On 10/12/2014 2:49 PM, Ilya Zverev wrote:
Hi! Yesterday I had a simple task: there is a rendering server, which
is not to be minutely updated. So its PostgreSQL database doesn't need
temporary "slim" tables. But I cannot import an extract without
creating those, so basically I can import 3 times less data than
possible.
It's worth noting that the --drop mode in osm2pgsql will delete the slim
tables when done, and also not create the indexes on those tables. As
those indexes are large, this is a substantial space savings in itself.
--drop also will create more compact indexes for the rendering tables.
The space saved is more than a simple inspection of table and indexes
sizes would cause you to expect. For an import, the key parameter is
peak disk usage, which normally happens partway through the clustering
and index creation phase. As the slim tables are dropped before this,
they do not contribute. --disable-parallel-indexing will also reduce
peak ram usage, though I am uncertain if the slim tables get dropped
early enough with both this and --drop.
--slim --drop --disable-parallel-indexing should have a database not
much larger than the final rendering tables.*
Of course, Zverev's technique has lesser space requirements, and
probably faster, at the cost of requiring a second server.
* The final space required is (rendering tables + rendering indexes)
while the largest space requirement is the maximum of (rendering tables
+ slim tables), (rendering tables + largest rendering table again -
largest rendering index)
_______________________________________________
dev mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/dev
