Re: WIP: dynahash replacement for buffer table

On Thu, Oct 16, 2014 at 12:26 PM, Ryan Johnson
<ryan.johnson@cs.utoronto.ca> wrote:
> The only metric where RCU loses to
> hazard pointers is if you have really tight timing constraints on resource
> reclamation.

I think we do have that problem.  It's certainly completely
unacceptable for a query to prevent buffer reclaim on any significant
number of buffers even until the end of the query, let alone the end
of the transaction.

But, hey, if somebody wants to try writing a patch different than the
one that I wrote and see whether it works better than mine, I'm
totally cool with that.  This is something I came up with, and we're
here to evaluate whether it works better than any other option that we
have now or that someone wants to develop.  I'm not saying this is the
best solution; it's just something I've got that seems to basically
work.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company