On 10/11/2012 04:06 PM, Tom Lane wrote:
> Yeah, if we decide to stick with the limitation, some documentation
> would be called for. I remember having run into this and having removed
> functionality from an rm_desc function rather than question the premise.
> But maybe the extra functionality is worth the cycles.
Well, I've been interested in getting it correct, and didn't care about
performance.
But I can certainly imagine someone enabling it on a production system
to get more debug info. In which case performance would matter more.
However, WAL_DEBUG being a #define, I bet only very few admins do that.
So I tend towards sacrificing performance for better debug info in the
WAL_DEBUG case. (Especially given that you can still disable it via GUC).
Regards
Markus