On 7/6/06, Brian Weaver <[EMAIL PROTECTED]> wrote:

Speaking as someone who just had a server failure and is *still*
trying to recover from it. Backup often and AUTOMATE it, and for the
love of all that's holy... do a test restore from your backups on a
semi regular basis!


yes. we should have automated backup. I have a dedicated backup server for
this.

I think I might have been happier if my box had a "catastrophic"
failure, but what happened was more insidious. The memory went south,
but not so bad that it would crashed the box. It introduced problems
into my database and potentially into the process of trying to extract
the data from the box. Since the memory wasn't bad enough to cause the
box to fail I didn't notice a problem for a while. So every time I did
something that saved information from main memory back to disk I was
potentially corrupting my data.


That was serious. I can't imagine if it ever happens to me.

As a result I was wondering what dbmail users think of this potential
enhancement: Add a column in the database for the physical message
blocks that is the MD5 or SHA hash of the block at insert. This could
be used by the dbmail utility to do a data validation and to catch
problems in the db that might be caused by either memory or disk
errors.


this i believe involve some overhead but I also think it should present as
an option.

Reply via email to