Re: How to make sure I don't write a corrupt page to db when I checkpoint?

From: Sune <sune_ahlgren_at_hotmail.com>
Date: Sun, 01 Jul 2007 18:54:06 -0700
Message-ID: <1183341246.142591.115060@i13g2000prf.googlegroups.com>

> checksums?
>
> As in, a process that modifies a page has to generate a checksum for the
> modified page, and the process that writes the pages (is there one -
> it's an in-memory database)

Yes, there will be a data store manager that performs fuzzy checkpoints periodically.

> has to lock the page and verify the checksum

Page lock is no option for concurrency reasons, I'm afraid.

> (and, maybe, worry about whether a process modified the page while it
> was locked by ignoring the lock).

Yes!

> In the context of a DBMS, paranoia is good. Just because you can't see
> that the processes attached to the database aren't out to make life
> difficult for you doesn't mean they aren't going to make difficult for
> you anyway.

Ok, I'm not mad (yet).

Is there a scheme that has proven itself through research and practice, that makes this process lightweight or at least reasonably efficient?

BRs
/Sune Received on Sun Jul 01 2007 - 20:54:06 CDT