PostgreSQL 9.0 Open Items

Project Planning

• 9.0.0 was released at 20 September 2010
• First commit fest for 9.1 began 15 July 2010

Issues

Remaining issues were moved to PostgreSQL 9.1 Open Items.

Resolved Issues

• Hot Standby needs to fully support VACUUM FULL
• BUG #5236: Aparent bug in ecpg
  • rhaas says: fixed by mmeskes
• AbortOutOfAnyTransaction within ProcessInterrupts is *utterly* unsafe
• Archive recovery broken on EXEC_BACKEND (win32)
• Poor handling of error cases for deferrable unique constraints
• Replace ALTER TABLE ... SET STATISTICS DISTINCT with a general attribute-options facility
• pg_listener entries deleted under heavy NOTIFY load only on Windows
• B-tree dead page removal not safe with read-only transactions
• Bogus handling of relcache init files during HS replay
• Large object ownership and permissions are dumped incorrectly
• Fix pg_hba.conf handling of "@"
• Fix GSS crash
• incorrect exit code from psql with single transaction + violation of deferred FK constraint
• Duplicative display of exclusion constraints in psql \d
• missing need_initialization = false in reloptions code
• Ragged latency log data in multi-threaded pgbench: patch, review with updated patch
• ALTER ROLE/DATABASE RESET ALL versus security: commit message
• Redefine smart shutdown in standby mode?
• Add the GUC parameter to specify the maximum number of log file segments held in pg_xlog directory to send to the standby server. Which is useful to avoid disk full in the primary.
• Starting recovery connections from a shutdown checkpoint
• Walreceiver and dblink are not interruptible on win32. - related thread
• Document changes to situations where WAL-logging can be skipped.
• Standalone backends run StartupXLOG in an incorrect environment
• Fix walsender to not assume template1 exists
• Improve behavior of SUSET GUC variables added by loadable modules
• Fix to_char('L') (localization) on Win32 when the operating system character encoding does not match the database encoding [1] [2] [3],applied patch.
• don't allow walsender to consume superuser_reserved_connection slots, or during shutdown
  • latest patch 2010-04-21
• Tuning of KnownAssignedXids module
• clarify HS error message when WAL settings are no good
  • rhaas says: issue should be addressed by Heikki's WAL-level patch
• Documentation overflows TeX capacity while building PDFs
• change "AuthenticationOK" to "AuthenticationMaybeOK"?
• Incrementally Updated Backups and restartpoints - doc change
• Doc update for Hot Standby - PMState
• Interesting near-bug in shared-dependency management - commit by tgl
• Type Mismatch Error in Set Returning Functions - change in user exposed behavior between versions
  • tgl says this is a deliberate change

Resolved after Beta1

• ERROR: btree index keys must be ordered by attribute - resolved commit
• ALTER ROLE/DATABASE RESET ALL versus security - fixed by this patch
• LD_LIBRARY_PATH versus rpath
• SQLSTATE for Hot Standby cancellation
• Assertion failure in walreceiver - fixed
• HS/SR and smart shutdown - fixed
• Patch for additional checking for SIGHUP - was Fuji-san's patch applied after this message?
• Hot Standby performance and deadlocking
• %u used in process listing - remove? - someone got there first
• Smart shutdown gets stuck - patch to fix from Fuji Masao, associated test test associated with fix
• Where should wal_level description and location be in postgresql.conf (no action required)
• keep_mumble or min_wal_segments naming (no consensus on a better name)
• invalid declspec for PG_MODULE_MAGIC - done
• vacuum_defer_cleanup_age listed on postgresql.conf.sample? (it's there now)
• Add SIGCHLD catch to psql
• psql \? \daS
• Performance problem in textanycat/anytextcat
• CLUSTER doc patch - something along these lines committed by rhaas
• Outstanding issues in HS/SR - Resolved? Simon thinks yes.
• Assertion failure twophase.c (2) (testing HS/SR) - Simon says this is resolved
• small exclusion constraints patch - suggestion was to just document we'd need to remove the sanity check if we ever did add support for such a case (<>), done
• Mergejoin null handling
• move 'long long' check to c.h
• Should we revert the default output format for bytea to the old style before shipping 9.0.0?
  • tgl thinks we should leave that there for awhile, though I agree it's likely that the final decision will be "no change".
• Crash in buildfarm for Mac OS X 10.6.3
  • the machine ran out of disk space, but we would like to find out exactly what happened because UpdateControlFile should always update in place.
• walsender in smart shutdown
• Patch for distinguishing normal shutdown from unexpected exit - patch being reviewed, Robert Haas to commit - but requests help with review
• GiST readme update - done
• improve plpgsql's EXECUTE 'select into' message with a hint - done

