[OSRM-talk] osrm-extract hangs on osmosis updated germany file
Florian Lohoff
f at zz.de
Mon Jun 29 17:15:21 UTC 2015
On Fri, Jun 26, 2015 at 07:31:39PM +0200, Florian Lohoff wrote:
> Hi,
>
> i am regularly (every 4 hours or so) generating a new germany file with
> osmosis from planet diff (updating + cutting) and then convert it
> with the default car profile.
>
> Now i switched from 0.3.9 to git current and
> osrm-extract now hangs, or takes infinite time.
>
> For testing i used a germany file from download.geofabrik.de which
> worked. Convert took 663.46s.
>
> Now i used my file again and it is running for >2 hours at the same
> file size.
> [extractor] Sorting edges by renumbered start ... ok, after 16.5565s
>
I found the live-lock of the osrm-extract - Its caused by commit
commit 2777d53a12ca6753880f1dc025d8a7cc34377bcb
Author: Patrick Niklaus <patrick.niklaus at student.kit.edu>
Date: Sun May 24 17:25:38 2015 +0200
Direct edges in contractor correctly and add better graph validation.
[...]
@@ -303,13 +308,18 @@ void ExtractionContainers::PrepareEdges()
std::cout << "ok, after " << TIMER_SEC(sort_edges_by_renumbered_start) << "s" << std::endl;
BOOST_ASSERT(all_edges_list.size() > 0);
- for (unsigned i = 1; i < all_edges_list.size();)
+ for (unsigned i = 0; i < all_edges_list.size();)
{
// only invalid edges left
if (all_edges_list[i].result.source == SPECIAL_NODEID)
{
break;
}
+ // skip invalid edges
+ if (all_edges_list[i].result.target == SPECIAL_NODEID)
+ {
+ continue;
+ }
The later continue can lead to be stuck in this for loop as "i" does not
get incremented.
https://github.com/Project-OSRM/osrm-backend/issues/1553
Flo
--
Florian Lohoff f at zz.de
We need to self-defense - GnuPG/PGP enable your email today!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 828 bytes
Desc: Digital signature
URL: <http://lists.openstreetmap.org/pipermail/osrm-talk/attachments/20150629/a36a2809/attachment.sig>
More information about the OSRM-talk
mailing list