On 01/05/13 12:36, Yang Zhang wrote:
> On Tue, Apr 30, 2013 at 7:21 PM, Julian Glass <tempura(at)internode(dot)on(dot)net> wrote:
>> On 01/05/13 09:55, Yang Zhang wrote:
>>> I would intuit that it's fine, but I just want to make sure there are
>>> no gotchas from a recovery point of view:
>>>
>>> If I were to lose my temp tablespace upon system crash, would this
>>> prevent proper crash recovery?
>>>
>>> Also, if I were to omit the temp tablespace from the base backup,
>>> would that prevent proper backup recovery?
>>>
>>> Thanks.
>>>
>>>
>> Yes I find this interesting. I wonder if you can snapshot the tablespace
>> filesystem upon initiation, then mount the snapshot backup before
>> restarting the server.
>>
>> Worth testing.
> This strikes me as the sort of thing that is dangerous to attempt to
> validate using strictly black-box testing, esp. by someone such as
> myself who has little knowledge of PG internals - just because it
> works for certain test cases of mine doesn't yield generalizable
> guarantees.
System disk failure is pretty serious, but I'm not sure if shutting down
the server would be required.
I have no experience yet with a tablespace failure on temporary objects
(tablespaces + MV is a future interest).
Is there a reason why dropping the temp objects, including the
tablespace and resetting temp_tablespaces to DEFAULT would not work?
Its not clear how you utilize this temp tablespace. (Assuming you are
using temp_tablespaces and not defining it withing the table definition.)
Regards,
Julians
(Sorry about missing the ML previously)