[pgpool-hackers: 3082] Re: Time to remove replication_mode ?
ishii at sraoss.co.jp
Sat Oct 20 11:21:12 JST 2018
Sorry for late reply.
> On 08/15/2018 10:21 PM, Tatsuo Ishii wrote:
>>> I know this is controversial, but is it time to remove
>>> replication_mode and friends from pgpool ?
>>> master_slave_mode has been around for a while, and by removing
>>> replication_mode from the configuration it will force people to setup
>>> streaming or logical replication instead.
>>> This will hopefully lead to fewer data corruption cases, since all
>>> community supported PostgreSQL versions has streaming replication now.
>>> I would be willing to write a patch for this, or at least a patch that
>>> let people know to look at master_slave_mode instead.
>>> This is of course for pgpool-4.
>>> Feedback most welcome.
>> I think we should not remove replication_mode for now.
>> PostgreSQL developers are now trying to implement replication/sharding
>> clusters using FDW. In the architecture, FDW sends queries to each
>> cluster node. But right now it lacks the feature of distributed
>> transaction manager, and it might be possible for Pgpool-II do
>> something in this area possibly with modified replication_mode.
>> Of course my idea is pretty vague at this point but at least I do not
>> want to remove replication_mode until the idea becomes clear and find
>> replication_mode is useless in this area.
> Ok, understood.
> Would the attached patch be ok to include ?
> Thanks for your feedback !
Just recommending master slave mode for a substituion of native
replication mode is not always appropreate. Replication mode could be
unconditionaly replaced with streaming replication mode iff:
1) PostgreSQL version is 10.0 or higher.
2) synchronous_commit is configured as remote_apply.
If those conditions are not satisfied, streaming replication users
have to be aware of its down side, e.g. replication delay.
I appreciate if you are going to consider those point in the patch.
SRA OSS, Inc. Japan
More information about the pgpool-hackers