Re: Speedup twophase transactions

> On Apr 2, 2016, at 3:14 AM, Michael Paquier <michael.paquier@gmail.com> wrote:
>
> On Fri, Apr 1, 2016 at 10:53 PM, Stas Kelvich <s.kelvich@postgrespro.ru> wrote:
>> I wrote:
>>> While testing the patch, I found a bug in the recovery conflict code
>>> path. You can do the following to reproduce it:
>>> 1) Start a master with a standby
>>> 2) prepare a transaction on master
>>> 3) Stop immediate on standby to force replay
>>> 4) commit prepare transaction on master
>>> 5) When starting the standby, it remains stuck here:
>>
>> Hm, I wasn’t able to reproduce that. Do you mean following scenario or am I missing something?
>>
>> (async replication)
>>
>> $node_master->psql('postgres', "
>>        begin;
>>        insert into t values (1);
>>        prepare transaction 'x';
>> ");
>> $node_slave->teardown_node;
>> $node_master->psql('postgres',"commit prepared 'x'");
>> $node_slave->start;
>> $node_slave->psql('postgres',"select count(*) from pg_prepared_xacts", stdout => \$psql_out);
>> is($psql_out, '0', "Commit prepared on master while slave is down.");
>
> Actually, not exactly, the transaction prepared on master created a
> table. Sorry for the lack of precisions in my review.

Sorry for delay.

Actually I can’t reproduce that again, tried with following tx:

begin;
    insert into t values(0);
    create table t1(id int);
    insert into t1 values(1);
    create table t2(id int);
    insert into t2 values(2);
    savepoint s1;
    drop table t1;
    select * from t for update;
    select * from t2 for share;
prepare transaction 'x’;



--
Stas Kelvich
Postgres Professional: http://www.postgrespro.com
Russian Postgres Company