Duplicate division features in July release

[skmoore]

I'm seeing duplicate division features in the July release. There are a few patterns, some of which may be expected.

The value for local_type is different, so perhaps this is expected? In this example the capital_of_divisions column is identical for both features, but the values are too long to include here

id	subtype	local_type	name
085e1b033fffffff0143e1f3681c0468	locality	suburb	Bratislava
085e1b033fffffff018e7dd51bb7f7c2	locality	city	Bratislava

\ Another example where local_type and capital_of_divisions have different values for each duplicate

id	subtype	local_type	name	capital_of_divisions
085cf0a87fffffff01899c602517d124	locality	city	Kingston	[{division_id=085d2436ffffffff018dac84a99372bd, subtype=country}, {division_id=085cf0acbfffffff0132b9b77708bb1e, subtype=region}]
08516260bfffffff010a9b6d0cbde49e	locality	town	Kingston	[{division_id=08516260bfffffff01d5cb63d73e2437, subtype=county}, {division_id=085a391c7fffffff01dcadc0d18a31d7, subtype=country}]
08520ed77fffffff01152a636d6595da	locality	hamlet	Kingston	[{division_id=08520e87bfffffff01f0d8333eaebd81, subtype=country}]

\ Others are basically exact matches of each other

id	subtype	local_type	name	capital_of_divisions
085b2cd0ffffffff01ec1a70e2268d2a	locality	city	Lefkoşa	[{division_id=085b39333fffffff01f145d7f9110d70, subtype=country}]
085b2cd0ffffffff012a65eb36fb45c1	locality	city	Levkosia	[{division_id=085b39333fffffff01f145d7f9110d70, subtype=country}]
085b2cd0ffffffff01861b0866ec3f98	locality	city	Levkosia	[{division_id=085b39333fffffff01f145d7f9110d70, subtype=country}]
085b2cd0ffffffff01aa2be6f778df73	locality	city	Levkosia	[{division_id=085b39333fffffff01f145d7f9110d70, subtype=country}]

[stepps00]

Thanks for the examples @skmoore.

Today, multiple place tags from OpenStreetMap - the local_type values you're seeing - are used to generate locality entities in the divisions theme. So for the Bratislava example, because those places are represented though multiple features in OSM with suburb and city place tags, multiple entities are generated in Overture. This is not ideal and is causing the duplicate and overlap issues you're seeing, so some changes are being planned to the ingestion pipeline as a fix.

The Kingston examples are actually legitimate entities - one in Jamaica, one in Tasmania, and one in Norfolk Island, hence they have unique capital_of_divisions values.

The issue with the Levkosia example is slightly different, as multiple entities were generated even though they all share the same place tags / local_type value. Running this query in duckdb

SELECT
    id,
    sources[1].dataset as dataset,
    sources[1].record_id as concordance_id
FROM
    read_parquet('s3://overturemaps-us-west-2/release/2024-07-22.0/theme=divisions/type=*/*', filename=true, hive_partitioning=1)
WHERE
    id in ('085b2cd0ffffffff01ec1a70e2268d2a','085b2cd0ffffffff01aa2be6f778df73','085b2cd0ffffffff01861b0866ec3f98','085b2cd0ffffffff012a65eb36fb45c1');

you'll see four unique OSM features

┌──────────────────────────────────┬───────────────┬────────────────┐
│                id                │    dataset    │ concordance_id │
│             varchar              │    varchar    │    varchar     │
├──────────────────────────────────┼───────────────┼────────────────┤
│ 085b2cd0ffffffff01ec1a70e2268d2a │ OpenStreetMap │ R16283715      │
│ 085b2cd0ffffffff01861b0866ec3f98 │ OpenStreetMap │ R2628520       │
│ 085b2cd0ffffffff01aa2be6f778df73 │ OpenStreetMap │ N1893015330    │
│ 085b2cd0ffffffff012a65eb36fb45c1 │ OpenStreetMap │ R2628521       │
└──────────────────────────────────┴───────────────┴────────────────┘

Ideally, a single entity would be maintained on Overture's end for this locality.

Both of these issues are related and similar to a discussion around localities here. There is no timeline for a fix yet, but once some action is taken, we can share a progress update. We're hoping to make some pipeline updates soon, so this should be corrected in one of the upcoming releases.

Feel free to add additional examples, they're very helpful.

[skmoore]

@stepps00 Thanks for the info