Sarah Hoffmann

Yes, I've recreated the database (which also felt like it was taking much longer but I have no numbers to back up that claim). If I read the hexdumps right,...

I've compared old and new database on gorwen and they use exactly the same block sizes and have approximately the same total sizes. So block sizes is not the issue...

Here are the greped write-commit lines from the transaction logs on gorwen: [write-commit-log-gowen.log](https://github.com/drolbr/Overpass-API/files/9949259/write-commit-log-gowen.log) I don't have the old database from before the 0.7.59 upgrade anymore, so can't go further back...

That would indicate that it is really the update process that is to slow. That is also in line with the observation that gorwen sees the same delays on weekends,...

I've now added a monitor for the diff size to the [Prometheus stats](https://prometheus.openstreetmap.org/d/_MGuv5SVk/overpass?orgId=1&refresh=1m).

Oh dear, you are right. There is an overwrite for the setting on gorwen. I wonder why areas are compressed then. We probably should try with a reimport and see...

gorwen has plenty of CPU to spare. I'll give the full z compression a try and report back.

I went with lz4 in the end because libz was not compiled in. That has pretty much solved the issue and gorwen is now able to keep up with minutelies....

The 0.7.57 database was exactly the same: uncompressed except for the area index. You can't compare the disk utilisation between July and October. Area processing wasn't running then. Only the...

Nominatim's parent concept is in general not exactly what you would expect. Most importantly: it does not create a full hierarchy of all places. So countries are not the only...