Comment 16 for bug 1042946

Revision history for this message
Aleksandr Khomyakov (toairo) wrote : Re: Memory leak on a filtered slave

my experiments with valgrind on test server, showed no leaks. server runs with 2.5GB and growing to 3.3Gb, this growth stops.
I tested with version
- 5.5.27-28.1 Percona Server
- original 5.5.28.

after stopping the server, valgrind shows no leaks.

I supposed that the growing internal buffers. seen in smaps.

for example i attached two state smaps. not from test server. this smaps from production small server that replicates only 2 tables. no customers besides monitoring and service receives the data through replication.

state after the start and after 18 hours.

and attach my.cnf for this instance. i attach them in single archive.
mysql 5.5.27-28.1 Percona Server
OS CentOS 5.5 x86 kernel 2.6.35.13

Volume RSS first and second smaps
[root@db25 airo]# grep "^Size:" /tmp/smaps_2250 |awk '{ SUM += $2} END { print SUM }'
552088

[root@db25 airo]# grep "^Size:" /tmp/smaps_1715 |awk '{ SUM += $2} END { print SUM }'
714600

Diff 162512Kb

internal smaps 3 suspicious buffers.

1. address start 020af000 - from 14.5mb to 46mb - diff 30.4Mb
2. address start 7fa99c000000 from 7Mb to 62Mb - diff 55Mb
3. additional buffer address start 7fa994000000 - 64Mb

Here is the difference between the initial and final states.

I do not know how to find those buffers.