Skip to content

README.md

Latest commit

 

History

History
1059 lines (834 loc) · 51 KB

README.md

File metadata and controls

1059 lines (834 loc) · 51 KB

cargo-pgrx

cargo-pgrx is a Cargo subcommand for managing pgrx-based Postgres extensions.

You'll want to use cargo pgrx during your extension development process. It automates the process of creating new Rust crate projects, auto-generating the SQL schema for your extension, installing your extension locally for testing with Postgres, and running your test suite against one or more versions of Postgres.

A video walkthrough of its abilities can be found here: https://www.twitch.tv/videos/684087991

Installing

Install via crates.io:

$ cargo install --locked cargo-pgrx

As new versions of pgrx are released, you'll want to make sure you run this command again to update it.

Note that some of the features of PGRX involve compiling C code, including cargo pgrx init, and as such you will also need a toolchain for doing so and potentially must provide various libraries. Normally, Rust requires a C toolchain anyways, but it does not require e.g. pkg-config.

Usage

$ cargo pgrx --help
Cargo subcommand for 'pgrx' to make Postgres extension development easy

Usage: cargo pgrx [OPTIONS] <COMMAND>

Commands:
  bench         Run in-process benchmarks for this crate
  init          Initialize pgrx development environment for the first time
  info          Provides information about pgrx-managed development environment
  start         Start a pgrx-managed Postgres instance
  stop          Stop a pgrx-managed Postgres instance
  status        Is a pgrx-managed Postgres instance running?
  new           Create a new extension crate
  install       Install the crate as an extension into the Postgres specified by `pg_config`
  package       Create an installation package directory
  schema        Generate extension schema files
  run           Compile/install extension to a pgrx-managed Postgres instance and start psql
  connect       Connect, via psql, to a Postgres instance
  test          Run the test suite for this crate
  get           Get a property from the extension control file
  cross         Commands having to do with cross-compilation. (Experimental)
  upgrade       Upgrade pgrx crate versions in `Cargo.toml`. Defaults to latest
  regress       Run the regression test suite for this crate
  help          Print this message or the help of the given subcommand(s)

Options:
  -v, --verbose...  Enable info logs, -vv for debug, -vvv for trace
  -h, --help        Print help
  -V, --version     Print version

Environment Variables

  • PGRX_HOME - Defaults to "${HOME}/.pgrx/" if not set.
  • PGRX_BUILD_FLAGS - If set during cargo pgrx run/test/install/package/bench, these additional flags are passed to cargo build while building the extension
  • PGRX_BUILD_VERBOSE - Set to true to enable verbose "build.rs" output -- useful for debugging build issues
  • HTTPS_PROXY - If set during cargo pgrx init, it will download the Postgres sources using these proxy settings. For more details refer to the env_proxy crate documentation.
  • PGRX_IGNORE_RUST_VERSIONS - Set to true to disable the rustc version check we have when performing schema generation (schema generation requires the same version of rustc be used to build cargo-pgrx as the crate in question).

First Time Initialization

$ cargo pgrx init
   Discovered Postgres v12.16, v13.12, v14.9, v15.4, v16.0
  Downloading Postgres v14.9 from https://ftp.postgresql.org/pub/source/v14.9/postgresql-14.9.tar.bz2
  Downloading Postgres v15.4 from https://ftp.postgresql.org/pub/source/v15.4/postgresql-15.4.tar.bz2
  Downloading Postgres v12.16 from https://ftp.postgresql.org/pub/source/v12.16/postgresql-12.16.tar.bz2
  Downloading Postgres v13.12 from https://ftp.postgresql.org/pub/source/v13.12/postgresql-13.12.tar.bz2
  Downloading Postgres v16.0 from https://ftp.postgresql.org/pub/source/v16.0/postgresql-16.0.tar.bz2
     Removing /home/you/.pgrx/12.16_unpack
     Removing /home/you/.pgrx/14.9_unpack
    Untarring Postgres v12.16 to /home/you/.pgrx/12.16_unpack
    Untarring Postgres v14.9 to /home/you/.pgrx/14.9_unpack
     Removing /home/you/.pgrx/15.4_unpack
    Untarring Postgres v15.4 to /home/you/.pgrx/15.4_unpack
     Removing /home/you/.pgrx/16.0_unpack
    Untarring Postgres v16.0 to /home/you/.pgrx/16.0_unpack
     Removing /home/you/.pgrx/13.12_unpack
    Untarring Postgres v13.12 to /home/you/.pgrx/13.12_unpack
     Removing /home/you/.pgrx/12.16
     Removing /home/you/.pgrx/14.9
     Renaming /home/you/.pgrx/12.16_unpack/postgresql-12.16 -> /home/you/.pgrx/12.16
  Configuring Postgres v12.16
     Renaming /home/you/.pgrx/14.9_unpack/postgresql-14.9 -> /home/you/.pgrx/14.9
  Configuring Postgres v14.9
     Removing /home/you/.pgrx/15.4
     Renaming /home/you/.pgrx/15.4_unpack/postgresql-15.4 -> /home/you/.pgrx/15.4
  Configuring Postgres v15.4
     Removing /home/you/.pgrx/16.0
     Renaming /home/you/.pgrx/16.0_unpack/postgresql-16.0 -> /home/you/.pgrx/16.0
  Configuring Postgres v16.0
     Removing /home/you/.pgrx/13.12
     Renaming /home/you/.pgrx/13.12_unpack/postgresql-13.12 -> /home/you/.pgrx/13.12
  Configuring Postgres v13.12
    Compiling Postgres v16.0
    Compiling Postgres v12.16
    Compiling Postgres v14.9
    Compiling Postgres v15.4
    Compiling Postgres v13.12
   Installing Postgres v12.16 to /home/you/.pgrx/12.16/pgrx-install
   Installing Postgres v13.12 to /home/you/.pgrx/13.12/pgrx-install
   Installing Postgres v14.9 to /home/you/.pgrx/14.9/pgrx-install
   Installing Postgres v15.4 to /home/you/.pgrx/15.4/pgrx-install
   Installing Postgres v16.0 to /home/you/.pgrx/16.0/pgrx-install
   Validating /home/you/.pgrx/12.16/pgrx-install/bin/pg_config
   Validating /home/you/.pgrx/13.12/pgrx-install/bin/pg_config
   Validating /home/you/.pgrx/14.9/pgrx-install/bin/pg_config
   Validating /home/you/.pgrx/15.4/pgrx-install/bin/pg_config
   Validating /home/you/.pgrx/16.0/pgrx-install/bin/pg_config

