PostgreSQL 10 Open Items
From PostgreSQL wiki
Jump to navigationJump to searchDecisions to Recheck Mid-Beta
- Which synchronous replication method, priority or quorum, should be chosen when neither FIRST nor ANY is specified in synchronous_standby_names?
- original commit: 3901fd7 (principal author: Masahiko Sawada; owner: Fujii Masao)
- Right now, a priority-based sync replication is chosen for keeping backward compatibility. However some hackers argued to change this decision so that a quorum commit is chosen because they think that most users prefer to a quorum.
- Cannot connect to alternative hosts when non-socket level connection error occurs
- patch exists
- This is a matter of prioritizing either of service continuity or configuration error detection. Continue considering what to do based on user feedback during beta period, other developers' opinions, and the solution using SQLSTATE.
Older Bugs
- Oddity in EXPLAIN for foreign/custom join pushdown plans
- postgres_fdw produces incorrect aliases for joining relations shown in EXPLAIN for some join pushdown queries
- low-priority issue; let's leave this for v10
- regression in v9.6, nothing changed from v9.6 beta to v10 beta
- standby can fail to reconnect even with replication slots
- patch exists
- present in 9.4 and later
- Crash recovery can leave everlasting empty pages
- The pages are actually empty but FSM says that they have no room for new data and ALL_FROZEN. ALL_FROZEN prevents autovacuum from fixing FSM.
- regression in v9.6, nothing new in v10
- parallel query locks up if worker process fails to launch
- code is broken since birth, but do we really want to ship v10 like this?
- Failure to transmit SET ROLE state to parallel workers correctly
- "I think it's reasonable to characterize this as "must fix for v10"." --tgl
- Fixed by commit 69125c883df30e6232412d35bf7c89b8c3ad1305 for 10.1.
- Replication status in logical replication
- notified
- patch exists
- present since logical decoding; not new under logical replication
Non-bugs
- CREATE/ALTER ROLE PASSWORD ('value' USING 'method'), extension of this DDL to enforce type of password without changing password_encryption.
- Not Open-Item worthy, let's do this in v11 if it still feels worthwhile then.
- Declarative partitioning vs. information_schema, etc.
- include partitions in information_schema.tables, pg_tables, and psql's \d listing, etc?
- firm support for status quo, lack of firm support for alternatives
- Document that foreign table's partition constraint is not enforced locally
- Patch exists
- documentation is correct, but this would add emphasis to a point
- libpq connection failover: Use more efficient mechanism instead of SHOW transaction_read_only, add the ability to connect to the standby
- original commit: 274bb2b3 (principal author: Robert Haas; owner: Robert Haas)
- Right now, libpq issues SHOW transaction_read_only upon each connection. This is inefficient, and we should use some GUC_REPORT variable to avoid this round-trip when connecting to PG10 or later. In addition, target_session_attrs=read-only should be available even in PG10 to meet the average expectation.
- Robert is skeptical about changing this now
- patch exists
- Add migration instructions encouraging migration from MD5 to SCRAM once we know the state of drivers; see post
- On Windows, ICU doesn't use sortsupport, even though there is no logical reason for that restriction. This is obviously an artifact of Windows libc. Windows isn't special from the point of view of the new ICU collation provider.
- Autovacuum makes use of the RangeVar of the parent relation for partitions for error reports
- An idea is to not show any information to users if the RangeVar is undefined, no information being better than misleading information.
- Problem with the stability of ICU locale identifier strings (pg_collation.collcollate) that are passed to ICU's ucol_open().
- ICU version downgrades are broken. Users will have problems with initdb ICU collations here (not just when CREATE COLLATION was used).
- We should consider fixing the issue by converting to and from pg_collation's BCP 47 format to the more widely supported legacy language tag format as needed (a patch to do this is available), and standardizing on ICU collations' pg_collation.collcollate being in BCP 47 format on all ICU version (not just ICU >= 54).
- Removing support for ICU versions 4.2 and 4.4 (versions for which support was only added in eccead9, back in August) will simplify resolving this open item. (Technically, we wouldn't even need to change how collcolate canonicalization is performed if we took this approach.)
