Subject | Re: [firebird-support] Symantec livestate recovery |
---|---|
Author | Ann W. Harrison |
Post date | 2005-03-26T16:36:27Z |
Somebody wrote:
middle, and end. Copying is done front to back, so new writes before
the current copy point are not reflected in the copy.
somebody starts a transaction, writes some data, and commits the
transaction. Even if the TIP and the data are all copied, the copy of
the header page - page 0 - is older than that newest transaction, so its
idea of the next available transaction number is actually a committed
transaction. Not at all healthy.
Regards,
Ann
>>>Firebird's concept of "carefull writes" should NEVER leave the database fileBecause the file is random access - pages are written to the front,
>>>in an inconsistent state at any point in time...
>>and if that is so then
>> why is everybody advising against copying the DB - file while
>>transactions are running ?
middle, and end. Copying is done front to back, so new writes before
the current copy point are not reflected in the copy.
> Daniel Rail wrote:...
> Here's an example of what could happen:
> These were just examples that I could think of. And, you couldThe most likely case is that you start a copy, and while its going on,
> probably come up with more examples yourself, and I'm sure Ann could
> think of more, unless she proves me wrong.
somebody starts a transaction, writes some data, and commits the
transaction. Even if the TIP and the data are all copied, the copy of
the header page - page 0 - is older than that newest transaction, so its
idea of the next available transaction number is actually a committed
transaction. Not at all healthy.
Regards,
Ann