Databases, network and secrets
Postgres and Redis in tests, tests that need the internet, secrets, build steps and generated-file checks.
Databases and services
A test that needs Postgres starts it the way you would on a laptop. Install it in .land/setup.sh, and start a fresh copy inside the test command. There is no services file to write.
#!/bin/sh # .land/setup.sh: install Postgres once, with the network on set -e apt-get update -qq DEBIAN_FRONTEND=noninteractive apt-get install -y -qq postgresql ln -sf "$(ls -d /usr/lib/postgresql/*/bin | tail -1)" /opt/pgbin
#!/bin/sh # scripts/with-pg.sh: a fresh Postgres for one test command, then the command, then stop set -e PG=/opt/pgbin; DATA="$TMPDIR/pg" $PG/initdb -D "$DATA" -A trust -U app >/dev/null $PG/pg_ctl -D "$DATA" -o "-k $TMPDIR -c listen_addresses=''" -w -s start trap '$PG/pg_ctl -D "$DATA" -m fast -w -s stop' EXIT export PGHOST="$TMPDIR" PGUSER=app PGDATABASE=postgres PATH="$PG:$PATH" "$@"
# .land/tests # name folder files command orders . - scripts/with-pg.sh pytest -q tests/test_orders.py api . tests/api/test_*.py scripts/with-pg.sh pytest -q {file}
- Every test command gets its own fresh database, in its own process and network namespace, so two commands running side by side never see each other's data or ports.
- The database runs inside the command, so land records what it reads too. A change to one migration reruns only the tests whose database read that migration, and the rest are carried.
- Commands run as an unprivileged user, which Postgres requires anyway. Keep the data directory and socket in
$TMPDIR, which is empty for every command.
Redis and others
Any service that runs as a normal process works the same way: install it in setup.sh, start it in the command.
# setup.sh apt-get install -y -qq redis-server # in the test command redis-server --port 0 --unixsocket "$TMPDIR/redis.sock" --daemonize yes REDIS_URL="unix://$TMPDIR/redis.sock" pytest -q tests/cache
Tests that need the internet
Commands run offline by default, which is what lets land carry them. A command that has to reach the internet says network after ;; on its line. It runs on every merge, because land can't record what it read from outside, and the verdict marks it as run online.
# .land/tests contract . - pytest -q tests/contract ;; network staging . - npm run e2e ;; secrets=STAGING_URL,STAGING_TOKEN
Secrets
secrets=A,B gives a command those secrets as environment variables (it implies network). The owner sets each one once. A secret is stored encrypted, given only to the lines that name it, and masked in output.
printf %s "$TOKEN" | land secret set STAGING_TOKEN # the value on stdin, never an argument land secret set < .env.staging # every NAME=value line of a .env file (land 0.3.5+)
Code an agent pushes into a command that names a secret can read that secret, exactly as in any CI. Name secrets only on the lines that need them.
Shared build steps
A line marked keep is a build step: what it writes is kept, and lines marked after=NAME see it. It is recorded and carried like a test, so the build reruns only when something it read changed.
# .land/tests build . - pnpm run build ;; keep web apps/web src/**/*.test.ts pnpm exec vitest run {files} ;; after=build
Generated files are up to date
Generated code (API clients, protobufs, ORM models) drifts when someone edits the source and forgets to regenerate. Add the check as one more line. land reruns it only when something the generator read changed.
# .land/tests: fails when someone changed the schema but not the generated code generated . - make generate && git diff --exit-code
What isn't supported
Docker inside a test command, docker compose and testcontainers don't work yet. Install the service in setup.sh and start it in the command instead, as above. See Limits.