TileCloud Chain

The goal of TileCloud Chain is to provide tools around tile generation on a chain like:

Source: WMS, Mapnik.

Optionally using an SQS queue, AWS host, SNS topic.

Destination in WMTS layout, on S3, on Berkley DB (bsddb), on MBTiles, or on local filesystem.

Features:

• Generate tiles.
• Drop empty tiles.
• Drop tiles outside a geometry or a bbox.
• Use MetaTiles.
• Generate the legend images.
• Generate GetCapabilities.
• Generate OpenLayers example page.
• Generate the Apache configuration.
• Obtain the hash of an empty tile.
• In the future, measure tile generation speed.
• Calculate cost and generation time.
• In the future, manage the AWS hosts that generate tiles.
• Delete empty tiles.
• Copy files between caches.
• Be able to use an SQS queue to dispatch the generation.
• Post processing the generated tiles.
• ...

Get it

With Docker

# Login to docker hub
docker login
docker pull camptocamp/tilecloud-chain

# Initialize the project
docker run -ti \
     --volume .:/project \
     camptocamp/tilecloud-chain \
     pcreate -s tilecloud_chain .

# Run the commands
DOCKER_ADRS=`ifconfig docker0 | head -n 2 | tail -n 1 | awk -F : '{print $2}' | awk '{print $1}'`
docker run -ti \
     --volume .:/project \
     --add-host=db:${DOCKER_ADRS} \
     --add-host=mapserver:${DOCKER_ADRS} \
     --env=USER_NAME=`whoami` \
     --env=USER_ID=`id -u` \
     --env=GROUP_ID=`id -g` \
     --env=UMASK=`umask` \
     camptocamp/tilecloud-chain \
     run <a command>

To share the home folder you should add the arguments:

--volume=${HOME}:${HOME} \
--env=HOME=${HOME} \

The image also contains some tools needed to render OSM data like: fonts-dejavu, node-carto.

With pip

Requirements:

pg_config and a build environment.

Install:

pip install tilecloud-chain
pcreate -s tilecloud_chain .

Edit your layers configuration in ./tilegeneration/config.yaml.

Default configuration file.

Support

Only the latest release is supported and version < 1.11 contains security issues.

Configure

Configure grids

The grid describes how the tiles are arranged.

Especially on s3 be careful to choose every of the grid settings before generating the tiles. If you change one of them you must regenerate all the tiles.

The resolutions in [px/m] describes all the resolutions available for this layer. For a raster layer, have a look to the maximum resolution of the source files. It is not needed to generate tiles at smaller resolutions than the sources, it is preferable to use the OpenLayers client zoom. Note that you can add a resolution in the end without regenerating all the tiles.

The bbox should match the resolution of the extent. CAREFUL: you will have big issue if you use this parameter to generate the tile on a restricted area: use the bbox on the layer instead.

The srs specifies the code of the projection.

The unit is the unit used by the projection.

The tile_size is the tile size in [px], defaults to 256.

The matrix_identifier is zoom by default and can also be set to resolution. It specifies how the z index is build to store the tiles, for example, for the resolutions [2, 1, 0.5] the used values are [0, 1, 2] based on the zoom and [2, 1, 0_5] based on the resolution. The second has the advantage of allowing to add a new resolution without regenerating all the tiles, but it does not work with MapCache.

Configure caches

The available tile caches are: s3, bsddb, mbtile and filesystem.

The best solution to store the tiles, s3, mbtiles and bsddb, have the advantage of using only one file per layer - style dimensions. To serve the mbtile and the bsddb see Distribute the tiles.

s3 needs a bucket and a folder (defaults to '').

mbtiles, bsddb and filesystem just need a folder.

On all the caches we can add some information to generate the URL where the tiles are available. This is needed to generate the capabilities. We can specify:

• http_url direct url to the tiles root.
• http_urls (array) urls to the tiles root.
• http_url and hosts (array), where each value of hosts is used to replace %(host)s in http_url.

In all case http_url or http_urls can include all attributes of this cache as %(attribute)s.

MBTiles vs Berkley DB (bsddb)

• Read performance: similar, eventually the MBTiles is 10% faster.
• Write performance: The Berkley DB is largely faster, about 10 times.
• List the tiles: the MBTiles is largely faster, but we usually don't need it.

Configure layers

First, all the attributes in layer_default are copied in all the layers to define the default values.

We have two type of layer: wms or mapnik.

To start the common attributes are:

min_resolution_seed the minimum resolution that is seeded, other resolutions are served by MapCache.

bbox used to limit the tiles generation.

px_buffer a buffer in px around the object area (geoms or extent).

WMTS layout

To generate the file paths and the WMTS capabilities we need additional information:

The mime_type of the tiles, it's also used by the WMS GetMap and to upload the tiles.

The wmts_style defaults to 'default'.

The extension is used to end the filename.

The dimensions (defaults to []) is an array of objects that have a name, a default value specified in the capabilities, a value to generate the tiles (it can be overwritten by an argument), and an array of values that contains all the possible values available in the capabilities.

For example if you generate the tiles and capabilities with the following configuration:

dimensions:
    -   name: DATE
        default: 2012
        value: 2012
        values: [2012]

then with the following configuration:

dimensions:
    -   name: DATE
        default: 2012
        value: 2013
        values: [2012, 2013]

We will have two set of tiles 2012 and 2013, both accessible by the capabilities, and by default we will see the first set of tiles.

Metatiles

The metatiles are activated by setting meta to on (by default it's off).

The metatiles are used for two things: first to generate multiple tiles with only one WMS query. By setting meta_size to 8 we will generate a square of 8 by 8 tiles in one shot.

The second usage of metatiles is prevent cut label names: this is solved by getting a bigger image and cutting the borders. The meta_buffer should be set to a bigger value than half the size of the longest label.

Configure hash

We can filter tiles and metatiles by using an hash.

The configuration of this hash is in the layer like this:

empty_metatile_detection:
    size: 740
    hash: 3237839c217b51b8a9644d596982f342f8041546
empty_tile_detection:
    size: 921
    hash: 1e3da153be87a493c4c71198366485f290cad43c

To easily generate this configuration we can use the following command:

generate_tiles --get-hash <z/x/y> -l <layer_name>

Where <z/x/y> should refer to an empty tile/metatile. Generally it's a good idea to use z as the maximum zoom, x and y as 0.

Configure geom/sql

We can generate the tiles only on some geometries stored in PostGis.

The configuration is in the layer like this:

geoms:
-   connection: user=www-data password=www-data dbname=<db> host=localhost
    sql: <column> AS geom FROM <table>
    min_resolution: <resolution> # included, optional, last win
    max_resolution: <resolution> # included, optional, last win

Example:

geoms:
-   connection: user=postgres password=postgres dbname=tests host=localhost
    sql: the_geom AS geom FROM tests.polygon
-   connection: user=postgres password=postgres dbname=tests host=localhost
    sql: the_geom AS geom FROM tests.point
    min_resolution: 10
    max_resolution: 20

It's preferable to use simple geometries, too complex geometries can slow down the generation.

Legends

To be able to generate legends with generate_controller --generate-legend-images you should have legend_mime and legend_extension in the layer configuration.

for example:

legend_mime: image/png
legend_extension: png

Then it will create a legend image per layer and per zoom level named .../1.0.0/{{layer}}/{{wmts_style}}/legend{{zoom}}.{{legend_extension}} only if she is different from the previous zoom level. If we have only one legend image it still stores in the file named legend0.{{legend_extension}}.

When we do generate_controller --generate-wmts-capabilities we will at first parse the legend images to generate a layer configuration like this:

legends:
-   mime_type: image/png
    href: http://host/tiles/layer/style/legend0.png
    min_resolution: 500 # optional, [m/px]
    max_resolution: 2000 # optional, [m/px]
    min_scale: # if define overwrite the min_resolution [m/m]
    max_scale: # if define overwrite the max_resolution [m/m]

If you define a legends array in the layer configuration it is directly used to generate the capabilities.

WMS layers

The additional value needed by the WMS is the URL of the server and the layers.

The previously defined mime_type is also used in the WMS requests.

To customize the request you also have the attributes params, headers and generate_salt. In params you can specify additional parameter of the WMS request, in headers you can modify the request headers. In version, you can change the WMS version. See the Proxy/cache issue for additional information.

Mapnik layers

We need to specify the mapfile path.

With Mapnik we have the possibility to specify a data_buffer then we should set the unneeded meta_buffer to 0.

And the output_format used for the Mapnik renderer, can be png, png256, jpeg, grid (grid_renderer).

Mapnik grid layers

With Mapnik we can generate UTFGrid tiles (JSON format that describes the tiles present on a corresponding tile) by using the output_format 'grid', see also: https://github.com/mapnik/mapnik/wiki/MapnikRenderers#grid_renderer.

Specific configuration:

We have a specific way to drop_empty_utfgrid by using the on value.

We should specify the pseudo pixel size [px] with the resolution.

And the layers_fields that we want to get the attributes. Object with the layer name as key and the values in an array as value.

In fact the Mapnik documentation says that's working only for one layer.

And don't forget to change the extension to json, and the mime_type to application/utfgrid and the meta to off (not supported).

Configuration example:

grid:
    type: mapnik
    mapfile: style.mapnik
    output_format: grid
    extension: json
    mime_type: application/utfgrid
    drop_empty_utfgrid: on
    resolution: 4
    meta: off
    data_buffer: 128
    layers_fields:
        buildings: [name, street]

Process

We can configure some tile commands to process the tiles. They can be automatically be called in the tile generation it we set the property post_process or pre_hash_post_process in the layer configuration.

The process is a set of names processes, and each one has a list of commands declared like this:

process:  # root process config
    optipng:  # the process command
    -   cmd: optipng %(args)s -q -zc9 -zm8 -zs3 -f5 -o %(out)s %(in)s  # the command line
        need_out: true  # if false the command rewrite the input file, default to false
        arg:  # argument used with the deferent log switches, all default to ''
            default: '-q' # the argument used by default
            quiet: '-q' # the argument used in quiet mode
            verbose: '-v' # the argument used in verbose mode
            debug: '-log /tmp/optipng.log' # the argument user in debug mode

The cmd can have the following optional argument:

• args the argument configured in the arg section.
• in, out the input and output files.
• x, y, z the tile coordinates.

Logging

Tile logs can be saved to a PostgreSQL database with this configuration:

..code:: yaml

logging:

:

database:

:   dbname: my\_db host: db port: 5432 table: tilecloud\_logs

PostgreSQL authentication can be specified with the PGUSER and PGPASSWORD environment variables. If the database is not reachable, the process will wait until it is.

Configure Apache

To generate the Apache configuration we use the command:

generate_controller --generate-apache-config

The Apache configuration look like this (default values):

apache:
    # Generated file
    config_file: apache/tiles.conf
    # Serve tiles location, default is /tiles
    location: /${instanceid}/tiles
    # Expires header in hours
    expires: 8

    # Headers added to the responses
    headers:
        Cache-Control: max-age=864000, public

If we use a proxy to access to the tiles we can specify a different URL to access to the tiles by adding the parameter tiles_url in the cache.

Configure MapCache

For the last zoom levels we can use MapCache.

To select the levels we switch from pre-generate the tiles to generate tiles on runtime by using MapCache 'min_resolution_seed' in the layer configuration.

The MapCache configuration look like this (default values):

mapcache:
    # The generated file
    config_file: apache/mapcache.xml
    # The memcache host
    memcache_host: localhost
    # The memcache port
    memcache_port: 11211
    # The mapcache location, default is /mapcache
    location: /${instanceid}/mapcache

To generate the MapCache configuration we use the command:

generate_controller --generate-mapcache-config

Tiles error file

If we set a file path in configuration file:

generation:
    error_file: <path>

The tiles that's in error will be append to the file, ant the tiles can be regenerated with generate_tiles --layer <layer> --tiles <path>.

The <path> can be /tmp/error_{layer}_{datetime:%Y-%m-%d_%H:%M:%S} to have one file per layer and per run.

The tiles file looks like:

# [time] some comments
z/x/y # [time] the error
z/x/y:+m/+m # [time] the error

The first line is just a comment, the second, is for an error on a tile, and the third is for an error on a metatile.

Proxy/cache issue

In general we shouldn't generate tiles throw a proxy, to do that you should configure the layers as this:

layers_name:
    url: http://localhost/wms
    headers:
        Host: the_host_name

The idea is to get the WMS server on localhost and use the Host header to select the right Apache VirtualHost.

To don't have cache we use the as default the headers:

headers:
    Cache-Control: no-cache, no-store
    Pragma: no-cache

And if you steal have issue you can add a SALT random argument by setting the layer parameter generate_salt to true.

Alternate mime type

By default TileCloud support only the image/jpeg and image/png mime type.

Amazon services

Authentication

To be authenticated by Amazon you should set those environments variable before running a command:

export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...

Configure S3

The cache configuration is like this:

s3:
    type: s3
    # the s3 bucket name
    bucket: tiles
    # the used folder in the bucket [default to '']
    folder: ''
    # for GetCapabilities
    http_url: https://%(host)s/%(bucket)s/%(folder)s/
    cache_control: 'public, max-age=14400'
    hosts:
    - wmts0.<host>

The bucket should already exists. If you don't use Amazon's S3, you must specify the host and the tiles_url configuration parameter.

Configure SQS

The configuration in layer is like this:

sqs:
    # The region where the SQS queue is
    region: eu-west-1
    # The SQS queue name, it should already exists
    queue: the_name

The queue should be used only by one layer.

To use the SQS queue we should first fill the queue:

generate_tiles --role master --layer <a_layer>

And then generate the tiles present in the SQS queue:

generate_tiles --role slave --layer <a_layer>

For the slave to keep listening when the queue is empty and be able to support more than one layer, you must enable the daemon mode and must not specify the layer:

generate_tiles --role slave --daemon

Configure SNS

SNS can be used to send a message when the generation ends.

The configuration is like this:

sns:
    topic: arn:aws:sns:eu-west-1:your-account-id:tilecloud
    region: eu-west-1

The topic should already exists.

Amazon tool

Amazon has a command line tool (homepage).

To use it, add in the setup.py:

• awscli as an install_requires,
• 'aws = awscli.clidriver:main', in the console_scripts.

Than install it:

pip install awscli

And use it:

aws help

For example to delete many tiles do:

aws s3 rm --recursive s3://your_bucket_name/folder

Other related configuration

Openlayers pink tiles

To avoid the OpenLayers red tiles on missing empty tiles we can add the following CSS rule:

.olImageLoadError {
    display: none;
}

To completely hide the missing tiles, useful for a transparent layer, or for an opaque layer:

.olImageLoadError {
    background-color: white;
}

Distribute the tiles

There two ways to serve the tiles, with Apache configuration, or with an internal server.

The advantage of the internal server are:

• Can distribute MBTiles or Berkley DB.
• Return 204 No Content HTTP code in place of 404 Not Found (or 403 Forbidden for s3).
• Can be used in KVP mode.
• Can have zone per layer where are the tiles, otherwise it redirect on mapcache.

To generate the Apache configuration we use the command:

generate_controller --generate-apache-config

The server can be configure as it:

server:
    layers: a_layer # Restrict to serve an certain number of layers [default to all]
    cache: mbtiles # The used cache [default use generation/default_cache]
    # the URL without location to MapCache, [default to http://localhost/]
    mapcache_base: http://localhost/
    mapcache_headers: # headers, can be used to access to an other Apache vhost [default to {}]
        Host: localhost
    geoms_redirect: true # use the geoms to redirect to MapCache [default to false]
    # allowed extension in the static path (default value), not used for s3.
    static_allow_extension: [jpeg, png, xml, js, html, css]

The minimal configuration is to enable it:

server: {}

You should also configure the http_url of the used cache, to something like https://%(host)s/${instanceid}/tiles or like https://%(host)s/${instanceid}/wsgi/tiles if you use the Pyramid view.

Pyramid view

To use the pyramid view use the following configuration:

config.get_settings().update({
    'tilegeneration_configfile': '<the configuration file>',
})
config.add_route('tiles', '/tiles/\*path')
config.add_view('tilecloud_chain.server:PyramidView', route_name='tiles')

Internal WSGI server

in production.ini:

[app:tiles]
use = egg:tilecloud_chain#server
configfile = %(here)s/tilegeneration/config.yaml

with the Apache configuration:

WSGIDaemonProcess tiles:${instanceid} display-name=%{GROUP} user=${modwsgi_user}
WSGIScriptAlias /${instanceid}/tiles ${directory}/apache/wmts.wsgi
<Location /${instanceid}/tiles>
    WSGIProcessGroup tiles:${instanceid}
    WSGIApplicationGroup %{GLOBAL}
</Location>

Commands

Available commands

• generate_controller generate the annex files like capabilities, legend, OpenLayers test page, MapCache configuration, Apache configuration.
• generate_tiles generate the tiles.
• generate_copy copy the tiles from a cache to an other.
• generate_process process the tiles using a configured process.
• generate_cost estimate the cost.
• import_expiretiles import the osm2pgsql expire-tiles file as geoms in the database.

Each commands have a --help option to give a full arguments help.

Generate tiles

Generate all the tiles:

generate_tiles

Generate a specific layer:

generate_tiles --layer <a_layer>

Generate a specific zoom:

generate_tiles --zoom 5

Generate a specific zoom range:

generate_tiles --zoom 2-8

Generate a specific some zoom levels:

generate_tiles --zoom 2,4,7

Generate tiles from an (error) tiles file:

generate_tiles --layer <a_layer> --tiles <z/x/y>

Generate tiles on a bbox:

generate_tiles --bbox <MINX> <MINY> <MAXX> <MAXY>

Generate a tiles near a tile coordinate (useful for test):

generate_tiles --near <X> <Y>

Generate a tiles in a different cache than the default one:

generate_tiles --cache <a_cache>

And don't forget to generate the WMTS Capabilities:

generate_controller --capabilities

OpenLayers test page

To generate a test page use:

generate_controller --openlayers

Explain cost

Configuration (default values):

cost:
    # [nb/month]
    request_per_layers: 10000000
    cloudfront:
        download: 0.12,
        get: 0.009
    request_per_layers: 10000000
    s3:
        download: 0.12,
        get: 0.01,
        put: 0.01,
        storage: 0.125
    sqs:
        request: 0.01

Layer configuration (default values):

cost:
    metatile_generation_time: 30.0,
    tile_generation_time: 30.0,
    tile_size: 20.0,
    tileonly_generation_time: 60.0

The following commands can be used to know the time and cost to do generation:

generate_controller --cost

Useful options

--quiet or -q: used to display only errors.

--verbose or -v: used to display info messages.

--debug or -d: used to display debug message, pleas use this option to report issue. With the debug mode we don't catch exceptions, and we don't log time messages.

--test <n> or -t <n>: used to generate only <n> tiles, useful for test.

The logging format is configurable in theconfig.yaml - generation/log_format, See.

Important remarks

Especially on S3 the grid name, the layer name, the dimensions, can't be changed (understand if we want to change them we should regenerate all the tiles).

By default we also can't insert a zoom level, if you think that you need it we can set the grid property matrix_identifier: resolution, bit it don't work with MapCache.

Please use the --debug to report issue.

From sources

Build it:

git submodule update --recursive
virtualenv --python=python3 .build/venv
.build/venv/bin/pip install -e .
.build/venv/bin/pip install -r dev-requirements.txt

Run the tests

Setup your environment:

docker build --tag camptocamp/tilecloud-chain:latest .
docker-compose -p tilecloud up

To run the tests:

docker-compose -p tilecloud exec test python setup.py nosetests --logging-filter=tilecloud,tilecloud_chain --attr '!'nopy3

Changelog

Release 1.4

1. Add optional metadata section to the config file. See the scaffolds for example.

Release 0.9

1. Correct some error with slash.
2. Better error handling.
3. Be able to have one error file per layer.

Release 0.8

1. Correct some error with slash.
2. Add pre_hash_post_process and post_process.
3. Add copy command.

Release 0.7

1. Support of deferent geoms per layers, requires configuration changes, old version:

   connection: user=www-data password=www-data dbname=<db> host=localhost
   sql: <column> AS geom FROM <table>
   

   to new version:

   connection: user=www-data password=www-data dbname=<db> host=localhost
   geoms:
   -   sql: <column> AS geom FROM <table>
   

   More informations in the Configure geom/sql chapter.

2. Update from optparse to argparse, and some argument refactoring, use --help to see the new version.

3. Add support of Blackbery DB (bsddb).

4. The tile server is completely rewrite, now it support all cache, REST and KVP interface, GetFeatureInfo request, and it can be used as a pyramid view or as a WSGI server. More informations in the istribute the tiles chapter.

5. Add three strategy to bypass the proxy/cache: Use the headers Cache-Control: no-cache, no-store, Pragma: no-cache (default). Use localhost in the URL and the header Host: <host_name> (recommended). Add a SALT random argument (if the above don't work). More informations in the Proxy/cache issue chapter.

6. Improve the dimensions usage by adding it ti the WMS requests, And add a --dimensions argument of generate_tiles to change the dimensions values.

7. Extract generate_cost and generate_amazon from generate_controler.

8. Now we can creates legends, see the Legends chapter.

9. Now the tiles generation display generation statistics at the ends.

10. The EC2 configuration is moved in a separate structure, see README for more informations.

Release 0.6

1. Now the apache configuration can be generated with .build/venv/bin/generate_controller --generate-apache-config, it support filesystem cache and MapCache.
2. Windows fixes.
3. Use console rewrite (r) to log generated tiles coordinates.
4. Now if no layers is specified in generation:default_layers we generate all layers by default.
5. Now bbox to be floats.
6. New --get-bbox option to get the bbox of a tile.
7. Add coveralls support (https://coveralls.io/r/camptocamp/tilecloud-chain).
8. Add an config option generation:error_file and a command option --tiles to store and regenerate errored tiles.

Release 0.5

1. SQS config change:

layers:
    layer_name:
        sqs:
            # The region where the SQS queue is
            region: eu-west-1
            # The SQS queue name, it should already exists
            queue: the_name

2. Add debug option (--debug), please use it to report issue.
3. Now the sql request can return a set of geometries in a column names geom but the syntax change a little bit => <column> AS geom FROM <table>