Re: [overpass] apply_osc_to_db.sh transactions

[Roland Olbricht <>]

Hi mmd,

> can you maybe also elaborate a bit what happens if: :copy_shadows_to_mains()
> gets aborted somewhere in the middle?

The steps are as follows:
- The lock file "osm_base_shadow" is created
- copy_shadows_to_mains() runs
- The lock file "osm_base_shadow" is deleted
- The .shadow files are deleted

The semantics are:
- if "osm_base_shadow" is present, then the ".shadow" files are authorative
- otherwise the standard index files are authorative. If .shadow files 
exist then they are only authorative for the write process.

In particular the database is never invalid. A file can have no other 
state then "present" or "absent". And the state of the (very shortlived) 
file "osm_base_shadow" determines which of the two sets of index files 
are valid.

From this one can conclude the right approach on restart:
- if "osm_base_shadow" is present then first copy_shadows_to_mains(), 
afterwards delete "osm_base_shadow"
- if .shadow files are (still) present, then delete them

The dispatcher does that for you. Hence, it should be impossible to get 
the database corrupted this way.

Best regards,

Roland