Getting specific POI data - and keeping them up to date.

There is currently no ready-made solution for what you want to do. If you are a programmer then the easiest way to accomplish what you want is to write a small parser for OSM/OSMChange files yourself (you will be able to re-use the code for both) and follow this logic:

• Get an initial planet file from OSM
• Extract the POIs that you want to be available using your own program and write to PostGIS
• Initialize a replication environment using --read-replication-interval-init
• Periodically, retrieve updates and process them.

The logic to apply when processing updates is:

• Change file contains node deletion - then delete the node on your side.
• Change file contains node creation - then check if the tags are of interest and create the node on your side if applicable.
• Change file contains node modification - check if the tags are of interest; if not, then delete the node in your local database if it is there (this means somebody has removed the tags of interest from the node); if the node has interesting tags, then update or create it on your side.

You could of course also amend Osmosis to do what you want.

This procedure has a disadvantage in that it is hard to cope with POIs that are modeled as ways in OSM (because you might theoretically receive an OsmChange file that says "way #1234 now has these tags" and you go "wow, I need that in my database", but you don't know about that way's geometry because you ignored it on import since it didn't have the right tags...)

One way to avoid programming, but at the cost of more processing overhead, is this:

• Get an initial planet file from OSM
• Extract the POIs that you want to be available using the osmosis --tag-filter option; save to new file (e.g. --write-pbf myfile.osm.pbf)
• Import new file into some kind of database using whatever process you fancy (could also be osm2pgsql or imposm, both of which may be faster than osmosis)
• Initialize a replication environment using --read-replication-interval-init
• Periodically call Osmosis to append an update to your local file and immediately tag-filter the result in the same step (i.e. --rri --simc --read-pbf myfile.osm.pbf --ac --tag-filter ... --write-pbf myfile-new.osm.pbf), then again make a full import of the resulting file into a database of your choice. Afterwards, rename myfile-new.osm.pbf to myfile.osm.pbf and use that as the basis onto which you apply the next update.

This wastes resources by always doing a full import, but the import will always be "clean" and only contain the things you really want. Plus, it has the capacity to work with way-POIs as well.

When using a local .osm.pbf file as your master database like in this example, it is advisable to use the --compress=none switch on --write-pbf which will speed up writing.