Repository files navigation

PostgreSQL pooler and proxy (like PgBouncer) with support for sharding, load balancing, failover and mirroring.

PgCat is stable and used in production to serve hundreds of thousands of queries per second.

Instacart	PostgresML	OneSignal

Some features remain experimental and are being actively developed. They are optional and can be enabled through configuration.

SeeDockerfile for example deployment using Docker. The pooler is configured to spawn 4 workers so 4 CPUs are recommended for optimal performance. That setting can be adjusted to spawn as many (or as little) workers as needed.

A Docker image is available fromdocker pull ghcr.io/postgresml/pgcat:latest. See ourGithub packages repository.

For quick local example, use the Docker Compose environment provided:

docker-compose up# In a new terminal:PGPASSWORD=postgres psql -h 127.0.0.1 -p 6432 -U postgres -c'SELECT 1'

SeeConfiguration.

The project is being actively developed and looking for additional contributors and production deployments.

1. Install Rust (latest stable will work great).
2. cargo build --release (to get better benchmarks).
3. Change the config inpgcat.toml to fit your setup (optional given next step).
4. Install Postgres and runpsql -f tests/sharding/query_routing_setup.sql (user/password may be required depending on your setup)
5. RUST_LOG=info cargo run --release You're ready to go!

When making substantial modifications to the protocol implementation, make sure to test them with pgbench:

pgbench -i -h 127.0.0.1 -p 6432 && \pgbench -t 1000 -p 6432 -h 127.0.0.1 --protocol simple && \pgbench -t 1000 -p 6432 -h 127.0.0.1 --protocol extended

Seesharding README for sharding logic testing.

Additionally, all features are tested with Ruby, Python, and Rust unit and integration tests.

Runcargo test to run Rust unit tests.

Run the following commands to run Ruby and Python integration tests:

cd tests/docker/docker compose up --exit-code-from main # This will also produce coverage report under ./cov/

You can open a Docker development environment where you can debug tests easier. Run the following command to spin it up:

./dev/script/console

This will open a terminal in an environment similar to that used in tests. In there, you can compile the pooler, run tests, do some debugging with the test environment, etc. Objects compiled inside the container (and bundled gems) will be placed indev/cache so they don't interfere with what you have on your machine.

In session mode, a client talks to one server for the duration of the connection. Prepared statements,SET, and advisory locks are supported. In terms of supported features, there is very little if any difference between session mode and talking directly to the server.

To use session mode, changepool_mode = "session".

In transaction mode, a client talks to one server for the duration of a single transaction; once it's over, the server is returned to the pool. Prepared statements,SET, and advisory locks are not supported; alternatives are to useSET LOCAL andpg_advisory_xact_lock which are scoped to the transaction.

This mode is enabled by default.

All queries are load balanced against the configured servers using either the random or least open connections algorithms. The most straightforward configuration example would be to put this pooler in front of several replicas and let it load balance all queries.

If the configuration includes a primary and replicas, the queries can be separated with the built-in query parser. The query parser, implemented with thesqlparser crate, will interpret the query and route allSELECT queries to a replica, while all other queries including explicit transactions will be routed to the primary.

The query parser will do its best to determine where the query should go, but sometimes that's not possible. In that case, the client can select which server it wants using this custom SQL syntax:

-- To talk to the primary for the duration of the next transaction:SET SERVER ROLE TO'primary';-- To talk to the replica for the duration of the next transaction:SET SERVER ROLE TO'replica';-- Let the query parser decideSET SERVER ROLE TO'auto';-- Pick any server at randomSET SERVER ROLE TO'any';-- Reset to default configured settingsSET SERVER ROLE TO'default';

The setting will persist until it's changed again or the client disconnects.

By default, all queries are routed to the first available server;default_role setting controls this behavior.

All servers are checked with a; (very fast) query before being given to a client. Additionally, the server health is monitored with every client query that it processes. If the server is not reachable, it will be banned and cannot serve any more transactions for the duration of the ban. The queries are routed to the remaining servers. If all servers become banned, the ban list is cleared: this is a safety precaution against false positives. The primary can never be banned.

The ban time can be changed withban_time. The default is 60 seconds.

We use thePARTITION BY HASH hashing function, the same as used by Postgres for declarative partitioning. This allows to shard the database using Postgres partitions and place the partitions on different servers (shards). Both read and write queries can be routed to the shards using this pooler.

To route queries to a particular shard, we use this custom SQL syntax:

-- To talk to a shard explicitlySET SHARD TO'1';-- To let the pooler choose based on a valueSET SHARDING KEY TO'1234';

The active shard will last until it's changed again or the client disconnects. By default, the queries are routed to shard 0.

For hash function implementation, seesrc/sharding.rs andtests/sharding/partition_hash_test_setup.sql.

classUser <ActiveRecord::Baseend# Metadata will be fetched from shard 0ActiveRecord::Base.establish_connection# Grab a bunch of users from shard 1User.connection.execute"SET SHARD TO '1'"User.take(10)# Using id as the sharding keyUser.connection.execute"SET SHARDING KEY TO '1234'"User.find_by_id(1234)# Using geographical shardingUser.connection.execute"SET SERVER ROLE TO 'primary'"User.connection.execute"SET SHARDING KEY TO '85'"User.create(name:"test user",email:"test@example.com",zone_id:85)# Let the query parser figure out where the query should go.# We are still on shard = hash(85) % shards.User.connection.execute"SET SERVER ROLE TO 'auto'"User.find_by_email("test@example.com")

-- Grab a bunch of users from shard 1SET SHARD TO'1';SELECT*FROM usersLIMIT10;-- Find by idSET SHARDING KEY TO'1234';SELECT*FROM USERSWHERE id=1234;-- Writing in a primary/replicas configuration.SET SHARDING ROLE TO'primary';SET SHARDING KEY TO'85';INSERT INTO users (name, email, zome_id)VALUES ('test user','test@example.com',85);SET SERVER ROLE TO'auto';-- let the query router figure out where the query should goSELECT*FROM usersWHERE email='test@example.com';-- shard setting lasts until set again; we are reading from the primary

Issuing queries to the pooler can cause additional latency. To reduce its impact, it's possible to include sharding information inside SQL comments sent via the query. This is reasonably easy to implement with ORMs likeActiveRecord andSQLAlchemy.

/* shard_id: 5 */ SELECT * FROM foo WHERE id = 1234;/* sharding_key: 1234 */ SELECT * FROM foo WHERE id = 1234;

PgCat can use thesqlparser crate to parse SQL queries and extract the sharding key. This is configurable with theautomatic_sharding_key setting. This feature is still experimental, but it's the ideal implementation for sharding, requiring no client modifications.

The stats are very similar to what PgBouncer reports and the names are kept to be comparable. They are accessible by querying the admin databasepgcat, andpgbouncer for compatibility.

psql -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW DATABASES'

Additionally, Prometheus statistics are available at/metrics via HTTP.

We also have abasic Grafana dashboard based on Prometheus metrics that you can import into Grafana and build on it or use it for monitoring.

The config can be reloaded by sending akill -s SIGHUP to the process or by queryingRELOAD to the admin database. All settings except thehost andport can be reloaded without restarting the pooler, including sharding and replicas configurations.

Mirroring allows to route queries to multiple databases at the same time. This is useful for prewarning replicas before placing them into the active configuration, or for testing different versions of Postgres with live traffic.

PgCat is free and open source, released under the MIT license.

Many thanks to our amazing contributors!