[Tile-serving] [osm2pgsql] OOM killed after processing phase (#351)

[Paul Norman]

cc @kevinkreiser

When porting to C++ I found the lockfree queue did help, but I don't think this was on a full planet, so memory usage was not an issue.

On a typical machine, the difference between all nodes in memory cache and having to hit disk is the biggest cause of performance differences outside of end of import indexing/clustering. On one with mechanical hard drives, this is even more so. So, besides being confusing and annoying, this bug is the biggest performance issue currently present.

On the machine I use for benchmarking, this is probably less so, as it has two SSDs in RAID0.

I'm running a benchmark to verify this.

---
Reply to this email directly or view it on GitHub:
https://github.com/openstreetmap/osm2pgsql/issues/351#issuecomment-117450401
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tile-serving/attachments/20150630/282f3396/attachment.html>