- original commit: eccfef8 (principal author: Peter Eisentraut; owner: Peter Eisentraut notified)
- (item is disputed/too late)
Resolved Issues
resolved before 10beta1
- PANIC in pg_commit_ts slru after crashes, 2PC restore code at fault
- Fixed by commit aa203e76004daaee3d70b19cc727ed17b87b3d3a
- Fixed by commit ee01f7092fb6430ad9bb9bb1f42f19d22bcb9329
- COMMENT ON COLUMN fails for partitioned tables
- Fixed by commit 51175f3638524231405e674e40bde159b0b76727
- Fix WAL receiver spinlock use
- Fixed by commit 572d6ee6d41b43b8871f42c7dbbef795523b2dbf
- Partitioning optimization broke ConvertRowtypeExpr
- Fixed by commit 3f902354b08ac788600f0ae54fcbfc1d4e3ea765.
- Integer overflow in enlargeStringInfo
- pg_upgrade is broken because of renaming of pg_resetxlog to pg_resetwal
- Fixed by commit b877761.
- PQsendQuery fails when it should not if target_session_attrs is set to read-write
- Fixed by commit 1de0a4e.
- Substantial bloat in postgres_fdw regression test runtime
- Fixed by commit aa7f593b1ffa9717bd5570174944c06c482d1c1f.
- Error detail shown when partition not found
- Fixed by commit 5a73e17317e91912b2755f7960d5bf31d374cf31.
- DROP SUBSCRIPTION and ROLLBACK
- Fixed by commit 272adf4f9cd67df323ae57ff3dee238b649d3b73.
- Dropping partitioned tables without CASCADE
- Fixed by commit 8b4d582d279d784616c228be58af1e39aa430402.
- pg_recvlogical.c doesn't build with --disable-integer-datetimes
- Fixed by commit b6aa17e0ae367afdcea07118e016111af4fa6bc3 and c29aff959dc64f7321062e7f33d8c6ec23db53d3
- pg_dump and PUBLICATIONS
- Fixed by commit 05227e0c345247c9e9ff91445850f414e2b0bb70
- simplehash performance regressions
- Fixed by commit d4c62a6b623d6eef88218158e9fa3cf974c6c7e5
- exposing wait events for non-backends
- Fixed by commit fc70a4b0df38bda6a13941f1581f25fbb643c7f3
- Partitioning vs INSERT ON CONFLICT
- Fixed by commit 8355a011a0124bdf7ccbada206a967d427039553
- Crash when leaf partition has an index
- Reverted by commit f05230752d53c4aa74cffa9b699983bbb6bcb118
- Bug in list partitioning tuple-routing
- Fixed by commit 7ecb7143589f38d679bb566311dfa9be1a650fd5
- segfault in hot_standby for hash indexes
- Fixed by commit c4c51541e22bf7f2da8ecf6986271687b0d7a90e
- attaching to DSA area that was already destroyed
- Fixed by commit fddf45b38097d14301d249fbeebca32e40233bd2
- postgres_fdw, partitioned tables and IMPORT SCHEMA
- postgres_fdw IMPORT SCHEMA and partitioned tables
- Fixed by commit f49bcd4ef3e9a75de210357a4d9bbe3e004db956
- Partitioning tables create a file on-disk, which remains empty and has no purpose. Those relations don't need any storage
- Fixed by commit c94e6942cefe7d20c5feed856e27f672734b1e2b
- Documentation improvements for partitioning
- Fixed by commit 8f18a880a5f138d4da94173d15514142331f8de6
- crashes due to setting max_parallel_workers=0
- Fixed by commit 25dc142a49c60c3107480c487cd8444dc83f9bdf
- Failed assertion in _hash_kill_items/MarkBufferDirtyHint
- Fixed by commit 93cd7684ee2bba227fa371daa81b88f25456dcb2
- inconsistent page found on STANDBY server
- Fixed by commit 75a1cbdc3cfca1e815da6dfa5d7e96d82a6b0725
- page inspect to show appropriate type of page
- Fixed by commit 633e15ea0f1bf2e1d70441fe9da8781befebd6e9
- Logical replication between differrent encodings fails
- Fixed by commit 6f1b9aaae35bfabe2654a8e44ce226c91e7d8bd9
- improve test coverage of parallel explain analyze
- Fixed by commit b2ff37d43cc81348fd8e9d9c5fcc9dfadf790763
