A table created USING pgcolumnar stores its data by column, with per-column
compression, chunk-group skipping, and a vectorized aggregate path. It is for
analytic workloads: large scans, aggregates, and column projections over
append-mostly data.
pgColumnar builds from one source tree on PostgreSQL 15 through 18, with 19
validated against 19beta2, and is
licensed under the MIT License. It is pre-release; the version marker
is 1.0-alpha5, recorded in VERSION.
The latest published pre-release is v1.0-alpha4. A table USING pgcolumnar is stored in the
native on-disk format, PGCN v1.
Documentation
The full documentation is published at
commandprompt.github.io/pgcolumnar. The
same pages are in this repository under docs/:
Quick start
Build with PGXS against the target server, add the library to
shared_preload_libraries, and restart:
make PG_CONFIG=/path/to/pg_config
make install PG_CONFIG=/path/to/pg_config
shared_preload_libraries = 'pgcolumnar'
Then, in a database:
CREATE EXTENSION pgcolumnar;
CREATE TABLE events (id bigint, ts timestamptz, kind int, payload text)
USING pgcolumnar;
INSERT INTO events
SELECT g, now(), g % 8, 'p' || g
FROM generate_series(1, 1000000) g;
SELECT count(*), avg(kind) FROM events WHERE kind = 3;
See the installation guide for requirements and the
user guide for loading and querying.
Independence
pgColumnar is an independent implementation. It is not derived from the source of
any other columnar project. Its on-disk format, its metadata catalog and its
SQL interface come from published column-store research and from the open Apache
Arrow, Parquet and ORC specifications. They are recorded in
design/NATIVE_FORMAT_AND_INTERFACE_SPEC.md.
The implementation is built from that specification and the public PostgreSQL
API, by the clean-room method described in PROVENANCE.md.
License
MIT. See LICENSE.
