Skip to content

Wöchentlicher PostgreSQL Newsletter - 27. März 2011

Der Originalartikel befindet sich unter:

== Wöchentlicher PostgreSQL Newsletter - 27. März 2011 ==

Greg Smith veröffentlichte eine Revision des Performance Farm Codes,
basierend auf Andrew Dunstans Buildfarm Code.

pgbr findet in Sao Paulo, Brazilien, am 3. und 4. November 2011 statt.

PG Session 2 über PostGIS findet am 23. Juni in Paris statt.
Der Call for Papers ist jetzt offen.

== PostgreSQL Produkt Neuigkeiten ==

Dumbo 0.50, ein Terminal Client für PostgreSQL, ist erschinenen.

MyJSQLView 3.26, ein GUI Werkzeug welches mit PostgreSQL verwendet
werden kann, ist erschienen.

Npgsql 2.0.12beta1, ein .NET Treiber für PostgreSQL, ist erschienen.

== PostgreSQL Jobs im März ==

== PostgreSQL Lokal ==

Es wird eine umfangreiche PostgreSQL Präsenz auf der MySQL Konferenz
und Expo am am 11-14. April 2011 in Santa Clara, Kalifornien, geben.

Das Open Database Camp findet vom 7. bis 9. Mai 2011 in Sardinien,
Italien statt.

PGCon findet am 19. und 20. Mai 2011 an der Universität
von Ottawa statt, vorher gibt es am 17. und 18. Mai
zwei Tage mit Trainings.

== PostgreSQL in den News ==

Planet PostgreSQL:

Dieser wöchentliche PostgreSQL Newsletter wurde erstellt von David Fetter.

Sende Neuigkeiten und Ankündigungen bis Sonntag, 15 Uhr Pazifischer
Zeit. Bitte sende englische Beiträge an, deutsche an, italienische an, spanische an

== Reviews ==

== Angewandte Patches ==

Heikki Linnakangas pushed:

- When two base backups are started at the same time with
  pg_basebackup, ensure that they use different checkpoints as the
  starting point. We use the checkpoint redo location as a unique
  identifier for the base backup in the end-of-backup record, and in
  the backup history file name.  Bug spotted by Fujii Masao.

Tom Lane pushed:

- Fix check_exclusion_constraint() to insert correct collations in

- Reimplement planner's handling of MIN/MAX aggregate optimization
  (again).  Instead of playing cute games with pathkeys, just build a
  direct representation of the intended sub-select, and feed it
  through query_planner to get a Path for the index access.  This is a
  bit slower than 9.1's previous method, since we'll duplicate most of
  the overhead of query_planner; but since the whole optimization only
  applies to rather simple single-table queries, that probably won't
  be much of a problem in practice.  The advantage is that we get to
  do the right thing when there's a partial index that needs the
  implicit IS NOT NULL clause to be usable.  Also, although this makes
  planagg.c be a bit more closely tied to the ordering of operations
  in grouping_planner, we can get rid of some coupling to lower-level
  parts of the planner.  Per complaint from Marti Raudsepp.

- Avoid potential deadlock in InitCatCachePhase2().  Opening a
  catcache's index could require reading from that cache's own
  catalog, which of course would acquire AccessShareLock on the
  catalog.  So the original coding here risks locking index before
  heap, which could deadlock against another backend trying to get
  exclusive locks in the normal order.  Because InitCatCachePhase2 is
  only called when a backend has to start up without a relcache init
  file, the deadlock was seldom seen in the field.  (And by the same
  token, there's no need to worry about any performance disadvantage;
  so not much point in trying to distinguish exactly which catalogs
  have the risk.) Bug report, diagnosis, and patch by Nikhil Sontakke.
  Additional commentary by me.  Back-patch to all supported branches.

- Throw error for indeterminate collation of an ORDER/GROUP/DISTINCT
  target.  This restores a parse error that was thrown (though only in
  the ORDER BY case) by the original collation patch.  I had removed
  it in my recent revisions because it was thrown at a place where
  collations now haven't been computed yet; but I thought of another
  way to handle it.  Throwing the error at parse time, rather than
  leaving it to be done at runtime, is good because a syntax error
  pointer is helpful for localizing the problem.  We can reasonably
  assume that the comparison function for a collatable datatype will
  complain if it doesn't have a collation to use.  Now the planner
  might choose to implement GROUP or DISTINCT via hashing, in which
  case no runtime error would actually occur, but it seems better to
  throw error consistently rather than let the error depend on what
  the planner chooses to do.  Another possible objection is that the
  user might specify a nondefault sort operator that doesn't care
  about collation ... but that's surely an uncommon usage, and it
  wouldn't hurt him to throw in a COLLATE clause anyway.  This change
  also makes the ORDER BY/GROUP BY/DISTINCT case more consistent with
  the UNION/INTERSECT/EXCEPT case, which was already coded to throw
  this error even though the same objections could be raised there.