- improve test coverage of parallel bitmap scan
- Fixed by commit 5a5931533edd2b70bde1f069609f58998dd26fef
- SASLprep
- Fixed by commit 60f11b87a2349985230c08616fa8a34ffde934c8
- dsa.c needs a visit from the message style police
- Fixed by commit 5c4488478b182983f290a61fc8cf2ec83548622b
- Logical replication + EXEC_BACKEND + ASLR is broken
- Fixed by commit 0ef26bb394abedb2745bd838c26ecb3131682bda
- Broken locking design for accesses to pg_subscription_rel
- Fixed by commit 521fd4795e3ec3d0b263b62e5eb58e1557be9c86
- Both launcher and worker don't handle SIGHUP signal and cannot reload the configuration
- Fixed by commit 26ad194cb0a6b955e155d44fb52a74212ce85759
- pg_dump and data durability, with addition of --no-sync option
- original commit: 96a7128 (principal author: Michael Paquier; owner: Andrew Dunstan notified)
- Regression tests should use --no-sync as much as possible
- Fixed by commit 3820c63da8d0e59e2bd4476e91968f03be5dd041
- RENAME RULE doesn't work with partitioned tables
- Fixed by commit 02af7857e5694b13c21401d1982ac21d31e27dee
- sepgsql and partitioned tables
- feature request, not bug
- implemented in 25542d77dd549940468d1a932809feb9959d717d
- Mishandling of non-parallel-safe initplans/subplans in a parallelized query
- original commit: 5e6d8d2bb
- Fixed by commit 16ebab68862bb5d3595b8c8df083f650d9d7cd20
- pg_dump and SUBSCRIPTIONS
- fixed by commits c31671f9b5f6eee9b6726baad2db1795c94839d1, a9254e675bde7dc2d976d207450c559d914c0dd6
- Letting the client choose the protocol to use during a SASL exchange
- fixed by commit 4f3b87ab780b95c2cc8a591259baefaff4852037
- Add overview of SCRAM to the FE/BE protocol documentation. Mention that SASLprep is used on all passwords, UTF-8 or not.
- fixed by commit 4f3b87ab780b95c2cc8a591259baefaff4852037
- add synchronous_commit control for logical apply
- fixed by commit 887227a1cc861d87ca0f175cf8bd1447554090eb
- logical replication worker and statistics
- fixed by commit 139eb9673cb84c76f493af7e68301ae204199746
- web site CSS issues
- fixed in pgweb
- publications vs inheritance
- fixed by commit 419a23b478ae760b797188341ddce5b41322684b
- extended stats not friendly towards ANALYZE with subset of columns
- original commit: 7b504eb
- fixed by commit bf2a691e02d7766f185d9d8e0f092222a5c0a129
- Parallel Query + EXEC_BACKEND + ASLR is broken
- Problem in 9.6 and later
- Fixed by commits 0ef26bb39 + 32470825d + a74740fbd
- Different table schema in logical replication crashes
- fixed by commit e6242c18a5bb08788e6c4cc773952fc8e2a6291a
- pg_hba.conf syntax
- fixed by commit c727f120ff50f624a1ee3abe700d995c18314a0b
- CREATE/DROP SUBSCRIPTION, query cancellations and slot handling
- fixed by commit dcb39c3
- synchronous_standby_names shows unused priority values
- original commit: 3901fd7 (principal author: Masahiko Sawada; owner: Fujii Masao notified)
- The priority value is assigned to each standby listed in s_s_names even in quorum commit though those priority values are not used at all. Users can see those priority values in pg_stat_replication. Isn't this confusing? If yes, it might be better to always assign 1 as the priority, for example.
- Not a bug. Consensus against current design; no consensus on which replacement to adopt. Need to resolve that debate.
- Fixed by commit 346199dcab4cfb2c023373fb3d859583b59810d7
- tablesync patch broke the assumption that logical rep depends on?
- fixed by commit de4389712206d2686e09ad8d6dd112dc4b6c6d42
- logical replication fixes
- fixed by commit e495c1683f2c243f6769b34a009cf9c28526b555 and following
- Crash when partition column specified twice
- fixed by commit 504c2205abc7de67386f9c95630f38ee15626f07
- Interval for launching the table sync worker
- fixed by commit e3cf708016ca6045dc1cd5a0768cfecf17caf3d1
- Dropping a partitioned table takes too long
- fixed by commit c1e0e7e1d790bf18c913e6a452dea811e858b554
- Removing ADD GENERATED for identity columns
- owner proposes closing this as a non-bug
- secondary issue fixed by commit e4fddfd49241dc8dfda354993bad8d5518df1873
- UPDATE/DELETE statement-level triggers don't fire on partitioned tables
- notified
- fixed by commit e180c8aa8caf5c55a273d4a8e6092e77ff3cff10
- some review comments on logical rep code
- fixed by several commits, last is 9414e41ea703ea5fcc288bcf7dc000e53306896b
- disabling a subscription need not wake the launcher
- fixed by commit a99448ab4515aaadc17647e53633f418893f5adf
- Incorrect application of unique-inner join logic
- correctness issue fixed by commit 2057a58d1629ebffce694e3cef7f714571a88dd7
- performance issue fixed by commit 92a43e4857d9682b93c9f755f453cc8fd7c66c81
- pg_dump and partitioned tables
- original commit: f0e4475 (principal author: Amit Langote; owner: Stephen Frost, previously Robert Haas)
- notified
- pg_dump should not emit ALTER TABLE ONLY for a partitioned table in case of attribute changes and inheritable constraints
- patch exists
- pg_dump TAP tests around partitioning
- fixed by commit 44c528810a1eca52a7888ed74c08353d45331b00
- pgoutput is not built with MSVC
- fixed by commit 28d1c8ccc87128f9b0b937eae277473027c36b7e
- Logical replication - TRAP: FailedAssertion in pgstat.c
- fixed by commit 9a591c1bccc5edeb06b979c59f39753982131181
- ALTER SUBSCRIPTION doesn't check connection string validity
- fixed by commit fe974cc5a69903e9f53b36d6e2709fd3de0a1ac7
- Having DROP SLOT option on DROP SUBSCRIPTION is broken by design
- fixed by commit 013c1178fd0adefa0f68d5ce2d84e7ae6f9613a1
- consider adding pg_dump --no-subscriptions
- fixed by commit 26aa1cf376f68b800b73c326edeea6d1996ec246
- Logical replication worker ApplyContext bloat
- fixed by commit 489b96e80b96c0eda02575347654e87968f2f5f4
- Transition tables for triggers on foreign tables and views are broken
- fixed by commit 9e6104c6672dc948a430d1ee269b0c31bf5bc974
- assertion failure when rescanning transition table
- fixed by commit 304007d9f1f66fd37e50e5a5aa6f17400f1239f8
- assertion failure for TRUNCATE triggers with transition tables
- fixed by commit 29fd3d9da0ff9e230ff051c1423871bd6eac377d
- Remove pre-10 compatibility code in hash index
- fixed by commit a5775991bb86d95939b3eb1173b88d8c5312962d
- consider revising syntax to avoid NO prefixes (NOCOPY, NOCONNECT, NOREFRESH, NODROP, NOPUBLISH)
- fixed by commit b807f59828fbc02fea612e1cbc0066c6dfa3be9b
- pg_dump --no-publications
- fixed by commit 96e1cb4c0fb45dfca2d8b0a58693b94cbdaabe11
- Rename wal location functions to "lsn"
- Backward compatibility is already broken for most of these functions during the xlog to wal rename, and it'll likely be too late to change this once beta1 is out.
