Message-ID: <8231183d-1077-5394-1ffa-0c9d8c3905b0@enterprisedb.com>
Date: Tue, 21 Dec 2021 01:53:02 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.2.0
Subject: Re: sequences vs. synchronous replication
Content-Language: en-US
From: Tomas Vondra <tomas.vondra@enterprisedb.com>
To: Peter Eisentraut <peter.eisentraut@enterprisedb.com>,
 Tom Lane <tgl@sss.pgh.pa.us>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
References: <712cad46-a9c8-1389-aef8-faf0203c9be9@enterprisedb.com>
 <1452362.1639803168@sss.pgh.pa.us>
 <38a6a779-ee8d-de3c-fbd7-709285be88fa@enterprisedb.com>
 <7869761b-fdc7-6f7d-dc95-9ba8315a8ecf@enterprisedb.com>
 <1637476.1639862832@sss.pgh.pa.us>
 <8e24a470-b092-3c9f-0614-acd9df3f724f@enterprisedb.com>
 <f4dd81e9-fc9e-3b71-0a8a-4a5ae345331d@enterprisedb.com>
 <a5303964-717e-9cb0-52cf-7a4b7d16adc8@enterprisedb.com>
In-Reply-To: <a5303964-717e-9cb0-52cf-7a4b7d16adc8@enterprisedb.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: 
 <https://www.postgresql.org/message-id/8231183d-1077-5394-1ffa-0c9d8c3905b0%40enterprisedb.com>
Precedence: bulk

On 12/20/21 17:40, Tomas Vondra wrote:
> On 12/20/21 15:31, Peter Eisentraut wrote:
>> On 18.12.21 22:48, Tomas Vondra wrote:
>>> What do you mean by "not caching unused sequence numbers"? Reducing 
>>> SEQ_LOG_VALS to 1, i.e. WAL-logging every sequence increment?
>>>
>>> That'd work, but I wonder how significant the impact will be. It'd 
>>> bet it hurts the patch adding logical decoding of sequences quite a bit.
>>
>> It might be worth testing.  This behavior is ancient and has never 
>> really been scrutinized since it was added.
>>
> 
> OK, I'll do some testing to measure the overhead, and I'll see how much 
> it affects the sequence decoding patch.
> 

OK, I did a quick test with two very simple benchmarks - simple select 
from a sequence, and 'pgbench -N' on scale 1. Benchmark was on current 
master, patched means SEQ_LOG_VALS was set to 1.

Average of 10 runs, each 30 seconds long, look like this:

1) select nextval('s');

      clients          1         4
     ------------------------------
      master       39497    123137
      patched       6813     18326
     ------------------------------
      diff          -83%      -86%

2) pgbench -N

      clients          1         4
     ------------------------------
      master        2935      9156
      patched       2937      9100
     ------------------------------
      diff            0%        0%


Clearly the extreme case (1) is hit pretty bad, while the much mure 
likely workload (2) is almost unaffected.


I'm not sure what conclusion to make from this, but assuming almost no 
one does just nextval calls, it should be acceptable.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company