Tom Lane wrote:
>Andreas Pflug <pgadmin@pse-consulting.de> writes:
>
>
>>Tom Lane wrote:
>>
>>
>>>Why is that a useful approach? You might as well shut down the
>>>postmaster and do a cold filesystem backup,
>>>
>>>
>>>
>>We're talking about *hot* backup, aren't we?
>>
>>
>
>Exactly. The approach you're sketching can't work for hot backup,
>because it effectively assumes that the database won't be changing.
>
>
Well in the first place my posting was to express my suspicion that WAL
replay relies on clog/pg_control being accurate, i.e. transactions
marked as flushed must be on disk. AFAICS this is the consequence of WAL
replay implementation. In case of hot backup, this means that data files
must not be older than clog. Do you agree? So PITR needs a mechanism to
insure this at backup time.
Next question would be: If the point in time I'd like to recover is that
very backup checkpoint time, do I need xlog at all?
Regards,
Andreas