Skip to content

Wöchentlicher PostgreSQL Newsletter - 20. Februar 2011

Der Originalartikel befindet sich unter:

== Wöchentlicher PostgreSQL Newsletter - 20. Februar 2011 ==

== PostgreSQL Produkt Neuigkeiten ==

pgpool-II 3.0.2, ein Connection Pooler und mehr,
ist erschienen.

PL/Proxy 2.2, ein schreib-skalierendes Clusteringsystem für
PostgreSQL, ist erschienen.

== PostgreSQL Jobs im Februar ==

== PostgreSQL Lokal ==

PGDay findet dieses Jahr während der südkalifornischen Linux Ausstellung
(SCALE) im LAX Hilton Hotel in Los Angeles, Kalifornien am Freitag dem
25. Februar 2011 statt.

PostgreSQL Conference East 2011: New York,
vom 22. bis 25. März

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 ==

Bruce Momjian pushed:

- Clarify documentation for libpq's PQescapeBytea to mention the new
  hex format.  Modify PQescapeStringConn() docs to be consisent with
  other escaping functions.  Add mention problems with pre-9.0
  versions of libpq using not understanding bytea hex format to the
  9.0 release notes.  Backpatch to 9.0 docs.

- Adjust pg_upgrade error message, array freeing, and add error check.

- Fix bug in 9.1 pg_upgrade processing of old/new relations;  adjust
  debug output.

- In pg_upgrade, no need to initialize global struct values;  they are
  always zeros.  Also no need to free memory before we exit.

- In documentation example, use concat_values() instead of concat()
  because concat() is a built-in function.  Erik Rijkers

- Document that plpgsql upsert example can fail if insert triggers
  return unique exception, or if not used in read committed
  transaction mode.

- Remove doc mention about read committed in upsert example.

- Typo fix.  Erik Rijkers

- Document that text search comparisons happen after text search
  configuration rules.  Erik Rijkers

- Improve text search prefix doc addition.

- Initialize variable to quiet compiler.

- Document that an SQL-standard-compliant query that expects a unique
  row from information_schema.referential_constraints might return
  several in Postgres because we allow duplicate-named constraints in
  the same schema.

- Documentation update for UTF16 surrogate pairs in UTF8 encoding.
  Marko Kreen

- Adjust documentation wording of window function ORDER BY to not
  mention avg() because it was inaccurate.

- Merge two documentation permission chapters into a single chapter.

Tom Lane pushed:

- Change the naming convention for extension files to use double
  dashes.  This allows us to have an unambiguous rule for
  deconstructing the names of script files and secondary control
  files, without having to forbid extension and version names from
  containing any dashes.  We do have to forbid them from containing
  double dashes or leading/trailing dashes, but neither restriction is
  likely to bother anyone in practice.  Per discussion, this seems
  like a better solution overall than the original design.

- Convert contrib modules to use the extension facility.  This isn't
  fully tested as yet, in particular I'm not sure that the
  "foo--unpackaged--1.0.sql" scripts are OK.  But it's time to get
  some buildfarm cycles on it.  sepgsql is not converted to an
  extension, mainly because it seems to require a very nonstandard
  installation process.  Dimitri Fontaine and Tom Lane.

- Avoid use of CREATE OR REPLACE FUNCTION in extension installation
  files.  It was never terribly consistent to use OR REPLACE (because
  of the lack of comparable functionality for data types, operators,
  etc), and experimentation shows that it's now positively pernicious
  in the extension world.  We really want a failure to occur if there
  are any conflicts, else it's unclear what the extension-ownership
  state of the conflicted object ought to be.  Most of the time,
  CREATE EXTENSION will fail anyway because of conflicts on other
  object types, but an extension defining only functions can succeed,
  with bad results.

- Assorted fixups for "unpackaged" conversion scripts.  From first
  pass of testing.  Notably, there seems to be no need for
  adminpack--unpackaged--1.0.sql because none of the objects that the
  old module creates would ever be dumped by pg_dump anyway (they are
  all in pg_catalog).

