Re: Patch for fail-back without fresh backup

From: Sawada Masahiko <sawada(dot)mshk(at)gmail(dot)com>
To: Pavan Deolasee <pavan(dot)deolasee(at)gmail(dot)com>
Cc: Fujii Masao <masao(dot)fujii(at)gmail(dot)com>, Samrat Revagade <revagade(dot)samrat(at)gmail(dot)com>, Peter Eisentraut <peter_e(at)gmx(dot)net>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Patch for fail-back without fresh backup
Date: 2013-10-04 04:46:14
Message-ID: CAD21AoCp65kw4m7mbLmQoQRTC5tcUEMd5DZ5-NS8hu2RoGaTaA@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Fri, Sep 27, 2013 at 6:44 PM, Sawada Masahiko <sawada(dot)mshk(at)gmail(dot)com> wrote:
> On Fri, Sep 27, 2013 at 5:18 PM, Pavan Deolasee
> <pavan(dot)deolasee(at)gmail(dot)com> wrote:
>> On Fri, Sep 27, 2013 at 1:28 PM, Sawada Masahiko <sawada(dot)mshk(at)gmail(dot)com>
>> wrote:
>>>
>>>
>>> >
>>>
>>> Thank you for comment. I think it is good simple idea.
>>> In your opinion, if synchronous_transfer is set 'all' and
>>> synchronous_commit is set 'on',
>>> the master wait for data flush eve if user sets synchronous_commit to
>>> 'local' or 'off'.
>>> For example, when user want to do transaction early, user can't do this.
>>> we leave the such situation as constraint?
>>>
>>
>> No, user can still override the transaction commit point wait. So if
>>
>> synchronous_transfer is set to "all":
>> - If synchronous_commit is ON - wait at all points
>> - If synchronous_commit is OFF - wait only at buffer flush (and other
>> related to failback safety) points
>>
>> synchronous_transfer is set to "data_flush":
>> - If synchronous_commit is either ON o OFF - do not wait at commit points,
>> but wait at all other points
>>
>> synchronous_transfer is set to "commit":
>> - If synchronous_commit is ON - wait at commit point
>> - If synchronous_commit is OFF - do not wait at any point
>>
>
> Thank you for explain. Understood.
> if synchronous_transfer is set 'all' and user changes
> synchronous_commit to 'off'( or 'local') at a transaction,
> the master server wait at buffer flush, but doesn't wait at commit
> points. Right?
>
> In currently patch, synchronous_transfer works in cooperation with
> synchronous_commit.
> But if user changes synchronous_commit at a transaction, they are not
> in cooperation.
> So, your idea might be better than currently behaviour of synchronous_transfer.

I attached the v11 patch which have fixed following contents.
- synchronous_transfer controls to wait at only data flush level,
synchronous_commit controls to wait at commit level. ( Based on
Pavan suggestion)
- If there are no sync replication standby name, both
synchronous_commit and synchronous_transfer
don't work.
- Fixed that we didn't support failback-safe standby.
Previous patch can not support failback-safe standby. Because the
patch doesn't wait at FlushBuffer
which is called by autovacuum.

So, if user want to do transaction early temporarily, user need to
change the synchronous_transfer value and reload
postgresql.conf.

Regards,

-------
Sawada Masahiko

Attachment Content-Type Size
synchronous_transfer_v11.patch application/octet-stream 31.5 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Fujii Masao 2013-10-04 05:19:27 Re: Compression of full-page-writes
Previous Message Amit Kapila 2013-10-04 04:19:04 Re: Suggestion: Issue warning when calling SET TRANSACTION outside transaction block