Thread: pgsql: Reorder 9.3 release note items
Reorder 9.3 release note items Typo fix from David Fetter. Branch ------ master Details ------- http://git.postgresql.org/pg/commitdiff/ef3defcf0d507abb75d9e9696f11b44f4fc09291 Modified Files -------------- doc/src/sgml/release-9.3.sgml | 138 ++++++++++++++++++++--------------------- 1 files changed, 67 insertions(+), 71 deletions(-)
On 21 April 2013 05:57, Bruce Momjian <bruce@momjian.us> wrote: > Reorder 9.3 release note items A few corrections, additions and clarifications. Corrections * Allow recovery.conf to be relocated using configuration variable recovery_config_directory Patch was revoked... * Need to mention the fast promote feature for standbys, which allows us to skip shutdown checkpoint, reducing failover time from minutes to seconds in busier databases * Correct predicate locking for DROP INDEX CONCURRENTLY - bugfix not a feature * Allow CREATE TABLE to succeed for a non-existent schema - should say CREATE TABLE IF NOT EXISTS in that para and the one below it * Previously functions already run in the current session ignored changes (what does that mean?? missing text) * Improve memory usage for in-memory sorts - IMHO this should read "Allow in-memory sorts to use their full memory allocation", which then explains why users may see higher memory usage and why they should think about altering settings Additions * I think it would be useful to say that pg_upgrade is now much faster because of the aggregate effect of the individual optimisations mentioned in the performance section. * "Require superuser privileges to set commit_delay" Meaning of commit_delay has changed and users should review their settings, if used * Add isolation tests for CREATE INDEX CONCURRENTLY - Abhijit Menon-Sen More info required... * Need to say why Pavan's work on VACUUM is related to performance * Need to say what effect Andres' patch to make HOT work on catalog tables will have * "Restructure WAL files to use a more compact storage format" - begs the question how much more compact? * Restructure WAL files to better handle timeline changes during recovery - need to explain "better" Thanks -- Simon Riggs http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
OK, all changes made and committed. Thanks. (You can see the details
in the git commit.)
For this item:
* Correct predicate locking for DROP INDEX CONCURRENTLY -
bugfix not a feature
did you have new text. We do mention bug fixes. I changed "Correct" to
"Fix" in a later commit.
---------------------------------------------------------------------------
On Sun, Apr 21, 2013 at 08:57:08AM +0100, Simon Riggs wrote:
> On 21 April 2013 05:57, Bruce Momjian <bruce@momjian.us> wrote:
>
> > Reorder 9.3 release note items
>
> A few corrections, additions and clarifications.
>
> Corrections
>
> * Allow recovery.conf to be relocated using configuration variable
> recovery_config_directory
> Patch was revoked...
>
> * Need to mention the fast promote feature for standbys, which allows
> us to skip shutdown checkpoint, reducing failover time from minutes to
> seconds in busier databases
>
> * Correct predicate locking for DROP INDEX CONCURRENTLY - bugfix not a feature
>
> * Allow CREATE TABLE to succeed for a non-existent schema - should say
> CREATE TABLE IF NOT EXISTS in that para and the one below it
>
> * Previously functions already run in the current session ignored
> changes (what does that mean?? missing text)
>
> * Improve memory usage for in-memory sorts - IMHO this should read
> "Allow in-memory sorts to use their full memory allocation", which
> then explains why users may see higher memory usage and why they
> should think about altering settings
>
> Additions
>
> * I think it would be useful to say that pg_upgrade is now much faster
> because of the aggregate effect of the individual optimisations
> mentioned in the performance section.
>
> * "Require superuser privileges to set commit_delay"
> Meaning of commit_delay has changed and users should review their
> settings, if used
>
> * Add isolation tests for CREATE INDEX CONCURRENTLY - Abhijit Menon-Sen
>
>
> More info required...
>
> * Need to say why Pavan's work on VACUUM is related to performance
>
> * Need to say what effect Andres' patch to make HOT work on catalog
> tables will have
>
> * "Restructure WAL files to use a more compact storage format" - begs
> the question how much more compact?
