PANIC: called `Option::unwrap()` on a `None` value in query-engine/core/src/interpreter/query_interpreters/nested_read.rs:231:50

Bug description

Hi, I’m suddenly getting this on our Prisma server:

Thrown with args: {}
Resolver info : { prev: undefined, key: 'me', typename: 'Query' }
Model : User
Error : PrismaClientRustPanicError2 [PrismaClientRustPanicError]:
Invalid `prisma.user.findUnique()` invocation:
  PANIC: called `Option::unwrap()` on a `None` value in query-engine/core/src/interpreter/query_interpreters/nested_read.rs:231:50
This is a non-recoverable error which probably happens when the Prisma Query Engine has a panic.

How to reproduce

1. Launch your prisma project
2. Leave it running for ~20 hours
3. See the above error

Expected behavior

As Douglas Adams wrote, “Dont panic!”

Prisma information

Using paljs and nexus

query { me { id } }

Environment & setup

• OS: Linux using Docker node:16-buster
• Database:

generator client {
  provider        = "prisma-client-js"
  previewFeatures = ["selectRelationCount", "filterJson", "referentialActions", "nApi"]
}

datasource db {
  provider = "mysql"
  url      = env("DATABASE_URL")
}

• Node.js version: 16

Prisma Version

"@prisma/client": "2.28.0",
"prisma": "2.28.0",

[andrewicarlson]

I can reproduce this issue. Please see this repo for a small reproduction. It seems as though this Rust Panic occurs when a relationship is created using a non-unique constraint. Some connectors check for this (SQLite) but others don’t and cause a runtime error (MySQL)

Notes:

1. When using the SQLite connector an error is thrown in the command line tooling, catching this error before it gets to production. MySQL throws no error and then the Rust engine Panics at runtime.

Error: P1012

error: Error validating: The argument `references` must refer to a unique criteria in the related model `A`. But it is referencing the following fields that are not a unique criteria: custom_id
  -->  schema.prisma:21

1. This reproduction uses v3.11.0. In >v3.12 an opaque error is thrown in the command line tooling that is difficult to debug.

An error occured while running the seed command:
Error: Command was killed with SIGKILL (Forced termination): ts-node prisma/seed.ts

[binary64]

My hunches are:

• In our express server, we have a global array of PrismaClients. Each customer has their own mysql database. Plus we have one master mysql which has the list of the customers. So in some resolvers, we even call both the master, and the customer PrismaClient.
• There is a memory issue where you do a large query that somehow starves or corrupts the arguments for the nested_join.rs function until the node is restarted.