This repo contains the app that was used to collect the data for ORM benchmarks.

You can learn more about the benchmark methodology and results in this blog post: Performance Benchmarks: Comparing Query Latency across TypeScript ORMs & Databases.

Clone the repo, navigate into it and install dependencies:

git clone git@github.com:prisma/orm-benchmarks.git
cd orm-benchmarks
npm install

Set the DATABASE_URL environment variable to your database connection string in a .env file.

First, create a .env file:

touch .env

Then open the .env file and add the following line:

DATABASE_URL="your-database-url"

For example:

DATABASE_URL="postgresql://user:password@host:port/db"

export DATABASE_URL="postgresql://user:password@host:port/db"

To create the database and the schema, run the prisma db push command by pointing it to the schema of your database.

If you use PostgreSQL, run:

npx prisma db push --schema ./prisma-pg/schema.prisma

This project uses pg-native for high-performance native bindings to PostgreSQL via libpq. This requires additional setup:

Requirements:

• Your local PostgreSQL version must match your target PostgreSQL server version (for pg_dump/pg_restore compatibility)
• PostgreSQL client libraries and tools must be installed locally

macOS setup:

# Install libpq (required for pg-native)
brew install libpq

# Install full PostgreSQL (includes pg_dump/pg_restore)
brew install postgresql

# Verify pg_config is in your PATH
pg_config --version

Alternative: Simpler Setup (No native dependencies)

If you prefer to avoid native dependencies or encounter build issues, you can use the Prisma-based seeder instead:

1. Remove pg-native dependency:

   npm uninstall pg-native

2. Update package.json start script - remove NODE_PG_FORCE_NATIVE=true from the start command

3. Switch to Prisma seeder in src/run-benchmarks-pg.ts:

   // Comment out the native seeder
   // import { preparePg } from "./lib/prepare-pg-native"; // seed via `pg_restore`
   
   // Uncomment the Prisma seeder
   import { preparePg } from "./lib/prepare-pg-prisma"; // seed via `createMany`

Note: The native approach (pg-native + pg_restore) is faster for large datasets, while the Prisma approach is simpler to set up but slower for seeding.

Note: We may add more databases in the future.

Note for PostgreSQL: Since the data preparation/seeding relies on pg_dump and pg_restore, the PostgreSQL versions of the machine that's executing the script must match the version of the target PostgreSQL server.

sh ./benchmark.sh -i 500 -s 1000

This executes the benchmark scripts with 500 iterations and a sample size of 1000 records per table. See below for the different options you can provide to any benchmark runs.

The results of the benchmark run will be stored in a folder called results/DB-SIZE-ITERATIONS-TIMESTAMP, e.g. results/postgresql-1000-500-1721027353940. This folder will have one .csv file per ORM, e.g.:

results/postgresql-1000-500-1721027353940
├── drizzle.csv
├── prisma.csv
└── typeorm.csv

To generate JSON output for website visualization from your benchmark results, run:

npm run website:output results/postgresql-1000-500-1721027353940

Replace results/postgresql-1000-500-1721027353940 with your actual results directory path. This will output a JSON structure that can be used for result visualization.

You can execute the benchmarks by running the benchmark.sh:

sh ./benchmark.sh [options]

You can provide the following options to the script:

For example:

sh ./benchmark.sh -i 500 -s 1000 --database-url postgresql://user:password@host:port/db

You can turn on two debug setting via the DEBUG environment variable:

• benchmarks:compare-results: Compare the results at the end of each benchmark run. Note that this approach will consume more memory because the results of all executed queries are collected.

• This repo contains an unfinished MySQL implementation.
• The final results that are published on benchmarks.prisma.io are based on the data in ./results-website.
• The script in ./src/lib/website-output.ts is used to generate the JSON structures that are the basis for the result visualisation in benchmarks.prisma.io.
• For Postgres, there are two ways how the seeding is implemented:
  • Option 1: Seed data with Prisma Client's createMany query (implemented in ./src/lib/prepare-pg-prisma.ts)
  • Option 2: Seed data with Prisma Client's createMany query only once, then store a DB dump in ./data/... and restore the dump upon subsequent invocations (implemented in ./src/lib/prepare-pg-native.ts)