New OpenStreetMap Carto release, v5.0.0

[pnorman]

A new version of OpenStreetMap Carto, v5.0.0, has been released.

This version requires a database reload

[tomhughes]

There seems to be something very wrong somewhere with this...

I setup our new render server (albi) with this, along with postgres 12 and postgis 3 and the initial database import went fine and took about 16 hours and I built the extra indexes.

Replication is insanely slow though - like less than real time. I think the first batch of changes, which covered about 8 hours, took about 14 hours to apply!

It seems to be the initial phase of osm2pgsql processing the nodes that is the problem - on other servers running the previous style that does like 4k nodes/s while the new server is on the second batch of changes and is basically doing 0/s on average - it has done 450k in about 7 hours.

[lonvia]

See osm2pgsql-dev/osm2pgsql#1045. Disabling jit and parallel threads in postgres should fix it.

[tomhughes]

Ha I actually checked to make sure the JIT was on and that we had the byte code for postgis!

[tomhughes]

Is the problem index only used by the import? The renderer won't touch it right?

[lonvia]

Yes, it is only used when applying an update. We probably should disable jit/parallel threads on osm2pgsql's connections to not interfere with the rest.

[tomhughes]

Yes that's what I was wondering, whether to do it in postgresql.conf or just via PGOPTIONS in the replication script.

I did the second and it is now doing 3.8k/s ;-)

[tomhughes]

The last of the render servers has now been reloaded so this is complete.