cargo pgrx init is required to be run once to properly configure the pgrx development environment.

As shown by the screenshot above, it downloads the latest releases of supported Postgres versions, configures them for debugging, compiles them with assertions, and installs them to "${PGRX_HOME}". These include all contrib extensions and tools included with Postgres. Other cargo pgrx commands such as run and test will manage and use these installations on your behalf.

pgrx is designed to support multiple Postgres versions in such a way that during development, you'll know if you're trying to use a Postgres API that isn't common across all versions. It is also designed to make testing your extension against these versions easy. This is why it enables debug symbols and the database assertions, and why it expects all versions of Postgres that your extension supports to be installed during development.

In cases when default ports pgrx uses to run PostgreSQL within are not available, one can specify custom values for these during initialization using --base-port and --base-testing-port options. One of the use cases for this is using multiple installations of pgrx (using different "$PGRX_HOME"s) when developing multiple extensions at the same time. These values can be later changed in "$PGRX_HOME/config.toml".

If you want to use your operating system's package manager to install Postgres, cargo pgrx init has optional arguments that allow you to specify where they're installed (see below). Be aware, this may result in different behavior than a database compiled by cargo-pgrx, where tests that would have been failed by an internal Postgres assertion instead successfully pass. This can be problematic if you are using pgrx-pg-sys directly, as those assertions are often the only thing that will catch directly misusing the Postgres extension API!

Each version is specified as "--pg${VER} and the full path to pg_config for that version. For any version you specify, cargo pgrx init will forego downloading/compiling/installing it. cargo-pgrx will then use that locally-installed version the same way it uses any version it compiles and installs itself.

However, if the "path to pg_config" is the literal string download, then cargo-pgrx will download and compile that version of Postgres for you. When any "--pg${VER}" option is specified, these are the only versions of Postgres that cargo-pgrx will manage for you.

You'll also want to make sure you have the "postgresql-server-dev" package installed for each version you want to manage yourself. If you need to customize the configuration of the Postgres build, you can use --configure-flag to pass optins to the configure script. For example, you could use --configure-flag=--with-ssl=openssl to enable SSL support or --configure-flag=--with-libraries=/path/to/libs to use a non-standard location for dependency libraries. This flag can be used multiple times to pass multiple configuration options.

Once complete, cargo pgrx init also creates "${PGRX_HOME}/config.toml" which describes where to find each version's pg_config tool.

If a new minor Postgres version is released in the future you can simply run cargo pgrx init [args] again, and your local version will be updated, preserving all existing databases and configuration.

$ cargo pgrx init -h
Initialize pgrx development environment for the first time

Usage: cargo pgrx init [OPTIONS]

Options:
      --pg13 <PG13>                            If installed locally, the path to PG13's `pgconfig` tool, or `download`
                                               to have pgrx download/compile/install it [env: PG13_PG_CONFIG=]
  -v, --verbose...                             Enable info logs, -vv for debug, -vvv for trace
      --pg14 <PG14>                            If installed locally, the path to PG14's `pgconfig` tool, or `download`
                                               to have pgrx download/compile/install it [env: PG14_PG_CONFIG=]
      --pg15 <PG15>                            If installed locally, the path to PG15's `pgconfig` tool, or `download`
                                               to have pgrx download/compile/install it [env: PG15_PG_CONFIG=]
      --pg16 <PG16>                            If installed locally, the path to PG16's `pgconfig` tool, or `download`
                                               to have pgrx download/compile/install it [env: PG16_PG_CONFIG=]
      --pg17 <PG17>                            If installed locally, the path to PG17's `pgconfig` tool, or `download`
                                               to have pgrx download/compile/install it [env: PG17_PG_CONFIG=]
      --base-port <BASE_PORT>                  Base port number
      --base-testing-port <BASE_TESTING_PORT>  Base testing port number
      --configure-flag <CONFIGURE_FLAG>        Additional flags to pass to the configure script
      --valgrind                               Compile PostgreSQL with the necessary flags to detect a good amount of memory errors when run under Valgrind
  -j, --jobs <JOBS>                            Allow N make jobs at once
  -h, --help                                   Print help (see more with '--help')
  -V, --version                                Print version```