>
> * Restructure WAL files to better handle timeline changes during
> recovery - need to explain "better"
>
> Thanks
>
> --
> Simon Riggs http://www.2ndQuadrant.com/
> PostgreSQL Development, 24x7 Support, Training & Services
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +
On 2013-04-21 09:06:51 -0400, Bruce Momjian wrote: > > OK, all changes made and committed. Thanks. (You can see the details > in the git commit.) > > For this item: > > * Correct predicate locking for DROP INDEX CONCURRENTLY - > bugfix not a feature > > did you have new text. We do mention bug fixes. I changed "Correct" to > "Fix" in a later commit. But we usually don't if they have been fixed in a previous point release in the back branches, right? Greetings, Andres Freund -- Andres Freund http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
On Sun, Apr 21, 2013 at 03:09:19PM +0200, Andres Freund wrote: > On 2013-04-21 09:06:51 -0400, Bruce Momjian wrote: > > > > OK, all changes made and committed. Thanks. (You can see the details > > in the git commit.) > > > > For this item: > > > > * Correct predicate locking for DROP INDEX CONCURRENTLY - > > bugfix not a feature > > > > did you have new text. We do mention bug fixes. I changed "Correct" to > > "Fix" in a later commit. > > But we usually don't if they have been fixed in a previous point release > in the back branches, right? Oh, that was backpatched? OK, removed. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + It's impossible for everything to be true. +
On 2013-04-21 04:57:22 +0000, Bruce Momjian wrote: > Reorder 9.3 release note items > > Typo fix from David Fetter. > > Branch > ------ > master > > Details > ------- > http://git.postgresql.org/pg/commitdiff/ef3defcf0d507abb75d9e9696f11b44f4fc09291 > > Modified Files > -------------- > doc/src/sgml/release-9.3.sgml | 138 ++++++++++++++++++++--------------------- > 1 files changed, 67 insertions(+), 71 deletions(-) > > > -- > Sent via pgsql-committers mailing list (pgsql-committers@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-committers * Use a binary heap for Merge-Append operations (Abhijit Menon-Sen) Implement a generic binary heap and use it for Merge-Append operations which previously used a ad-hoc version. * Dramatically reduce System V shared memory usage (Robert Haas) s/eleminates/reduces/ in "This eliminates the requirement of increasing System V shared memory kernel parameters.". There are still quite some usecases where some of those need to be touched. * Improve performance of streaming replication log shipping ... with synchronous_commit = off * Store WAL in a continuous stream, rather than skipping the last 16MB segment every 4GB (Heikki Linnakangas) Should be marked as a potential backward compatibility break, some tools might need to adapt their logic. * Add IF NOT EXISTS clause to CREATE SCHEMA (FabrÃzio de Royes Mello) another broken name :( * Add pg_dump --jobs to dump in parallel (Joachim Wieland) Should we mention that this is only working for directory mode dumps? Some users might be a bit confused otherwise. * Add infrastructure to better support plug-in background worker processes (Álvaro Herrera) ... configurable via shared_preload_libraries * Split out WAL reading as an independent facility Should mention Heikki as primary author instead of me. Greetings, Andres Freund -- Andres Freund http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
On Sun, Apr 21, 2013 at 03:22:38PM +0200, Andres Freund wrote:
> * Use a binary heap for Merge-Append operations (Abhijit Menon-Sen)
> Implement a generic binary heap and use it for Merge-Append operations
> which previously used a ad-hoc version.
OK, text updated to:
Implement a generic binary heap and use it for Merge-Append
operations (Abhijit Menon-Sen)
> * Dramatically reduce System V shared memory usage (Robert Haas)
> s/eleminates/reduces/ in "This eliminates the requirement of increasing
> System V shared memory kernel parameters.". There are still quite some
> usecases where some of those need to be touched.
Well, you still might need to increase the number of semaphors, but not
shared memory.
> * Improve performance of streaming replication log shipping
> ... with synchronous_commit = off
New text is, based on Simon's request about adding failover speedup:
Improve performance of streaming replication log shipping and
standby promotion (Andres Freund, Simon Riggs)
It is worth breaking those out into two items?
> * Store WAL in a continuous stream, rather than skipping the last 16MB
> segment every 4GB (Heikki Linnakangas)
> Should be marked as a potential backward compatibility break, some tools
> might need to adapt their logic.
OK, added a marker about that.
>
> * Add IF NOT EXISTS clause to CREATE SCHEMA (FabrÃzio de Royes Mello)
> another broken name :(
Fixed (found mention in 9.2 release notes with SGML entities).
> * Add pg_dump --jobs to dump in parallel (Joachim Wieland)
>
> Should we mention that this is only working for directory mode dumps?
> Some users might be a bit confused otherwise.
Yes, added mention.
> * Add infrastructure to better support plug-in background worker processes
> (Álvaro Herrera)
> ... configurable via shared_preload_libraries
I am going to link to our docs so they can see such details.
> * Split out WAL reading as an independent facility
> Should mention Heikki as primary author instead of me.
Done.
Changes applied.
