On 3/3/15 11:48 AM, Andres Freund wrote:
> On 2015-03-03 11:43:46 -0600, Jim Nasby wrote:
>> >It's certainly better than now, but why put DBAs through an extra step for
>> >no reason?
> Because it makes it more complicated than it already is? It's nontrivial
> to capture the output, escape it to somehow fit into a delimited field,
> et al. I'd rather have a committed improvement, than talks about a
> bigger one.
I don't see how this would be significantly more complex, but of course
that's subjective.
>> >Though, in the case of multiple errors perhaps it would be best
>> >to just report a count and point them at the log.
> It'll be confusing to have different interfaces in one/multiple error cases.
If we simply don't want the code complexity then fine, but I just don't
buy this argument. How could it possibly be confusing? Either you'll get
the actual error message or a message that says "Didn't work, check the
log." And the error will always be in the log no matter what. I really
can't imagine that anyone would be unhappy that we just up-front told
them what the error was if we could. Why make them jump through needless
hoops?
> I'm saying that you'll need a way to notice that a reload was processed> or not. And that can't really be the message
itself,there has to be> some other field; like the timestamp Tom proposes.
Ahh, right. We should make sure we don't go brain-dead if the system
clock moves backwards. I assume we couldn't just fstat the file...
--
Jim Nasby, Data Architect, Blue Treble Consulting
Data in Trouble? Get it in Treble! http://BlueTreble.com