## Creating a new Extension

```rust
$ cargo pgrx new example
$ ls example/
Cargo.toml  example.control  sql  src

cargo pgrx new <extname> is an easy way to get started creating a new extension. It's similar to cargo new <name>, but does the additional things necessary to support building a Rust Postgres extension.

If you'd like to create a "background worker" instead, specify the --bgworker argument.

cargo pgrx new does not initialize the directory as a git repo, but it does create a .gitignore file in case you decide to do so.

Workspace users: cargo pgrx new $NAME will create a $NAME/.cargo/config.toml, you should move this into your workspace root as .cargo/config.toml.

If you don't, you may experience unnecessary rebuilds using tools like Rust-Analyzer, as it will use the wrong rustflags option.

$ cargo pgrx new --help
cargo-pgrx-new 0.5.0
PgCentral Foundation, Inc. <contact@pgcentral.org>
Create a new extension crate

USAGE:
    cargo pgrx new [OPTIONS] <NAME>

ARGS:
    <NAME>    The name of the extension

OPTIONS:
    -b, --bgworker    Create a background worker template
    -h, --help        Print help information
    -v, --verbose     Enable info logs, -vv for debug, -vvv for trace
    -V, --version     Print version information

Managing Your Postgres Installations

$ cargo pgrx status all
Postgres v12 is stopped
Postgres v13 is stopped
Postgres v14 is stopped
Postgres v15 is stopped
Postgres v16 is stopped

$ cargo pgrx start all
    Starting Postgres v12 on port 28812
    Starting Postgres v13 on port 28813
    Starting Postgres v14 on port 28814
    Starting Postgres v15 on port 28815
    Starting Postgres v16 on port 28816

$ cargo pgrx status all
Postgres v12 is running
Postgres v13 is running
Postgres v14 is running
Postgres v15 is running
Postgres v16 is running

$ cargo pgrx stop all
    Stopping Postgres v12
    Stopping Postgres v13
    Stopping Postgres v14
    Stopping Postgres v15
    Stopping Postgres v16

cargo pgrx has three commands for managing Postgres installations: start, stop, and status. Additionally, cargo pgrx run (see below) will automatically start its target Postgres instance if not already running. Note that when directed to use "all" Postgres instances, this is interpreted in terms of an extension's pg{MAJOR} features in its Cargo.toml, except for cargo pgrx status.

When starting a Postgres instance, pgrx starts it on port 28800 + PG_MAJOR_VERSION, so Postgres 15 runs on 28815, 16 on 28816, etc. Additionally, the first time any of these are started, it will initialize PGDATA directories in "${PGRX_HOME}"/data-{12,13,14,15,16}. Doing so allows pgrx to manage either Postgres versions it installed or ones already on your computer, and ensure that the pgrx managed versions don't interfere with what might already be running. The locale of the instance is C.UTF-8 (or equivalently, a locale of C with a ctype of UTF8 on macOS), or C if the C.UTF-8 locale is unavailable.

pgrx doesn't tear down these instances. While PGRX_HOME is by default a hidden directory, pgrx considers these important and permanent database installations.

Once started, you can connect using psql (if available) like so: psql -p 28816. However, you probably just want the cargo pgrx run command.

Compiling and Running Your Extension

$ cargo pgrx run pg13
building extension with features ``
"cargo" "build" "--message-format=json-render-diagnostics"
    Finished dev [unoptimized + debuginfo] target(s) in 0.06s

installing extension
     Copying control file to /home/ana/.pgrx/13.5/pgrx-install/share/postgresql/extension/strings.control
     Copying shared library to /home/ana/.pgrx/13.5/pgrx-install/lib/postgresql/strings.so
    Building for SQL generation with features ``
    Finished dev [unoptimized + debuginfo] target(s) in 0.07s
 Discovering SQL entities
  Discovered 6 SQL entities: 0 schemas (0 unique), 6 functions, 0 types, 0 enums, 0 sqls, 0 ords, 0 hashes, 0 aggregates
     Writing SQL entities to /home/ana/.pgrx/13.5/pgrx-install/share/postgresql/extension/strings--0.1.0.sql
    Finished installing strings
    Starting Postgres v13 on port 28813
    Re-using existing database strings
psql (13.5)
Type "help" for help.

strings=# DROP EXTENSION strings;
ERROR:  extension "strings" does not exist
strings=# CREATE EXTENSION strings;
CREATE EXTENSION
strings=# \df strings.*
                                      List of functions
 Schema  |     Name      | Result data type |           Argument data types            | Type
---------+---------------+------------------+------------------------------------------+------
 strings | append        | text             | input text, extra text                   | func
 strings | return_static | text             |                                          | func
 strings | split         | text[]           | input text, pattern text                 | func
 strings | split_set     | SETOF text       | input text, pattern text                 | func
 strings | substring     | text             | input text, start integer, "end" integer | func
 strings | to_lowercase  | text             | input text                               | func
(6 rows)

strings=# select strings.to_lowercase('PGRX');
 to_lowercase
--------------
 pgrx
(1 row)

cargo pgrx run <pg13 | pg14 | pg15 | pg16 | pg17> is the primary interface into compiling and interactively testing/using your extension during development.

The very first time you execute cargo pgrx run pgXX, it needs to compile not only your extension, but pgrx itself, along with all its dependencies. Depending on your computer, this could take a bit of time (pgrx is nearly 200k lines of Rust when counting the generated bindings for Postgres). Afterwards, however (as seen in the above screenshot), it's fairly fast.

cargo pgrx run compiles your extension, installs it to the specified Postgres installation as described by its pg_config tool, starts that Postgres instance using the same process as cargo pgrx start pgXX, and drops you into a psql shell connected to a database, by default, named after your extension. From there, it's up to you to create your extension and use it.

This is also the stage where pgrx automatically generates the SQL schema for your extension via the sql-generator binary.

When you exit psql, the Postgres instance continues to run in the background.

