Subject: Re: Tid scan improvements
To: David Rowley <david.rowley@2ndquadrant.com>,
 Edmund Horner <ejrh00@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
References: 
 <CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com>
 <CAMyN-kCknpsgHt8eP=jeYMGMU=suRAoZ38CoRJB2fxKTKa52gw@mail.gmail.com>
 <CAKJS1f8FyS51b6j++qShP_av5Kq1oq83t16ALPftPm4F0BbqkQ@mail.gmail.com>
 <CAMyN-kCCjuLR=guOHBbD66BE4EuiDe8R5mDrmBJOySntdUS6Xg@mail.gmail.com>
 <CAKJS1f-VsLvQa45yPSSSagFq7VoG-MREDaddJiLt=82gX2WQcg@mail.gmail.com>
 <CAMyN-kBDks3CaALvhCd-KokpygQjO6-1pQ+vnv6Yq6EZUwccUg@mail.gmail.com>
 <CAMyN-kAO-bc5pqgRp1vZ=gFmVjbEQfM7Anikkm6PPDDAekH2Nw@mail.gmail.com>
 <CAKJS1f8Ap6cFKeBbTnHXpOcgq5iXV1UhwA7ZiP4fxxx1LArbwQ@mail.gmail.com>
 <CAMyN-kBRYy=DV3oQ0TD6pJYvNg1oHSTXdCTe94XJGOpJK1GeAA@mail.gmail.com>
 <CAMyN-kB6+KTyH3TQqoLLbNtj2t4AZr8RMHUU0yzxp429o5PyDw@mail.gmail.com>
 <CAKJS1f9+etR1VGkqp=hkGWn3EofWJVFnjKaTpHZva2PKhALdng@mail.gmail.com>
 <CAMyN-kCLxCo=rbysFRM=Ouarp+xd_b=oP26+tr1yXe3C9k89OQ@mail.gmail.com>
 <CAMyN-kDNyqy2RUGn53GnoS1shdB5TYhif9OpGFFkOb7z=za2eQ@mail.gmail.com>
 <CAKJS1f9TGbBMNsDK+wXtGpVL2=73cmB9=NBxUk9Wp9OH59kg7Q@mail.gmail.com>
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Message-ID: <0aad6687-037c-5f2f-4bbe-50f68e5c4cf6@2ndquadrant.com>
Date: Thu, 22 Nov 2018 19:03:21 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.3.0
MIME-Version: 1.0
In-Reply-To: 
 <CAKJS1f9TGbBMNsDK+wXtGpVL2=73cmB9=NBxUk9Wp9OH59kg7Q@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 11/22/18 8:41 AM, David Rowley wrote:
 >
> ...
> 
> 3. I'd rather see EnsureTidRangeSpace() keep doubling the size of the
> allocation until it reaches the required size. See how
> MakeSharedInvalidMessagesArray() does it.  Doing it this way ensures
> we always have a power of two sized array which is much nicer if we
> ever reach the palloc() limit as if the array is sized at the palloc()
> limit / 2 + 1, then if we try to double it'll fail.  Of course, it's
> unlikely to be a problem here, but... the question would be how to
> decide on the initial size.
> 

I think it kinda tries to do that in some cases, by doing this:

     *numAllocRanges *= 2;

     ...

     tidRanges = (TidRange *)
         repalloc(tidRanges,
                  *numAllocRanges * sizeof(TidRange));

The problem here is that what matters is not numAllocRanges being 2^N, 
but the number of bytes allocated being 2^N. Because that's what ends up 
in AllocSet, which keeps lists of 2^N chunks.

And as TidRange is 12B, so this is guaranteed to waste memory, because 
no matter what the first factor is, the result will never be 2^N.

regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services