- Support replacing MODULE_PATHNAME during extension script file
  execution.  This avoids the need to find a way to make PGXS' rule insert the right thing.  We'll just deprecate
  use of that hack for extensions.

- More fixups for "unpackaged" conversion scripts.

- Remove no-longer-needed special case hacks in MSVC build scripts.

- Fix obsolete references to old-style contrib installation methods.

- Small improvements to external-projects documentation.

- Rearrange extension-related views as per recent discussion.  The
  original design of pg_available_extensions did not consider the
  possibility of version-specific control files.  Split it into two
  views: pg_available_extensions shows information that is generic
  about an extension, while pg_available_extension_versions shows all
  available versions together with information that could be
  version-dependent.  Also, add an SRF pg_extension_update_paths() to
  assist in checking that a collection of update scripts provide sane
  update path sequences.

- Fix MSVC build scripts for recent extension-related changes.
  Untested, but we'll soon see if the buildfarm likes this.

- Rethink naming of contrib/intagg extension.  Initially it was called
  int_aggregate after the old SQL file, but since the documentation
  just says "intagg" and that's also the directory name, let's conform
  to that instead.

- Fix obsolete comment.  Comment about MaxAllocSize was not updated
  when the TOAST-header macros were replaced in 8.3 "varvarlena"
  changes.  Per report from Frederik Ramm.

- Bring hstore's comment into line with style of other contrib
  comments.  All the other ones that are primarily a new datatype say
  "data type for <purpose>", so make this one similar.

- Add CheckTableNotInUse calls in DROP TABLE and DROP INDEX.  Recent
  releases had a check on rel->rd_refcnt in heap_drop_with_catalog,
  but failed to cover the possibility of pending trigger events at
  DROP time.  (Before 8.4 we didn't even check the refcnt.)  When the
  trigger events were eventually fired, you'd get "could not open
  relation with OID nnn" errors, as in recent report from strk.
  Better to throw a suitable error when the DROP is attempted.  Also
  add a similar check in DROP INDEX.  Back-patch to all supported

- Fix corner case for binary upgrade: extension functions in
  pg_catalog.  Normally, pg_dump summarily excludes functions in
  pg_catalog from consideration.  However, some extensions may create
  functions in pg_catalog (adminpack already does that, and extensions
  for procedural languages will likely do it too).  In binary-upgrade
  mode, we have to dump such functions, or the extension will be
  incomplete after upgrading.  Per experimentation with adminpack.

- Add FOREACH IN ARRAY looping to plpgsql.  (I'm not entirely sure
  that we've finished bikeshedding the syntax details, but the
  functionality seems OK.) Pavel Stehule, reviewed by Stephen Frost
  and Tom Lane.

- Make a no-op ALTER EXTENSION UPDATE give just a NOTICE, not ERROR.
  This seems a bit more user-friendly.

- Add backwards-compatible declarations of some core GIN support
  functions.  These are needed to support reloading dumps of 9.0
  installations containing contrib/intarray or contrib/tsearch2.
  Since not only regular dump/reload but binary upgrade would fail, it
  seems worth the trouble to carry these stubs for awhile.  Note that
  the contrib opclasses referencing these functions will still work
  fine, since GIN doesn't actually pay any attention to the declared
  signature of a support function.

- Fix bogus test for hypothetical indexes in
  get_actual_variable_range().  That function was supposing that
  indexoid == 0 for a hypothetical index, but that is not likely to be
  true in any non-toy implementation of an index adviser, since
  assigning a fake OID is the only way to know at EXPLAIN time which
  hypothetical index got selected.  Fix by adding a flag to
  IndexOptInfo to mark hypothetical indexes.  Back-patch to 9.0 where
  get_actual_variable_range() was added.  Gurjeet Singh

- Fix blatantly uninitialized variable in recent commit.  Doesn't
  anybody around here pay attention to compiler warnings?

