Re: [HACKERS] Another nasty cache problem

It seems that I am still tracking problems, but each time they turn out to
have a different cause: A slight variant on the select that caused memory
to run out gives

newnham=# select crsids.surname, "tblPerson"."Surname" from crsids,"tblPerson" where crsids.usn="tblPerson"."USN"::int4;
pqReadData() -- backend closed the channel unexpectedly.
This probably means the backend terminated abnormally
before or while processing the request.
The connection to the server was lost. Attempting reset: Failed.

Nested Loop (cost=66496.62 rows=34359 width=40)
-> Seq Scan on tblPerson (cost=157.62 rows=2625 width=24)
-> Seq Scan on crsids (cost=25.27 rows=584 width=16)

this is the table I based the memory hog on (2600*600). The backend closes
instantly ie., no memory usage! And, as before, it is hard to find a test case
that will do the same as repeatably (ie., test case never crashes, the
above case crashes every single time). "tblPerson", as its strange
capitalisation suggests, was imported from M$ access via ODBC.

select test.txt,test2.var from test,test2 where test2.i=test.var::int4;

Nested Loop (cost=63504.80 rows=2600 width=40)
-> Seq Scan on test2 (cost=24.80 rows=600 width=16)
-> Seq Scan on test (cost=105.80 rows=2600 width=24)

works fine.

Any thoughts on where to look?

Cheers,

Patrick