Instead of using [Douglas-Peucker algorithm](https://postgis.net/docs/ST_SimplifyPreserveTopology.html) which is using for generalized tables [imposm](https://imposm.org/docs/imposm3/latest/mapping.html#generalized-tables) it is used [Visvalingam-Whyatt algorithm](https://postgis.net/docs/ST_SimplifyVW.html).
Solution:
remove imposm generalized tables (during `import-osm`) and create generalization in `import-sql` step for zooms 7-14. Zooms 0-6 are from Natural Earth data.
Upper zoom levels (7 and 8) are ok to merge with no big impact on creation speed. In Canada from z9 it took too long. Between zoom levels 10 - 13 there can be union polygons with less than 300 edge points (empirical number based on a test on Canada forest, can be discussed). Polygons with more than 300 edge points are just simplified. Zoom 14 is from the original dataset `osm_landcover_polygon`.
There is also removed `osm_id` which is never used and not passed into vector tiles.
https://github.com/openmaptiles/openmaptiles/issues/999
- Point 1 solved only partially - there is `brunnel` value added into zooms 9, 10, 11. For zooms 8 and smaller it would need change in osm_transportation_merge_ tables - I am going to try it but maybe its SQL costs will be to high.
- Point 2 solved.
The ocean layer (table osm_ocean_polygon insert by `make import-data`) is made from squares that have a 10m buffer.
This creates extra geometry in vector tiles as:
Union of full ocean squares should decrease the size of the water layer.
there are 8042 polygons (squares, 5 points, more then 100km<sup>2</sup>), which was reduced to 22 polygons - 1184kB vs 40kB respectively.
In mbtiles should be the size reduction even more significant, due to the creation of more polygons based on a 10m buffer of each square.
The index `osm_highway_linestring_highway_idx` is not used. There is also a used partial index `osm_highway_linestring_highway_partial_idx`.
I cross check the usage of the index from postgres query stats and all SQL query code where the table `osm_highway_linestring` is refered.
This index is relatively big.
Add salt pond as water.
Most of salt pond are already tagged with natural=water even if it's not required. So the load added by this tag is very light.
Current implementation of osm_important_waterway_linestring have two bugs:
* The distinct on is_old keep the oldest version of the old object, and the oldest version of the new object, but need the last version of the new object.
* Delete the old version of the object and rebuild the using the new version of the object. But we need to remove matchings the old and the new version of the object, then rebuild the two.
Then only delete and update using first and last version of the object, intermediate versions are ignored.
Similar implementation of what is done in #996.
Replacing materialized view by a tables with update from trigger on change only.
Differential update of `osm_transportation_name_linestring`, `osm_transportation_name_linestring` and `osm_transportation_name_linestring_genX`).
At the end of the transaction the dependent rows are updated.
The goal is to update more quickly the content of derivated tables by just updating the changing content. It replaces the update of materialized view because their need a full recompute (with lock issue).
It is the last part of the replacement of materialized view for the transportation layer.
It addresses #814 and a part of #809.
When OSM roads in the `transportation_name` layer are stitched together, their grouping does not consider all `name:*` tags.
As a result, roads with different `name:*` tags may be stitched together.
The `waterway` layer performs the grouping properly, for the same purpose:
1685eaccbd/layers/waterway/update_important_waterway.sql (L34)
Co-authored-by: Frédéric Rodrigo <fred.rodrigo@gmail.com>
Reduce the size of the materialized view osm_transportation_name_network by moving the weare clause.
osm_transportation_name_network is only used for osm_transportation_name_linestring.
Co-authored-by: zstadler <zeev.stadler@gmail.com>
Replacing materialized view by a tables with update from trigger on change only.
The osm_id of object changing on table osm_route_member and osm_highway_linestring are logged.
At the end of the transaction the dependent row of osm_transportation_name_network are updated.
The goal is to update more quickly the content of derivated tables by just updating the changing content. It replaces the update of materialized view because their need a full recompute (with lock issue).
Note, it is only a part of the materialized view of the transportation layer, I am also working on the other parts, other parts are in #892 (omaterialized view on sm_transportation_name_linestring and osm_transportation_name_linestring_genX).
It addresses #814 and a part of #809.
Clean and do refactoring on `update_route_member.sql` as first step to replace materialized view osm_transportation_name_network by tables with diff update. See #892.
* Change NE 50m by NE 10m state provinces dataset
* update documentation
Fixes#977
This PR implements the change discussed at #977 to replace Natural Earth 10m states and provinces by the 50m version, since the 10m only covers state borders for US, Canada, Brazil, and Australia.
A few remarks:
* This change only affects zooms 1 to 4, not making any change in borders based in the `OSM Borders` source.
* The `min_zoom<=7` shows the first level regions (level 4 in OSM) as you can check for example for France, Italy, or Spain.
* I don't think it's necessary to simplify the geometries, it would mean adding a new materialized view, or a slower tile generation process.
* This change does not need any change in the styles, even someone could argue it could be interesting to add the country code in this layer as a new field, allowing style editors to filter features by country, but I would prefer to leave that change for another PR.
`disputed_name` is filled for disputed lines of NaturalEarth sources - it contains `ogc_fid` as suggested in https://github.com/openmaptiles/openmaptiles/issues/964. This way it is possible to filter disputed lines in style.
* Remove state labels zoom, rank, and country conditions from layer function
* Don't add labels for zooms 0 and 1
If we check Bright, Positron, or Dark styles, they all show a huge lack of data about states and regions. At zoom 2, for big countries like Canada, Brazil, China or Australia there are big patches of white space (aka cartographer's *horror vacui* 😅)
Replacing update on the whole table with an update only on changed rows.
