Postgres FDW for Clickhouse

This project aims to provide federated access to Clickhouse cluster from Postgres, with the ability to write Foreign Data Wrapper(FDW) API functions in Go.

Here be dragons!

The thing about this FDW is that, it is written in Go leveraging CGo.

See design.md to make your own mind :)

Status

This project should be considered experimental.

There is no active developement and usage within MessageBird but feel free to fork/extend this project.

We hope that the code can serve as a reference for a working FDW and to some extent a sample usage of Golang/C interop.

The FDW has been tested to work with Postgres 13 and Clickhouse 20.3.19.4 on Ubuntu 20.04.

But should work on any unix/linux OS that has PG 13 library headers and Clickhouse.

The FDW doesn't rely on any specific features of Clickhouse(CH) so it should work with any version of CH available.

Getting Started

Clone this repository.

Spin up a container from the Dockerfile present in the project.

• docker build --pull -f Dockerfile -t ch_fdw:$SOME_VERSION .

• docker run -it --rm $IMAGE_ID

• ./setup_test_env.sh # see the source of script to manually build the extension

Prerequisites for development

Postgres development library headers(PG version 13), go 1.15 and Clickhouse.

See Dockefile for verbose installation instructions.

For local development, install PG on mac with

    brew install postgres

Build and testing env are most easily available on ubuntu (either docker images or native).

To locate the postgres library sources on either mac/linux, you can do:

pg_config --includedir-server
# or rather to see all the options, just do
pg_config

Installing

This FDW uses the Postgres extension build infrastructure

Makefile should take care of builing and installing the FDW.

make clean
# There is an extra step of building `c-archive` from the go code.
# `make go` will build the archive library that should be linked with extension c code.
make go
make
make install # or make installcheck

Sample Usage

After installation of extensions, restart PG server and enter into psql shell.

Make sure that corresponding table also exists in CH (provide CH table name in options) and have matching column datatypes and names.

shell# su postgres -c psql
psql (11.1 (Ubuntu 11.1-1.pgdg18.04+1))
Type "help" for help.

postgres=# create extension ch_fdw ;
postgres=# create server ch_fdw foreign data wrapper ch_fdw options(host '127.0.0.1', port '9000');
postgres=# create foreign table gotest (name text not null, value text not null, num bigint not null, val numeric, val2 numeric, val3 real, val4 numeric, val5 date, val6 timestamptz) server ch_fdw options (table 'gotest');
postgres=# SELECT * FROM gotest;
INFO:  passed columns [name value num val val2 val3 val4 val5 val6]
INFO:  passed table gotest
 name  | value | num | val  | val2 | val3 | val4 |    val5    |             val6             
-------+-------+-----+------+------+------+------+------------+------------------------------
 hello | world | 110 | 1101 |   11 |  1.1 |    1 | 01-08-2019 | Tue Jan 08 01:11:34 2019 PST
(1 row)

postgres=# -- check that target columns are minimized
postgres=# SELECT name, val, val3 FROM gotest;
INFO:  passed columns [name val val3]
INFO:  passed table gotest
 name  | val  | val3 
-------+------+------
 hello | 1101 |  1.1
(1 row)

postgres=# -- check timestamp conversions
postgres=# SELECT val6 AT TIME ZONE 'CET' FROM gotest ;
INFO:  passed columns [val6]
INFO:  passed table gotest
         timezone         
--------------------------
 Tue Jan 08 10:11:34 2019
(1 row)

postgres=# SELECT val6 AT TIME ZONE 'UTC' FROM gotest ;
INFO:  passed columns [val6]
INFO:  passed table gotest
         timezone         
--------------------------
 Tue Jan 08 09:11:34 2019
(1 row)

postgres=# -- convert CH float to numeric in PG , shows that not all columns necessarily need to be mapped, it's only needs to be correct.
postgres=# create foreign table gotest2 (name text not null, val3 numeric) server ch_fdw options (table 'gotest');
SELECT * FROM gotest2;
INFO:  passed columns [name val3]
INFO:  passed table gotest
 name  | val3 
-------+------
 hello |  1.1
(1 row)

Features

See ch_fdw.out for details on features supported.

Supports

• WHERE clauses pushdown
• Aggregate pushdown