Обсуждение: Parallel VACUUM in feature matrix on website

Поиск
Список
Период
Сортировка

Parallel VACUUM in feature matrix on website

От
Peter Geoghegan
Дата:
It's not really parallel VACUUM, is it? So why call it parallel VACUUM?

--
Regards,
Peter Geoghegan


Re: Parallel VACUUM in feature matrix on website

От
Peter Geoghegan
Дата:
Also, I think "WAL Buffer auto-tuning" should be cut. It doesn't
warrant inclusion here. There are one or two other items that should
be pruned too, but less obviously so.

--
Regards,
Peter Geoghegan


Re: Parallel VACUUM in feature matrix on website

От
Thom Brown
Дата:
On 27 September 2015 at 20:33, Peter Geoghegan
<peter.geoghegan86@gmail.com> wrote:
> It's not really parallel VACUUM, is it? So why call it parallel VACUUM?

It allows multiple vacuum processes to be initiated from vacuumdb at
the same time (i.e. in parallel), although not for the same tables.
What do you propose would be more appropriate?

Thom


Re: Parallel VACUUM in feature matrix on website

От
Peter Geoghegan
Дата:
On Sun, Sep 27, 2015 at 12:39 PM, Thom Brown <thom@linux.com> wrote:
> It allows multiple vacuum processes to be initiated from vacuumdb at
> the same time (i.e. in parallel), although not for the same tables.
> What do you propose would be more appropriate?

"Parallel vacuumdb jobs" seems better. I have seen cases where the
existing wording caused confusion.

I think that we will probably have actual parallel vacuum within the
next couple of releases (probably vacuuming multiple indexes at the
same time, using a list of TIDs in shared memory). The messaging here
is important.

--
Regards,
Peter Geoghegan


Re: Parallel VACUUM in feature matrix on website

От
Guillaume Lelarge
Дата:
2015-09-27 21:39 GMT+02:00 Thom Brown <thom@linux.com>:
On 27 September 2015 at 20:33, Peter Geoghegan
<peter.geoghegan86@gmail.com> wrote:
> It's not really parallel VACUUM, is it? So why call it parallel VACUUM?

It allows multiple vacuum processes to be initiated from vacuumdb at
the same time (i.e. in parallel), although not for the same tables.
What do you propose would be more appropriate?


Sounds more like a parallel vacuumdb, than a parallel VACUUM, to me :)


--

Re: Parallel VACUUM in feature matrix on website

От
Thom Brown
Дата:
On 27 September 2015 at 20:43, Peter Geoghegan
<peter.geoghegan86@gmail.com> wrote:
> On Sun, Sep 27, 2015 at 12:39 PM, Thom Brown <thom@linux.com> wrote:
>> It allows multiple vacuum processes to be initiated from vacuumdb at
>> the same time (i.e. in parallel), although not for the same tables.
>> What do you propose would be more appropriate?
>
> "Parallel vacuumdb jobs" seems better. I have seen cases where the
> existing wording caused confusion.
>
> I think that we will probably have actual parallel vacuum within the
> next couple of releases (probably vacuuming multiple indexes at the
> same time, using a list of TIDs in shared memory). The messaging here
> is important.

Okay, I've changed it to "Parallel vacuumdb jobs".

Thom


Re: Parallel VACUUM in feature matrix on website

От
Peter Geoghegan
Дата:
On Sun, Sep 27, 2015 at 1:06 PM, Thom Brown <thom@linux.com> wrote:
> Okay, I've changed it to "Parallel vacuumdb jobs".

Thanks.


--
Regards,
Peter Geoghegan


Re: Parallel VACUUM in feature matrix on website

От
Josh Berkus
Дата:
On 09/27/2015 12:39 PM, Peter Geoghegan wrote:
> Also, I think "WAL Buffer auto-tuning" should be cut. It doesn't
> warrant inclusion here. There are one or two other items that should
> be pruned too, but less obviously so.
>

I disagree here; I think that anything which eliminates a need for
manual tuning is a significant feature.  Expecially if you're looking at
the chart and trying to remember "hey, do I need to tune checkpoint
segments on this version?"

--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com


Re: Parallel VACUUM in feature matrix on website

От
Darren Duncan
Дата:
On 2015-09-28 10:22 AM, Josh Berkus wrote:
> On 09/27/2015 12:39 PM, Peter Geoghegan wrote:
>> Also, I think "WAL Buffer auto-tuning" should be cut. It doesn't
>> warrant inclusion here. There are one or two other items that should
>> be pruned too, but less obviously so.
>
> I disagree here; I think that anything which eliminates a need for
> manual tuning is a significant feature.  Expecially if you're looking at
> the chart and trying to remember "hey, do I need to tune checkpoint
> segments on this version?"

I totally agree with Josh.  Especially for newer or less-savvy users, this means
it just became easier to use Postgres in a performant way, and ease of use gets
customers. -- Darren Duncan