Multiple runs of osrm-partition lead to OOM kills

[oxidase]

With the current master running osrm-partition on the already partitioned graph leads to incorrect behavior:

• for the first run weight_offset + 1 equals 4052896 for bayern-latest.osm

• for the second run weight_offset + 1 equals 3654768480 and osrm-partition is killed by OOM

Rerunning osrm-extract solves the issue.

@TheMarex Could it be a regression after #4089 or it is expected and osrm-extract run is required before osrm-partition?

[TheMarex]

Seems like a regression. We probably don't renumber one of the input files of osrm-partition that references the NBG ids we re-write.

[miccolis]

I'm pretty sure this was resolved. @oxidase please re-open if there is anything still to do here.

[oxidase]

@miccolis it is still an issue. To reproduce

./osrm-extract -p ../profiles/car.lua bayern-latest.osm.pbf && \
./osrm-partition bayern-latest.osrm && \
./osrm-partition bayern-latest.osrm && \
./osrm-customize bayern-latest.osrm && \
./osrm-routed bayern-latest.osrm

...
[error] [exception] std::bad_alloc
[error] Please provide more memory or consider using a larger swapfile

[deniskoronchik]

@oxidase in current master i got:

./osrm-extract -p ../profiles/car.lua bayern-latest.osm.pbf
./osrm-partition bayern-latest.osrm 
./osrm-partition bayern-latest.osrm 
./osrm-customize bayern-latest.osrm
[info] Loaded edge based graph: 10356818 edges, 2795182 nodes
[info] Loading partition data took 28.0876 seconds
Segmentation fault (core dumped)

Compiled in Debug mode.

update: Crashed there https://github.com/Project-OSRM/osrm-backend/blob/master/include/customizer/cell_customizer.hpp#L85

[oxidase]

@deniskoronchik it is expected, there is an issue with the second renumbering in osrm-partition

[oxidase]

Fixed via #4602