-
Notifications
You must be signed in to change notification settings - Fork 9
Npgsql: Use NpgsqlPoint .NET type for marshalling GEO_POINT types. Explore communicating and marshalling GeoJSON types.
#782
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
This comment was marked as resolved.
This comment was marked as resolved.
| public async Task InsertGeoJsonTyped() | ||
| { | ||
| /*** | ||
| * Verify Npgsql PostGIS/GeoJSON Type Plugin with CrateDB. | ||
| * https://www.npgsql.org/doc/types/geojson.html | ||
| * | ||
| * TODO: Does not work yet, because CrateDB communicates GEO_SHAPE as string? | ||
| * The error message is: | ||
| * | ||
| * System.NotSupportedException : The NpgsqlDbType 'Geometry' isn't present in your | ||
| * database. You may need to install an extension or upgrade to a newer version. | ||
| */ | ||
| Console.WriteLine("Running InsertGeo"); | ||
|
|
||
| // Insert single data point. | ||
| await using (var cmd = new NpgsqlCommand(""" | ||
| INSERT INTO testdrive.example ( | ||
| "geoshape" | ||
| ) VALUES ( | ||
| @geoshape | ||
| ); | ||
| """, conn)) | ||
| { | ||
| var point = new Point(new Position(85.43, 66.23)); | ||
| cmd.Parameters.AddWithValue("geoshape", NpgsqlDbType.Geometry, point); | ||
| cmd.ExecuteNonQuery(); | ||
| } | ||
|
|
||
| // Flush data. | ||
| await RefreshTable(); | ||
| } |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It would be sweet to gain typed GeoJSON support, like the Npgsql PostGIS/GeoJSON Type Plugin might be providing it when talking to PostGIS. I don't know why it isn't working, the error message when running this code is:
System.NotSupportedException : The NpgsqlDbType 'Geometry' isn't present in your
database. You may need to install an extension or upgrade to a newer version.Maybe it does not work, because CrateDB communicates GEO_SHAPE as exclusively as string when using the PostgrSQL wire protocol? Please advise if you see any options for improvements here.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is PostGIS' Geometry equivalent to CrateDB's geo_shape? Would a simple alias in the server do the trick?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
When talking about PostGIS, and looking at compatibility concerns, it is not just about types, but also, and mostly, about operations on them.
PostGIS unlocks GDAL, while CrateDB unlocks JTS. Those are technically different animals, while they are still living in the same habitat. In this spirit, I figure that a simple alias will probably not be applicable, even if it would also be my dearest wish.
This doesn't mean we should not explore this area closer, how we could provide downstream compatibility, or at least a reasonable feature parity, possibly by other means.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would a simple alias in the server do the trick?
Maybe @seut has more insights into that. I will be so happy to also learn more about those details, and if they have been parts of any discussions in the past already.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
From reading the PostGIS docs, the generic geometry type standalone looks like it could be mapped to our GEO_SHAPE type as it serves as a generic type for all concrete spatial types the same like our geo_shape does. But as far as I read, the PostGIS also allows to specify a concrete geometry subset on table definition, e.g. geometry(LINESTRING). This isn't possible at CrateDB, we cannot limit the allowed concrete shape of a geometry value.
So we maybe could alias the PostGIS geometry type without a concrete spatial type definition, but even so we'd need to do some (extensive) testing to ensure that this works as expected (in terms of SQL and PG compatibility), especially, the query/filter behaviour.
I suggest to open a feature request at CrateDB to implement the geometry data type with the alias of geo_shape as a possible solution.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@amotl @seut I have created crate/crate#17187
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you. 👍
| var point = new Point(new Position(85.43, 66.23)); | ||
| var poly = new Polygon([ | ||
| new LineString([ | ||
| new Position(longitude: 5.0, latitude: 5.0), | ||
| new Position(longitude: 5.0, latitude: 10.0), | ||
| new Position(longitude: 10.0, latitude: 10.0), | ||
| new Position(longitude: 10.0, latitude: 5.0), | ||
| new Position(longitude: 5.0, latitude: 5.0), | ||
| ]) | ||
| ]); | ||
| // TODO: Can GEO_SHAPE types be directly marshalled to a .NET GeoJSON type? | ||
| // Currently, `InsertGeoJsonTyped` does not work yet. | ||
| cmd.Parameters.AddWithValue("geoshape", NpgsqlDbType.Json, JsonConvert.SerializeObject(point)); | ||
| cmd.ExecuteNonQuery(); | ||
|
|
||
| cmd.Parameters.Clear(); | ||
|
|
||
| cmd.Parameters.AddWithValue("geoshape", NpgsqlDbType.Json, JsonConvert.SerializeObject(poly)); | ||
| cmd.ExecuteNonQuery(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What works is to communicate GeoJSON data using the NpgsqlDbType.Json type, but it needs manual marshalling like JsonConvert.SerializeObject(point).
