Leaking using Python

[monetdb-team]

Date: 2012-06-18 16:39:34 +0200
From: @skinkie
To: SQL devs <>
Version: 11.11.5 (Jul2012)
CC: @njnes

Last updated: 2012-07-17 13:46:38 +0200

Comment 17370

Date: 2012-06-18 16:39:34 +0200
From: @skinkie

User-Agent: Mozilla/5.0 (X11; Linux i686) AppleWebKit/537.1 (KHTML, like Gecko) Chrome/21.0.1171.0 Safari/537.1
Build Identifier:

For some reason a connection by Python, and a few hunderd queries over the MAPI interface result in significant leakage over time.

Reproducible: Always

Steps to Reproduce:

1. use the pyhon dbapi
2. make queries
3. observe leakage

Comment 17372

Date: 2012-06-18 16:42:22 +0200
From: @skinkie

Created attachment 116
MonetDB Valgrind

Valgrind output.

Attached file: monetdb-valgrind.txt.gz (application/gzip, 72051 bytes)
Description: MonetDB Valgrind

Comment 17374

Date: 2012-06-18 17:48:07 +0200
From: @skinkie

Reproduce with:

/tmp/test.sql;
select 1;

while [ 1 ]; do /opt/monetdb-mei/bin/mclient -i /tmp/test.sql 1>/dev/null 2>&1; done

So the bug is much broader. This specific example shows the issues with SQL scenario. Likely within the scenario something going on - and I presume Python is not making a new connection per query.

Comment 17409

Date: 2012-07-04 15:40:24 +0200
From: @njnes

this leak is/was in the new groups optimizer

Comment 17410

Date: 2012-07-04 15:46:26 +0200
From: @skinkie

You know that select 1; was merely an example query? I have a much larger valgrind trace now, if you are interested in it, I can also post it.

Comment 17472

Date: 2012-07-17 13:46:38 +0200
From: @grobian

Fix delivered in Jul2012 release