--On Freitag, Mai 25, 2007 00:02:06 +0000 Jaime Casanova
<systemguards@gmail.com> wrote:
> the original patch is from Albert Cervera =)
Ah, missed that, thanks ;)
>>
>
> sounds good. can we see the new patch?
Attached tablespace.c.diff shows my current changes to use an OID lookup
list.
>
> the reason for those messages is that the tablespace can get full or
> can be dropped before use, so we throw the message for the dba to take
> actions. if no one thinks is a good idea the message can be removed.
>
I could imagine that this could irritate DBA's (at least, that is what
happened to me during testing). It's okay that someone could drop a
tablespace concurrently to other transactions, but i have concerns that
with temp_tablespaces this could happen during _queries_. Do queries
delete/recreate temp files during execution, maybe within sorts so that the
used temp tablespace looks empty for a certain period of time?
>> The silent
>> mechanism to drop a tablespace during temporary usage makes me a little
>> bit uncomfortable about its robustness.
>>
>
> maybe using the list you put in TopMemoryContext we can deny the
> ability to drop the tablespace until it's removed from the list of
> temp tablespaces.
That would mean we have to share this information between backends. This
looks complicated since every user could have its own temp_tablespaces
GUC....
--
Thanks
Bernd