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +
On Sun, Apr 21, 2013 at 12:57 AM, Simon Riggs <simon@2ndquadrant.com> wrote: > * Improve memory usage for in-memory sorts - IMHO this should read > "Allow in-memory sorts to use their full memory allocation", which > then explains why users may see higher memory usage and why they > should think about altering settings +1. As I believe you pointed out at the time of the development of this feature, it has the potential to greatly increase the amount of memory allocated by tuplesort, but only within the actual bounds specified by the DBA. There needs to be a clear warning of that - it's easy to imagine that DBAs are specifying a high value of work_mem to compensate for the previous implementation's tendency to under provision. > * "Require superuser privileges to set commit_delay" > Meaning of commit_delay has changed and users should review their > settings, if used +1 -- Peter Geoghegan
On 21 April 2013 15:09, Bruce Momjian <bruce@momjian.us> wrote: > New text is, based on Simon's request about adding failover speedup: > > Improve performance of streaming replication log shipping and > standby promotion (Andres Freund, Simon Riggs) Um, it's hard to see how that text relates to what I requested. It's an important thing for High Availability that failover is now fast; a feature we have been waiting for some time. The point is not that the performance is improved, the point is we can now achieve 5 nines availability, whereas before we couldn't. > It is worth breaking those out into two items? Yes, they are completely unrelated. -- Simon Riggs http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
On Sun, Apr 21, 2013 at 08:22:38PM +0100, Simon Riggs wrote:
> On 21 April 2013 15:09, Bruce Momjian <bruce@momjian.us> wrote:
>
> > New text is, based on Simon's request about adding failover speedup:
> >
> > Improve performance of streaming replication log shipping and
> > standby promotion (Andres Freund, Simon Riggs)
>
> Um, it's hard to see how that text relates to what I requested.
>
> It's an important thing for High Availability that failover is now
> fast; a feature we have been waiting for some time. The point is not
> that the performance is improved, the point is we can now achieve 5
> nines availability, whereas before we couldn't.
>
> > It is worth breaking those out into two items?
>
> Yes, they are completely unrelated.
OK, new text:
<listitem>
<para>
Improve performance of non-synchronous streaming log shipping (Andres Freund)
</para>
</listitem>
<listitem>
<para>
Allow much faster fail-over to a standby server (Simon Riggs)
</para>
</listitem>
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +
On 2013-04-21 21:04:17 -0400, Bruce Momjian wrote: > On Sun, Apr 21, 2013 at 08:22:38PM +0100, Simon Riggs wrote: > > On 21 April 2013 15:09, Bruce Momjian <bruce@momjian.us> wrote: > > > > > New text is, based on Simon's request about adding failover speedup: > > > > > > Improve performance of streaming replication log shipping and > > > standby promotion (Andres Freund, Simon Riggs) > > > > Um, it's hard to see how that text relates to what I requested. > > > > It's an important thing for High Availability that failover is now > > fast; a feature we have been waiting for some time. The point is not > > that the performance is improved, the point is we can now achieve 5 > > nines availability, whereas before we couldn't. > > > > > It is worth breaking those out into two items? > > > > Yes, they are completely unrelated. > > OK, new text: > > <listitem> > <para> > Improve performance of non-synchronous streaming log shipping (Andres Freund) > </para> > </listitem> Its about synchronous_commit = off of which is more than just non-synchronous replication. How about * Greatly reduce replication delay in streaming replication when synchronous_commit=off (or asynchronous commits) are used locally > <listitem> > <para> > Allow much faster fail-over to a standby server (Simon Riggs) > </para> > </listitem> I think this should also mention Kyotaro Horiguchi (c.f. fd4ced5230162b50a5c9d33b4bf9cfb1231aa62e) Greetings, Andres Freund -- Andres Freund http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
On Mon, Apr 22, 2013 at 10:15:58AM +0200, Andres Freund wrote:
> > <listitem>
> > <para>
> > Improve performance of non-synchronous streaming log shipping (Andres Freund)
> > </para>
> > </listitem>
>
> Its about synchronous_commit = off of which is more than just
> non-synchronous replication. How about
> * Greatly reduce replication delay in streaming replication when
> synchronous_commit=off (or asynchronous commits) are used locally
>
>
> > <listitem>
> > <para>
> > Allow much faster fail-over to a standby server (Simon Riggs)
> > </para>
> > </listitem>
>
> I think this should also mention Kyotaro Horiguchi
> (c.f. fd4ced5230162b50a5c9d33b4bf9cfb1231aa62e)
Thanks, updated:
Improve performance of streaming log shipping with
synchronous_commit disabled (Andres Freund)
Allow much faster promotion of a streaming standby to primary (Simon
Riggs, Kyotaro Horiguchi)
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +