A PostGIS-only tile server in Go. Strip away all the other requirements, it just has to take in HTTP tile requests and form and execute SQL. In a sincere act of flattery, the API looks a lot like that of the Martin tile server.
Builds of the latest code:
export DATABASE_URL=postgresql://username:password@host/dbname
./pg_tileserv
SET DATABASE_URL=postgresql://username:password@host/dbname
pg_tileserv.exe
To get more information about what is going on behind the scenes, run with the --debug
commandline parameter on, or turn on debugging in the configuration file:
./pg_tileserv --debug
The configuration file will be automatically read from the following locations, if it exists:
- In the system configuration directory, at
/etc/pg_tileserv.toml
- Relative to the directory from which the program is run,
./config/pg_tileserv.toml
- In a root volume at
/config/pg_tileserv.toml
If you want to pass a path directly to the configuration file, use the --config
commandline parameter to pass in a pull path to configuration file. When using the --config
option, configuration files in other locations will be ignored.
./pg_tileserv --config /opt/pg_tileserv/pg_tileserv.toml
In general the defaults are fine, and the program autodetects things like the server name.
# Database connection
DbConnection = "user=you host=localhost dbname=yourdb"
# Close pooled connections after this interval
DbPoolMaxConnLifeTime = "1h"
# Hold no more than this number of connections in the database pool
DbPoolMaxConns = 4
# Look to read html templates from this directory
AssetsPath = "/usr/share/pg_tileserv/assets"
# Accept connections on this subnet (default accepts on all)
HttpHost = "0.0.0.0"
# Accept connections on this port
HttpPort = 7800
# Advertise URLs relative to this server name
# default is to looke this up from incoming request headers
# UrlBase = "http://yourserver.com/"
# Resolution to quantize vector tiles to
DefaultResolution = 4096
# Padding to add to vector tiles
DefaultBuffer = 256
# Limit number of features requested (-1 = no limit)
MaxFeaturesPerTile = 10000
# Advertise this minimum zoom level
DefaultMinZoom = 0
# Advertise this maximum zoom level
DefaultMaxZoom = 22
# Allow any page to consume these tiles
CORSOrigins = *
# Output extra logging information?
Debug = false
The purpose of pg_tileserv
is to turn a set of spatial records into tiles, on the fly. The tile server reads two different layers of data:
- Table layers are what they sound like: tables in the database that have a spatial column with a spatial reference system defined on it.
- Function layers hide the source of data from the server, and allow the HTTP client to send in optional parameters to allow more complex SQL functionality. Any function of the form
function(z integer, x integer, y integer, ...)
that returns an MVTbytea
result can serve as a function layer.
After start-up you can connect to the server and explore the published tables and functions in the database via a web interface at:
A list of layers is available in JSON at:
The index JSON just returns the minimum information about each layer.
{
"public.ne_50m_admin_0_countries" : {
"name" : "ne_50m_admin_0_countries",
"schema" : "public",
"type" : "table",
"id" : "public.ne_50m_admin_0_countries",
"description" : "Natural Earth country data",
"detailurl" : "http://localhost:7800/public.ne_50m_admin_0_countries.json"
}
}
The detailurl
provides more detailed metadata for table and function layers.
The description
field is read from the comment
value of the table. To set a comment on a table, use the COMMENT
command.
COMMENT ON TABLE ne_50m_admin_0_countries IS 'This is my comment';
By default, pg_tileserv
will provide access to only those spatial tables:
- that your database connection has
SELECT
privileges for; - that include a geometry column;
- that declare a geometry type; and,
- that declare an SRID (spatial reference ID)
For example:
CREATE TABLE mytable (
geom Geometry(Polygon, 4326),
pid text,
address text
);
GRANT SELECT ON mytable TO myuser;
To restrict access to a certain set of tables, use database security principles:
- Create a role with limited privileges
- Only grant
SELECT
to that role for tables you want to publish - Only grant
EXECUTE
to that role for functions you want to publish - Connect
pg_tileserv
to the database using that role
In the detail JSON, each layer declares information relevant to setting up a map interface for the layer.
