Prisma2 Application Potentially Creating DB Connections per API Request

I have had a prisma app running in GCP that connects to a postgres instance also running in GCP for about a day.

All API calls started failing to the app with:

2020-01-24 21:15:52.056 CSTError querying the database: db error: FATAL: remaining connection slots are reserved for non-replication superuser connections
2020-01-24 21:15:52.056 CST at PhotonFetcher.request (/app/node_modules/@prisma/photon/index.js:62:23)
2020-01-24 21:15:52.056 CST at runMicrotasks (<anonymous>)
2020-01-24 21:15:52.056 CST at processTicksAndRejections (internal/process/task_queues.js:97:5)
2020-01-24 21:15:52.058 CSTGET500395 B44 msChrome 79 https://flag-api-vykaoik56q-uc.a.run.app/switches

Looks like the app is holding open connections. I have basically the boiler plate from the prisma init:

import { GraphQLServer } from 'graphql-yoga';
import { schema } from './schema';
import { createContext } from './context';
import getSwitchesWithAuthHeader from './controllers/switches';
import * as cors from 'cors';

const server = new GraphQLServer({ schema, context: createContext });
server.express.use(cors());
server.express.get('/switches', getSwitchesWithAuthHeader);

server.start(
  { port: 8080 },

  () =>
    console.log(
      `🚀 Server ready at: http://localhost:8080\n⭐️ See sample queries: http://pris.ly/e/ts/graphql#5-using-the-graphql-api`,
    ),
);

Versions:

    "@prisma/photon": "2.0.0-alpha.538",
    "basic-auth": "^2.0.1",
    "cors": "^2.8.5",
    "dotenv": "^8.2.0",
    "graphql": "14.5.8",
    "graphql-yoga": "1.18.3",
    "nexus": "0.12.0-rc.5",

Restarting my database or the application fixes the issue temporarily.

[timsuchanek]

Thanks for reporting everyone! Can you please show me, how you instantiate the Prisma Client? You should normally just instantiate one client and reuse it for all requests.

[pimeys]

I’m trying to replicate this. Now just working with the rust core without the js layer. Having a local database with the following uri postgresql://user:password@localhost:5432/postgres?schema=sql_load_test&connection_limit=5.

Starting the Query Engine against the database and querying PostgreSQL about the number of open connections:

postgres=# SELECT sum(numbackends) FROM pg_stat_database;
-[ RECORD 1 ]
sum | 6

Which is my psql client and 5 connections for the pool. I try to keep this open, query every now and then and see if the async pool has a problem with leaked connections, which might be possible due to their nature, but might also be something else.

@timsuchanek suggested the JS client also might not close the QE cleanly keeping the connections open. We’ll put some effort now to replicate the issue here.