On Fri, Jan 20, 2012 at 11:34 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On Fri, Jan 20, 2012 at 12:54 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
>> Thanks for the review!
>>
>> On Fri, Jan 20, 2012 at 8:15 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
>>> I'm looking at this patch and wondering why we're doing so many
>>> press-ups to ensure full_page_writes parameter is on. This will still
>>> fail if you use a utility that removes the full page writes, but fail
>>> silently.
>>>
>>> I think it would be beneficial to explicitly check that all WAL
>>> records have full page writes actually attached to them until we
>>> achieve consistency.
>>
>> I agree that it's worth adding such a safeguard. That can be a self-contained
>> feature, so I'll submit a separate patch for that, to keep each patch small.
>
> Maybe, but you mean do this now as well? Not sure I like silent errors.
If many people think the patch is not acceptable without such a safeguard,
I will do that right now. Otherwise, I'd like to take more time to do
that, i.e.,
add it to 9.2dev Oepn Items.
I've not come up with good idea. Ugly idea is to keep track of all replays of
full_page_writes for every buffer pages (i.e., prepare 1-bit per buffer page
table and set the specified bit to 1 when full_page_writes is applied),
and then check whether full_page_writes has been already applied when
replaying normal WAL record... Do you have any better idea?
Regards,
--
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center
