HI Tomas,
>> No, at least in the current version. The next version (9.2) will support
>> checksums, but it's meant mostly as a protection against failures at the
>> I/O level. It might catch some memory issues, but it certainly won't be
>> 100% protection.
>
>Oh, I see - it was bumped to 9.3 and I've missed that.
Glad to see there is work going on in the integrity area.
> There are unofficial tools (e.g. pg_check @ github, written by me) that
> perform some checking when requested, but it's not (and never will be)
> automatic.
>
> Moreover, in many cases it's impossible to identify hw-level corruption,
> unless you take the mainframe approach (running the task on multiple
> devices and check if they produce the same result).
Sure, but checksumming in combination with a structural integrity
check should give at least some confidence everything is ok.
> (3) use good hw (ECC memory, ...) / test it thoroughly etc.
Thats the problem - because of cost constraints I have to deploy
postgresql on non-ECC boxes.
So I am looking forward to the checksum feature and hope no bit will toogle ;)
Thanks again, Clemens