W dniu 08.10.2017 o 21:19, Darafei "Komяpa" Praliaskouski pisze:
> PostGIS 2.4 brings support for ST_AsMVT for Mapbox Vector Tile export.
> You don't need backend anymore to connect your cool vector rendering
> Mapbox GL to your database.
According to this blog entry:
"We also found out that MVT’s are good for rendering client side, but
not so good when you want to render tiles with a lot of features.
However, it is not all about speed. More importantly, having MVT’s in
the database as a building block lets you build the SQL on top, which
gives you a lot of flexibility."
On 10/8/2017 11:03 AM, Sven Geggus wrote:
> Debian stable 9.x comes with PostgreSQL 9.6 and Postgis 2.3.
> However the now released PostgreSQL 10.0 and Postgis 2.4 are an easy
> Will the newer version provide any advantage for a osm2pgsql database use case?
Yes, but nothing game-changing for a osm2pgsql database and a standard
production workload according to the release notes. There's a lot of
improvements related to replication. The following are likely to help
- Faster GiST index inserts and updates
- Faster GIN index vacuuming
- General performance improvements
The parallel b-tree index usage could help updates, but this is seldom
an issue, and may even be undesirable under heavy rendering load. I've
also seen no data on if it helps or not for the particular queries
In short, I'd recommend the upgrade for any new deployments, but it's
not likely to be worth an out of schedule reload.
If you're doing analysis with an osm2pgsql database, it's likely to be
more useful. Query-level parallelism is more useful there because
there's generally not so many connections running in parallel at the
application level. This makes the parallelism improvements useful, if
they apply to the types of queries that are being run.
The multi-column statistics have the potential to be interesting for
If upgrading to 10.x, make sure to set the parallel configuration
options if you want to see those performance gains.
PostGIS 2.4 is a similar situation. Stylesheets are unlikely to require
it yet, so it won't enable anything new for rendering servers for some
time. If a new function helps with your analysis use-case, it will be