Contrary to that, as mentioned above, the Npgsql PostGIS/GeoJSON Type Plugin enables to communicate .NET's GeoJSON types natively.
@simonprickett: Please let me know if you find any way do that already, which I might not have discovered yet. Thanks!
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
| // Query back data. | ||
| await using (var cmd = new NpgsqlCommand("SELECT * FROM testdrive.example", conn)) | ||
| await using (var reader = cmd.ExecuteReader()) | ||
| { | ||
| reader.Read(); | ||
| // TODO: Can GEO_SHAPE types be directly marshalled to a .NET GeoJSON type? | ||
| // Currently, `InsertGeoJsonTyped` does not work yet. | ||
| var obj = reader.GetFieldValue<JsonDocument>("geoshape"); | ||
| var geoJsonObject = JsonConvert.DeserializeObject<Point>(obj.RootElement.ToString()); | ||
| return (Point?) geoJsonObject; | ||
| } |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Dito: Manual procedures are currently needed when working with .NET's native GeoJSON types.
Here, the code uses reader.GetFieldValue<JsonDocument> for retrieval, and JsonConvert.DeserializeObject<Point>(...) for unmarshalling and type casting.
|
Do you have any objections to merge this patch, @simonprickett and @kneth? |
|
|
||
| // Enable PostGIS/GeoJSON Type Plugin. | ||
| // https://www.npgsql.org/doc/types/geojson.html | ||
| // dataSourceBuilder.UseGeoJson(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Isn't this sending GEOJson (so a map encoded in JSON) to CrateDB?
If so, this could work when explicitly casting the insert value to a map/object:
insert into geom (geo) values('{"coordinates": [8.308903076149363, 47.05038385401457], "type": "Point"}'::object);
or
insert into geom (geo) select '{"coordinates": [8.308903076149363, 47.05038385401457], "type": "Point"}'::object;
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for your suggestions, here and below. I will investigate them and report back.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've tried to amend the corresponding INSERT statement, but found that it isn't even executed. The program already fails before, probably based on a type introspection query included in those bunch of statements I've traced using ctk tail.
Npgsql introspection SQL
SELECT version();
SELECT ns.nspname,
t.oid,
t.typname,
t.typtype,
t.typnotnull,
t.elemtypoid
FROM
(-- Arrays have typtype=b - this subquery identifies them by their typreceive and converts their typtype to a
-- We first do this for the type (innerest-most subquery), and then for its element type
-- This also returns the array element, range subtype and domain base type as elemtypoid
SELECT typ.oid,
typ.typnamespace,
typ.typname,
typ.typtype,
typ.typrelid,
typ.typnotnull,
typ.relkind,
elemtyp.oid AS elemtypoid,
elemtyp.typname AS elemtypname,
elemcls.relkind AS elemrelkind,
CASE
WHEN elemproc.proname='array_recv' THEN 'a'
ELSE elemtyp.typtype
END AS elemtyptype ,
typ.typcategory
FROM
(SELECT typ.oid,
typnamespace,
typname,
typrelid,
typnotnull,
relkind,
typelem AS elemoid,
CASE
WHEN proc.proname='array_recv' THEN 'a'
ELSE typ.typtype
END AS typtype,
CASE
WHEN proc.proname='array_recv' THEN typ.typelem
WHEN typ.typtype='r' THEN rngsubtype
WHEN typ.typtype='m' THEN
(SELECT rngtypid
FROM pg_range
WHERE rngmultitypid = typ.oid)
WHEN typ.typtype='d' THEN typ.typbasetype
END AS elemtypoid ,
typ.typcategory
FROM pg_type AS typ
LEFT JOIN pg_class AS cls ON (cls.oid = typ.typrelid)
LEFT JOIN pg_proc AS proc ON proc.oid = typ.typreceive
LEFT JOIN pg_range ON (pg_range.rngtypid = typ.oid)) AS typ
LEFT JOIN pg_type AS elemtyp ON elemtyp.oid = elemtypoid
LEFT JOIN pg_class AS elemcls ON (elemcls.oid = elemtyp.typrelid)
LEFT JOIN pg_proc AS elemproc ON elemproc.oid = elemtyp.typreceive) AS t
JOIN pg_namespace AS ns ON (ns.oid = typnamespace)
WHERE (typtype IN ('b',
'r',
'm',
'e',
'd')
OR -- Base, range, multirange, enum, domain
(typtype = 'c'
AND relkind='c')
OR -- User-defined free-standing composites (not table composites) by default
(typtype = 'p'
AND typname IN ('record',
'void',
'unknown'))
OR -- Some special supported pseudo-types
(typtype = 'a'
AND (-- Array of...
elemtyptype IN ('b',
'r',
'm',
'e',
'd')
OR -- Array of base, range, multirange, enum, domain
(elemtyptype = 'p'
AND elemtypname IN ('record',
'void'))
OR -- Arrays of special supported pseudo-types
(elemtyptype = 'c'
AND elemrelkind='c')-- Array of user-defined free-standing composites (not table composites) by default
)))
ORDER BY CASE
WHEN typtype IN ('b',
'e',
'p') THEN 0 -- First base types, enums, pseudo-types
WHEN typtype = 'c' THEN 1 -- Composites after (fields loaded later in 2nd pass)
WHEN typtype = 'r' THEN 2 -- Ranges after
WHEN typtype = 'm' THEN 3 -- Multiranges after
WHEN typtype = 'd'
AND elemtyptype <> 'a' THEN 4 -- Domains over non-arrays after
WHEN typtype = 'a' THEN 5 -- Arrays after
WHEN typtype = 'd'
AND elemtyptype = 'a' THEN 6 -- Domains over arrays last
END;
-- Load field definitions for (free-standing) composite types
SELECT typ.oid,
att.attname,
att.atttypid
FROM pg_type AS typ
JOIN pg_namespace AS ns ON (ns.oid = typ.typnamespace)
JOIN pg_class AS cls ON (cls.oid = typ.typrelid)
JOIN pg_attribute AS att ON (att.attrelid = typ.typrelid)
WHERE (typ.typtype = 'c'
AND cls.relkind='c')
AND attnum > 0
AND -- Don't load system attributes
NOT attisdropped
ORDER BY typ.oid,
att.attnum;
-- Load enum fields
SELECT pg_type.oid,
enumlabel
FROM pg_enum
JOIN pg_type ON pg_type.oid=enumtypid
ORDER BY oid,
enumsortorder;|
|
||
| // GEO_SHAPE | ||
| // While `GEO_POINT` is transparently marshalled as `NpgsqlPoint`, | ||
| // `GEO_SHAPE` is communicated as scalar `string` type, using WKT or GeoJSON format. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nitpick: CrateDB won't communicate this as string but as a JSON type, see also https://github.com/crate/crate/blob/master/server/src/main/java/io/crate/protocols/postgres/types/PGTypes.java#L66.
Maybe my previously commented workaround by casting the insert to an object will work?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry, it didn't work.
| * TODO: Does not work yet, because CrateDB communicates GEO_SHAPE as string? | ||
| * The error message is: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
See previous comment about a possible workaround.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🤷
| var obj = reader.GetFieldValue<JsonDocument>("geoshape"); | ||
| var geoJsonObject = JsonConvert.DeserializeObject<Point>(obj.RootElement.ToString()); | ||
| return (Point?) geoJsonObject; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reading data is not directly related to how data is ingested. Shouldn't the marshalling work when enabling GeoJSON by dataSourceBuilder.UseGeoJson();? Afaik the output JSON should be a valid GeoJSON format, at least for simple structures like Points.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Unfortunately, the marshalling currently does not work, it looks like CrateDB would need to identify itself as PostGIS, see conversation here.
identify itself as PostGIS
Sorry for the sloppy phrasing, I didn't investigate further, but will be happy to do it through a separate iteration.
|
I removed requested reviewers for now, and you can add new - or same - reviewers when the PR moved out of draft. |
WalkthroughMade BasicPoco.Equals nullable-safe; introduced DemoProgram.GetDataSource to centralize NpgsqlDataSource creation; added GeoJSON insert/read methods and GeoJSON.Net dependency; updated tests to use the shared data source and to validate GeoJSON flows. Changes
Sequence DiagramsequenceDiagram
autonumber
participant Test as DemoProgramTest
participant Demo as DemoProgram
participant Types as DatabaseWorkloadsTypes
participant DB as PostgreSQL
Test->>Demo: GetDataSource(connString)
Demo-->>Demo: Build NpgsqlDataSource (enable dynamic JSON)
Demo->>DB: OpenConnection() via dataSource
Demo-->>Test: Return NpgsqlDataSource
Test->>Types: GeoJsonTypesExample()
Types->>DB: CREATE/SETUP table
Types->>Types: InsertGeoJsonString() / InsertGeoJsonTyped()
Types->>DB: INSERT geoshape (JSON string or geometry param)
DB-->>Types: INSERT OK
Types->>DB: SELECT geoshape
DB-->>Types: Row with geoshape (JsonDocument or geometry)
Types->>Types: Parse/convert to GeoJSON Point
Types-->>Test: Return Point? for assertions
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45-75 minutes
Poem
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Currently, this has to be conducted "manually".
About
Explore CrateDB+Npgsql type support for geometry and geospatial types.
Status
NpgsqlPointworks well forGEO_POINTtypes. Thanks, @simonprickett.Trivia
Please note 762b488 includes some chore adjustments to remedy warnings, which are unrelated to the main topic of the patch.
Footnotes
... because
GEO_SHAPEtypes are communicated as strings? I don't actually know the reason, but would like to investigate why it doesn't work fluently, optimally/optionally using the PostGIS/GeoJSON Type Plugin? ↩