Matching differs slightly from OSM-hosted instance

[tpalts]

Hello!

For some addresses, the matching process between our local instance and the OSM-hosted instance seems to differ.

Simple example:
https://nominatim.openstreetmap.org/ui/search.html?street=VIA+NAZIONALE+DELLE+PUGLIE+KM+840&city=CASORIA&country=IT

The OSM-hosted instance returns several results, the first of which is osm_id = W908701685.

When I run a search on our instance with the exact same GET parameters, I get no results. However, running the details request locally confirms that this Node DOES exist in our DB: GET /details?osmtype=W&osmid=908701685.

Our local instance starts returning results when we wither remove the house number 840 from the street parameter, or when we remove the token KM.

What could be causing this inconsistency between the two instances?

Thanks for your help :) Awesome project, by the way 👏

Details:

• We are running v4.0 with the Europe data extract from https://download.geofabrik.de/

[mtmail]

In general 'km 840' is not supported and not data mapped in OpenStreetMap so it's always better to remove it from the query.

https://nominatim.openstreetmap.org uses the master branch of https://github.com/osm-search/nominatim and is ahead of the stable version 4.0.1 by two months.

The tokenizer got completely rewritten and osm-search/Nominatim#2588 improves the house-number matching (not clear if that explains the difference for this query though as KM 840 is not a house number).

You can try the new tokenizer with the NOMINATIM_TOKENIZER variable https://nominatim.org/2021/11/03/release-40.html

I should note at the same time https://nominatim.openstreetmap.org is also behind, e.g. reindexing OSM data, so it can happen results on your local fresh installed instance are better in some regions of the world.

[leonardehrenfried]

@mtmail Does nominatim not pick up regular env variables? Just the stuff in .env?

If this is so, then we also would merge in a PR that makes this configurable, @tpalts.

[mtmail]

Nominatim uses dotenv and should pick up regular env variables as well https://nominatim.org/release-docs/latest/customize/Settings/#configuring-nominatim

[tpalts]

I finally managed to set up the instance with NOMINATIM_TOKENIZER=icu. However, the issue still remains - my instance provides no results for the above-mentioned search, while the OSM-hosted instance does.

What other explanations may there be for this discrepency?

[leonardehrenfried]

Are you using the latest master?

[tpalts]

Yes. Although some very minor tweaks were needed to make it run with the new tokenizer. See my fork here.