The goal is to update more quickly by just updating the changing content.
The update now focus on osm_id of changed rows, it use index. Add a where clause tags != update_tags(tags, geometry) to ensure only update when changed.
It requires one more trigger and a table to store changed osm_id.
The UPDATE is keep in a function to be reusable for initial setup and trigger update.
It is a based on the already merged https://github.com/openmaptiles/openmaptiles/pull/944
It is a separated PR as less obvious than previous. It replaces the reset of the `rank` field to NULL by missing value resulting of `LEFT JOIN`. It avoid triggering a new update on the table by reset the value then re-seting it to initial or new value.
It addresses #814.
Thanks @frodrigo
Replacing update on the whole table with an update only on changed rows.
The goal is to update more quickly by just updating the changing content.
The update now focus on osm_id of changed rows, it use index. Add a where clause tags != update_tags(tags, geometry) to ensure only update when changed.
It requires one more trigger and a table to store changed osm_id.
The UPDATE is keep in a function to be reusable for initial setup and trigger update.
It is a based on the already merged https://github.com/openmaptiles/openmaptiles/pull/944
It is a separated PR as less obvious than previous. It replaces the reset of the `rank` field to NULL by missing value resulting of `LEFT JOIN`. It avoid triggering a new update on the table by reset the value then re-seting it to initial or new value.
It addresses #814.
Thanks @frodrigo
Replacing update on the whole table with an update only on changed rows.
The goal is to update more quickly by just updating the changing content.
The update now focus on osm_id of changed rows, it use index. Add a where clause tags != update_tags(tags, geometry) en ensure only update when changed.
It requires one more trigger and a table to store changed osm_id.
The UPDATE is keep in a function to be reusable for initial setup and trigger update.
I try many code layout before done it in this way with the goal to keep the code for initial pass and for update. It should have low impact on initial data load. Better performance for row update can be achieve with BEFORE UPDATE, but require to duplicate the logic.
It is not based on the already merged https://github.com/openmaptiles/openmaptiles/pull/896 because calling and update within a function for each updated row was not efficient for larger table (like housenumber).
It addresses #814.
* Remake update_peak_point use incremental update #814
* Make update_aerodrome_label_point use incremental update #814
* Make housenumber_centroid use incremental update #814
* Make update_continent_point use incremental update #814
* Make update_island_point use incremental update #814
* Make update_island_polygon use incremental update #814
* Remove dead code in update_state_point.sql
* Make update_state_point use incremental update #814
* Remove dead code in update_country_point.sql
* Make update_country_point use incremental update #814
* Make osm_poi_polygon use incremental update #814
Thanks @frodrigo
Improve 97216c5c191bf0df3705134cff234ed980f8ac78 and #853
In case of replay update it may fails because of already existing primary key on osm_id.
Add a on conflict clause to make it fail safe.
* All functions that access database must be declared as `STABLE`, not `IMMUTABLE` -- because database can change at any moment, e.g. during an update
* there are a few functions that could be made `STRICT` -- passing `NULL` as a parameter will always result in a `NULL`, but for some reason that causes a significant decrease in perf.
* tagged one function as parallel safe
NOTE: somehow `ST_AsMVT()` method of tile generation is showing 70-90% slowdown with this patch. I am not sure of why this is happening. If the reason is the `IMMUTABLE` -> `STABLE` change, we may have to dig deeper into PG optimization
I would like to reformat all of our SQL to have a concise coding style.
This makes it far easier to understand the code for a casual contributor,
and lets us spot errors more easily.
Most importantly, it makes it much easier to grep (search) the code because it is more likely to be in the same syntax
Some key changes:
* SQL keywords are always UPPERCASE, e.g. `SELECT WHEN AS END ...`
* types, variables, aliases, and field names (identifiers) are always lower case
* `LANGUAGE 'plpgsql'` is now `LANGUAGE plpgsql` (no quotes)
* a few minor spacing/semicolon cleanups
P.S. Per @TomPohys request, `TABLE` is spelled using upper case despite being a type for consistency with PG Docs. Same for `LANGUAGE SQL` vs `LANGUAGE plpgsql`.
Change the feature_id of objects in the tiles from arbitrary number to osm_id.
From discussion here https://github.com/openmaptiles/openmaptiles/pull/725#issuecomment-619059931
Note the multiploygon building are load with negative relation osm_id. But the use in feature_id do an overflow (eg -7980888 -> 18446744073701570000). We can keep as is or not.
No stats done for yet. Trying to let the CI do it.
`nullif(as_numeric(height),-1)` is doing a double conversion -- null into -1 followed by -1 back into null. Remove that, and delete unused `as_numeric`
layer_building uses osm_all_buildings, which uses a
non-parallel-safe as_numeric. Marking it as non-parallel safe
for now until we switch to CleanNumeric later.
Replacing materialized view by a tables with update from trigger on change only.
Start with the most simple cases.
Just replicate the change on:
* `osm_water_polygon` to `osm_water_lakeline`,
* `osm_water_polygon` to `osm_water_point`.
Use a view to factorize the `osm_water_lakeline` and `osm_water_point_view` definition and reuse it in the trigger.
The update of `osm_important_waterway_linestring` is more complex, as it is a merge of `osm_waterway_linestring`. It not done in the same way. At the end of the transaction we remove impacted and recompute them.
The goal is to update more quickly the content of derivated table by just updating the changing content. It replaces the update of materialized view because their need a full recompute (with lock issue).
Note, an advanced version of differential update over materialized view as already implemented in the building cluster PR #725.
It addresses #814 and a part of #809.
