On 6/11/19 11:15 AM, Karsten Hilbert wrote:
>>> The problem I hope to protect against with this approach: the
>>> CREATE DATABASE might untaint corrupted data from a bad disk
>>> block into a good disk block virtue of doing a file level
>>> copy.
>>>
>>> I hope my reasoning isn't going astray.
>>
>> As I understand it checksums are done on the page level using a hash(for
>> details: https://doxygen.postgresql.org/checksum__impl_8h_source.html).
>> I am not sure how a page could get un-corrupted by virtue of a file copy.
>
> Ah, no, I did not explain myself well.
>
> Let's assume a corrupted, bad (but readable at the hardware
> level) disk block B. A filesystem level copy (as in CREATE
> DATABASE) would successfully read that disk block B and
> copy the corrupted content into a good disk block G elsewhere
> on the disk. Verifying the checksum of the page sitting on
> block B before doing the database cloning would
> reveal the corruption before it got cloned.
>
> Does that make sense ?
Yes.
>
> Karsten
>
--
Adrian Klaver
adrian.klaver@aklaver.com