Hi

I'm using slony1-1.2.10, Linux spec-2 2.6.22-gentoo-r5,
postgresql-8.2.4-r1, libpq-8.2.4, libpqxx-2.6.8

If I follow the steps in "Replicating Your First Database", and then
check for the memory usage of the slon processes using ps I notice that
the RSS of each one (the worker process for pgbench and the worker
process for pgbenchslave) increases in 4 kilobytes about every 2
minutes.

I ran the test again with valgrind and the results show in the summary
for the slon process for the "master" node:
(...)
==00:00:02:02.047 2024== 2,280 bytes in 19 blocks are definitely lost in
loss record 16 of 17
==00:00:02:02.047 2024==    at 0x4022998: malloc
(in /usr/lib/valgrind/x86-linux/vgpreload_memcheck.so)
==00:00:02:02.047 2024==    by 0x403158A: PQmakeEmptyPGresult
(in /usr/lib/libpq.so.5.0)
==00:00:02:02.047 2024==    by 0x403A1EC: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.047 2024==    by 0x4030623: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.047 2024==    by 0x40321B5: PQgetResult
(in /usr/lib/libpq.so.5.0)
==00:00:02:02.047 2024==    by 0x403235A: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.047 2024==    by 0x804F936: remoteListenThread_main
(remote_listen.c:262)
==00:00:02:02.047 2024==    by 0x4051FFA: start_thread
(in /lib/libpthread-2.5.so)
==00:00:02:02.047 2024==    by 0x412195D: clone (in /lib/libc-2.5.so)

In the summary for the slon process for the "slave" node:
(...)
==00:00:02:02.080 2036== 30,824 (4,200 direct, 26,624 indirect) bytes in
35 blocks are definitely lost in loss record 19 of 20
==00:00:02:02.080 2036==    at 0x4022998: malloc
(in /usr/lib/valgrind/x86-linux/vgpreload_memcheck.so)
==00:00:02:02.080 2036==    by 0x403158A: PQmakeEmptyPGresult
(in /usr/lib/libpq.so.5.0)
==00:00:02:02.081 2036==    by 0x403A21A: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.081 2036==    by 0x4030623: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.081 2036==    by 0x40321B5: PQgetResult
(in /usr/lib/libpq.so.5.0)
==00:00:02:02.081 2036==    by 0x403235A: (within /usr/lib/libpq.so.5.0)
==00:00:02:02.081 2036==    by 0x8058E4F: remoteWorkerThread_main
(remote_worker.c:2938)
==00:00:02:02.081 2036==    by 0x4051FFA: start_thread
(in /lib/libpthread-2.5.so)
==00:00:02:02.081 2036==    by 0x412195D: clone (in /lib/libc-2.5.so)

The line remote_listen.c:262 is
	res = PQexec(dbconn, dstring_data(&query1));
for which PQclear() is only called if (PQresultStatus(res) !=
PGRES_COMMAND_OK)

And the line remote_worker.c:2938 is
	res4 = PQexec(loc_dbconn, dstring_data(&query2));
for which again PQclear() is only called if (PQresultStatus(res4) !=
PGRES_COMMAND_OK)

Shouldn't PQClear() be also called when the query is successful?

I tried adding the PQClear() calls and running the test again. The
results were that the 4 kilobytes increases in memory usage are now more
unfrequent (about every 10 minutes). But I don't know whether doing that
is correct or not. So my doubts are:
a) Are those really a memory leaks?
b) Is that the correct way to fix them?

Thanks in advance,

-- 
Damián Quiroga
Integratech S.A