For Postgres installations which are already on your computer, cargo pgrx run will need write permissions to the directories described by pg_config --pkglibdir and pg_config --sharedir. It's up to you to decide how to make that happen. While a single Postgres installation can be started multiple times on different ports and different data directories, it does not support multiple "extension library directories".

With the --install-only option, pgrx will compile and install your extension but will skip all subsequent steps. This might be useful if you have an alternative testing approach and don't need an interactive shell.

$ cargo pgrx run --help
Compile/install extension to a pgrx-managed Postgres instance and start psql

Usage: cargo pgrx run [OPTIONS] [PG_VERSION] [DBNAME]

Arguments:
  [PG_VERSION]  Do you want to run against pg13, pg14, pg15, pg16, pg17, or pg18? [env: PG_VERSION=]
  [DBNAME]      The database to connect to (and create if the first time).  Defaults to a database with the same name as the current extension name

Options:
  -p, --package <PACKAGE>              Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
  -r, --release                        Compile for release mode (default is debug)
      --profile <PROFILE>              Specific profile to use (conflicts with `--release`)
      --all-features                   Activate all available features
      --no-default-features            Do not activate the `default` feature
  -F, --features <FEATURES>            Space-separated list of features to activate
      --pgcli                          Use an existing `pgcli` on the $PATH [env: PGRX_PGCLI=]
      --install-only                   Install without running
  -h, --help                           Print help
  -V, --version                        Print version

Connect to a Database

$ cargo pgrx connect
    Re-using existing database strings
psql (13.5)
Type "help" for help.

strings=# select strings.to_lowercase('PGRX');
 to_lowercase
--------------
 pgrx
(1 row)

strings=#

If you'd simply like to connect to a managed version of Postgres without re-compiling and installing your extension, use cargo pgrx connect <pg13 | pg14 | pg15 | pg16 | pg17>.

This command will use the default database named for your extension, or you can specify another database name as the final argument.

If the specified database doesn't exist, cargo pgrx connect will create it. Similarly, if the specified version of Postgres isn't running, it'll be automatically started.

$ cargo pgrx connect --help
Connect, via psql, to a Postgres instance

Usage: cargo pgrx connect [OPTIONS] [PG_VERSION] [DBNAME]

Arguments:
  [PG_VERSION]  Do you want to run against pg13, pg14, pg15, pg16, pg17, or pg18? [env: PG_VERSION=]
  [DBNAME]      The database to connect to (and create if the first time).  Defaults to a database with the same name as the current extension name [env: DBNAME=]

Options:
  -p, --package <PACKAGE>              Package to determine default `pg_version` with (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
      --pgcli                          Use an existing `pgcli` on the $PATH [env: PGRX_PGCLI=]
  -h, --help                           Print help
  -V, --version                        Print version

Installing Your Extension Locally

$ cargo pgrx install
building extension with features ``
"cargo" "build" "--message-format=json-render-diagnostics"
    Finished dev [unoptimized + debuginfo] target(s) in 0.06s

installing extension
     Copying control file to /usr/share/postgresql/13/extension/strings.control
     Copying shared library to /usr/lib/postgresql/13/lib/strings.so
    Building for SQL generation with features ``
    Finished dev [unoptimized + debuginfo] target(s) in 0.06s
 Discovering SQL entities
  Discovered 6 SQL entities: 0 schemas (0 unique), 6 functions, 0 types, 0 enums, 0 sqls, 0 ords, 0 hashes, 0 aggregates
     Writing SQL entities to /usr/share/postgresql/13/extension/strings--0.1.0.sql
    Finished installing strings

If for some reason cargo pgrx run <PG_VERSION> isn't your style, you can use cargo pgrx install to install your extension to the Postgres installation described by the pg_config tool currently on your $PATH.

You'll need write permissions to pg_config --pkglibdir and pg_config --sharedir. If this is problematic, consider using cargo pgrx install --sudo which compiles the extension as the current user and copies the extension files to their proper location using sudo, prompting you for your password.

By default, cargo pgrx install builds your extension in debug mode. Specifying --release changes that.

$ cargo pgrx install --help
Install the extension from the current crate to the Postgres specified by whatever `pg_config` is currently on your $PATH

Usage: cargo pgrx install [OPTIONS]

Options:
  -p, --package <PACKAGE>              Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
  -r, --release                        Compile for release mode (default is debug)
      --profile <PROFILE>              Specific profile to use (conflicts with `--release`)
      --test                           Build in test mode (for `cargo pgrx test`)
  -c, --pg-config <PG_CONFIG>          The `pg_config` path (default is first in $PATH)
  -s, --sudo                           Use `sudo` to install the extension artifacts
      --all-features                   Activate all available features
      --no-default-features            Do not activate the `default` feature
  -F, --features <FEATURES>            Space-separated list of features to activate
  -h, --help                           Print help
  -V, --version                        Print version

Testing with Unit Tests

$ cargo pgrx test
"cargo" "test" "--features" " pg_test"
    Finished test [unoptimized + debuginfo] target(s) in 0.07s
     Running unittests (target/debug/deps/spi-312296af509607bc)

running 2 tests
building extension with features ` pg_test`
"cargo" "build" "--features" " pg_test" "--message-format=json-render-diagnostics"
    Finished dev [unoptimized + debuginfo] target(s) in 0.06s

installing extension
     Copying control file to /home/ana/.pgrx/13.5/pgrx-install/share/postgresql/extension/spi.control
     Copying shared library to /home/ana/.pgrx/13.5/pgrx-install/lib/postgresql/spi.so