- Fix contrib/pg_trgm to have smoother updates from 9.0.  Take care of
  some loose ends in the update-from-unpackaged script, and apply some
  ugly hacks to ensure that it produces the same catalog state as the
  fresh-install script.  Per discussion, this seems like a safer plan
  than having two different catalog states that both call themselves
  "pg_trgm 1.0", even if it's not immediately clear that the subtle
  differences would ever matter.  Also, fix the stub function
  gin_extract_trgm() so that it works instead of just bleating.
  Needed because this function will get called during a regular dump
  and reload, if there are any indexes using its opclass.  The user
  won't have an opportunity to update the extension till later, so
  telling him to do so is unhelpful.

- Fix upgrade of contrib/btree_gist from 9.0.  The initial version of
  the update-from-unpackaged script neglected to include the <>
  operators that were added to the opclasses during 9.1.  To make this
  script produce the same final state as the regular install script,
  use the same ALTER OPERATOR FAMILY trick as in pg_trgm.

- Fix upgrade of contrib/intarray and contrib/unaccent from 9.0.  Take
  care of a couple of discrepancies between what you get from a fresh
  install and what the first-draft update-from-unpackaged scripts

- Fix upgrade of contrib/xml2 from 9.0.  Update script was being
  sloppy about two functions that have been changed since 9.0.

- Fix tsmatchsel() to account properly for null rows.  ts_typanalyze.c
  computes MCE statistics as fractions of the non-null rows, which
  seems fairly reasonable, and anyway changing it in released versions
  wouldn't be a good idea.  But then ts_selfuncs.c has to account for
  that.  Failure to do so results in overestimates in columns with a
  significant fraction of null documents.  Back-patch to 8.4 where
  this stuff was introduced.  Jesper Krogh

- One more hack to make contrib upgrades from 9.0 match fresh 9.1
  installs.  intarray and tsearch2 both reference core support
  functions in their GIN opclasses, and the signatures of those
  functions changed for 9.1.  We added backwards-compatible pg_proc
  entries for the functions in order to allow 9.0 dump files to be
  restored at all, but that hack leaves the opclasses pointing at
  pg_proc entries different from what they'd point to if the contrib
  modules were installed fresh in 9.1.  To forestall any possibility
  of future problems, fix the opclasses to match fresh installs via
  the expedient of direct UPDATEs on pg_amproc in the
  update-from-unpackaged scripts.  (Yech ... but the alternatives are
  worse, or require far more effort than seems justified right now.)
  Note: updating pg_amproc is sufficient because there will be no
  pg_depend entries corresponding to these dependencies, since the
  referenced functions are all pinned.

- Fix parallel pg_restore to handle comments on POST_DATA items
  correctly.  The previous coding would try to process all
  SECTION_NONE items in the initial sequential-restore pass, which
  failed if they were dependencies of not-yet-restored items.  Fix by
  postponing such items into the parallel processing pass once we have
  skipped any non-PRE_DATA item.  Back-patch into 9.0; the original
  parallel-restore coding in 8.4 did not have this bug, so no need to
  change it.  Report and diagnosis by Arnd Hannemann.

- Un-break building with BTREE_BUILD_STATS.  This has been broken for
  awhile, but not clear it's worth back-patching.  Euler Taveira de

- Create the catalog infrastructure for foreign-data-wrapper handlers.
  Add a fdwhandler column to pg_foreign_data_wrapper, plus HANDLER
  WRAPPER commands, plus pg_dump support for same.  Also invent a new
  pseudotype fdw_handler with properties similar to language_handler.
  This is split out of the "FDW API" patch for ease of review; it's
  all stuff we will certainly need, regardless of any other details of
  the FDW API.  FDW handler functions will not actually get called
  yet.  In passing, fix some omissions and infelicities in
  foreigncmds.c.  Shigeru Hanada, Jan Urbanski, Heikki Linnakangas

- Implement an API to let foreign-data wrappers actually be
  functional.  This commit provides the core code and documentation
  needed.  A contrib module test case will follow shortly.  Shigeru
  Hanada, Jan Urbanski, Heikki Linnakangas

