On 12.12.2011 13:16, Matteo Beccati wrote:
> On 12/12/2011 09:29, Heikki Linnakangas wrote:
>> On 12.12.2011 08:26, Tom Lane wrote:
>>> Andres Freund<andres@anarazel.de> writes:
>>>> In 3b8161723c645853021b57330dd2ea0484ec6131 Heikki made DatumGetInetP
>>>> unpack
>>>> toasted values. Unfortunately the btree support functions for the
>>>> inet type
>>>> didn't free memory which they have to do in contrast to about
>>>> everything else.
>>>
>>>> I fixed a few more functions than strictly necessary but I guess
>>>> thats ok.
>>>
>>> Seems like the correct fix is to revert these functions to the former
>>> behavior, ie they should be using the PP macros not the unpacking ones.
>>
>> Agreed, there's no need to unpack here. Fixed, thanks for the report!
>
> Just to clarify, am I correct assuming that the issue does not affect
> tables which have non-indexed inet fields?
Hmm, I think it might also affect queries that do large merge joins on
inet fields. Not sure about that. In any case, it's a memory leak that
gets cleared at the end of statement, so it can only become a problem
when dealing with large data sets.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com
