pg_embedded_setup_unpriv

pg_embedded_setup_unpriv prepares a postgresql_embedded data directory while you are running as root, dropping privileges to nobody only for the filesystem mutations that must occur as the target user. The binary is useful when you need to initialize PostgreSQL assets inside build pipelines or CI images where direct access to the nobody account is required.

Prerequisites

Linux host with sudo or direct root access.
Rust toolchain (matching rust-toolchain.toml) and cargo installed.
Ability to download crates from crates.io.

Configuration model

Configuration is discovered via environment variables, files, and CLI arguments thanks to ortho_config. All fields are optional; when omitted the defaults from postgresql_embedded::Settings::default() are used.

PG_VERSION_REQ – SemVer requirement such as ^17 or =16.4.0.
PG_PORT – TCP port for the server to listen on. Defaults to 5432.
PG_SUPERUSER – Administrator account name.
PG_PASSWORD – Password for the superuser.
PG_DATA_DIR – Directory where PostgreSQL data files should live.
PG_RUNTIME_DIR – Directory for the downloaded distribution and binaries.
PG_LOCALE – Locale passed to initdb.
PG_ENCODING – Cluster encoding (for example UTF8).
PG_SHUTDOWN_TIMEOUT_SECS – Optional number of seconds to wait for PostgreSQL to stop during teardown. Defaults to 15 seconds and accepts values between 1 and 600.

You may also provide these values through a configuration file named pg.toml, pg.yaml, or pg.json5 (depending on enabled features) located in any path recognised by ortho_config, or through CLI flags if you wrap the binary inside your own launcher.

Running the setup helper

Ensure the desired directories exist or can be created. They will be owned by nobody after the tool completes.

Export any configuration overrides, for example:

export PG_VERSION_REQ="^17"
export PG_DATA_DIR="/var/lib/postgres/data"
export PG_RUNTIME_DIR="/var/lib/postgres/runtime"

Execute the binary as root so it can chown the directories before dropping privileges:
```
sudo -E cargo run --release --bin pg_embedded_setup_unpriv
```
The -E flag preserves any exported PG_* variables for the run.
On success the command exits with status 0. The PostgreSQL payload is downloaded into PG_RUNTIME_DIR, initialized into PG_DATA_DIR, and both paths are owned by nobody. Any failure emits a structured error via color-eyre to standard error and the process exits with status 1.

Troubleshooting

Must be run as root – The helper aborts when the effective UID is not 0. Re-run the command using sudo or inside a root shell.
Directory permission errors – Confirm the paths specified in PG_DATA_DIR and PG_RUNTIME_DIR are writable by root so ownership can be transferred to nobody.
PostgreSQL download issues – Ensure outbound network access is available to fetch the PostgreSQL distribution used by postgresql_embedded.
Invalid TimeZone parameter – The embedded cluster requires access to the system timezone database. Install your distribution's tzdata (or equivalent) package inside the container or VM running the tool.

Integration testing with `rstest`

The crate ships an rstest fixture, test_support::test_cluster, so test modules can request a ready TestCluster without invoking constructors manually. Bring the fixture into scope and declare a parameter named test_cluster to opt into automatic setup and teardown.

use pg_embedded_setup_unpriv::{test_support::test_cluster, TestCluster};
use rstest::rstest;

#[rstest]
fn migrates_schema(test_cluster: TestCluster) {
    let url = test_cluster.connection().database_url("postgres");
    assert!(url.starts_with("postgresql://"));
}

Because the fixture handles environment preparation, tests stay declarative and can focus on behaviours instead of bootstrap plumbing. When a bootstrap failure occurs the fixture panics with a SKIP-TEST-CLUSTER prefix, so higher-level behaviour tests can convert known transient errors into soft skips.

Fast test isolation with template databases

For test suites where per-test cluster bootstrap is too slow, use shared_test_cluster to share a single cluster across all tests and clone template databases for isolation:

use pg_embedded_setup_unpriv::{test_support::shared_test_cluster, TestCluster};
use rstest::rstest;

#[rstest]
fn fast_isolated_test(shared_test_cluster: &'static TestCluster) {
    // Clone a pre-migrated template (milliseconds, not seconds)
    shared_test_cluster
        .create_database_from_template("my_test_db", "migrated_template")
        .unwrap();

    let url = shared_test_cluster.connection().database_url("my_test_db");
    // ... run test queries ...

    shared_test_cluster.drop_database("my_test_db").unwrap();
}

Key features:

shared_cluster() – process-global cluster initialized once per binary
create_database_from_template() – clone databases via PostgreSQL's TEMPLATE clause (filesystem copy, completes in milliseconds)
ensure_template_exists() – concurrency-safe template creation with per-template locking
hash_directory() – generate content-based template names that auto-invalidate when migrations change

See docs/users-guide.md for complete examples and performance guidance.

Behaviour-driven diagnostics

Behavioural coverage relies on rstest-bdd (Behaviour-Driven Development, BDD) v0.1.0, which now bundles Fluent localization files. The test suite includes tests/localized_diagnostics.rs, a Dutch Gherkin scenario that switches diagnostics to French via rstest_bdd::select_localizations and fails if the embedded assets are missing. Run make test (or the focused cargo test localized_diagnostics) in CI to ensure every target platform loads the lazy localization payload correctly.

Next steps

After the bootstrap completes you can start PostgreSQL with postgresql_embedded (or another supervisor of your choice) using the same directories and superuser credentials established by the helper.

pg-embed-setup-unpriv 0.3.0