Skip to content

Latest commit

 

History

History
518 lines (372 loc) · 19.9 KB

TESTING.md

File metadata and controls

518 lines (372 loc) · 19.9 KB

Testing

Local databases for tests

To run tests requiring a database, start the test databases using Docker, see Docker.

Environment variables

These are located in the .db.env file which is loaded automatically by the test commands.

Optionally, if you want the environment variables to always be accessible, you can install direnv.

Jest tips

  1. Install/update the dependencies, run the following command in the root directory:

    pnpm i
  2. We use the Jest test framework. Its CLI is powerful and removes the need for npm scripts mostly. For most cases this is what you need to know:

    Note: the following command pnpm run test can be used inside the packages folders like packages/client/. In the base folder you can only run pnpm run test without extra arguments.

    pnpm run test <fileNamePattern> -t <testNamePattern>

    and to update snapshots use the -u option

    pnpm run test <fileNamePattern> -u
  3. In packages/integration-tests/ Jest's each feature is used. If you only want to run a subset of the test cases, simply leverage the -t flag on the command line (see above point). Here is an example to run only the findOne where PK cases for SQLite:

    pnpm run jest integration.sqlite -t 'findOne where PK'

Where should I find and write tests?

Something is broken? You built a new feature? It's time to write a test! But where?

Everything related to working with specific frameworks like Next.js or using external resources like deploying to Netlify should be covered by an Ecosystem Test.

Everything that is more basic functionality like a specific query or feature, that doesn't need a platform specific test (yet) should get a test in the prisma/prisma repo.

Rule of thumb: If you can write a test in prisma/prisma, prefer that over a test in prisma/ecosystem-tests.

In the prisma/prisma repository we have a few places where you can write tests:

  • cli
    • ./packages/cli/src/__tests__/ - Tests for prisma studio, prisma version, prisma format, prisma generate, prisma generate, loading .env file loading...
  • client
  • debug
    • Unit tests for debug package
  • generator-helper
    • Integration tests for generator interface implementation
  • migrate
    • Unit and integration tests for migrate and db commands
  • internals
    • Convert credentials to connection string and back
    • Dotenv expansion
    • Engine commands (getDMMF, getConfig) (snapshots)
    • getGenerators (central function for generation)
    • introspection (snapshots)
  • integration-tests (Prisma Client & Introspection)
    • Integration tests for basic query and mutation functionality
    • All databases that we support are covered here:
      • mariadb
      • mssql (SQL Server)
      • mysql
      • postgresql
      • sqlite
    • While these tests also test the client itself, they're rather only a base to make sure that basic query engine functionality actually works in the Prisma Client
    • When you want to test very specific queries for a new feature, you can write a functional test in the client package, as that's usually easier

Prisma CLI

Testing a namespace

See how db namespace is tested in DbCommand.test.ts When creating a new namespace, a <command>Command.test.ts file must be created and filled with unit tests.

Testing a command

See how the init command is tested in Init.test.ts When creating a new command, a <command>.test.ts file must be created and filled with unit tests.

Running tests using Jest

If the tests you want to run require a database, see Docker.

For running all tests

pnpm run test

For running tests for a single command use pnpm run test <name>

pnpm run test init

Prisma Client: functional tests

Functional tests in the client package are testing that all aspects of client and query engine work correctly. They strive to be as close as possible to the way client will be used in real project: they generate an actual client, talk to a real database, perform the type checks and generally test the client through its public API.

Creating a new functional test

To create new test, run following command

pnpm new-test

You'll then be asked for the name of your test and list of providers you want to run this test on. If you opt out of testing any of the providers, you'll also have to specify the reason. New test will be created under test/functional/<name of the test> directory.

Structure of the functional test setup

Test consists of the 3 files:

  • test matrix _matrix.ts
  • schema template prisma/_schema.ts
  • test suite tests.ts

Nesting tests

If you have related tests but different schemas, you can nest directories. For example, composite indexes have the following file structure:

composite-index/

  1. list/

    • test.ts
    • _matrix.ts
    • prisma/
      • _schema.ts
  2. named/

    • test.ts
    • _matrix.ts
    • prisma/
      • _schema.ts

Test matrix

_matrix.ts file defines parameters for generating test suites. It can have as many parameters as necessary, but at minimum, it should define at least one provider.

