Deploy and Host PostGIS on Railway

PostGIS turns PostgreSQL into a spatial database: geometry and geography columns, GiST indexes, ST_Intersects / ST_DWithin / ST_Distance, raster, topology, and the tiger geocoder. This template runs PostgreSQL 18.4 with PostGIS 3.6.4, and adds the three extensions a real spatial project reaches for next — pgRouting 4.0.1 for shortest path and driving distance, H3 4.2.3 for hexagonal indexing, and pgvector 0.8.6 for embeddings — already installed and already enabled in your database.

About Hosting PostGIS

The database is exposed over a Railway TCP proxy, so any client — psql, QGIS, GDAL, Prisma, PostgREST, a GIS worker in another project — connects with an ordinary postgresql:// URL. The DATABASE_URL and DATABASE_PRIVATE_URL variables are filled in for you, the password is generated per deploy, and everything durable lives on a volume mounted at /var/lib/postgresql, the parent of the data directory, so a redeploy keeps your data.

Three things are handled that a stock postgis/postgis container does not handle on Railway:

The server is sized from the container's own cgroup limits at boot. PostgreSQL ships 128 MB shared_buffers, 4 MB work_mem, 64 MB maintenance_work_mem and a 4 GB effective_cache_size that describes no container in particular. Those are the settings a spatial workload feels first: GiST index builds are maintenance_work_mem-bound and spatial joins spill to disk on 4 MB. On an 8 GB instance this deploy comes up with 1907 MB shared_buffers and 119 MB work_mem instead; measured on the same 5-million-point ST_Intersects join in a 2 GB / 2 vCPU container, 2.20 s becomes 1.23 s. Every value is passed as a start-up argument, so your own ALTER SYSTEM still overrides it.
TLS uses a key pair generated for your deploy, on your volume. The usual shortcut is to point ssl_key_file at Debian's ssl-cert-snakeoil files, which are created when the image is built — so every container started from that public image shares one private key that anybody can pull and read. This deploy generates its own on first boot and keeps it across redeploys.
The listener is pinned to 5432. Railway injects PORT into every service and PostgreSQL reads it, which would move the server off the port the TCP proxy dials.

Why Deploy PostGIS?

Managed spatial databases are priced as a premium tier, and the free tiers cap rows or area. Self-hosting on Railway gives you the full extension set with no row limits, usage billing instead of a fixed plan, and private networking to the rest of your project, so your API talks to the database over the internal network at no egress cost.

Running it here rather than on a VPS also means no apt-get install postgresql-18-postgis, no initdb tuning pass, no certificate to renew, and no server to patch.

Common Use Cases

Store locator / nearest-neighbour search — ST_DWithin with a GiST index over points of interest, with the trigram and fuzzy-string extensions already present for the name search beside it.
Delivery and field-service routing — pgRouting over an OpenStreetMap extract for shortest path, driving distance and the travelling-salesperson solver.
Geospatial analytics and dashboards — H3 hex aggregation over event data, raster overlays, and pg_stat_statements for finding the query that costs the most.
RAG over places — pgvector embeddings and geometry in the same row, so a single query can filter by similarity and by distance.
A GIS backend for QGIS, GeoServer, PostGREST or Prisma — connect over the public TCP proxy with sslmode=require.

Dependencies for PostGIS

The image ghcr.io/bon5co/postgis-railway, built from postgis/postgis:18-3.6 (PostgreSQL 18.4, PostGIS 3.6.4, GEOS 3.13.1, PROJ 9.6.0, GDAL 3.10.3) with pgRouting, H3 and pgvector added.
A Railway volume for the data directory and the TLS material.

Deployment Dependencies

No external service is required. The database runs alone.
POSTGRES_PASSWORD is generated for you; DATABASE_URL and DATABASE_PRIVATE_URL are derived from it, so nothing has to be filled in on the deploy form.
Source: PostGIS · pgRouting · H3 · pgvector

Extensions enabled on first boot

postgis, postgis_topology, postgis_raster, postgis_tiger_geocoder, fuzzystrmatch, pgrouting, h3, h3_postgis, vector, hstore, pg_trgm, btree_gist, uuid-ossp, pg_stat_statements — in your database and in template_postgis, so every database you create from that template starts spatial.