Add a top level config parameter -- the size of cache memory (in MB) to
use for caching tiles and PMT directories, defaulting to 512, and 0 to
disable. This also removes the `pmtiles.dir_cache_size_mb` parameter (it
will be ignored, but will give a warning)
```
cache_size_mb: 512
```
The new cache will contain all tiles as provided by the source. So if
PostgreSQL returns a non-compressed tile, the cache will contain the
uncompressed variant, and will be compressed for each response. This
will be fixed in the later releases.
Note that fonts and sprites are not cached at this time, and are still a
TODO.
* Make it possible to have configuration specific to the file-based
config sections: pmtiles, mbtiles, and sprites.
* Implement PMTiles directory cache shared between all pmtiles (both
http and local), with configurable max cache size (in MB), or 0 to
disable. Defaults to 32MB (?)
* PMTiles now share web client instance, which optimizes connection
reuse in case multiple pmtiles reside on the same host
* Major refactoring to allow modular reuse, enabling the following build
features:
* **postgres** - enable PostgreSQL/PostGIS tile sources
* **pmtiles** - enable PMTile tile sources
* **mbtiles** - enable MBTile tile sources
* **fonts** - enable font sources
* **sprites** - enable sprite sources
* Use justfile in the CI
Fixes#1093
PMTiles is a web-optimized format, allowing the actual file to be read
with HTTP range requests. Supporting this use case instantly allows
Martin to function as a lambda executable accessing PMTiles, but without
any significant investment into devops or hosting large file.
PMTiles config now also allows `http` and `https` protocol.
```
# Publish PMTiles files
pmtiles:
paths:
# specific pmtiles file will be published as mypmtiles source
# (use last portion of the URL without extension)
- http://example.org/path/to/mypmtiles.pmtiles
sources:
# named source matching source name to a single file
pm-src1: https://example.org/path/to/some_pmtiles.pmtiles
```
fixes#884
---------
Co-authored-by: Kyle Slugg-Urbino <35903887+kyleslugg@users.noreply.github.com>
* BREAKING: `martin-cp` will now set `format=pbf` instead of `mvt`. This
is what QGIS and possibly others expect, and this is what tools like
tilelive generates.
* `martin-cp` sets `minzoom` and `maxzoom` metadata values based on the
zoom parameters
* Add `mbtiles meta-update` command to refresh zoom levels based on the
present tiles.
Partially addresses items in #1081
Figure out and fix the bug with `tile-grid` crate or remove the crate
and implement it ourself.
---------
Co-authored-by: Yuri Astrakhan <yuriastrakhan@gmail.com>
Make sure all values in the `tiles` table or view are correct:
* zoom_level is between 0 and 30 (max allowed zoom)
* the x,y values are within the limits for the corresponding zoom level
* the column type of z,x,y are all integers
* the `tile_data` is a NULL or a BLOB
Mathematics should be consolidated in the utils crate. Eventually, I
hope we will move it to [tile-grid](https://crates.io/crates/tile-grid)
or [utiles](https://crates.io/crates/utiles) crates and consolidate our
efforts there.
Also, modify earth circumference by a few centimeters (to match various
other implementations): from `.7` to `.685_578_5`. This modified a few
unit tests.
* Now by default mbtiles will not print internal warnings (e.g. that
something is taking too long to execute)
* remove mbtiles --help test - not really needed, but adds complexity
because it has to be updated every time a version changes.
When a SQL comment is set on a table or a function to customize
tilejson, and that tbl/func is pre-configured as part of the config
file, the comment was silently ignored. Now both table and function
cases are handled correctly.
Also, update docs to not include function parameters - makes SQL example
a bit simpler.
Thanks @jjcfrancisco for reporting!
Fixes: #1044
* Use gzip compression for MVT tiles by default
* Allow user to set compression with `--encoding` parameter (same as
browser's header, e.g. use `br,gzip' to encode with brotli as first
choice, unless already encoded as gzip)
* Allow user to pass a query to the Postgres functions with
`--url-query`
* A bit of a cleanup for `mbtiles summary` output
This adds a new utility called `martin-cp` that allows copying any
number of tiles from a single source (or a composite source) to an
mbtiles file.
`martin-cp` is a tool for generating tiles in bulk, and save retrieved
tiles into a new or an existing MBTiles file. It can be used to generate
tiles for a large area or multiple areas. If multiple areas overlap, it
will generate tiles only once. `martin-cp` supports the same
configuration file and CLI arguments as Martin server, so it can support
all sources and even combining sources.