- patch exists
- fixed by commit d10c626de47d8b048b663471c7785603a2ec8641
- incorrect multi-column range partition constraint
- fixed by commit f8bffe9e6d700fd34759a92e47930ce9ba7dcbd5; see also 1848b73d4576e30c89ba450ad9f169774a6819bf
- Time based lag tracking for logical replication
- fixed by commit 024711bb544645c8b1061e9f02b261e2e336981d
- snapshot builder has bugs
- fixed by commits 2bef06d51646058c6bb480fcdbffb1f0cc914fed, 56e19d938dd1457ae078304df1b9903509a0a2bf, 955a684e0401954a58e956535107bc4b7136d952 and 524dbc14335cde0b18745f05a9112436d212f061
- statistics objects fail to cope with ALTER COLUMN TYPE
- fixed by commit b5b0db19b895f033ada35bc7c337183be7356977
- ALTER SEQUENCE RESTART is not concurrent-safe
- fixed by commit f8dc1985fd390774aab4ab0ba71036d6d5e631a9
resolved before 10beta2
- autovacuum launcher complaining on SIGINT because DSA re-attachment
- fixed by commit a4f06606a328664d79f34b6041e816908d93e063
- Improve use of TwoPhaseStateLock in 2PC redo code paths
- fixed by commit e90ceeaa495f5f40f224bcf84d2b0700eae8d7a3
- synchronous_commit missing from tab completion of CREATE SUBSCRIPTION
- fixed by commit 0fbfb65d4bfd429651dc28faf3ad1846c965e18d
- we should disallow subscribing a foreign table
- fixed by commit 944dc0f9cec7c3d33648f41aaecfbd976106e975
- Server Crashes if try to provide slot_name='none' at the time of creating subscription
- fixed by commit 62345698513cbcb3c48a6dae414abf0f24fd163a
- Event triggers + table partitioning cause server crash in current master
- fixed by commit 3ec76ff1f2cf52e9b900349957b42d28128b7bc7
- Partitioned table Appends not coping with being referenced in subplans
- fixed by commit b522759508dae17535f8cd20598a50a409a97f4d
- If recovery.conf has target_session_attrs=read-write, the standby fails to start.
- fixed by commit aa41bc794c51a4d5c364cca014c199b1a00d26aa
- connect_timeout behavior differs between the documentation and the program
- notified
- fixed by commit 5f374fe7a83304fd339789da22599bd102dac9e5
- Improvement in log message of logical replication worker
- fixed by commit 92ecb148e517704ec945dce513db71fee7790cfd
- Server ignores SASL mechanism name defined in SASLInitialResponse (Author: Heikki Linnakangas; Owner: Heikki Linnakangas)
- fixed by commit 505b5d2f8672f13c98dd744a6d421da14f59cd39
- Partition definition for money type looks broken (principal author: Amit Langote ; owner: Robert Haas)
- fixed by commit 76a3df6e5e621928fbf0cddf347e16a62e9433ec
- PATCH: recursive json_populate_record() has bugs
- fixed by commits e45c5be99d08d7bb6708d7bb1dd0f5d99798c6aa and 68cff231e3a3d0ca9988cf1fde5a3be53235b3bb
- lack of location support in outfuncs/readfuncs for partitioning node types
- fixed by commit 80f583ffe930b21d6e5c47be4342356e57851a9a
- ALTER PUBLICATION documentation: referenced first variant no longer exists
- fixed by commit 185364b161512806d23ca390f5b615666079699b
- pg_basebackup generates non-unique slot names
- fixed by commit 2712da8b64b4e399a2666cce2c25329f4f834f2d
- "create publication..all tables" ignore 'partition not supported' error
- fixed by commit 2d460179baa8744e9e2a183a5121306596c53fba
- Create subscription with `create_slot=false` and incorrect slot name should ERROR
- fixed by commits 420a0392ef8fdac3eb6f0a616c136215f7454674, de492c17f064ea3ddcb73d9529f3e30a1483ffa5
- Dropping subscription may stuck if done during tablesync
- fixed by several patches June 2/3
- DDL queries for subcriptions not working in single mode
- fixed by commit 06bfb801c73c89e66f44c1cf693386793e98b637
- Background worker display in pg_stat_activity (logical replication especially)
- fixed by commit 41a21bf9b4a2449eddc8ea2b29597e835eea9bfd
- libpq misses cleanup of SASL context with sslmode=prefer
- fixed by commit e6c33d594a004a2c831cdff1a16276347d30f703
- BR triggers might break a partition constraint during tuple-routing
- fixed by commit 15ce775faa428dc91027e4e2d6b7a167a27118b5
- logical replication and PANIC during shutdown checkpoint in publisher
- fixed in 2017-06-05 Freund commits
- Also affects v9.4-v9.6 but v10 can trigger it from more places
- major part fixed by commit 086221cf6b1727c2baed4703c582f657b7c5350e, but that commit had weaknesses
- namely, HOT pruning during logical decoding might generate WAL
- alternative patches exist
- walsender termination error messages worse in v10
- fixed by commit 5fd56b9f5b4a007a4122c313a184f78f1647c4ab
- ALTER SUBSCRIPTION REFRESH and table sync worker
- fixed by commit ddd7b22b225ae41d16ceb218b387645cb9becfdc
- Allow ICU builds in MSVC
- fixed by commit 03c396080ddc77b188a11dd54aa2a075ab2718e0
