On Mon, Apr 30, 2018 at 3:13 PM, Andres Freund <andres@anarazel.de> wrote:
> On 2018-04-30 14:59:31 +1200, Thomas Munro wrote:
>> In EXPLAIN (BUFFERS), there are two kinds of cache misses that show up
>> as "reads" when in fact they are not reads at all:
>>
>> 1. Relation extension, which in fact writes a zero-filled block.
>> 2. The RBM_ZERO_* modes, which provoke neither read nor write.
>
> Just for understanding: 2) can never happen for buffers showing up in
> EXPLAIN, right?
>
> I'm not saying you shouldn't fix the accounting...
Maybe the hash AM can reach that in _hash_getinitbuf() while adding
overflow pages to bucket chains? Admittedly case 2 is obscure and
rare if not unreachable and probably no one would care too much about
that in practice (whereas case 1 can be seen by simply inserting stuff
into a new empty table). Other patches I'm working on for later
proposal do it more often (think accessing known-empty pages in a kind
of preallocated extent), and it occurred to me that it's clearly a bug
on principle, hence this patch.
--
Thomas Munro
http://www.enterprisedb.com