Postgres Pro
Downloads
Home   >   mailing lists

Thread: pgsql: Update item description: * Consider having the background writer

pgsql: Update item description: * Consider having the background writer

From
momjian@postgresql.org (Bruce Momjian)
Date:
Log Message:
-----------
Update item description:

* Consider having the background writer update the transaction status
  hint bits before writing out the page

  Implementing this requires the background writer to have access to system
  catalogs and the transaction status log.

Modified Files:
--------------
    pgsql/doc:
        TODO (r1.2230 -> r1.2231)
        (http://developer.postgresql.org/cvsweb.cgi/pgsql/doc/TODO?r1=1.2230&r2=1.2231)
    pgsql/doc/src/FAQ:
        TODO.html (r1.739 -> r1.740)
        (http://developer.postgresql.org/cvsweb.cgi/pgsql/doc/src/FAQ/TODO.html?r1=1.739&r2=1.740)

Re: pgsql: Update item description: * Consider having the background writer

From
Simon Riggs
Date:
On Mon, 2007-12-17 at 01:40 +0000, Bruce Momjian wrote:
> Log Message:
> -----------
> Update item description:
>
> * Consider having the background writer update the transaction status
>   hint bits before writing out the page
>
>   Implementing this requires the background writer to have access to system
>   catalogs and the transaction status log.

Not very sure why you've added this one to the TODO list?

I raised it as not-a-good idea and I think Tom and Alvaro agreed. i.e. I
mentioned it in order to kill it. It's a ton of work for what very well
might be overhead rather than gain.

--
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com

Re: pgsql: Update item description: * Considerhaving the background writer

From
Heikki Linnakangas
Date:
Simon Riggs wrote:
> On Mon, 2007-12-17 at 01:40 +0000, Bruce Momjian wrote:
>> Log Message:
>> -----------
>> Update item description:
>>
>> * Consider having the background writer update the transaction status
>>   hint bits before writing out the page
>>
>>   Implementing this requires the background writer to have access to system
>>   catalogs and the transaction status log.
>
> Not very sure why you've added this one to the TODO list?
>
> I raised it as not-a-good idea and I think Tom and Alvaro agreed. i.e. I
> mentioned it in order to kill it. It's a ton of work for what very well
> might be overhead rather than gain.

I actually do feel it might be valuable. All our opinions are based on
gut-feeling, though, so would need benchmarking to see if it's really
worth it or not.

I think we could get away without accessing system catalogs in bgwriter,
if we flagged heap pages as such in ReadBuffer. Just a thought..

--
   Heikki Linnakangas
   EnterpriseDB   http://www.enterprisedb.com

Re: pgsql: Update item description: * Considerhaving the background writer

From
Simon Riggs
Date:
On Mon, 2007-12-17 at 22:46 +0000, Heikki Linnakangas wrote:

> I actually do feel it might be valuable. All our opinions are based on
> gut-feeling, though, so would need benchmarking to see if it's really
> worth it or not.
>
> I think we could get away without accessing system catalogs in bgwriter,
> if we flagged heap pages as such in ReadBuffer. Just a thought..

Well, if you or Bruce or anybody else thinks it should be on the TODO
then a case should be made for it. I have no doubt you could write it,
but the point is why would you? Or why would you want somebody else to
if you're not going to?

Everybody that spoke on the original thread was against it and I
specifically mentioned it to kill the idea. Its just plain weird that
multiple people have given arguments as to why its a bad idea and yet it
goes straight onto the TODO. LOL.

If that criteria is used to add stuff to the TODO then it would be
bursting with crappy ideas...

--
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com

Re: pgsql: Update item description: * Considerhaving the background writer

From
Bruce Momjian
Date:
Simon Riggs wrote:
> On Mon, 2007-12-17 at 22:46 +0000, Heikki Linnakangas wrote:
>
> > I actually do feel it might be valuable. All our opinions are based on
> > gut-feeling, though, so would need benchmarking to see if it's really
> > worth it or not.
> >
> > I think we could get away without accessing system catalogs in bgwriter,
> > if we flagged heap pages as such in ReadBuffer. Just a thought..
>
> Well, if you or Bruce or anybody else thinks it should be on the TODO
> then a case should be made for it. I have no doubt you could write it,
> but the point is why would you? Or why would you want somebody else to
> if you're not going to?
>
> Everybody that spoke on the original thread was against it and I
> specifically mentioned it to kill the idea. Its just plain weird that
> multiple people have given arguments as to why its a bad idea and yet it
> goes straight onto the TODO. LOL.
>
> If that criteria is used to add stuff to the TODO then it would be
> bursting with crappy ideas...

I saw people saying it could be complex, not that it was a bad idea, and
I just added a description --- the idea was already on the TODO list.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://postgres.enterprisedb.com

  + If your life is a hard drive, Christ can be your backup. +