Build Postgres compatible access layer for your data service.
This library implements PostgreSQL Wire Protocol, and provide essential APIs to write PostgreSQL compatible servers and clients. If you are interested in related topic, you can check project ideas to build on top of this library.
- Message format
- Frontend-Backend protocol messages
- Streaming replication protocol
- Logical streaming replication protocol message
- Backend TCP/TLS server on Tokio
- Frontend-Backend interaction over TCP
- SSL Request and Response
- PostgreSQL 17 direct SSL negotiation
- Startup
- No authentication
- Clear-text password authentication
- Md5 Password authentication
- SASL SCRAM authentication (optional feature
server-api-scram-ring
orserver-api-scram-aws-lc-rs
)- SCRAM-SHA-256
- SCRAM-SHA-256-PLUS
- Simple Query and Response
- Extended Query and Response
- Parse
- Bind
- Execute
- Describe
- Sync
- Termination
- Cancel
- Error and Notice
- Copy
- Notification
- SSL Request and Response
- Streaming replication over TCP
- Logical streaming replication over TCP
- Data types
- Text format
- Binary format, implemented in
postgres-types
- APIs
- Startup APIs
- AuthSource API, fetching and hashing passwords
- Server parameters API, ready but not very good
- Simple Query API
- Extended Query API
- QueryParser API, for transforming prepared statement
- ResultSet builder/encoder API
- Query Cancellation API
- Error and Notice API
- Copy API
- Copy-in
- Copy-out
- Copy-both
- Streaming replication over TCP
- Logical streaming replication server API
- Startup APIs
Postgres Wire Protocol is a relatively general-purpose Layer-7 protocol. There are 6 parts of the protocol:
- Startup: client-server handshake and authentication.
- Simple Query: The text-based query protocol of postgresql. Query are provided as string, and server is allowed to stream data in response.
- Extended Query: A new sub-protocol for query which has ability to cache the query on server-side and reuse it with new parameters. The response part is identical to Simple Query.
- Copy: the subprotocol to copy data from and to postgresql.
- Replication
- Logical Replication
Also note that Postgres Wire Protocol has no semantics about SQL, so literally you can use any query language, data formats or even natural language to interact with the backend.
The response are always encoded as data row format. And there is a field description as header of the data to describe its name, type and format.
Jelte Fennema-Nio's on talk on PgConf.dev 2024 has a great coverage of how the wire protocol works: https://www.youtube.com/watch?v=nh62VgNj6hY
To use pgwire
in your server application, you will need to implement two key
components: startup processor and query processor. For query
processing, there are two kinds of queries: simple and extended. By adding
SimpleQueryHandler
to your application, you will get psql
command-line tool
compatibility. And for more language drivers and additional prepared statement,
binary encoding support, ExtendedQueryHandler
is required.
Examples are provided to demo the very basic usage of pgwire
on server side:
examples/sqlite.rs
: uses an in-memory sqlite database at its core and serves it with postgresql protocol. This is a full example with both simple and extended query implementation.cargo run --features _sqlite_ --example sqlite
examples/duckdb.rs
: similar to sqlite example but with duckdb backend. Note that not all data types are implemented in this example.cargo run --features _duckdb_ --example duckdb
examples/gluesql.rs
: uses an in-memory gluesql at its core and serves it with postgresql protocol.examples/server.rs
: demos a server that always returns fixed results.examples/secure_server.rs
: demos a server with ssl support and always returns fixed results.examples/scram.rs
: demos how to configure more secure authentication mechanism: SCRAMexamples/datafusion.rs
: Now moved to datafusion-postgres
I think in most case you do not need pgwire to build a postgresql client, existing postgresql client like rust-postgres should fit your scenarios. Please rise an issue if there is a scenario.
- GreptimeDB: Cloud-native time-series database
- risinglight: OLAP database system for educational purpose
- PeerDB Postgres first ETL/ELT, enabling 10x faster data movement in and out of Postgres
- CeresDB CeresDB is a high-performance, distributed, cloud native time-series database from AntGroup.
- dozer a real-time data platform for building, deploying and maintaining data products.
- restate Framework for building resilient workflow
This library is released under MIT/Apache dual license.