- Dropping partitioned table drops a previously detached partition
- fixed by commit ee252f074b88e34ff7ac2b45a73d3cee12b1c671
- Transactional sequence stuff breaks pg_upgrade
- proximate cause is probably commit 3d79013b9
- fixed by commit a475e46634dc7abde1d5a6fc7aaa708219383004
- const-simplification of SRF calls now has user-visible effects
- proposed solution is to start throwing an error for SRF inside CASE or COALESCE
- fixed by commit 0436f6bde8848b7135f19dd7f8548b8c2ae89a34
- Query fails when SRFs are part of FROM clause
- original commit: 69f4b9c (principal author: Tom Lane; owner: Andres Freund notified)
- owner proposes classifying this as a non-bug
- Really this is a variant of the SRF-inside-CASE issue, and was resolved by decision to treat that as an error
- hence, fixed by commit 0436f6bde8848b7135f19dd7f8548b8c2ae89a34
- RTE_NAMEDTUPLESTORE, enrtuples and comments
- fixed by commit f32d57fd7088a558dadbe21b9859b09d2f877c19
- psql's probe for a table's publications sometimes fails
- fixed by commit a2141c42f9ebc51b4501a4fafea9dd3fb7eda23d
- initdb has become much too trusting of locale -a output
- fixed by commits 0b13b2a7712b6f91590b7589a314240a14520c2f and ddb5fdc068635d003a0d1c303cb109d1cb3ebeb1
- transition table behavior with inheritance appears broken
- fixed by commit 501ed02cf6f4f60c3357775eb07578aebc912d3a
- transition tables get confused by wCTEs
- fixed by commit c46c0e5202e8cfe750c6629db7852fdb15d528f3
- transition tables' interaction with INSERT ... ON CONFLICT is questionable
- fixed by commit 8c55244ae379822d8bf62f6db0b5b1f7637eea3a
- Fix WAL receiver spinlock use
- Fixed by commit 572d6ee6d41b43b8871f42c7dbbef795523b2dbf
- Buffer overrun hazards in ICU patch
- Fixed by commit d8b3c81335600ad3487ca9bd642ef354d62919dc
- get_random_uuid() and get_random_bytes() should behave similarly with --disable-strong-random
- Fixed by commit bf723a274cbb00c7fba66c66312a77940af13d79
- "ERROR: tuple concurrently updated" when changing subscription state
- Fixed by commit cb9079cd51a2df677dc182aec72d88383b9c2a79
- Transitions tuples not captured for COPY when using heap_multi_insert and there is an index
- Fixed by commit 1add0b15f117769f619af12720bea2f73d4f7359
- retry shm attach for Windows to mitigate the impact of ASLR
- code is in the same state as previous releases, but it seems advisable to fix it sooner rather than later; eventually Microsoft will force the issue
- Fixed by commit 45e004fb4e3937dbdabf6d5c1706f1a02fdceb94
resolved before 10beta3
- Problems defining mutli-column range partition bounds
- original commit: f0e4475 (principal author: Amit Langote; owner: Dean Rasheed, previously Robert Haas)
- Patch exists
- fixed by commit d363d42bb9a4399a0207bd3b371c966e22e06bd3
- Error'ed tuple's slot descriptor not set correctly when constraint on a partitioned table fails
- fixed by commit 4132dbec69dd4d437e132e57a74a98a40cdcf776
- Race conditions in logrep worker restart
- fixed by commit 9915de6c1cb2c9b87f5f504c97832cdf3a809753
- pg_stop_backup's wait_for_archive is ignored on standbys
- original commit: 017e4f2 (principal author: David Steele; owner: Stephen Frost notified)
- patch exists
- fixed by commit 52f8a59dd953c6820baf153e97cf07d31b8ac1d6
resolved before 10beta4
- Assertion failure in clog.c
- fixed by commit ec99dd5aee8b831760046d43098c2d1cf23157ed
- Bug in free space percent calculation in pgstathashindex
- fixed by commit 0b7ba3d6474b8f58e74dba548886df3250805cdf
- Fix documentation format of table for wait events in pg_stat_activity
- original commit: 030273b7 (principal author and owner: Alvaro Herrera)
- fixed by commit e88928c50dfe2623c899f82b54aad69da248ad07
- error handling for AutovacuumWorkItem is busted, and it doesn't work if dynamic_shared_memory_type=none
- Fixed by commits 870da1e1546c9563c5ae45918392cf7bbc7e5b0e and f2f9fcb3030ba67b4347a44b18b201f84d23997b
- initdb creates some ICU pg_collation entries whose names could be unstable across pg_upgrade/ICU version upgrade
- original commit: eccfef8 (principal author: Peter Eisentraut; owner: Peter Eisentraut notified)
- Item is probably fixable by making sure that all ICU locales get pg_collation entries; not just all distinct ICU collations. The main issue here is that locales may or may not share a collation, which is essentially an ICU/CLDR implementation detail subject to change. The BCP 47 language tag format makes stability guarantees about locale names [1], not collations. Those are guarantees that we ought to be relying on. In practice, CLDR locale names are completely stable aside from changes needed due to cultural or geopolitical developments.
