Hi,
Last weekend, I update 2 customers servers with the latest Postgresql
8.3.5 vrsion. The databases in each customer have +/- 50GB of data.
I changed some ODBCs versions in clients machines from 7.3.0208 to 8.3.0400.
In this machines, when I run some reports (thats run million of small
queries), the Postgresql server crashes because this processes use all
physical and swap memory in the server.
If I disable both 'Use declare/fetch' and 'Updatable Cursor', the error
does not occurs. I don't need the 'Updatable Cursors', but the 'Use
Declare Fetch' is crucial to me.
In ODBC 7.3.0208, with this option enabled, the error doesn't occurs.
I think that use of memory grows in sort operations (If I reduce the
work_mem, the memory grows not so fast). But the queries are too fast
(average 0.005 ms). I think that after the query is executed, the work
memory should be released.
Is SAVEPOINT/RELEASE statements generated by ODBC 8.3.0400 (and not
generated by 7.3) the cause of the problem ???
Best regards,
Alexandre
Server 1:
Debian 4.0 etch
Postgresql 8.3.5 compiled from sources.
16 Gb RAM
shared buffers = 2GB
work_mem = 64MB
Server 2:
CentOS 5
Postgresql 8.3.5 compiled from sources.
8 Gb RAM
shared buffers = 1GB
work_mem = 32MB