> One of the developers deleted a directory from postgress's data
> directory. (He mistakenly believed this was "just like mysql" where
> he could drop the database like that)
>
> At any rate, now postgres just dies when it's attempting to
> start, how can I fix this?
I am not 100% sure whether that was a database or a table, but it
helped me once to set up a decoy. I just copied a similar file (or
a directory -- I am sorry, it's been too long ago). I could then start
the postmaster and later fix the problem by deleting the orphaned
records from the system catalogs.
That was when the schema still had some resemblance of the one
inherited from Berkeley
(http://www.postgresql.org/devel-corner/docs/programmer/system-catalogs.htm
-- this one is now four or more years out of date)
But I failed the same stunt recently, after going through a dozen or so
system tables without completely understanding them. Maybe I just
didn't want it too bad, but I am still wondering: would it be possible
for people of knowledge to go over this diagram, mark the changes in
pencil, scan it and post it here? I could then review, verify and
build a new diagram -- if that's a problem.
I often hear people argue that perl code doesn't need comments since
perl is self-documenting. A couple of times, I heard a similar
argument in regards to a database schema. C'mon, have mercy on the old
man! I feel myself like a blind electrician when I hack in the schema
without a diagram.
> I've never done anything this stupid
This is how the blind electrician got zapped:
su root
cd /
ls
rm -rf /tmp *
(automatically inserting blanks on the word boundaries and hitting CR
at the end of a paragraph is a well-known typing habit: it's cheaper
to occasionally erase extra space or a CR than stop and think
whether you need one every time. Watch out!)
--Gene