- Separately, it's clear that initdb should not generate variant keyword entries (e.g. -emoji subtag variants).
- Also, it doesn't seem ideal that CREATE COLLATION will not reject ICU collation names no matter how far they are from a valid BCP 47 language tag.
- fixed by commits a79fb8e0c452a9b88206e2abd4add2b432a2596b 958ffb8c286d93d1bfced17e6300d13f9634b431
- Fix drop replication slot blocking instead of returning error
- Fixed by commit be7161566db
- More replication race conditions for logical decoding tests
- Fixed by commit 4f27c674fd9
- Fix race conditions recovery test dedicated to two-phase commit
- Fixed by commit 89c59b742a
resolved before 10rc1
- GatherMerge is not hooked up to ExecReScan
- original commit: 355d399 (principal author: Rushabh Lathia; owner: Robert Haas notified)
- Fixed by commits 1295a777882b8ed180a854e249f91be565a99422 and a2b70c89ca1a5fcf6181d3c777d82e7b83d2de1b
- Fixed some more by commits 7df2c1f8daeb361133ac8bdeaf59ceb0484e315a, 41b0dd987d44089dc48e9c70024277e253b396b7
- Dropped column issue with transition tables
- original commit: 18ce3a4 (principal author: Kevin Grittner;)
- Fixed by commit: 483882905a9a5dc72c9487ceee12320b9630ba2b
- Warnings "unrecognized node type" for some DDLs with log_statement = 'ddl'
- original commit: 7b504eb2 (principal author: Tomas Vondra; owner: Alvaro Herrera)
- fixed by commit 81276fdd3931d286e62b86b2512a517de2ba2de8
- Partition bounds representation issue
- fixed by commit e8b65986ba0de2daeb5bcedc02fb936b04fe464c
- Trigger transition tables vs RI cascades
- crash fixed by commit 5c11717185bc24a2d0a20b38815e182ed99101ce
- but there's still a SQL spec compliance issue
- fixed by commits 54d4d0ff6cd40638d026c01e46deb102e7951ba6 and 5cc23493195cbd2205d5e476e725657c0a29ea34
- Wait event documentation missing two entries for LWLock
- Fixed by commits e9c9ba7845 and 885cab5811
resolved before 10.0
- !USE_WIDE_UPPER_LOWER compile errors in v10+
- fixed by commit c08c98df3d30c0d773d5624860145fb4215b84fb
- "inconsistent page found" with checksum and wal_consistency_checking enabled; see post
- fixed by commit 1a44df007c9b9adc5e6082fc90fe68e615d38ecd
- ALTER TYPE ADD VALUE improvement fails in a common pg_restore scenario
- fixed by commit 93a1af0b3f63838774a9e524589344c3d44c867d
- Shaky coding for vacuuming partitioned relations
- original commit: f0e4475 (principal author: Amit Langote; owner: Robert Haas notified)
- Cache lookup error with elog messages can easily show up, this can be solved with a higher lock taken when scanning all relations.
- patch available
- fixed by commit 2aab70205be012d06f7d077dd1fa5e6afea9d19c
Important Dates
Current schedule:
- feature freeze: April 7th
- beta1: wrap May 15th, announce May 18th
- beta2: wrap July 10th, announce July 13th
- beta3: wrap August 7th, announce August 10th
- beta4: wrap August 28, announce August 31
- rc1: wrap September 18, announce September 21
- 10.0: wrap October 2, announce October 5
References:
