Date: 2012-08-21 15:37:35 +0200
From: info
To: MonetDB5 devs <>
Version: -- development
CC: arkr17997, hotekabop, @yzchang

Last updated: 2019-10-31 13:35:49 +0100

Comment 17634

Date: 2012-08-21 15:37:35 +0200
From: info

Created attachment 140
Valgrind logging

monetdb --version
MonetDB Database Server Toolkit v1.0 (Jul2012-SP1)

There seems to be a memory leak when making a connection. The amount of bytes leaked increase with each connection. We have tested this with both mclient and mapi(C-code). Mclient shows an additional leak compared to connecting with mapi. The first connection made to a database also seems to have some extra leakage.

I've added an archive with valgrind logging.

MAPI:
For mapi we used the following C-program:
Mapi dbh = NULL;
dbh = mapi_connect("localhost", 50000, "monetdb", "monetdb", "sql", "connectionmapi");
mapi_destroy(dbh);

Test: No connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmapi >& output_mserver5_mapi.txt
Terminal1: \q
==26903== definitely lost: 216 bytes in 3 blocks

Test: 1st connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmapi >& output_mserver5_mapi_1st.txt
Terminal2: ./Program
Terminal1: \q
==26955== definitely lost: 237,648 bytes in 116 blocks

Test: 2nd connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmapi >& output_mserver5_mapi_2nd.txt
Terminal2: ./Program
Terminal1: \q
==27001== definitely lost: 40 bytes in 1 blocks

Test: 3rd connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmapi >& output_mserver5_mapi_3rd.txt
Terminal2: ./Program
Terminal1: \q
==27034== definitely lost: 40 bytes in 1 blocks

Test: 4th + 5th connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmapi >& output_mserver5_mapi_4th_5th.txt
Terminal2: ./Program
Terminal2: ./Program
Terminal1: \q
==27071== definitely lost: 80 bytes in 2 blocks

MCLIENT:
Test: No connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmclient >& output_mserver5_mclient.txt
Terminal1: \q
==26105== definitely lost: 216 bytes in 3 blocks

Test: 1st connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmclient >& output_mserver5_mclient_1st.txt
Terminal2: mclient -dconnectionmclient
Terminal2: \q
Terminal1: \q
==26147== definitely lost: 238,824 bytes in 117 blocks

Test: 2nd connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmclient >& output_mserver5_mclient_2nd.txt
Terminal2: mclient -dconnectionmclient
Terminal2: \q
Terminal1: \q
==26229== definitely lost: 1,216 bytes in 2 blocks

Test: 3rd connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmclient >& output_mserver5_mclient_3rd.txt
Terminal2: mclient -dconnectionmclient
Terminal2: \q
Terminal1: \q
==26293== definitely lost: 1,216 bytes in 2 blocks

Test: 4th + 5th connection
Terminal1: valgrind --leak-check=full mserver5 --set gdk_alloc_map=no --dbname=connectionmclient >& output_mserver5_mclient_4th_5th.txt
Terminal2: mclient -dconnectionmclient
Terminal2: \q
Terminal2: mclient -dconnectionmclient
Terminal2: \q
Terminal1: \q
==26415== definitely lost: 2,432 bytes in 4 blocks

Attached file: connection_leak.tar.bz2 (application/x-bzip, 254089 bytes)
Description: Valgrind logging

Comment 17650

Date: 2012-08-24 12:33:32 +0200
From: @sjoerdmullender

Changeset 1a1256bc8d44 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=1a1256bc8d44

Changeset description:

Fix memory leak.
The lifespan info wasn't used in the groups optimizer, and what's
worse, it wasn't freed.
This fixes (part of) bug #3144.

Comment 17651

Date: 2012-08-24 13:56:19 +0200
From: @sjoerdmullender

Changeset 1db88d19e874 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=1db88d19e874

Changeset description:

Plug memory leak.
Part of fixing bug #3144.

Comment 17652

Date: 2012-08-24 14:37:11 +0200
From: @sjoerdmullender

Changeset 293d29d8bec2 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=293d29d8bec2

Changeset description:

Plug some more memory leaks.
Fixes for bug #3144.

Comment 17653

Date: 2012-08-24 14:48:08 +0200
From: @sjoerdmullender

With the three sets of fixes I now get 0 definitely lost bytes of memory when I replay the mclient tests.
This'll be in Jul2012-SP2.

Comment 18049

Date: 2012-11-27 13:19:34 +0100
From: @yzchang

Don't know how to add test for this bug