    Building for SQL generation with features ` pg_test`
    Finished test [unoptimized + debuginfo] target(s) in 0.07s
 Discovering SQL entities
  Discovered 11 SQL entities: 1 schemas (1 unique), 8 functions, 0 types, 0 enums, 2 sqls, 0 ords, 0 hashes, 0 aggregates
     Writing SQL entities to /home/ana/.pgrx/13.5/pgrx-install/share/postgresql/extension/spi--0.0.0.sql
    Finished installing spi
test tests::pg_test_spi_query_by_id_direct ... ok
test tests::pg_test_spi_query_by_id_via_spi ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.61s

Stopping Postgres

cargo pgrx test ${VERSION} runs your #[test] and #[pg_test] annotated functions using cargo's test system. Note that if you request cargo pgrx test all, this will only run tests for versions configured for control via cargo-pgrx AND set in your extension's Cargo.toml for the pgrx library.

During the testing process, pgrx starts a temporary instance of Postgres with its PGDATA directory in ./target/pgrx-test-data-PGVER/. This Postgres instance is stopped as soon as the test framework has finished. The locale of the temporary instance is C.UTF-8 (or equivalently, a locale of C with a ctype of UTF8 on macOS), or C if the C.UTF-8 locale is unavailable.

The output is standard "cargo test" output along with some Postgres log output. In the case of test failures, the failure report will include any Postgres log messages generated by that particular test.

Rust #[test] functions behave normally, while #[pg_test] functions are run inside the Postgres instance and have full access to all of Postgres internals. All tests are run in parallel, regardless of their type.

Additionally, a #[pg_test] function runs in a transaction that is aborted when the test is finished. As such, any changes it might make to the database are not preserved.

An administrative note is that the --runas and --pgdata options can be used to control the operating-system user used to run the separate postmaster process for test execution. Likely, if --runas is used, then --pgdata will also need to be set to a base directory that is readable and writable by that user -- the default PGDATA directory at ./target/pgrx-test-pgdata will have the permissions of the user running cargo pgrx test and won't be chown-able to the --runas user.

$ cargo pgrx test --help
Run the test suite for this crate

Usage: cargo pgrx test [OPTIONS] [PG_VERSION] [TESTNAME]

Arguments:
  [PG_VERSION]  Do you want to run against pg13, pg14, pg15, pg16, pg17, pg18, or all? [env: PG_VERSION=]
  [TESTNAME]    If specified, only run tests containing this string in their names

Options:
  -p, --package <PACKAGE>              Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
  -r, --release                        compile for release mode (default is debug)
      --profile <PROFILE>              Specific profile to use (conflicts with `--release`)
  -n, --no-schema                      Don't regenerate the schema
      --runas <USER>                   Use `sudo` to initialize and run the Postgres test instance as this system user
      --pgdata <DIR>                   Initialize the test database cluster here, instead of the default location.  If used with `--runas`, then it must be writable by the user
      --all-features                   Activate all available features
      --no-default-features            Do not activate the `default` feature
  -F, --features <FEATURES>            Space-separated list of features to activate
  -h, --help                           Print help
  -V, --version                        Print version

Benchmarking with In-Process Benches

cargo pgrx bench ${VERSION} runs #[pg_bench] benchmarks inside Postgres backends using the same managed Postgres instance that cargo pgrx run uses.

Unlike cargo pgrx test, this is not an ephemeral test harness. Benchmarks are installed into a managed benchmark database, by default named $extname_benches, and their historical results are stored there as well.

The intended authoring model is:

#[cfg(feature = "pg_bench")]
#[pg_schema]
mod benches {
    use pgrx::prelude::*;
    use pgrx_bench::{Bencher, black_box};

    #[pg_bench]
    fn bench_hello(b: &mut Bencher) {
        b.iter(|| {
            black_box(crate::hello());
        });
    }
}

cargo pgrx bench automatically enables the pg_bench feature, builds in --release mode by default, installs the extension artifacts using the normal install pipeline, refreshes the extension in the benchmark database, and then runs each benchmark in-process.

Key behaviors:

  • benchmark functions run inside a backend, not over an external SQL client loop
  • benchmark side effects are rolled back after each benchmark invocation
  • benchmark results are stored in a runner-owned pgrx_bench schema so history survives extension refreshes
  • the CLI records environment metadata, pg_settings, run groups, and comparison targets
  • the CLI prints the benchmark backend PID before benchmark execution starts
  • the CLI prints which benchmark is currently running and the effective settings for that benchmark

You can list benchmarks and their settings with:

$ cargo pgrx bench --list
bench_normalize_phrase [transaction=shared, setup=none, sample_size=100, warm_up=3000ms, measurement=5000ms, nresamples=100000, noise_threshold=0.01, significance_level=0.05]
bench_spi_insert_batch [transaction=subtransaction_per_batch, setup=prepare_spi_fixture, sample_size=50, warm_up=3000ms, measurement=2000ms, nresamples=100000, noise_threshold=0.01, significance_level=0.05]

To run a single benchmark:

$ cargo pgrx bench pg16 bench_normalize_phrase

To print the backend PID and wait before benchmark execution starts, which is handy when attaching samply, perf, lldb, or another external profiler/debugger:

$ cargo pgrx bench --wait 10

--wait is measured in seconds, is honored after the backend PID is printed, and the CLI tells you when it has started waiting.

To name the current run group and compare it against another group:

$ cargo pgrx bench --group-name before-rewrite
$ cargo pgrx bench --group-name after-rewrite --compare-group before-rewrite

To emit the final summary as JSON:

$ cargo pgrx bench --json

To render a read-only history report from the benchmark database, using the last 10 groups per benchmark and leaving the managed Postgres instance running:

$ cargo pgrx bench --report
