Re: profiling connection overhead

On Wed, Nov 24, 2010 at 11:33 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Wed, Nov 24, 2010 at 10:25 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Or make it execute only in assert-enabled mode, perhaps.
>
>> But making the check execute only in assert-enabled more
>> doesn't seem right, since the check actually acts to mask other coding
>> errors, rather than reveal them.  Maybe we replace the check with one
>> that only occurs in an Assert-enabled build and just loops through and
>> does Assert(PrivateRefCount[i] == 0).
>
> Yeah, that would be sensible.  There is precedent for this elsewhere
> too; I think there's a similar setup for checking buffer refcounts
> during transaction cleanup.
>
>> I'm not sure exactly where this
>> gets called in the shutdown sequence, though - is it sensible to
>> Assert() here?
>
> Assert is sensible anywhere.

OK, patch attached.  Here's what oprofile output looks like with this applied:

3505     10.4396  libc-2.11.2.so           memset
2051      6.1089  libc-2.11.2.so           memcpy
1686      5.0217  postgres                 AllocSetAlloc
1642      4.8907  postgres                 hash_search_with_hash_value
1247      3.7142  libc-2.11.2.so           _int_malloc
1096      3.2644  libc-2.11.2.so           fread
855       2.5466  ld-2.11.2.so             do_lookup_x
723       2.1535  ld-2.11.2.so             _dl_fixup
645       1.9211  ld-2.11.2.so             strcmp
620       1.8467  postgres                 MemoryContextAllocZero

Somehow I don't think I'm going to get much further with this without
figuring out how to get oprofile to cough up a call graph.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company