The Relation Analyzer for OpenStreetMap keeps a copy of all relations to provide a simple search function. You can easily search for the relation name or the type. Also some special attributes like the route or operator are available.

Since the early days of the Relation Analyzer I copied from time to time the planet file and then started a script that would parse the file and extract all the relations from it. This process would take several hours and create a heavy load on my server. I wasn’t happy with this solution and this was the main reason why I never added this script as a cronjob so it would run automatically. Instead I waited until a user would send me an e-mail that he cannot find his relation in my Relation Analyzer. This was usually the time to launch the update script by hand.

During the last few weeks I was moving all of my projects to a new, more powerful server and in the process of doing so I decided to create a sync function that would automatically update the relations database with the OSM server. I always knew that OSM provides minutely, hourly and daily updates to its database and I even looked at the OSM-Change format a few months back. The format is quite simple: for every object (node, way, relation) in the OSM database you get the info if it was created, modified or deleted. Every dump gets a unique nine-digit sequence number. This sequence number is split into three parts where every part becomes a part of the directory structure. The latest sequence number can be obtained from a static URL. The details of this process can be found here.

Now, the idea to get into sync with OSM is simple: get a planet file and start downloading and merging the diff files somewhere before the planet file was created. This makes sure that you get all the updates and that you have a complete database. The only thing you have to make sure is not to add the same object twice. Usually this is accomplished by setting a unique index on the id row of the object.

Since most of my OSM functions are now part of the osm-tools library I also added this sync framework to this library. It can be found in the osm-tools-process module in the package org.osmtools.osmchange. To use the framework you have to do three things. First decide how and where you want to sore your current sequence number. For this I created the SequenceHandler interface with a SimpleFileSequenceHandler default implementation. This implementation stores your current sequence number in a file. By creating your own implementation you can also store it in the database if you want.

The second thing is to create an implementation of the OsmChangeService interface. This interface contains a method that is called with every OsmChange object that is fetched from the OSM server. Since this task is quite similar in all implementations I’ve already created a dummy implementation that you can extend and override the process methods you are interested in. For example if you only want to listen for new relations you would extend the AbstractOsmChangeService and override the createRelation method.

The third thing is to wire everything up. In my case this looks like this:

public static void main(String[] args) {
	SequenceHandler sequenceHandler = new SimpleFileSequenceHandler(new File(args[0]));
	OsmChangeService osmChangeService = new RelationOsmChangeService();
	RestOperations restOperations = new RestTemplate();
	OsmChangeSyncService osmChangeSyncService = new OsmChangeSyncService(Granularity.hour, restOperations, osmChangeService, sequenceHandler);;

There is one more thing. Since the OSM-Change files (osc) are packed with GZip some extra wiring is needed. As you can see I’m using the RestTemplate from the Spring Framework. The RestTemplate can automatically handle lots of content types. Unfortunately it cannot handle gzipped XML files (application/x-gzip). For this I created the ReadOnlyGzipJaxb2HttpMessageConverter. As the name says this converter is only able to read gzipped XML files. It will use the JAXB framework to parse the OSM-Change file and create an object model. To wire it up you have to create the RestTemplate like this:

RestTemplate restTemplate = new RestTemplate();
restTemplate.getMessageConverters().add(new ReadOnlyGzipJaxb2HttpMessageConverter());

With all this knowledge you can now package everything up and plan your cronjob to run at the defined intervals. In my case I decided to update the database once an hour. This process takes about 2-3 sec and is acceptable as server load.

I hope this simple overview will encourage you to take a look at the osm-tools library and help you implement your own sync functions for OSM. You can find the binary release of the osm-tools library in maven central. Please use at least version 1.0.M7.

blog comments powered by Disqus