- Improve reporting of run-time-detected indeterminate-collation
  errors.  pg_newlocale_from_collation does not have enough context to
  give an error message that's even a little bit useful, so move the
  responsibility for complaining up to its callers.  Also, reword
  ERRCODE_INDETERMINATE_COLLATION error messages in a less jargony,
  more message-style-guide-compliant fashion.

- Make initdb ignore locales for client-only encodings.  While putting
  such entries into pg_collation is harmless (since backends will
  ignore entries that don't match the database encoding), it's also

- Fix ancient typo in user-defined-aggregates documentation.  The
  description of the initcond value for the built-in avg(float8)
  aggregate has been wrong since it was written.  Noted by Disc

- Improve user-defined-aggregates documentation.  On closer
  inspection, that two-element initcond value seems to have been a
  little white lie to avoid explaining the full behavior of
  float8_accum.  But if people are going to expect the examples to be
  exactly correct, I suppose we'd better explain.  Per comment from
  Thom Brown.

- Clean up handling of COLLATE clauses in index column definitions.
  Ensure that COLLATE at the top level of an index expression is
  treated the same as a grammatically separate COLLATE.  Fix bogus
  reverse-parsing logic in pg_get_indexdef.

- Fix handling of collation in SQL-language functions.  Ensure that
  parameter symbols receive collation from the function's resolved
  input collation, and fix inlining to behave properly.  BTW, this
  commit lays about 90% of the infrastructure needed to support use of
  argument names in SQL functions.  Parsing of parameters is now done
  via the parser-hook infrastructure ... we'd just need to supply a
  column-ref hook ...

- Fix collation handling in plpgsql functions.  Make plpgsql treat the
  input collation as a polymorphism variable, so that we cache
  separate plans for each input collation that's used in a particular
  session, as per recent discussion.  Propagate the input collation to
  all collatable input parameters.  I chose to also propagate the
  input collation to all declared variables of collatable types, which
  is a bit more debatable but seems to be necessary for
  non-astonishing behavior.  (Copying a parameter into a separate
  local variable shouldn't result in a change of behavior, for
  example.) There is enough infrastructure here to support declaring a
  collation for each local variable to override that default, but I
  thought we should wait to see what the field demand is before adding
  such a feature.  In passing, remove exec_get_rec_fieldtype(), which
  wasn't used anywhere.  Documentation patch to follow.

- Document collation handling in SQL and plpgsql functions.  This is
  pretty minimal but covers the bare facts.

- Fix failure to propagate collation in negate_clause().  Turns out it
  was this, and not so much plpgsql, that was at fault in Stefan
  Huehner's collation-error-in-a-trigger bug report of a couple weeks

- Pass collation to makeConst() instead of looking it up internally.
  In nearly all cases, the caller already knows the correct collation,
  and in a number of places, the value the caller has handy is more
  correct than the default for the type would be.  (In particular,
  this patch makes it significantly less likely that
  eval_const_expressions will result in changing the exposed collation
  of an expression.)  So an internal lookup is both expensive and

- Clean up a few failures to set collation fields in expression nodes.
  I'm not sure these have any non-cosmetic implications, but I'm not
  sure they don't, either.  In particular, ensure the CaseTestExpr
  generated by transformAssignmentIndirection to represent the base
  target column carries the correct collation, because parse_collate.c
  won't fix that.  Tweak lsyscache.c API so that we can get the
  appropriate collation without an extra syscache lookup.

- More collations cleanup, from trawling for missed collation
  assignments.  Mostly cosmetic, though I did find that
  generateClonedIndexStmt failed to clone the index's collations.

- Clean up cruft around collation initialization for tupdescs and
  scankeys.  I found actual bugs in GiST and plpgsql; the rest of this
  is cosmetic but meant to decrease the odds of future bugs of

- Fix plpgsql to release SPI plans when a function or DO block is
  freed.  This fixes the gripe I made a few months ago about DO blocks
  getting slower with repeated use.  At least, it fixes it for the
  case where the DO block isn't aborted by an error.  We could try
  running plpgsql_free_function_memory() even during error exit, but
  that seems a bit scary since it makes a lot of presumptions about
  the data structures being in good shape.  It's probably reasonable
  to assume that repeated failures of DO blocks isn't a
  performance-critical case.

