On Sun, Sep 11, 2016 at 12:39 AM, Andres Freund <andres@anarazel.de> wrote:
> On 2016-09-10 17:23:21 +0530, Amit Kapila wrote:
>> >
>>
>> I may be missing something here, but why would it contend on a lock,
>> as per locking scheme proposed by Alvaro, access to sequence object
>> will need a share lock on buffer page.
>
> To make checkpointing/bgwriter work correctly we need exclusive locks on
> pages while writing..., or some new lock level preventing the page from
> being written out, while "shared dirtying" locks are being held.
>
Right and I think you have a very valid concern, but if we think that
storing multiple sequences on a same page is a reasonable approach,
then we can invent some locking mechanism as indicated by you such
that two writes on same page won't block each other, but they will be
blocked with bgwriter/checkpointer.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com