

I think I've resolved the issue.

I started up a new project following the "Getting Started" instructions verbatim, and there was no issue (as we would expect). I started looking for differences between the fresh project and my existing project and the first thing I spotted was that the fresh project had slightly newer dependency versions, so I update the dependencies on the existing project to match and that seems to have done the trick.

So to summarize, for anyone else who might encounter this issue, I updated the following dependencies:

  • @apollo/server: 4.9.5 => 4.10.0
  • graphql: 16.6.0 => 16.8.1
  • neo4j-driver: 5.8.0 => 5.16.0

I'll review the change history on those packages to see if my issue was related to a known bug that's been fixed recently.

1 Like
12

Hi Simon,

So glad to hear you've found the solution!

Best regards,

Michael