Simon Riggs pushed:

- Update docs to say you need fsync to make sync rep work fast.

- Make FKs valid at creation when added as column constraints.  Bug
  report from Alvaro Herrera.

- Prevent intermittent hang in recovery from bgwriter interaction.
  Startup process waited for cleanup lock but when hot_standby = off
  the pid was not registered, so that the bgwriter would not wake the
  waiting process as intended.

- Minor changes to recovery pause behaviour.  Change location LOG
  message so it works each time we pause, not just for final pause.
  Ensure that we pause only if we are in Hot Standby and can connect
  to allow us to run resume function. This change supercedes the code
  to override parameter recoveryPauseAtTarget to false if not
  attempting to enter Hot Standby, which is now removed.

- Additional test for each commit in sync rep path to plug minute
  possibility of race condition that would effect performance only.
  Requested by Robert Haas.  Re-arrange related comments.

Peter Eisentraut pushed:

- Cosmetic capitalization fix

Robert Haas pushed:

- Remove synchronous_replication/max_wal_senders cross-check.  This is
  no longer necessary, and might result in a situation where the
  configuration file is reloaded (and everything seems OK) but a
  subsequent restart of the database fails.  Per an observation from
  Fujii Masao.

- Edits to 9.1 release notes.  Add some new items and some additional
  details to existing items, mostly by cribbing from the 9.1alpha
  notes.  Some additional clarifications and corrections elsewhere,
  and a few typo fixes.

- Add post-creation hook for extensions, consistent with other object
  types.  KaiGai Kohei

- Make walreceiver send a reply after receiving data but before
  flushing it.  It originally worked this way, but was changed by
  commit a8a8a3e0965201df88bdfdff08f50e5c06c552b7, since which time
  it's been impossible for walreceiver to ever send a reply with
  write_location and flush_location set to different values.

- Remove alpha release notes.  Temporarily move some of the alpha
  release note disclaimers into the regular release notes, for the
  sake of alpha5.

- Update release notes for latest commits.  In addition, correct the
  entry for pg_object_description, and some other minor cleanup.

- Remove disclaimer stating that fsync=off slows down sync rep.  The
  underlying problem that caused this phenomenon was fixed by commit

- Correct "characters" to "bytes" in createdb docs.  Susanne Ebrecht.

- Typo fix in SSI docs.  Kevin Grittner

Michael Meskes pushed:

- Documented some ecpg command line options that were missing: -r
  no_indicator, -r prepare, and -r questionsmarks

Alvaro Herrera pushed:

- Fix broken markup, and remove tabs.

Gregory Stark pushed:

- fix up a couple non-prototypes of the form foo() to be foo(void) --
  found using -Wstrict-prototypes

== Abgelehnte Patches (bis jetzt) ==

Gurjeet Singh's patch to add a missing semicolon to parser's gram.y.
Similar patch already applied.

== Eingesandte Patches ==

Piyush Newe sent in another revision of the patch intended to fix
to_date()'s behavior to be more consistent with Oracle's.

Bernd Helmle sent in a patch to make psql's \dt use the
pg_table_size() function, which includes things like TOAST tables.

Heikki Linnakangas sent in a patch to fix a bug in initdb on Windows
in Chinese locales.

Fujii Masao sent in a patch to enable standby servers to connect
during smart shutdown in order to wake up backends waiting for sync

Heikki Linnakangas sent in another revision of the replication server
timeout patch.

Gianni Ciolli sent in a doc patch to clarify the limits in size of the

Radoslaw Smogura sent in another revision of the patch to add a second
level of buffer cache.

Robert Haas sent in another revision of the crash-safe visibility map

Robert Haas sent in a patch to allow the safe removal of write
location from sync rep.

Peter Eisentraut sent in a patch to add a feature to "make
maintainer-check" to check for tabs in SGML.

Matthew Draper sent in a WIP patch to allow SQL language functions to
reference parameters by parameter name.

Simon Riggs sent in a patch to add more sync rep settings.


No Trackbacks


Display comments as Linear | Threaded

No comments

Add Comment

Enclosing asterisks marks text as bold (*word*), underscore are made via _word_.
E-Mail addresses will not be displayed and will only be used for E-Mail notifications.
To leave a comment you must approve it via e-mail, which will be sent to your address after submission.
Form options