Resolved after Beta2

• Standby instead of "slave" in documentation - done
• fillfactor gets set to zero for toast tables - done
• Doc patch for _PG_Init and PLs, Update from Josh Tolley
• ALTER TABLE .. DISABLE/ENABLE TRIGGER are out of date patch committed
• EXPLAIN: some constructs that are legal YAML may come through as the wrong type
• should we rename restartpoint_command to archive_cleanup_command? - done
• checkpoint_segments during recovery
  • proposed patch from Fujii Masao
• .pgpass entry for SR
  • patch from Fujii Masao
• parser handling of large object OIDs, with patch
• autovacuum doesn't know that analyze can also collect stats for inheritance trees
  • rhaas says: moving this back from the long-term issues bucket; this is a new issue in 9.0
  • rhaas says: proposed doc patch
• Improve the document about pg_last_xlog_receive_location()
  • Patch from Fujii Masao
• Streaming Replication can't sent WAL that has not been fsync'd on master
• revert changes to expose plperl subs as directly callable? Now reverted.
• Mention 'create or replace function' no longer allows renaming parameters in the incompatibilities section of the 9.0 release notes.
• supply alternate hstore operator for equals-greater in preparation for later user in function parameter assignment
• pg_upgrade issues on Windows
• Streaming Replication needs to detect death of master
• missing "else" in postmaster.c?
• Attempts to create both PLpythonU and PLpython3U fail
• max_standby_delay thread
• normalize use of LDFLAGS
• Fix log_temp_files units

Resolved after Beta3

• crash-recovery replay of CREATE TABLESPACE is broken
• regclass columns go boom with pg_upgrade
• pg_dump is broken for user mappings
• failed to fetch tuple for EvalPlanQual recheck
• page corruption after moving tablespace
• Documentation fails to build as PDF

Resolved after Beta4

• name and comment of XLogSetAsyncCommitLSN() should be changed
• RecordTransactionCommit does not respect wal_level GUC
• trace_recovery_messages inconsistency
• Backup procedure is wrong?
• BUG #5595: Documentation is not installs from VPATH build.

Resolved after RC1

• pg_subtrans keeps bloating up in the standby
• Walsender doesn't process options passed in the startup packet
• Join removal messes up constant-false qual conditions

Resolved in newer versions

• tsearch profiling - czech environment - take 55MB
  • Reduce the memory requirement for large ispell dictionaries
• DML Permission Checks - for SE-PostgreSQL integration in 9.1.
  • New module contrib/sepgsql, an SE-Linux integration for PostgreSQL

Long-term issues

These items are not 9.0-specific or brand new features for 9.1. They should be fixed or integrated eventually, but not for now.

Known bugs and limitations

• COPY statements with paths containing non-ASCII characters under WinXP
  • Todo#Miscellaneous Other: Deal with encoding issues for filenames in the server filesystem
• BUG #4769: xmlconcat produces invalid xml values
  • latest patch, but still has some problems.
  • Better to leave it alone until we have a full fix in 9.1.
• Thread safety and libxml2, patch included.
• Memory leak in SRF It seems the consensus on this thread is that this won't get fixed for now, because it's a small issue
• Weird page allocation pattern
• Streaming replication and pg_xlogfile_name() - just throw an error in 9.0. better solution planned in 9.1.
• infinite repeat of warning message in standby
  • rhaas says: to the extent that there's an issue here, it seems we don't know what to do about it
• ExplainOnePlan handles snapshots differently than ProcessQuery
  • rhaas says: should fix this for 9.1

New features

• xpath improvement V2
• Alternative return value for pg_cancel_backend
• pg_controldata - machine readable? - planned for 9.1. contrib module version is in github.
• add signaling between walreceiver and startup process
• change ruleutils.c to print the parameter expressions at the call site
• Idle in transaction query cancellation
• WAL optimization for Move tablespace

Other issues

• CVS filtering of DOS newlines - git can solve the issue with "git config core.autocrlf input".
• don't rename index columns behavior has already broken JDBC
  • Not our bug, but we needed to communicate to third-party authors.
• Bizarre buildfarm failure on baiji: can't find pg_class_oid_index
  • rhaas says: It would be nice to fix this, but we don't know how to investigate it, and it's not a release blocker.
• BUG #5574: PANIC on hot-standby: heap_update_redo
  • tgl says: apparently just pilot error
  • sriggs says: yes, user panic, not code PANIC