{
"id" : "public.ne_50m_admin_0_countries",
"geometrytype" : "MultiPolygon",
"name" : "ne_50m_admin_0_countries",
"description" : "Natural Earth countries",
"schema" : "public",
"bounds" : [
-180,
-89.9989318847656,
180,
83.599609375
],
"center" : [
0,
-3.19966125488281
],
"tileurl" : "http://localhost:7800/public.ne_50m_admin_0_countries/{z}/{x}/{y}.pbf",
"properties" : [
{
"name" : "gid",
"type" : "int4",
"description" : ""
},{
"name" : "featurecla",
"description" : "",
"type" : "varchar"
},{
"description" : "",
"type" : "varchar",
"name" : "name"
},{
"type" : "varchar",
"description" : "",
"name" : "name_long"
}
],
"minzoom" : 0,
"maxzoom" : 22
}
id
,name
andschema
are the fully qualified, table and schema name of the database table.bounds
andcenter
give the extent and middle of the data collection, in geographic coordinates. The order of coordinates in bounds is [minlon, minlat, maxlon, maxlat]. The order of coordinates in center is [lon, lat].tileurl
is the standard substitution pattern URL consumed by map clients like Mapbox GL JS and OpenLayers.properties
is a list of columns in the table, with their data types and descriptions. The columndescription
field can be set using theCOMMENT
SQL command, for example:COMMENT ON COLUMN ne_50m_admin_0_countries.name_long IS 'This is the long name';
Most developers will just use the tileurl
as is, but it possible to add some parameters to the URL to customize behaviour at run time:
limit
controls the number of features to write to a tile, the default is 50000.resolution
controls the resolution of a tile, the default is 4096 units per side for a tile.buffer
controls the size of the extra data buffer for a tile, the default is 256 units.properties
is a comma-separated list of properties to include in the tile. For wide tables with large numbers of columns, this allows a slimmer tile to be composed.
For example:
http://localhost:7800/public.ne_50m_admin_0_countries/{z}/{x}/{y}.pbf?limit=100000&properties=name,long_name
For property names that include commas (why did you do that?) URL encode the comma in the name string before composing the comma-separated string of all names.
By default, pg_tileserv
will provide access to only those functions:
- that have
z integer, x integer, y integer
as the first three parameters; - that return a
bytea
, and - that your database connection has
EXECUTE
privileges for.
In addition, hopefully obviously, for the function to actually be useful it does actually have to return an MVT inside the bytea
return.
Functions can also have additional parameters to control the generation of tiles: in fact, the whole reason for function layers is to allow novel dynamic behaviour.
In the detail JSON, each function declares information relevant to setting up a map interface for the layer. Because functions generate tiles dynamically, the system cannot auto-discover things like extent or center, unfortunately. However, the custom parameters and defaults can be read from the function definition and exposed in the detail JSON.
{
"name" : "parcels_in_radius",
"id" : "public.parcels_in_radius",
"schema" : "public",
"description" : "Given the click point (click_lon, click_lat) and radius, returns all the parcels in the radius, clipped to the radius circle.",
"minzoom" : 0,
"arguments" : [
{
"default" : "-123.13",
"name" : "click_lon",
"type" : "double precision"
},
{
"default" : "49.25",
"name" : "click_lat",
"type" : "double precision"
},
{
"default" : "500.0",
"type" : "double precision",
"name" : "radius"
}
],
"maxzoom" : 22,
"tileurl" : "http://localhost:7800/public.parcels_in_radius/{z}/{x}/{y}.pbf"
}
description
can be set usingCOMMENT ON FUNCTION
SQL command.id
,schema
andname
are the fully qualified name, schema and function name, respectively.minzoom
andmaxzoom
are just the defaults, as set in the configuration file.arguments
is a list of argument names, with the data type and default value.
This simple example returns just a filtered subset of a table (ne_50m_admin_0_countries EPSG:4326). The filter in this case is the first letters of the name. Note that the name_prefix
parameter includes a default value: this is useful for clients (like the preview interface for this server) that read arbitrary function definitions and need a default value to fill into interface fields.