$ cargo pgrx bench bench_normalize_phrase --report

--report starts the managed Postgres instance if needed, skips build/install/extension refresh, reads the persisted pgrx_bench history, and renders colored ASCII bars. Rows marked with * have broad drift from that benchmark's recorded baseline (for example profile, Postgres version, cargo features, or nondefault pg_settings).

cargo pgrx bench is meant for in-process performance work on extension code. It is not a replacement for cargo pgrx test or cargo pgrx regress, and it is not a client-side SQL load testing tool.

$ cargo pgrx bench --help
Usage: cargo pgrx bench [OPTIONS] [ARGS]...

Arguments:
  [ARGS]...  Positional arguments: [pgXX] [benchname] [env: PG_VERSION=]

Options:
      --dbname <DBNAME>
          If specified, use this database name instead of `$extname_benches`
      --group-name <GROUP_NAME>
          Unique name for this benchmark group
  -v, --verbose...
          Enable info logs, -vv for debug, -vvv for trace
      --compare-group <COMPARE_GROUP>
          Named benchmark group to compare against
      --resetdb
          Recreate the benchmark database before running
      --cascade
          Use CASCADE when dropping the extension during refresh
      --list
          List discovered benchmark wrappers and exit
      --report
          Render a read-only history report from the benchmark database
      --json
          Emit the final summary as JSON
      --wait <SECONDS>
          Sleep for this many seconds after printing the backend PID and before starting benchmarks
          [default: 0]
  -p, --package <PACKAGE>
          Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>
          Path to Cargo.toml
      --debug
          Compile for debug mode instead of the default release mode
      --profile <PROFILE>
          Specific profile to use (conflicts with `--debug`)
      --all-features
          Activate all available features
      --no-default-features
          Do not activate the `default` feature
  -F, --features <FEATURES>
          Space-separated list of features to activate
      --target <TARGET>
      --postgresql-conf <POSTGRESQL_CONF>
          Custom `postgresql.conf` settings in the form of `key=value`
  -h, --help
          Print help
  -V, --version
          Print version

Testing with Regression Tests

pgrx supports a regression test system very similar to the one prescribed by Postgres' pg_regress tool. In fact, pgrx uses pg_regress to run the regression tests.

cargo pgrx regress is used to run the regression tests. It has a number of options similar to cargo pgrx test.

Key flags:

Flag Purpose
--resetdb Drop and recreate the test database (recommended for reproducible runs)
--add <name> Bootstrap a new test — implies --resetdb, runs setup.sql, promotes output to expected/
-t / --test-filter <string> Only run tests whose names contain this string
-a / --auto Overwrite expected output for failed tests with actual output
-v / --verbose Print regression diffs to stderr on failure
--dry-run Print what would happen without doing it
--repeat <N> Run the entire configuration N times (default: 1)
-p / --package <name> Package to build (auto-detected in workspaces with a single pgrx extension)
[PG_VERSION] Postgres version (e.g., pg18). Optional — defaults to Cargo.toml's default feature

Regression tests are split into *.sql files and *.out files. The files themselves are organized into separate directories rooted at ./tests/pg_regress.

For example, using our range example, the directory structure looks like this:

$ tree       
.
├── Cargo.lock
├── Cargo.toml
├── README.md
├── tests
│   └── pg_regress
│       ├── sql                    # these are the individual regression test scripts
│       │   ├── make_range.sql
│       │   ├── setup.sql
│       │   └── store_ranges.sql
│       └── expected               # these are the corresponding test output files
│           ├── make_range.out
│           ├── setup.out
│           └── store_ranges.out
├── range.control
└── src
    ├── bin
    │   └── pgrx_embed.rs
    └── lib.rs

setup.sql is a special test in that it's run first, by itself, whenever the test database is first created, or reset using the --resetdb argument.

When creating a new test, first make the .sql file in ./tests/pg_regress/sql/ and then use --add to bootstrap it:

$ echo "SELECT 1;" > ./tests/pg_regress/sql/example.sql
$ cargo pgrx regress --add example

This will:

  1. Drop and recreate the test database (--add implies --resetdb)
  2. Run setup.sql to establish schema/data
  3. Run the new test
  4. Copy its output to ./tests/pg_regress/expected/example.out
  5. git add the new expected output file

After bootstrapping, verify the test passes by running the full suite:

$ cargo pgrx regress --resetdb

Tests without expected output are skipped by default during normal runs. If you use --test-filter to target a test that hasn't been bootstrapped yet, you'll get a clear error directing you to use --add first.

Updating expected output after code changes

When code changes alter query output or EXPLAIN plans, use --auto to accept the new output:

$ cargo pgrx regress --auto --resetdb

--auto overwrites expected output for failed tests with their actual output. Review the changes with git diff to verify they match the intended behavioral change. Never manually edit .out files — always let --auto generate them.

When tests fail, the path to the regression.diffs file is always shown. Use -v to also print the full diff content inline.

Things to Know

  • setup.sql is only executed when tests are run for the first time, or the --resetdb argument is used. This includes when running a single test with --resetdb — setup.sql will always run first to establish the database schema and data before the filtered test executes.
  • The point of setup.sql is to perform some heavy-weight database object creation/data-loading only when the test regression database is created.
  • tests are executed in alphabetical order
  • .sql files without a corresponding expected/*.out file are skipped during normal runs — use --add <name> to bootstrap new tests
  • pgrx creates a database named $extname_regress unless --dbname is used
  • Postgres' documentation for pg_regress begins here. While pgrx does not support every knob and dial, its organization is largely compatible (PRs welcome to enhance features)
