Fixes#1066
This PR adds a new field `access` in the transportation layer, which will be set to `no` if the `access` tag is either `no` or `private`. While `private` is the more popular value by a 17:1 ratio, I went with `no` because it's smaller in the tiles. In addition, the text `no` opens up the future possibility of other text-based access values such as `destination`, `customers`, or `permit`.
The screenshot below shows an example of access tagging for a road on a military base:
This PR updates the `transportation` layer creation scripts to simplify the SQL and remove unneeded sub-selects, checks and conditionals, fix indexes, and improve inline documentation.
Currently, there are two sequences of materialized view creations. There is one from zoom 11 through 9, and a second one for zoom 8 through 4. This PR removes that break in the sequence of transportation table materialized view creations, in favor of one in which high-zoom views are created first, and then each lower zoom is created from the zoom above.
Instead, the current generalized zoom 8 transportation table is built directly from `osm_transportation_linestring` rather than being built from the zoom 9 transportation table. This means that when building the zoom 8 table, it must scan the entire transportation network rather than just selecting from the pre-filtered zoom 9 table.
This PR removes an unneeded sub-select in the build of the zoom 8 table, which appears to be a leftover from an old version of the SQL that did some sort of merge. Once this PR is implemented all zooms from 11 through 4 will be linked via a progressive series of materialized views.
Lastly, this adds in missing materialized view refreshes for zooms 9-11, which appear to have been entirely missing, and as far as I can tell aren't getting updated in the current version of this code.
In addition, the following optimizations were added as part of this commit:
1. Updates the `osm_highway_linestring_highway_partial_idx` partial index to match the `SELECT..WHERE` clause actually present in `transportation/update_route_member.sql`, which is where it appears to be actually used, and update inline documentation to reflect this.
2. Collapses unnecessary sub-select block in `osm_transportation_merge_linestring_gen_z11`, and removes unnecessary ST_IsValid() call, which already provided in `mapping.yaml`, and update inline documentation to reflect these assumptions.
3. Updates `WHERE` blocks to remove unnecesary checks and further document assumptions. The `highway=construction` check is unnecessary in all cases, because it is sufficient to check the `construction` key alone. If `construction=*` is set, then `highway=construction` is implied.
4. Two indexes were added to `layers/transportation/update_route_member.sql` to improve route population performance.
In testing locally, I'm seeing performance improvements around 10% in the generation of the `transportation` layer, based on modifying `openmaptiles.yaml` to generate only the transportation layer and then repeatedly running `time make import-sql`, however, this timing might be impacted by docker, so I would ask for confirmation of acceptable performance.
In addition, this PR shortens the length of the transportation update SQL file by 30 lines, which makes it easier for contributors to work with.
PR #1168 removed several `WHERE` clauses in the `transportation_merge_*` table series. With those removed, it appears that `osm_transportation_merge_linestring_gen_z8` and `osm_transportation_merge_linestring` are nearly identical, with the former simply adding an `ST_Simplify()` operation.
A grep of the codebase indicates that the _only_ use for `osm_transportation_merge_linestring` is to hold the intermediate result of an `ST_Dump(geometry))` before it is fed into `ST_Simplify()`. Therefore, it appears that we're holding an entire zoom 8 copy of the transportation layer (all motorway/trunk/primary roads) in a materialized view for absolutely no reason at all.
This PR removes the `osm_transportation_merge_linestring` intermediate table and changes the definition of `osm_transportation_merge_linestring_gen_z8` to perform the `ST_Dump()` and `ST_Simplify()` transformations directly from the `osm_highway_linestring` table in a single operation.
This PR removes several redundant/unnecessary WHERE clauses from the transportation layer. Specifically:
The table `osm_transportation_merge_linestring` is a view of `osm_highway_linestring` which exposes only motorway/trunk/primary roads:
9e4be3e3b0/layers/transportation/update_transportation_merge.sql (L122-L123)
However, the create statement for the table `osm_transportation_merge_linestring_gen_z8`, which is a view of `osm_transportation_merge_linestring`, also contains a `WHERE` clause which selects down to motorway/trunk/primary roads. This `WHERE` is unnecessary:
9e4be3e3b0/layers/transportation/update_transportation_merge.sql (L144-L145)
This unneeded `WHERE` clause is similarly present in the create statement for `osm_transportation_merge_linestring_gen_z7`, which is a view of `osm_transportation_merge_linestring_gen_z8`:
9e4be3e3b0/layers/transportation/update_transportation_merge.sql (L163-L164)
Likewise, there is a similar redundant `WHERE` clause in the `osm_transportation_merge_linestring_gen_z5` and `osm_transportation_merge_linestring_gen_z6` tables, both of which select down to `motorway` and `trunk`. This `WHERE` clause is only needed on the z6 table, and is redundant on the z5 table.
I am not sure what the performance penalty is for these redundant `WHERE` clauses, but there does not appear to be any reason to keep them, and they may incur a performance cost.
To avoid discontinuous transportation lines between zooms 9 and 11.
- Originally limit geometry by length for z9 - z11 (`ST_Length(geometry) > ZRes(11)`)
- highway z9 to z11 was generalized during import-osm
- now just create a filtered and generalized z11 table
- then merge segments in the same way as from (full-featured) osm_highway_linestring and used this merged z11 for mat.view z10 and z9
Close#1107
On creation of materialized views `osm_transportation_merge_linestring_gen_zX`, an index is created each time the new view and not used else where than the next generalization step.
It does not worth it to to a scan on the table to create index that will be used only once, while the next generalisation step can also be done in one full scan.
So, removing these indices.
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.
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`.
Tag all SQL materialized views with a machine-readable comment
to indicate that this materialized view can be created without
data:
/* DELAY_MATERIALIZED_VIEW_CREATION */
In the next version of tools this comment can be optionally
replaced with the "WITH NO DATA" parameter, thus allowing
a much faster execution of the SQL script. All materialized
viewes will be populated with data in parallel afterwards
using the `refresh-views` tools script.
Brian Sperlongano