CREATE OR REPLACE
FUNCTION public.countries_name(
z integer, x integer, y integer,
name_prefix text default 'B')
RETURNS bytea
AS $$
WITH
bounds AS (
SELECT ST_TileEnvelope(z, x, y) AS geom
),
mvtgeom AS (
SELECT ST_AsMVTGeom(ST_Transform(t.geom, 3857), bounds.geom) AS geom,
t.name
FROM ne_50m_admin_0_countries t, bounds
WHERE ST_Intersects(t.geom, ST_Transform(bounds.geom, 4326))
AND upper(t.name) LIKE (upper(name_prefix) || '%')
)
SELECT ST_AsMVT(mvtgeom, 'public.countries_name') FROM mvtgeom;
$$
LANGUAGE 'sql'
STABLE
PARALLEL SAFE;
COMMENT ON FUNCTION public.countries_name IS 'Filters the countries table by the initial letters of the name using the "name_prefix" parameter.';
Some notes about this function:
- The
ST_AsMVT()
function uses the function name ("public.countries_name") as the MVT layer name. This is not required, but for clients that self-configure, it allows them to use the function name as the layer source name. - In the filter portion of the query (in the
WHERE
clause) the bounds are transformed to the spatial reference of the table data (4326) so that the spatial index on the table geometry can be used. - In the
ST_AsMVTGeom()
portion of the query, the table geometry is transformed into web mercator (3857) to match the bounds, and the de facto expectation that MVT tiles are delivered in web mercator projection. - The
LIMIT
is hard-coded in this example. If you want a user-defined limit you need to add another parameter to your function definition. - The function "volatility" is declared as
STABLE
because within one transaction context, multiple runs with the same inputs will return the same outputs. It is not marked asIMMUTABLE
because changes in the base table can change the outputs over time, even for the same inputs. - The function is declared as
PARALLEL SAFE
because it doesn't depend on any global state that might get confused by running multiple copies of the function at once. - The
ST_TileEnvelope()
function used here is a utility function available in PostGIS 3.0 and higher. For earlier versions, you will probably want to add a custom function to emulate the behavior.CREATE OR REPLACE FUNCTION ST_TileEnvelope(z integer, x integer, y integer) RETURNS geometry AS $$ DECLARE size float8; zp integer = pow(2, z); gx float8; gy float8; BEGIN IF y >= zp OR y < 0 OR x >= zp OR x < 0 THEN RAISE EXCEPTION 'invalid tile coordinate (%, %, %)', z, x, y; END IF; size := 40075016.6855784 / zp; gx := (size * x) - (40075016.6855784/2); gy := (40075016.6855784/2) - (size * y); RETURN ST_SetSRID(ST_MakeEnvelope(gx, gy, gx + size, gy - size), 3857); END; $$ LANGUAGE 'plpgsql' IMMUTABLE STRICT PARALLEL SAFE;
This example clips a layer of parcels EPSG:26910 using a radius and center point, returning only the parcels in the radius, with the boundary parcels clipped to the center.
CREATE OR REPLACE
FUNCTION public.parcels_in_radius(
z integer, x integer, y integer,
click_lon float8 default -123.13,
click_lat float8 default 49.25,
radius float8 default 500.0)
RETURNS bytea
AS $$
WITH
args AS (
SELECT
ST_TileEnvelope(z, x, y) AS bounds,
ST_Transform(ST_SetSRID(ST_MakePoint(click_lon, click_lat), 4326), 26910) AS click
),
mvtgeom AS (
SELECT
ST_AsMVTGeom(
ST_Transform(
ST_Intersection(
p.geom,
ST_Buffer(args.click, radius)),
3857),
args.bounds) AS geom,
p.site_id
FROM parcels p, args
WHERE ST_Intersects(p.geom, ST_Transform(args.bounds, 26910))
AND ST_DWithin(p.geom, args.click, radius)
LIMIT 10000
)
SELECT ST_AsMVT(mvtgeom, 'public.parcels_in_radius') FROM mvtgeom
$$
LANGUAGE 'sql'
STABLE
PARALLEL SAFE;
COMMENT ON FUNCTION public.parcels_in_radius IS 'Given the click point (click_lon, click_lat) and radius, returns all the parcels in the radius, clipped to the radius circle.';
Notes:
- The parcels are stored in a table with spatial reference system 3005, a planar projection.
- The click parameters are longitude/latitude, so in building a click geometry (
ST_MakePoint()
) to use for querying, we transform the geometry to the table spatial reference. - To get the parcel boundaries clipped to the radius, we build a circle in the native spatial reference (26910) using the
ST_Buffer()
function on the click point, then intersect that circle with the parcels.
So far all our examples have used simple SQL functions, but using the more procedural PL/PgSQL language we can create much more interactive examples.
CREATE OR REPLACE
FUNCTION public.squares(z integer, x integer, y integer, depth integer default 2)
RETURNS bytea
AS $$
DECLARE
result bytea;
sq_width float8;
tile_xmin float8;
tile_ymin float8;
bounds geometry;
BEGIN
-- Find the tile bounds
SELECT ST_TileEnvelope(z, x, y) AS geom INTO bounds;
-- Find the bottom corner of the bounds
tile_xmin := ST_XMin(bounds);
tile_ymin := ST_YMin(bounds);
-- We want tile divided up into depth*depth squares per tile,
-- so what is the width of a square?
sq_width := (ST_XMax(bounds) - ST_XMin(bounds)) / depth;
WITH mvtgeom AS (
SELECT
-- Fill in the tile with all the squares
ST_AsMVTGeom(ST_MakeEnvelope(
tile_xmin + sq_width * (a-1),
tile_ymin + sq_width * (b-1),
tile_xmin + sq_width * a,
tile_ymin + sq_width * b), bounds),
-- Each square gets a property that shows
-- what tile it is a part of and what its sub-address
-- in that tile is
Format('(%s.%s,%s.%s)', x, a, y, b) AS tilecoord
-- Drive the square generator with a two-dimensional
-- generate_series setup
FROM generate_series(1, depth) a, generate_series(1, depth) b
)
SELECT ST_AsMVT(mvtgeom.*, 'public.squares')
-- Put the query result into the result variale.
INTO result FROM mvtgeom;
-- Return the answer
RETURN result;
END;
$$
LANGUAGE 'plpgsql'
IMMUTABLE -- Same inputs always give same outputs
STRICT -- Null input gets null output
PARALLEL SAFE;
COMMENT ON FUNCTION public.squares IS 'For each tile requested, generate and return depth*depth polygons covering the tile. The effect is one of always having a grid coverage at the appropriate current scale.';
Hexagonal tilings are popular with data visualization experts because they can be used to summarize point data without adding a visual bias to the output via different summary area sizes. They also have a nice "non-pointy" shape, while still providing a complete tiling of the plane.
When you want to provide a hexagonal summary of a data set at multiple scales, you have an implementation problem: do you need to create a pile of hexagon tables, solely for the purpose of summary visualization?
No, you don't have to, you can generate your hexagons dynamically based on the scale of the requested map tiles.
The first challenge is that a hexagon tile set cannot be perfectly inscribed into a powers-of-two square tile set. That means that any given tile will contain some odd combination of full and partial hexagons. In order for the hexagons that straddle tile boundaries to match up, we need a hexagon tiling that is uniform over the whole plane.
So, our first function takes a "hexagon grid coordinate" and generates a hexagon for that coordinate. The size and location of that hexagon are controlled by the hexagon edge length for this particular tiling.
-- Given coordinates in the hexagon tiling that has this
-- edge size, return the built-out hexagon
CREATE OR REPLACE
FUNCTION hexagon(i integer, j integer, edge float8)
RETURNS geometry
AS $$
DECLARE
h float8 := edge*cos(pi()/6.0);
cx float8 := 1.5*i*edge;
cy float8 := h*(2*j+abs(i%2));
BEGIN
RETURN ST_MakePolygon(ST_MakeLine(ARRAY[
ST_MakePoint(cx - 1.0*edge, cy + 0),
ST_MakePoint(cx - 0.5*edge, cy + -1*h),
ST_MakePoint(cx + 0.5*edge, cy + -1*h),
ST_MakePoint(cx + 1.0*edge, cy + 0),
ST_MakePoint(cx + 0.5*edge, cy + h),
ST_MakePoint(cx - 0.5*edge, cy + h),
ST_MakePoint(cx - 1.0*edge, cy + 0)
]));
END;
$$
LANGUAGE 'plpgsql'
IMMUTABLE
STRICT
PARALLEL SAFE;
SELECT ST_AsText(hexagon(2, 2, 10.0));
POLYGON((20 34.6410161513775,25 25.9807621135332,
35 25.9807621135332,40 34.6410161513775,
35 43.3012701892219,25 43.3012701892219,
20 34.6410161513775))
Now we need a function that, given a square input (a map tile) can figure out all the hexagon coordinates that fall within the tile. Again, the edge size of the hexagon tiling determines the overall geometry of the hex tiling. More than one hexagon will be required, most times, so this is a set-returning function.
-- Given a square bounds, find all the hexagonal cells
-- of a hex tiling (determined by edge size)
-- that might cover that square (slightly over-determined)
CREATE OR REPLACE
FUNCTION hexagoncoordinates(bounds geometry, edge float8,
OUT i integer, OUT j integer)
RETURNS SETOF record
AS $$
DECLARE
h float8 := edge*cos(pi()/6);
mini integer := floor(st_xmin(bounds) / (1.5*edge));
minj integer := floor(st_ymin(bounds) / (2*h));
maxi integer := ceil(st_xmax(bounds) / (1.5*edge));
maxj integer := ceil(st_ymax(bounds) / (2*h));
BEGIN
FOR i, j IN
SELECT a, b
FROM generate_series(mini, maxi) a,
generate_series(minj, maxj) b
LOOP
RETURN NEXT;
END LOOP;
END;
$$
LANGUAGE 'plpgsql'
IMMUTABLE
STRICT
PARALLEL SAFE;
SELECT * FROM hexagoncoordinates(ST_TileEnvelope(15, 1, 1), 1000.0);
i | j
--------+-------
-13358 | 11567
-13358 | 11568
-13357 | 11567
-13357 | 11568
-13356 | 11567
-13356 | 11568
Next, a function that puts the two parts together. With tile coordinates and edge size as input, generate the set of all the hexagons that cover the tile. The output here is basically a spatial table: a set of rows, each row containing a geometry (hexagon) and some properties (hexagon coordinates). Just the input we need for a spatial join.
-- Given an input ZXY tile coordinate, output a set of hexagons
-- (and hexagon coordinates) in web mercator that cover that tile
CREATE OR REPLACE
FUNCTION tilehexagons(z integer, x integer, y integer, step integer,
OUT geom geometry(Polygon, 3857), OUT i integer, OUT j integer)
RETURNS SETOF record
AS $$
DECLARE
bounds geometry;
maxbounds geometry := ST_TileEnvelope(0, 0, 0);
edge float8;
BEGIN
bounds := ST_TileEnvelope(z, x, y);
edge := (ST_XMax(bounds) - ST_XMin(bounds)) / pow(2, step);
FOR geom, i, j IN
SELECT ST_SetSRID(hexagon(h.i, h.j, edge), 3857), h.i, h.j
FROM hexagoncoordinates(bounds, edge) h
LOOP
IF maxbounds ~ geom AND bounds && geom THEN
RETURN NEXT;
END IF;
END LOOP;
END;
$$
LANGUAGE 'plpgsql'
IMMUTABLE
STRICT
PARALLEL SAFE;
The function that the tile server actually calls looks like all other tile server functions: tile coordinates and optional parameter input; bytea
MVT output.
-- Given an input tile, generate the covering hexagons,
-- spatially join to population table, summarize
-- population in each hexagon, and generate MVT
-- output of the result. Step parameter determines
-- how many hexagons to generate per tile.
CREATE OR REPLACE
FUNCTION public.hexpopulationsummary(z integer, x integer, y integer, step integer default 4)
RETURNS bytea
AS $$
WITH
bounds AS (
-- Convert tile coordinates to web mercator tile bounds
SELECT ST_TileEnvelope(z, x, y) AS geom
),
rows AS (
-- Summary of populated places grouped by hex
SELECT Sum(pop_max) AS pop_max, Sum(pop_min) AS pop_min, h.i, h.j, h.geom
-- All the hexes that interact with this tile
FROM TileHexagons(z, x, y, step) h
-- All the populated places
JOIN ne_50m_populated_places n
-- Transform the hex into the SRS (4326 in this case)
-- of the table of interest
ON ST_Intersects(n.geom, ST_Transform(h.geom, 4326))
GROUP BY h.i, h.j, h.geom
),
mvt AS (
-- Usual tile processing, ST_AsMVTGeom simplifies, quantizes,
-- and clips to tile boundary
SELECT ST_AsMVTGeom(rows.geom, bounds.geom) AS geom,
rows.pop_max, rows.pop_min, rows.i, rows.j
FROM rows, bounds
)
-- Generate MVT encoding of final input record
SELECT ST_AsMVT(mvt, 'public.hexpopulationsummary') FROM mvt
$$
LANGUAGE 'sql'
STABLE
STRICT
PARALLEL SAFE;
COMMENT ON FUNCTION public.hexpopulationsummary IS 'Hex summary of the ne_50m_populated_places table. Step parameter determines how approximately many hexes (2^step) to generate per tile.';
A basic "just hexes" layer that skips the spatial join step is even simpler.
-- Given an input tile, generate the covering hexagons Step parameter determines
-- how many hexagons to generate per tile.
CREATE OR REPLACE
FUNCTION public.hexagons(z integer, x integer, y integer, step integer default 4)
RETURNS bytea
AS $$
WITH
bounds AS (
-- Convert tile coordinates to web mercator tile bounds
SELECT ST_TileEnvelope(z, x, y) AS geom
),
rows AS (
-- All the hexes that interact with this tile
SELECT h.i, h.j, h.geom
FROM TileHexagons(z, x, y, step) h
),
mvt AS (
-- Usual tile processing, ST_AsMVTGeom simplifies, quantizes,
-- and clips to tile boundary
SELECT ST_AsMVTGeom(rows.geom, bounds.geom) AS geom,
rows.i, rows.j
FROM rows, bounds
)
-- Generate MVT encoding of final input record
SELECT ST_AsMVT(mvt, 'public.hexagons') FROM mvt
$$
LANGUAGE 'sql'
STABLE
STRICT
PARALLEL SAFE;
COMMENT ON FUNCTION public.hexagons IS 'Hex coverage dynamically generated. Step parameter determines how approximately many hexes (2^step) to generate per tile.';
The basic principle of security is to connect your tile server to the database with a user that has just the access you want it to have, and no more. To support different access patterns, create different users with access to different tables/functions, and run multiple services, connecting with those different users.
CREATE USER tileserver;
Start with a blank user. A blank user will have no select privileges on tables it does not own. It will have execute privileges on functions. However, any the user will have no select privileges on tables accessed by functions, so effectively the user will still have no access to data.
If your tables are in a schema other than public, you will have to also grant "usage" on that schema to your user.
GRANT USAGE ON SCHEMA myschema TO tileserver;
You can then grant access to the user one table at a time.
GRANT SELECT ON TABLE myschema.mytable TO tileserver;
Or grant access to all the tables at once.
GRANT SELECT ON ALL TABLES IN SCHEMA myschema TO tileserver;
As noted above, functions that access table data effectively are restricted by the access levels the user has to the tables the function reads. However, if you want to completely restrict access to the function, including visibility in the user interface, you can strip execution privileges from the function.
-- All functions grant execute to 'public' and all roles are
-- part of the 'public' group, so public has to be removed
-- from the executors of the function
REVOKE EXECUTE ON FUNCTION myschema.myfunction FROM public;
-- Just to be sure, also revoke execute from the user
REVOKE EXECUTE ON FUNCTION myschema.myfunction FROM tileserver;