This example matrix defines 2 test suites: one for the SQLite provider and one for MongoDB, while also providing some test data to use later for both providers:

import { defineMatrix } from '../_utils/defineMatrix'

export default defineMatrix(() => [
  [
    {
      provider: Providers.SQLITE
      testEmail: '[email protected]',
    },
    {
      provider: Providers.MONGODB
      testEmail: '[email protected]',
    },
  ],
])

If the matrix has multiple dimensions, test suites will be generated for all permutations of the parameters. For example, the following matrix:

import { defineMatrix } from '../_utils/defineMatrix'

export default defineMatrix(() => [
  [
    {
      provider: Providers.SQLITE,
    },
    {
      provider: Providers.POSTGRESQL,
    },
  ],
  [
    {
      providerFeatures: '',
    },

    {
      providerFeatures: 'improvedQueryRaw',
    },
  ],
])

Will generate following test suites:

  • { provider: Providers.SQLITE providerFeatures: '' }
  • { provider: Providers.SQLITE providerFeatures: 'improvedQueryRaw' }
  • { provider: Providers.POSTGRESQL providerFeatures: '' }
  • { provider: Providers.POSTGRESQL providerFeatures: 'improvedQueryRaw' }

You can also optionally exclude certain combinations from matrix by using second argument of defineMatrix function:

import { defineMatrix } from '../_utils/defineMatrix'

export default defineMatrix(
  () => [
    [
      {
        provider: Providers.SQLITE,
      },
      {
        provider: Providers.POSTGRESQL,
      },
    ],
    [
      {
        providerFeatures: '',
      },

      {
        providerFeatures: 'improvedQueryRaw',
      },
    ],
  ],
  {
    exclude: ({ provider, providerFeatures }) =>
      provider === Providers.SQLITE && providerFeatures === 'improvedQueryRaw',
  },
)

This will generate the same test suites as the previous example, except sqlite/improvedQuery combination

Schema template

prisma/_schema.ts will be used for generating an actual schema for the test suite:

import { idForProvider } from '../../_utils/idForProvider'
import testMatrix from '../_matrix'

export default testMatrix.setupSchema(({ provider }) => {
  return /* Prisma */ `
    generator client {
      provider = "prisma-client-js"
    }
    
    datasource db {
      provider = "${provider}"
      url      = env("DATABASE_URI_${provider}")
    }
    
    model User {
      id ${idForProvider(provider)}
    }
  `
})

setupSchema callback receives all parameters from the matrix for a particular test suite as an argument. idForProvider is a helper function which returns a correct primary key definition for each of the supported providers.

Test suite

tests.ts contains actual tests for the suite:

import { Providers } from '../_utils/providers'
import testMatrix from './_matrix'

// @ts-ignore at the moment this is necessary for typechecks
declare let prisma: import('@prisma/client').PrismaClient

testMatrix.setupTestSuite(
  (suiteConfig, suiteMeta) => {
    test('create', async () => {
      await prisma.user.create({
        data: {
          // `testEmail` was defined in the example matrix before
          // you can also use a constant here if you don't need
          // unique email per provider
          email: suiteConfig.testEmail,
        },
      })
    })
  },
  {
    optOut: {
      // if you are skipping tests for certain providers, you
      // have to list them here and specify the reason
      from: [Providers.MONGODB],
      reason: 'The test is for SQL databases only',
    },
    skipDataProxy: {
      // similarly, you can opt out of testing with the Data Proxy
      // client (either completely or for certain runtimes) and
      // specify the reason
      runtimes: ['node', 'edge'],
      reason: "This test doesn't work with Data Proxy",
    },
  },
)

This test will run for every permutation of the parameters from the matrix. Currently used combination is available via suiteConfig parameter. Each suite will start with a clean already set up database, generated and initialized client, available via prisma global. After suite is finished, database will be dropped and client instance will disconnect automatically.

Running functional tests

  • pnpm test:functional:code generates and runs the test suites, defined by test matrix. It does no typechecking, but prepares all necessary files for it.
  • pnpm test:functional:types runs typechecking on all the suites, generated by pnpm test:functional:code command. If it reports any errors, you might want to examine generated test suite under tests/functional/<your test name>/.generated directory to get a better diagnostic.
  • pnpm test:functional will run tests and then perform type checks.

Add --data-proxy CLI flag to any of these commands to generate the Data Proxy client and run the tests under the local Data Proxy simulator called Mini-Proxy.

When using --data-proxy, you can additionally pass the following CLI flags:

  • --no-mini-proxy-server: don't start the Mini-Proxy server in node_modules automatically. This requires you to start the Mini-Proxy server and prepare the environment to make the root certificate trusted by Node.js in the current shell with eval $(mini-proxy env) before starting the tests.
  • --mini-proxy-debug: enable debug logs when using the default Mini-Proxy server.

Conditionally skipping type tests

Sometimes different test matrix parameters will generate different TS types so, and as a result, correct code for one suite config won't pass type checks for a different one. In that case, you can use @ts-test-if magical comment to conditionally skip type checks. For example:

// @ts-test-if: provider !== Providers.MONGODB
prisma.$queryRaw`...`

If JS expression, following @ts-test-if: evaluates to truthy value, @ts-expect-error will be inserted in its place during the generation. All parameters from the test matrix can be used within that expression.

Memory tests

This suite tests client for memory leaks. It works by repeatedly running test code in a loop and monitoring V8 heap usage after every iteration. If it detects that memory usage grows with a rate higher than the threshold, it will report a memory leak and fail the test.

