Test Partitioning

Tests run in parallel partitions (default: 3). Each partition uses its own database.

check --partitions 4  # Run with 4 partitions

Database setup

Each partition needs its own database to avoid deadlocks and connection limit issues. In config/runtime.exs, use MIX_TEST_PARTITION to create per-partition database URLs:

# config/runtime.exs
partition = System.get_env("MIX_TEST_PARTITION", "")

test_database_url =
  System.get_env("TEST_DATABASE_URL")
  |> URI.parse()
  |> then(fn uri ->
    db_name = String.trim_leading(uri.path || "", "/")
    Map.put(uri, :path, "/#{db_name}#{partition}")
  end)
  |> URI.to_string()

config :my_app, MyApp.Repo, url: test_database_url

This creates databases like my_app_test1, my_app_test2, etc.

Avoiding DB connection limits

With N partitions, your database needs at least N * pool_size connections. Increase max_connections in PostgreSQL if you see connection errors:

# PostgreSQL config (or docker run command)
postgres -c max_connections=400 -c shared_buffers=2GB

Consider reducing pool_size per partition in config/test.exs:

config :my_app, MyApp.Repo, pool_size: 10

With 3 partitions and pool_size 10, you need at least 30 connections.

Scheduler tuning

AlCheck automatically limits BEAM schedulers per partition to avoid CPU contention: schedulers_online / partitions. With 10 cores and 3 partitions, each gets ~3 schedulers.

Partition capping

If --dir points to a directory with fewer test files than the partition count, AlCheck automatically caps partitions to the file count and warns:

Not enough test files for 3 partitions (found 1), using 1

Managing partition databases

Set up or drop databases for all partitions in parallel:

check --setup-db                  # mix ecto.setup for each partition
check --drop-db                   # mix ecto.drop for each partition
check --for-partitions 'mix ecto.reset'   # any command across partitions

See Workflows for details.