- Add contrib/file_fdw foreign-data wrapper for reading files via
  COPY.  This is both very useful in its own right, and an important
  test case for the core FDW support.  This commit includes a small
  refactoring of copy.c to expose its option checking code as a
  separately callable function.  The original patch submission
  duplicated hundreds of lines of that code, which seemed pretty
  unmaintainable.  Shigeru Hanada, reviewed by Itagaki Takahiro and
  Tom Lane

- Minor logic fix for new levenshtein implementation.  Alexander

Simon Riggs pushed:

- Add version-sensitive SQL for psql when constraints NOT VALID.  Bug
  report and fix by Andres Freund.

- PITR can stop at a named restore point when recovery target = time
  though must not update the last transaction timestamp.  Plus comment
  and message cleanup for recent named restore point.  Fujii Masao,
  minor changes by me.

- Additional required docs for pg_constraint catalog entry.  Noted by
  Bernd Helmle

- Hot Standby feedback for avoidance of cleanup conflicts on standby.
  Standby optionally sends back information about oldestXmin of
  queries which is then checked and applied to the WALSender's
  proc->xmin.  GetOldestXmin() is modified slightly to agree with
  GetSnapshotData(), so that all backends on primary include WALSender
  within their snapshots.  Note this does nothing to change the
  snapshot xmin on either master or standby. Feedback piggybacks on
  the standby reply message.  vacuum_defer_cleanup_age is no longer
  used on standby, though parameter still exists on primary, since
  some use cases still exist.  Simon Riggs, review comments from Fujii
  Masao, Heikki Linnakangas, Robert Haas

- Separate messages for standby replies and hot standby feedback.
  Allow messages to be sent at different times, and greatly reduce the
  frequency of hot standby feedback. Refactor to allow additional
  message types.

- Make a hard state change from catchup to streaming mode.  More
  useful state change for monitoring purposes, plus a required change
  for synchronous replication patch.

Robert Haas pushed:

- Delete stray word from comment.

- Avoid a few more SET DATA TYPE table rewrites.  When the new type is
  an unconstrained domain over the old type, we don't need to rewrite
  the table.  Noah Misch and Robert Haas

- Rename max_predicate_locks_per_transaction.  The new name,
  max_pred_locks_per_transaction, is shorter.  Kevin Grittner, per

- Assorted corrections to the patch to add WAL receiver replies.  Per
  reports from Fujii Masao.

- pg_ctl promote.  Fujii Masao, reviewed by Robert Haas, Stephen
  Frost, and Magnus Hagander.

- WAL receiver shouldn't try to send a reply when dying.  Per report
  from, and discussion with, Fujii Masao.

- Remove rule to build sepgsql-regtest.pp.  Instead, document how to
  build it manually.  Per discussion.  KaiGai Kohei, with some
  wordsmithing by me.

- Typo fixes.  Thom Brown

- Add some words of caution to elog.c.  Stephen Frost, somewhat
  rewritten by me

- Add client_hostname field to pg_stat_activity.  Peter Eisentraut,
  reviewed by Steve Singer, Alvaro Herrera, and me.

- Prevent possible compiler warnings.  Simon Riggs reports that
  rnode.dbNode and rnode.spcNode were generating unused variable
  warnings on gcc 4.4.3 with CFLAGS=-O1

Peter Eisentraut pushed:

- Allow make check in PL directories.  Also add make check-world
  target, and refactor pg_regress invocation code in makefiles a bit.

- Better support for thread-support flag detection with clang.  When
  testing the stderr produced by various thread-support flags, also
  run a compilation in addition to a link, because clang warns on
  certain flags when compiling but not when linking.

- Also process psqlrc when running psql -l.  This was previously not
  very useful, but with many people customizing the linestyle, it is
  nice for a consistent appearance.

- Fix for warnings-free compilation with Python 3.2.  The first
  argument of PyEval_EvalCode() was changed from PyCodeObject* to
  PyObject* because of PEP 384.