  • to regenerate the expected test output, delete the ./tests/pg_regress/expected/TEST_NAME.out file and re-add it with cargo pgrx regress --add TEST_NAME
  • pg_regress uses psql to run each test and literally diffs the output against the expected output file. pgrx does two things to help eliminate noise in the test output. The first is it sets client_min_messages=warning when starting the Postgres instance and it also passes -v VERBOSITY=terse through to psql.
  • In a workspace with a single pgrx extension crate, --package is auto-detected — no need to specify it manually

Diffing psql Output?

Yes, Postgres' pg_regress tool plays each test_name.sql file through psql, captures the full output, and diffs that output against the test's corresponding test_name.out file. It's a good idea for your tests to avoid variability in their output.

Avoiding variability can mean some simple things like

  • ensuring the results of SELECT statements that return multiple rows are always sorted in a predictable/repeatable manner
  • avoiding outputting "random" values such as the result of the random() function, txid_current(), and others

Be Kind to Yourself

While always good advice, in the context of individual regression tests, this means that a test should not leave anything behind or should be able to tolerate leftover database objects from previous runs.

Each test should be written to "DROP IF EXISTS ... ; CREATE ..." or instead "DROP ..." every database object at the end of the test.

You can ignore this advice if you always run the regression tests with --resetdb as you'll have a clean database each time

An example of an unkind test might be:

$ echo "CREATE TABLE foo();" > ./tests/pg_regress/sql/bad.sql
$ cargo pgrx regress --add bad

Then on a second run (without --resetdb), the CREATE TABLE will fail because the table already exists. You'd see the regression diffs printed inline:

$ diff -U3  ~/_work/pgrx/tests/pgrx-examples/range/tests/pg_regress/expected/bad.out  ~/_work/pgrx/tests/pgrx-examples/range/tests/pg_regress/results/bad.out
---  ~/_work/pgrx/tests/pgrx-examples/range/tests/pg_regress/expected/bad.out	2025-05-07 12:15:10.759010127 -0400
+++  ~/_work/pgrx/tests/pgrx-examples/range/tests/pg_regress/results/bad.out	2025-05-07 12:15:10.775009912 -0400
@@ -1 +1,2 @@
 CREATE TABLE foo();
+ERROR:  relation "foo" already exists

What you wanted in the .sql file is either

DROP TABLE IF EXISTS foo;  
CREATE TABLE foo();

or

CREATE TABLE foo(); 
-- ... 
DROP TABLE foo;

Either be resilient in the face of existing objects or cleanup when the test is finished.

Building an Installation Package

$ cargo pgrx package
building extension with features ``
"cargo" "build" "--release" "--message-format=json-render-diagnostics"
    Finished release [optimized] target(s) in 0.07s

installing extension
     Copying control file to target/release/spi-pg13/usr/share/postgresql/13/extension/spi.control
     Copying shared library to target/release/spi-pg13/usr/lib/postgresql/13/lib/spi.so
    Building for SQL generation with features ``
    Finished release [optimized] target(s) in 0.07s
 Discovering SQL entities
  Discovered 8 SQL entities: 0 schemas (0 unique), 6 functions, 0 types, 0 enums, 2 sqls, 0 ords, 0 hashes, 0 aggregates
     Writing SQL entities to target/release/spi-pg13/usr/share/postgresql/13/extension/spi--0.0.0.sql
    Finished installing spi

cargo pgrx package [--debug] builds your extension, in --release mode, to a directory structure in ./target/[debug | release]/extension_name-PGVER using the Postgres installation path information from the pg_config tool on your $PATH.

The intent is that you'd then change into that directory and build a tarball or a .deb or .rpm package.

The directory structure cargo pgrx package creates starts at the root of the filesystem, as a package-manager installed version of Postgres is likely to split pg_config --pkglibdir and pg_config --sharedir into different base paths.

(In the example screenshot above, cargo pgrx package was used to build a directory structure using my manually installed version of Postgres 12.)

This command could be useful from Dockerfiles, for example, to automate building installation packages for various Linux distributions or MacOS Postgres installations.

$ cargo pgrx package --help
Create an installation package directory

Usage: cargo pgrx package [OPTIONS]

Options:
  -p, --package <PACKAGE>              Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
  -d, --debug                          Compile for debug mode (default is release)
      --profile <PROFILE>              Specific profile to use (conflicts with `--debug`)
      --test                           Build in test mode (for `cargo pgrx test`)
  -c, --pg-config <PG_CONFIG>          The `pg_config` path (default is first in $PATH)
      --out-dir <OUT_DIR>              The directory to output the package (default is `./target/[debug|release]/extname-pgXX/`)
      --all-features                   Activate all available features
      --no-default-features            Do not activate the `default` feature
  -F, --features <FEATURES>            Space-separated list of features to activate
  -h, --help                           Print help
  -V, --version                        Print version

Inspect your Extension Schema

If you just want to look at the full extension schema that pgrx will generate, use cargo pgrx schema.

$ cargo pgrx schema --help
Generate extension schema files

Usage: cargo pgrx schema [OPTIONS] [PG_VERSION]

Arguments:
  [PG_VERSION]  Do you want to run against pg13, pg14, pg15, pg16, pg17, or pg18?

Options:
  -p, --package <PACKAGE>              Package to build (see `cargo help pkgid`)
      --manifest-path <MANIFEST_PATH>  Path to Cargo.toml
  -v, --verbose...                     Enable info logs, -vv for debug, -vvv for trace
      --test                           Build in test mode (for `cargo pgrx test`)
  -r, --release                        Compile for release mode (default is debug)
      --profile <PROFILE>              Specific profile to use (conflicts with `--release`)
  -c, --pg-config <PG_CONFIG>          The `pg_config` path (default is first in $PATH)
      --all-features                   Activate all available features
      --no-default-features            Do not activate the `default` feature
  -F, --features <FEATURES>            Space-separated list of features to activate
  -o, --out <OUT>                      A path to output a produced SQL file (default is `stdout`)
  -d, --dot <DOT>                      A path to output a produced GraphViz DOT file
      --skip-build                     Skip building a fresh extension shared object
  -h, --help                           Print help
  -V, --version                        Print version

Extension Version Upgrade Scripts

When creating a pgrx extension using cargo pgrx new foo, the new extension template directory tree includes a directory named ./sql:

$ tree
.
├── Cargo.toml
├── blah.control
├── sql
└── src
    └── lib.rs

2 directories, 3 files

It is in this directory that you would manually create extension version upgrade scripts. The files you create should be named in the manner prescribed by the Postgres Extension Updates documentation. Generally that format is foo--oldver--newver.sql. For example, foo--1.0.0--1.0.1.sql.

When a user runs ALTER EXTENSION foo UPDATE; in a database with the foo extension, Postgres will build a graph of upgrade scripts to run, starting with the currently installed version and ending with the default_version defined in the extensions .control file. Postgres will then execute the scripts along the shortest path.

It is your responsibility to hand-write these extension upgrade scripts in whatever manner would allow Postgres to update your extension from one version to the next. pgrx has no ability to auto-generate these scripts.

While pgrx does not generate these upgrade scripts, it does know about them and all pgrx commands (cargo pgrx test/run/install/package/bench) that generate extension artifacts will automatically copy these files, and only these files, from the ./sql directory to their final destination as dictated by pg_config.

Information about pgrx-managed development environment

$ cargo pgrx info --help
Provides information about pgrx-managed development environment

Usage: cargo pgrx info [OPTIONS] <COMMAND>

Commands:
  path       Print path to a base version of Postgres build
  pg-config  Print path to pg_config for a base version of Postgres
  version    Print specific version for a base Postgres version
  help       Print this message or the help of the given subcommand(s)

Options:
  -v, --verbose...  Enable info logs, -vv for debug, -vvv for trace
  -h, --help        Print help
  -V, --version     Print version

cargo pgrx info helps retrieving information about pgrx-managed development environment (such as managed Postgres installations)

EXPERIMENTAL: Versioned shared-object support

pgrx experimentally supports the option to produce a versioned shared library. This allows multiple versions of the extension to be installed side-by-side, and can enable the deprecation (and removal) of functions between extension versions. There are some caveats which must be observed when using this functionality. For this reason it is currently experimental.

Activation

Versioned shared-object support is enabled by removing the module_pathname configuration value in the extension's .control file.

Concepts

Postgres has the implicit requirement that C extensions maintain ABI compatibility between versions. The idea behind this feature is to allow interoperability between two versions of an extension when the new version is not ABI compatible with the old version.

The mechanism of operation is to version the name of the shared library file, and to hard-code function definitions to point to the versioned shared library file. Without versioned shared-object support, the SQL definition of a C function would look as follows:

CREATE OR REPLACE FUNCTION "hello_extension"() RETURNS text /* &str */
STRICT
LANGUAGE c /* Rust */
AS 'MODULE_PATHNAME', 'hello_extension_wrapper';

MODULE_PATHNAME is replaced by Postgres with the configured value in the .control file. For pgrx-based extensions, this is usually set to <extension-name>.

When using versioned shared-object support, the same SQL would look as follows:

CREATE OR REPLACE FUNCTION "hello_extension"() RETURNS text /* &str */
STRICT
LANGUAGE c /* Rust */
AS 'extension-0.0.0', 'hello_extension_wrapper';

Note that the versioned shared library is hard-coded in the function definition. This corresponds to the extension-0.0.0.so file which pgrx generates.

It is important to note that the emitted SQL is version-dependent. This means that all previously-defined C functions must be redefined to point to the current versioned-so in the version upgrade script. As an example, when updating the extension version to 0.1.0, the shared object will be named <extension-name>-0.1.0.so, and cargo pgrx schema will produce the following SQL for the above function:

CREATE OR REPLACE FUNCTION "hello_extension"() RETURNS text /* &str */
STRICT
LANGUAGE c /* Rust */
AS 'extension-0.1.0', 'hello_extension_wrapper';

This SQL must be used in the upgrade script from 0.0.0 to 0.1.0 in order to point the hello_extension function to the new shared object. pgrx does not do any magic to determine in which version a function was introduced or modified and only place it in the corresponding versioned so file. By extension, you can always expect that the shared library will contain all functions which are still defined in the extension's source code.

This feature is not designed to assist in the backwards compatibility of data types.

@MODULE_PATHNAME@ Templating

In case you are already providing custom SQL definitions for Rust functions, you can use the @MODULE_PATHNAME@ template in your custom SQL. This value will be replaced with the path to the actual shared object.

The following example illustrates how this works:

#[pg_extern(sql = r#"
    CREATE OR REPLACE FUNCTION tests."overridden_sql_with_fn_name"() RETURNS void
    STRICT
    LANGUAGE c /* Rust */
    AS '@MODULE_PATHNAME@', '@FUNCTION_NAME@';
"#)]
fn overridden_sql_with_fn_name() -> bool {
    true
}

Caveats

There are some scenarios which are entirely incompatible with this feature, because they rely on some global state in Postgres, so loading two versions of the shared library will cause trouble.

These scenarios are:

  • when using shared memory
  • when using query planner hooks