Re: effective_cache_size
On 16/10/2018 17:38, Bruce Momjian wrote:
> diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
> index 2317e8b..e471d7f 100644
> --- a/src/backend/utils/misc/guc.c
> +++ b/src/backend/utils/misc/guc.c
> @@ -2987,10 +2987,9 @@ static struct config_int ConfigureNamesInt[] =
>
> {
> {"effective_cache_size", PGC_USERSET, QUERY_TUNING_COST,
> - gettext_noop("Sets the planner's assumption about the
> size of the disk cache."),
> - gettext_noop("That is, the portion of the kernel's disk
> cache that "
> - "will be used for PostgreSQL
> data files. This is measured in disk "
> - "pages, which are normally 8
> kB each."),
> + gettext_noop("Sets the planner's assumption about the
> size of the data cache."),
> + gettext_noop("That is, the size of the cache used for
> PostgreSQL data files. "
> + "This is measured in disk
> pages, which are normally 8 kB each."),
> GUC_UNIT_BLOCKS,
> },
> &effective_cache_size,
This change completely loses the context that this is the kernel's/host
system's memory size. What is "data cache"? I think this is a bad
change. I know it's confusing, but the old description at least had
some basis in terms that are known to the user.
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
Re: effective_cache_size
On Mon, Nov 5, 2018 at 12:50:01PM +0100, Peter Eisentraut wrote:
> On 16/10/2018 17:38, Bruce Momjian wrote:
> > diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
> > index 2317e8b..e471d7f 100644
> > --- a/src/backend/utils/misc/guc.c
> > +++ b/src/backend/utils/misc/guc.c
> > @@ -2987,10 +2987,9 @@ static struct config_int ConfigureNamesInt[] =
> >
> > {
> > {"effective_cache_size", PGC_USERSET, QUERY_TUNING_COST,
> > - gettext_noop("Sets the planner's assumption about the
> > size of the disk cache."),
> > - gettext_noop("That is, the portion of the kernel's disk
> > cache that "
> > -"will be used for PostgreSQL
> > data files. This is measured in disk "
> > -"pages, which are normally 8
> > kB each."),
> > + gettext_noop("Sets the planner's assumption about the
> > size of the data cache."),
> > + gettext_noop("That is, the size of the cache used for
> > PostgreSQL data files. "
> > +"This is measured in disk
> > pages, which are normally 8 kB each."),
> > GUC_UNIT_BLOCKS,
> > },
> > &effective_cache_size,
>
> This change completely loses the context that this is the kernel's/host
> system's memory size. What is "data cache"? I think this is a bad
> change. I know it's confusing, but the old description at least had
> some basis in terms that are known to the user.
Well, the change as outlined in the email is that effective_cache_size
is a combination of shared_buffers and kernel cache size, which I think
the docs now make clear. Do you have better wording for the GUC?
--
Bruce Momjian http://momjian.us
EnterpriseDB http://enterprisedb.com
+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +
Re: effective_cache_size
On Mon, Nov 5, 2018 at 3:01 PM Bruce Momjian wrote:
> > > {"effective_cache_size", PGC_USERSET, QUERY_TUNING_COST,
> > > - gettext_noop("Sets the planner's assumption about
> the size of the disk cache."),
> > > - gettext_noop("That is, the portion of the kernel's
> disk cache that "
> > > -"will be used for
> PostgreSQL data files. This is measured in disk "
> > > -"pages, which are
> normally 8 kB each."),
> > > + gettext_noop("Sets the planner's assumption about
> the size of the data cache."),
> > > + gettext_noop("That is, the size of the cache used
> for PostgreSQL data files. "
> > > +"This is measured in disk
> pages, which are normally 8 kB each."),
>
...
> Well, the change as outlined in the email is that effective_cache_size
> is a combination of shared_buffers and kernel cache size, which I think
> the docs now make clear. Do you have better wording for the GUC?
>
Maybe it's better to use this phrase, "a combination of shared_buffers and
kernel cache size"?
Or: "a combination of shared_buffers and estimated kernel cache size".
The phrase "the size of the cache" might be very confusing indeed – it
sounds like it's about
some single cache, while it's about the combination of two.
Maybe it's also worth to mention that the fact that some pages might be
cached twice –
in OS cache + in Postgres shared buffers – should be ignored, when choosing
the proper value
for effective_cache_size? I think this would finally eliminate the
possibility of confusion.
I see now, the docs chapter "19.7. Query Planning" has "some data might
exist in both places"
– this is great, since confusion here is not uncommon. It's worth to
propagate this change
everywhere where effective_cache_size is explained.
Nik
Re: effective_cache_size
On Mon, Nov 5, 2018 at 03:23:03PM +0300, Nikolay Samokhvalov wrote:
>
>
> On Mon, Nov 5, 2018 at 3:01 PM Bruce Momjian wrote:
>
> > > {"effective_cache_size", PGC_USERSET, QUERY_TUNING_COST,
> > > - gettext_noop("Sets the planner's assumption about
> the size of the disk cache."),
> > > - gettext_noop("That is, the portion of the kernel's
> disk cache that "
> > > - "will be used for
> PostgreSQL data files. This is measured in disk "
> > > - "pages, which are
> normally
> 8 kB each."),
> > > + gettext_noop("Sets the planner's assumption about
> the size of the data cache."),
> > > + gettext_noop("That is, the size of the cache used
> for PostgreSQL data files. "
> > > + "This is measured in disk
> pages, which are normally 8 kB each."),
>
> ...
>
> Well, the change as outlined in the email is that effective_cache_size
> is a combination of shared_buffers and kernel cache size, which I think
> the docs now make clear. Do you have better wording for the GUC?
>
>
> Maybe it's better to use this phrase, "a combination of shared_buffers and
> kernel cache size"?
> Or: "a combination of shared_buffers and estimated kernel cache size".
Well, here are the lines in guc.c:
gettext_noop("Sets the planner's assumption about the size of the
data cache."),
gettext_noop("That is, the size of the cache used for PostgreSQL
data files. "
"This is measured in disk pages, which are normally 8
kB each."),
The first line is already on the long side compared to other entries, as
is the second gettext_noop(). How long can we make it without being too
long? Is this really the place to explain this distinction? I thought
about these issues when I wrote this patch? You can come to a different
conclusion, but please consider these issues.
--
Bruce Momjian http://momjian.us
EnterpriseDB http://enterprisedb.com
+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +
Re: effective_cache_size
On 2018-Nov-05, Bruce Momjian wrote:
> Well, here are the lines in guc.c:
>
> gettext_noop("Sets the planner's assumption about the size of the
> data cache."),
> gettext_noop("That is, the size of the cache used for PostgreSQL
> data files. "
> "This is measured in disk pages, which are normally
> 8 kB each."),
I suggest "the size of data caches", plural, in the first line (two
letters shorter, since I lost the article). And in the second line, use
"... the size of the combined caches used for Pg data files, including
both the kernel cache and shared buffers" -- a few words longer, which
seems worth it to me.
--
Álvaro Herrerahttps://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