- Set psql client encoding from locale by default.  Add a new libpq
  connection option client_encoding (which includes the existing
  PGCLIENTENCODING environment variable), which besides an encoding
  name accepts a special value "auto" that tries to determine the
  encoding from the locale in the client's environment, using the
  mechanisms that have been in use in initdb.  psql sets this new
  connection option to "auto" when running from a terminal and not
  overridden by setting PGCLIENTENCODING.  original code by Heikki
  Linnakangas, with subsequent contributions by Jaime Casanova, Peter
  Eisentraut, Stephen Frost, Ibrar Ahmed

- Invalidate PL/Python functions with composite type argument when the
  type changes.  The invalidation will cause the type information to
  be refetched, and everything will work.  Jan Urbański, reviewed by
  Alex Hunsaker

Alvaro Herrera pushed:

- Make pg_upgrade compile again

- Cleanup ClusterInfo initialization in pg_upgrade

- Use $INDENT rather than indent throughout the pgindent code.  This
  allows the user to change the path to be used more easily.  Also,
  change URL in README.

- Convert Postgres arrays to Perl arrays on PL/perl input arguments.
  More generally, arrays are turned in Perl array references, and row
  and composite types are turned into Perl hash references.  This is
  done recursively, in a way that's natural to every Perl programmer.
  To avoid a backwards compatibility hit, the string representation of
  each structure is also available if the function requests it.
  Authors: Alexey Klyukin and Alex Hunsaker.  Some code cleanups by

- pgindent run on plperl.c

- Use $INDENT instead of `which` to find the indent binary.  Per
  discussion after my commit o yesterday.

ITAGAKI Takahiro pushed:

- Export the external file reader used in COPY FROM as APIs.  They are
  expected to be used by extension modules like file_fdw.  There are
  no user-visible changes.  Itagaki Takahiro.  Reviewed and tested by
  Kevin Grittner and Noah Misch.

- Add transaction-level advisory locks.  They share the same locking
  namespace with the existing session-level advisory locks, but they
  are automatically released at the end of the current transaction and
  cannot be released explicitly via unlock functions.  Marko Tiikkaja,
  reviewed by me.

- Fix an uninitialized field in DR_copy.  Shigeru HANADA

Magnus Hagander pushed:

- Properly initialize variables.  Kevin Grittner

== Abgelehnte Patches (bis jetzt) ==

No one was disappointed this week :-)

== Eingesandte Patches ==

ITAGAKI Takahiro sent in another revision of the patch to implement

Stephen Frost and Robert Haas traded revisions of the patch to allow
logging the current role.

Marti Raudsepp sent in another revision of the patch which makes it so
a key lock is only acquired for columns that can be referenced.

Heikki Linnakangas sent in another revision of the patch to add a FDW

Thom Brown sent in a patch to make array_cat consistent with the ||
operator for arrays with respect to NULLs.

ITAGAKI Takahiro sent in another revision of the patch to implement
transaction-scope advisory locks.

Kevin Grittner sent in two revisions of a patch to fix an issue with
uninitialized structures from the SSI patch.

ITAGAKI Takahiro sent in another revision of the COPY ENCODING patch.

Per feedback from Heikki Linnakangas, Simon Riggs sent in a patch to
add a server_name parameter, plus mechanism to send info from
standby to master.  This will help with synchronous replication.

YAMAMOTO Takashi sent in a patch to fix an issue with SSI.

Tom Lane sent in two versions of a patch to fix pg_tgrm's
update-from-unpackaged script.

Mark Kirkwood sent in a WIP patch to allow the backend to constrain
temporary file space.

Tom Lane sent in a WIP patch to update KNN-GiST.

Magnus Hagander sent in another revision of the patch to include

Robert Haas sent in a patch to fix an issue with an assertion failure

Simon Riggs sent in a patch to add server_name for synchronous

Simon Riggs sent in anothre revision of the patch to allow for
synchronous replication.

Radoslaw Smogura sent in a patch to add void_send and void_receive,
used in COPY ... BINARY.


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