# MBTiles
* New `--agg-hash (update|verify|off)` flag replaces
`--update-agg-tiles-hash` (still supported, but not shown in the help
screen). This allows bypassing aggregate hash validation entirely,
without either updating or validating it.
* Simplify MBTiles SQL generation
* MBTiles now uses faster `1 << zoom` everywhere, and a dedicated TMS
inversion fn
* split up metadata insert and delete into separate fn
* consolidated schema initialization
* ensure db settings (like pragma) are always reset on new files
# Other
* Always sort JSON-serialized keys for consistency
* this affects `/catalog` key ordering, but content is the same
* Minor code cleanup
Partial fix of #1002
* [x] Move all summary code from `mbtiles/src/mbtiles.rs` to
`mbtiles/src/summary.rs`
* [x] Move Metadata and Validation function to separate files
* [x] Remove `filename` String from the summary stats - not much point
because it is accessible from the mbtiles struct itself
---------
Co-authored-by: Yuri Astrakhan <YuriAstrakhan@gmail.com>
After some thinking, it seems `mbtiles summary` (aliased as `mbtiles
info`) would be a bit better than stats. I renamed and adjusted
documentation, consolidating it in one doc page.
Other changes:
* use file system's file size, reporting 'unknown' if needed
* report page count
* moved bbox computation into a separate function
* inlined a number of things for readability
This implements dynamic font protobuf generation, allowing users to
request font ranges on the fly, and combining them in any order, e.g.
`Font1,Font2,Font3`, same as with sprites and tiles
This is a first iteration, without any multithreading support. In
theory, this could be done far faster by generating SDFs with multiple
threads.
### Current process
* during init, figure out all glyphs available in each font, and store
them as a bitset
* during request:
* combine requested bitsets to figure out which glyph should come from
which font file
* load those glyphs from files (using a single instance of the freetype
lib)
* convert them to SDFs and package them into a protobuf
---------
Co-authored-by: Lucas <zhangyijunmetro@hotmail.com>
* Remove `--disable-bounds` flag and `disable_bounds` config parameters.
* Add `--auto-bounds` / `-b` CLI parameter and `auto_bounds` config
value:
* `quick`: Compute table geometry bounds, but abort if it takes longer
than 5 seconds (default)
* `calc`: Compute table geometry bounds. The startup time may be
significant. Make sure all GEO columns have indexes
* `skip`: Skip bounds calculation. The bounds will be set to the whole
world
* `-b` is now mapped to `--auto-bounds` param, but it will fail if used
by itself because it now requires a value.
Fixes#955
The `/catalog` now shows available sprites, which also paves way for the
future font support.
Lots of small refactorings to streamline tile source management. Now
each tile source can produce its own catalog entry, making the whole
thing much simpler.
Fixes#949
* `mbtiles apply-diff` is now `apply-patch` (old name is still
supported)
* `mbtiles copy` can now take `--apply-patch <file>` to apply the patch
while copying from source to destination. This way, the source file will
remain unmodified.
* Fix metadata copying
* Introduce a new metadata field `agg_tiles_hash_after_apply` for diff
files
* Added a lot of new info and debug logging
* Simplified Copying interface - not much value in having all the
complex builder pattern here it seems, might as well use a simple
object.
## Testing
* Generate SQLite DBs in memory on the fly to validate just what we need
* Use `insta` for validating DB content
There is now a function `dump(connection) -> Vec<Entry>` to dump the
content of the entire SQLite DB into text with `serde`. At many steps
through the testing, the DB content is validated with the corresponding
.snap file with `insta` crate (which makes this process mega-simple,
including a simple way to "bless" (update) any changes).
## Discovered bugs
* Seems like normalized files do not get copied properly - they contain
extras that should be removed.
`postgres.auto_publish.tables` now lets users customize `clip_geom`, `buffer`, and `extent` instead of using their default values.
See also #872
---------
Co-authored-by: Yuri Astrakhan <YuriAstrakhan@gmail.com>
* Rename `global_hash` to `agg_tiles_hash`
This is still a big sticking point: what should be the name for the
metadata key for this value? The value represents the hash of all
`z,x,y,tile` over all rows of the `tiles` table (or view). Should it
include `md5` in its name, or should the hash be auto-detected by its
length? (details in #856)
* Generate it based on `tiles` table/view
* validate or generate, but not both (it will always fail otherwise)
* break up logic for per-tile, total, and integrity checks
* delete unused sqlx prep file
