Re: Lock problem with autovacuum truncating heap

On Sun, Mar 27, 2011 at 2:13 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Mar 26, 2011, at 4:16 PM, Jan Wieck <JanWieck@Yahoo.com> wrote:
>> That was what I meant. Go in steps of 16-64MB backwards and scan from there to the current end in forward direction to find a nondeletable block. In between these steps, release and reacquire the exclusive lock so that client transactions can get their work done.
>
> Well, VACUUM uses a 16MB ring buffer, so anything that size or smaller should hit shared_buffers most of the time.
>
> I wonder though if this might defeat read-behind on operating systems that do have a working implementation.  With our current approach each read will end at the point the previous read started, which might be an algorithm somebody is using to detect a backward scan.

Good point. That means the last 16MB of buffers will be in
shared_buffers. Anything more than that will definitely not be,
because we wrote them out ourselves.

So we should truncate in 16MB chunks also.

-- 
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services