To create a memory test you need 2 files:

  • schema, located in tests/memory/<your-test-name>/prisma/schema.prisma file, which will be used for setting up the database.
  • test code, tests/memory/<your-test-name>/test.ts

Writing memory test

Tests are created using createMemoryTest function:

import { createMemoryTest } from '../_utils/createMemoryTest'

//@ts-ignore
type PrismaModule = typeof import('./.generated/node_modules/@prisma/client')

void createMemoryTest({
  async prepare({ PrismaClient }: PrismaModule) {
    const client = new PrismaClient()
    await client.$connect()
    return client
  },
  async run(client) {
    await client.user.findMany()
  },
  async cleanup(client) {
    await client.$disconnect()
  },
  iterations: 1500,
})

createMemoryTest accepts following options:

  • prepare(prismaModule) is used for any setup code, needed by the test. It is executed only once. Return value of this function will be passed into run and cleanup methods.
  • run(prepareResult) - actual test code. It is expected that after executing this function, memory usage does not increase. This function is repeatedly executed in a loop while the test runs.
  • cleanup(prepareResult) used for cleaning up any work, done in prepare function.
  • iterations specifies the number of executions of run function, for which memory measurement will be done, default is 1000. On top of that, test setup also adds 100 iterations for warming up.

Running memory tests

  • pnpm test:memory for running whole test suite
  • pnpm test:memory <test name> for running single test

How to use custom engines

By default, you get the engines that are downloaded on postinstall in @prisma/engines thanks to @prisma/engines-version.

However, you may want to use a custom engine via from a branch in prisma/prisma-engines, or one that you've built locally.

Prerequisites

You will need to have installed the Rust toolchain and a few extra dependencies. See Building Prisma Engines.

Using custom engines

  1. Edit ./packages/fetch-engine/package.json
  • Either, add a branch property to enginesOverride

    This will git pull the branch and build the engines from there.

      "enginesOverride": {
    +    "branch": "feat/column-comparison",
      }
  • Or, add a folder property to enginesOverride

    This will copy the engines from the folder where you've built them.

      "enginesOverride": {
    +    "folder": "/home/john/dev/prisma/prisma-engines/target/release"
      }
  1. Run pnpm install again to propagate the new engines.

On CI

For open pull request, can also add /engine-branch branchName command into PR body and re-run the pipeline. Engine from corresponding branch will be checked out and built before running any tests on CI.

CI - Continuous Integration

By creating a Pull Request the following pipelines will be triggered

CI will need to be successful before merging ("flaky" tests might show up and might be ignored).

By default, some tests are tested only during daily builds (e.g. binary engine). If you need to run all of them for your PR leave a ci test all comment on the PR and re-run the workflow.

Publishing all the packages to npm on the integration tag

If a branch name starts with integration/ like integration/fix-all-the-things the GitHub Actions - npm - release to dev/integration pipeline will be triggered. This workflow will directly publish (without running tests) the packages to npm on the integration tag with a version like 5.3.0-integration-fix-all-the-things.1 (where 5.3.0- is the current dev version prefix, integration- is statically added, fix-all-the-things is from the branch name and .1 indicates the first version published from this branch)

To make a Pull Request which will release a version to the integration tag automatically, the name of the branch of the PR would need to start with integration/. Alternatively, add /integration in the Pull Request description:

  • If this is added before opening the Pull Request, a release will happen automatically.
  • If this is added after the creation of the PR, the Detect jobs to run job from the CI workflow needs to be re-triggered to get a release.

The GitHub Actions - npm - release to dev/integration workflow can also be manually triggered to run on any branch. Example:

Once published to npm the version will need to be installed with the exact version like:

npm install -D [email protected]

# or executed with npx like
npx [email protected]

(Note that npm version upgrades or the update notifier in Prisma CLI might behave weird and unexpectedly with these integration versions.)

Internal note: You can check the #feed-prisma-releases channel on our private Slack to get notified when versions are published.

Once the integration version is published on npm:

  • The check-for-update workflow, which runs continuously, will find the new version, update the package.json and do a commit on the integration branch
  • The tests workflows will then run for that commit and will be visible here

How to trigger artificial panics for our engines

Sometimes it may be useful to trigger a panic in the Rust binaries or libraries used by Prisma under the hood. Most of the Rust artifacts are shipped as binaries, whereas query-engine is shipped both as a library (by default) and as a binary (on demand). To change the default Rust artifacts' type used under the hood, you can set the PRISMA_CLI_QUERY_ENGINE_TYPE environment variable to either library or binary.

Setup

  • mkdir artificial-panics && cd artificial-panics
  • npx prisma init --datasource-provider sqlite

Trigger panic in Schema Engine

  • run FORCE_PANIC_SCHEMA_ENGINE=1 npx prisma migrate dev

Trigger panic in Formatter

  • run FORCE_PANIC_PRISMA_SCHEMA=1 npx prisma format

Trigger panic in Query Engine - Get DMMF

  • run FORCE_PANIC_QUERY_ENGINE_GET_DMMF=1 npx prisma validate

Trigger panic in Query Engine - Get Config

  • run FORCE_PANIC_QUERY_ENGINE_GET_CONFIG=1 npx prisma validate