Using Stetl
This section explains how to use Stetl for your ETL. It assumes Stetl is installed and you are able to run the examples. It may be useful to study some of the examples, especially the core ones found in the examples/basics directory. These examples start numbering from 1, building up more complex ETL cases like (INSPIRE) transformation using Jinja2 Templating.
In addition there are example cases like the Dutch Topo map (Top10NL) ETL in the examples/top10nl directory .
The core concepts of Stetl remain pretty simple: an input resource like a file or a database table is mapped to an output resource (also a file, a database, a remote HTTP server etc) via one or more filters. The input, filters and output are connected in a pipeline called a processing chain or Chain. This is a bit similar to a current in electrical engineering: an input flows through several filters, that each modify the current. In our case the current is (geospatial) data. Stetl design follows the so-called Pipes and Filters Architectural Pattern.
Stetl Config
Stetl components (Inputs, Filters, Outputs) and their interconnection (the Pipeline/Chain)
are specified in a Stetl config file. The file format follows the Python .ini
file-format.
To illustrate, let’s look at the example 2_xslt.
This example takes the input file input/cities.xml
and transforms this file to a valid GML file called
output/gmlcities.gml
. The Stetl config file looks as follows.
[etl]
chains = input_xml_file|transformer_xslt|output_file
[input_xml_file]
class = stetl.inputs.fileinput.XmlFileInput
file_path = input/cities.xml
[transformer_xslt]
class = stetl.filters.xsltfilter.XsltFilter
script = cities2gml.xsl
[output_file]
class = stetl.outputs.fileoutput.FileOutput
file_path = output/gmlcities.gml
Most of the sections in this ini-file specify a Stetl component: an Input, Filter or Output component.
Each component is specified by its (Python) class and per-component specific parameters.
For example [input_xml_file]
uses the class stetl.inputs.fileinput.XmlFileInput
reading and parsing the
file input/cities.xml
specified by the file_path
property. [transformer_xslt]
is a Filter that
applies XSLT with the script file cities2gml.xsl
that is in the same directory. The [output_file]
component specifies the output, in this case a file.
These components are coupled in a Stetl Chain using the special .ini section [etl]
. That section specifies one
or more processing chains. Each Chain is specified by the names of the component sections, their interconnection using
a the Unix pipe symbol “|”.
So the above Chain is input_xml_file|transformer_xslt|output_file
. The names
of the component sections like [input_xml_file]
are arbitrary.
Note: since v1.1.0 a datastream can be split (see below) to multiple Outputs
using ()
like :
[etl]
chains = input_xml_file|transformer_xslt|(output_gml_file)(output_wfs)
Or multiple Input streams can be combined/merged like:
[etl]
chains = (input_http_api_1) (input_http_api_2) | data_transformer | output_db
It is even possible to have both Splitting and Merging together with filtering:
[etl]
chains = (input_http_api_1 | cleaner_filter) (input_http_api_2) | data_transformer | (output_db) (output_file)
Note: since version 2 of stetl it is required that the call to stetl components actually start with stetl. This is not necessary when you write your own components (see example 7)
Configuring Components
Most Stetl Components, i.e. inputs, filters, outputs, have properties that can be configured within their
respective [section] in the config file. But what are the possible properties, values and defaults?
This is documented within each Component class using the @Config
decorator much similar to the standard Python
@property
, only with
some more intelligence for type conversions, defaults, required presence and documentation.
It is loosely based on https://wiki.python.org/moin/PythonDecoratorLibrary#Cached_Properties and Bruce Eckel’s
http://www.artima.com/weblogs/viewpost.jsp?thread=240845 with a fix/hack for Sphinx documentation.
See for example the stetl.inputs.fileinput.FileInput
documentation.
For class authors: this information is added
via the Python Decorators much similar to @property
. The stetl.component.Config
is used to define read-only properties for each Component instance. For example,
class FileInput(Input):
"""
Abstract base class for specific FileInputs, use derived classes.
"""
# Start attribute config meta
# Applying Decorator pattern with the Config class to provide
# read-only config values from the configured properties.
@Config(ptype=str, default=None, required=False)
def file_path(self):
"""
Path to file or files or URLs: can be a dir or files or URLs
or even multiple, comma separated. For URLs only JSON is supported now.
"""
pass
@Config(ptype=str, default='*.[gxGX][mM][lL]', required=False)
def filename_pattern(self):
"""
Filename pattern according to Python ``glob.glob`` for example:
'\\*.[gxGX][mM][lL]'
"""
pass
@Config(ptype=bool, default=False, required=False)
def depth_search(self):
"""
Should we recurse into sub-directories to find files?
"""
pass
# End attribute config meta
def __init__(self, configdict, section, produces):
Input.__init__(self, configdict, section, produces)
# Create the list of files to be used as input
self.file_list = Util.make_file_list(self.file_path, None, self.filename_pattern, self.depth_search)
This defines three configurable properties for the class FileInput.
Each @Config
has three parameters: ptype
, the Python type (str
, list
, dict
, bool
, int
),
default
(default value if not present) and required
(if property in mandatory or optional).
Within the config one can set specific config values like,
[input_xml_file]
class = inputs.fileinput.XmlFileInput
file_path = input/cities.xml
This automagically assigns file_path
to self.file_path
without any custom code and assigns the
default value to filename_pattern
. Automatic checks are performed: if file_path
(required=True
) is present, if its type is string.
In some cases type conversions may be applied e.g. when type is dict
or list
. It is guarded that the value is not
overwritten and the docstrings will appear in the auto-generated documentation, each entry prepended with a CONFIG
tag.
Running Stetl
The above ETL spec can be found in the file etl.cfg
. Now Stetl can be run, simply by typing
stetl -c etl.cfg
Stetl will parse etl.cfg
, create all Components by their class name and link them in a Chain and execute
that Chain. Of course this example is very trivial, as we could just call XSLT without Stetl. But it becomes interesting
with more complex transformations.
Suppose we want to convert the resulting GML to an ESRI Shapefile. As we cannot use GDAL ogr2ogr
on the input
file, we need to combine XSLT and ogr2ogr. See example
3_shape. Now we replace the output
by using outputs.ogroutput.Ogr2OgrOutput, which can execute any ogr2ogr command, converting
whatever it gets as input from the previous Filter in the Chain.
[etl]
chains = input_xml_file|transformer_xslt|output_ogr_shape
[input_xml_file]
class = stetl.inputs.fileinput.XmlFileInput
file_path = input/cities.xml
[transformer_xslt]
class = stetl.filters.xsltfilter.XsltFilter
script = cities2gml.xsl
# The ogr2ogr command-line. May be split over multiple lines for readability.
# Backslashes not required in that case.
[output_ogr_shape]
class = stetl.outputs.ogroutput.Ogr2OgrOutput
temp_file = temp/gmlcities.gml
ogr2ogr_cmd = ogr2ogr
-overwrite
-f "ESRI Shapefile"
-a_srs epsg:4326
output/gmlcities.shp
temp/gmlcities.gml
Using Docker
The most convenient way to run Stetl is via Docker. See the installation instructions at Install with Docker. A full example can be viewed in the Smart Emission project: https://github.com/Geonovum/smartemission/tree/master/etl.
In the simplest case you run a Stetl Docker container from your own built image or the Dockerhub-provided one, geopython/stetl:<version> stetl as follows (latest version):
sudo docker run -v <host dir>:<container dir> -w <work dir> geopython/stetl:latest stetl <any Stetl arguments>
For example within the current directory you may have an etl.cfg
Stetl file:
WORK_DIR=`pwd`
sudo docker run -v ${WORK_DIR}:${WORK_DIR} -w ${WORK_DIR} geopython/stetl:latest stetl -c etl.cfg
# or leaner
sudo docker run --rm -v $(pwd):/work -w /work geopython/stetl:latest stetl -c etl.cfg
A more advanced setup would be (network-)linking to a PostGIS Docker image like kartoza/postgis:
# First run Postgis, remains running,
sudo docker run --name postgis -d -t kartoza/postgis:9.4-2.1
# Then later run Stetl
STETL_ARGS="-c etl.cfg -a local.args"
WORK_DIR="`pwd`"
sudo docker run --name stetl --link postgis:postgis -v ${WORK_DIR}:${WORK_DIR} -w ${WORK_DIR} geopython/stetl:latest stetl ${STETL_ARGS}
The last example is used within the SmartEmission project. Also with more detail and keeping all dynamic data (like PostGIS DB), your Stetl config and results, and logs within the host. For PostGIS see: https://github.com/Geonovum/smartemission/tree/master/services/postgis and Stetl see: https://github.com/Geonovum/smartemission/tree/master/etl.
Even better is to use docker-compose.
Stetl Integration
Note: one can also run Stetl via its main ETL class: stetl.etl.ETL
.
This may be useful for integrations in for example Python programs
or even OGC WPS servers (planned).
Reusable Stetl Configs
What we saw in the last example is that it is hard to reuse this etl.cfg when we have for example a different input file or want to map to different output files. For this Stetl supports config parameter substitution.
Dynamic or secret (e.g. database credentials) parameters in etl.cfg are declared symbolically and substituted at runtime via the commandline or the OS environment.
A variable is declared between curly brackets like {out_xml}. See example 6_cmdargs.
[etl]
chains = input_xml_file|transformer_xslt|output_file
[input_xml_file]
class = stetl.inputs.fileinput.XmlFileInput
file_path = {in_xml}
[transformer_xslt]
class = stetl.filters.xsltfilter.XsltFilter
script = {in_xsl}
[output_file]
class = stetl.outputs.fileoutput.FileOutput
file_path = {out_xml}
Note the symbolic input, xsl and output files. We can now perform the ETL using the stetl -a option in two basic ways. One, passing the arguments on the commandline, like
stetl -c etl.cfg -a "in_xml=input/cities.xml in_xsl=cities2gml.xsl out_xml=output/gmlcities.gml"
Two, passing the arguments in a properties file, here called etl.args (the name of the suffix .args is not significant, could be .env as well).
stetl -c etl.cfg -a etl.args
Where the content of the etl.args properties file is:
# Arguments in properties file
in_xml=input/cities.xml
in_xsl=cities2gml.xsl
out_xml=output/gmlcities.gml
It is also possible to specify multiple -a arguments. This provides for situations where a default.args contains all default arguments and a my.args or explicit -a settings that override the default values in default.args. Overriding is determined by the order of the -a arguments. Examples:
stetl -c etl.cfg -a default.args -a my.args
stetl -c etl.cfg -a default.args -a "db_user=docker db_password=pass"
stetl -c etl.cfg -a default.args -a db_user=docker -a db_password=pass
It is also possible to pass these key/value pairs via OS Environment variables. This is especially handy in Docker-based deployments like Docker Compose and Kubernetes. In this case the variable names need to be prepended with STETL_ or stetl_ as to not mix-up with other non-related OS-env vars. A mixture of commandline args (file) and environment vars is possible. The rule is that OS Environment variables always override/overrule arguments specified with -a option(s).
For example, the above args could also be passed as follows:
export stetl_in_xml="input/cities.xml"
export stetl_in_xsl="cities2gml.xsl"
export stetl_out_xml="output/gmlcities.gml"
stetl -c etl.cfg
or only override the input file name in_xml from etl.args:
export stetl_in_xml="input/cities2.xml"
stetl -c etl.cfg -a etl.args
or even with multiple -a args:
export stetl_in_xml="input/cities2.xml"
stetl -c etl.cfg -a etl.args -a my.args
This makes an ETL chain highly reusable. A very elaborate Stetl config with parameter substitution can be seen in the Top10NL ETL.
Connection Compatibility
During ETL Chain processing Components typically pass data to a next stetl.component.Component
.
A stetl.filter.Filter
Component both consumes and produces data, Inputs produce data and
Outputs only consume data.
Data and status flows as stetl.packet.Packet
objects between the Components. The type of the data in these Packets needs
to be compatible only between two coupled Components.
Stetl does not define one unifying data structure, but leaves this to the Components themselves.
Each Component provides the type of data it consumes (Filters, Outputs) and/or produces (Inputs, Filters). This is indicated in its class definition using the consumes and produces object constructor parameters. Some Components can produce and/or consume multiple data types, like a single stream of records or a record array. In those cases the produces or consumes parameter can be a list (array) of data types.
During Chain construction Stetl will check for compatible formats when connecting Components. If one of the formats is a list of formats, the actual format is determined by:
explicit setting: the actual input_format and/or output_format is set in the Component .ini configuration
no setting provided: the first format in the list is taken as default
Stetl will only check if these input and output-formats for connecting Components are compatible when constructing a Chain.
The following data types are currently symbolically defined in the stetl.packet.FORMAT
class:
any
- ‘catch-all’ type, may be any of the types below.etree_doc
- a complete in-memory XML DOM structure using thelxml
etreeetree_element
- each Packet contains a single DOM Element (usually a Feature) inlxml
etree formatetree_feature_array
- each Packet contains an array of DOM Elements (usually Features) inlxml
etree formatgeojson_feature
- asstruct
but following naming conventions for a single Feature according to the GeoJSON spec: http://geojson.orggeojson_collection
- asstruct
but following naming conventions for a FeatureCollection according to the GeoJSON spec: http://geojson.orggdal_vsi_path
- a single file path in the GDAL Virtual File System (VSI) format (via GDAL Python bindings)ogr_feature
- a single Feature object from an OGR source (via GDAL Python bindings)ogr_feature_array
- a Python list (array) of a single Feature objects from an OGR sourcerecord
- a Pythondict
(hashmap)record_array
- a Python list (array) ofdict
string
- a general stringstruct
- a JSON-like generic tree structurexml_doc_as_string
- a string representation of a complete XML documentxml_line_stream
- each Packet contains a line (string) from an XML file or string representation (DEPRECATED)
Many components, in particular Filters, are able to transform data formats.
For example the XmlElementStreamerFileInput can produce an
etree_element, a subsequent XmlAssembler can create small in-memory etree_doc s that
can be fed into an XsltFilter, which outputs a transformed etree_doc. The type any is a catch-all,
for example used for printing any object to standard output in the stetl.packet.Component
.
An etree_element may also be interesting to be able to process single features.
Starting with Stetl 1.0.7 a new stetl.filters.formatconverter.FormatConverterFilter
class provides a Stetl Filter
to allow almost any conversion between otherwise incompatible Components.
TODO: the Packet typing system is still under constant review and extension. Soon it will be possible to add new data types and converters. We have deliberately chosen not to define a single internal datatype like a “Feature”, both for flexibility and performance reasons.
Multiple Chains
Usually a complete ETL will require multiple steps/commands. For example we need to create a database, maybe tables and/or making tables empty. Also we may need to do postprocessing, like removing duplicates in a table etc. In order to have repeatable/reusable ETL without any manual steps, we can specify multiple Chains within a single Stetl config. The syntax: chains are separated by commas (steps are sill separated by pipe symbols).
Chains are executed in order. We can even reuse the specified components from within the same file. Each will have a separate instance within a Chain.
For example in the Top10NL example we see three Chains:
[etl]
chains = input_sql_pre|schema_name_filter|output_postgres,
input_big_gml_files|xml_assembler|transformer_xslt|output_ogr2ogr,
input_sql_post|schema_name_filter|output_postgres
Here the Chain input_sql_pre|schema_name_filter|output_postgres sets up a PostgreSQL schema and creates tables. input_big_gml_files|xml_assembler|transformer_xslt|output_ogr2ogr does the actual ETL and input_sql_post|schema_name_filter|output_postgres does some PostgreSQL postprocessing.
Chain Splitting
In some cases we may want to split processed data to multiple Filters
or Outputs
.
For example to produce output files in multiple formats like GML, GeoJSON etc
or to publish converted (Filtered) data to multiple remote services (SOS, SensorThings API)
or just for simple debugging to a target Output
and StandardOutput
.
See issue https://github.com/geopython/stetl/issues/35 and the Chain Split example.
Here the Chains are split by using ()
in the ETL Chain definition:
# Transform input xml to valid GML file using an XSLT filter and pass to multiple outputs.
# Below are two Chains: simple Output splitting and splitting to 3 sub-Chains at Filter level.
[etl]
chains = input_xml_file | transformer_xslt |(output_file)(output_std),
input_xml_file | (transformer_xslt|output_file) (output_std) (transformer_xslt|output_std)
[input_xml_file]
class = stetl.inputs.fileinput.XmlFileInput
file_path = input/cities.xml
[transformer_xslt]
class = stetl.filters.xsltfilter.XsltFilter
script = cities2gml.xsl
[output_file]
class = stetl.outputs.fileoutput.FileOutput
file_path = output/gmlcities.gml
[output_std]
class = stetl.outputs.standardoutput.StandardOutput
Chain Merging
In some cases we may want to merge (combine, join) multiple input streams.
For example to harvest data from multiple HTTP REST APIs, or to realize a Filter that integrates data from two data-sources.
See issue https://github.com/geopython/stetl/issues/59 and the Chain Merge example.
Here the Chains are merged by using ()
notation in the ETL Chain definition, possibly even combined with Splitting
Outputs:
# Merge two inputs into single Filter.
[etl]
chains = (input_1) (input_2)|transformer_xslt|output_std,
(input_1) (input_2)|transformer_xslt|(output_file)(output_std)
[input_1]
class = stetl.inputs.fileinput.XmlFileInput
file_path = input1/cities.xml
[input_2]
class = stetl.inputs.fileinput.XmlFileInput
file_path = input2/cities.xml
[transformer_xslt]
class = stetl.filters.xsltfilter.XsltFilter
script = cities2gml.xsl
[output_file]
class = stetl.outputs.fileoutput.FileOutput
file_path = output/gmlcities.gml
[output_std]
class = stetl.outputs.standardoutput.StandardOutput