mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-30 11:32:03 +00:00
9.7 KiB
9.7 KiB
ClickHouse release v20.9.4.76-stable FIXME as compared to v20.9.3.45-stable
Improvement
- Backported in #16145: Now it's allowed to execute
ALTER ... ON CLUSTER
queries regardless of the<internal_replication>
setting in cluster config. #16075 (alesapin). - Backported in #16312: Add allow_nondeterministic_optimize_skip_unused_shards (to allow non deterministic like rand() or dictGet() in sharding key). #16105 (Azat Khuzhin).
Bug Fix
- Backported in #15618: Throw an error when a single parameter is passed to ReplicatedMergeTree instead of ignoring it. #15516 (nvartolomei).
- Backported in #16201: Decrement the
ReadonlyReplica
metric when detaching read-only tables. This fixes #15598. #15592 (sundyli). - Backported in #16229: Fixed bug with globs in S3 table function, region from URL was not applied to S3 client configuration. #15646 (Vladimir Chebotarev).
- Backported in #15870: Fix error
Cannot add simple transform to empty Pipe
which happened while reading fromBuffer
table which has different structure than destination table. It was possible if destination table returned empty result for query. Fixes #15529. #15662 (Nikolai Kochetov). - Backported in #15774: Fixed too low default value of
max_replicated_logs_to_keep
setting, which might cause replicas to become lost too often. Improve lost replica recovery process by choosing the most up-to-date replica to clone. Also do not remove old parts from lost replica, detach them instead. #15701 (Alexander Tokmakov). - Backported in #15798: Fix some cases of queries, in which only virtual columns are selected. Previously
Not found column _nothing in block
exception may be thrown. Fixes #12298. #15756 (Anton Popov). - Backported in #15868: Fix
select count()
inaccuracy for MaterializeMySQL. #15767 (Alexander Tokmakov). - Backported in #15901: Fix exception
Block structure mismatch
inSELECT ... ORDER BY DESC
queries which were executed afterALTER MODIFY COLUMN
query. Fixes #15800. #15852 (alesapin). - Backported in #15924: Now exception will be thrown when
ALTER MODIFY COLUMN ... DEFAULT ...
has incompatible default with column type. Fixes #15854. #15858 (alesapin). - Backported in #15919: Fix possible deadlocks in RBAC. #15875 (Vitaly Baranov).
- Backported in #16167: Fix incorrect empty result for query from
Distributed
table if query hasWHERE
,PREWHERE
andGLOBAL IN
. Fixes #15792. #15933 (Nikolai Kochetov). - Backported in #16356: Fixed
DROP TABLE IF EXISTS
failure withTable ... doesn't exist
error when table is concurrently renamed (for Atomic database engine). Fixed rare deadlock when concurrently executing some DDL queries with multiple tables (likeDROP DATABASE
andRENAME TABLE
) FixedDROP/DETACH DATABASE
failure withTable ... doesn't exist
when concurrently executingDROP/DETACH TABLE
. #15934 (Alexander Tokmakov). - Backported in #15971: Fix a crash when database creation fails. #15954 (Winter Zhang).
- Backported in #16022: Fix ambiguity in parsing of settings profiles:
CREATE USER ... SETTINGS profile readonly
is now considered as using a profile namedreadonly
, not a setting namedprofile
with the readonly constraint. This fixes #15628. #15982 (Vitaly Baranov). - Backported in #16218: Fix rare segfaults when inserting into or selecting from MaterializedView and concurrently dropping target table (for Atomic database engine). #15984 (Alexander Tokmakov).
- Backported in #16026: Prevent replica hang for 5-10 mins when replication error happens after a period of inactivity. #15987 (filimonov).
- Backported in #16091: Allow to use direct layout for dictionaries with complex keys. #16007 (Anton Popov).
- Backported in #16361: Fix collate name & charset name parser and support
length = 0
for string type. #16008 (Winter Zhang). - Backported in #16140: Fix
ALTER MODIFY ... ORDER BY
query hang forReplicatedVersionedCollapsingMergeTree
. This fixes #15980. #16011 (alesapin). - Backported in #16079: Fixes #15780 regression, e.g. indexOf([1, 2, 3], toLowCardinality(1)) now is prohibited but it should not be. #16038 (Mike Kot).
- Backported in #16298: Fix dictGet in sharding_key (and similar places, i.e. when the function context is stored permanently). #16205 (Azat Khuzhin).
- Backported in #16232: Fix the case when memory can be overallocated regardless to the limit. This closes #14560. #16206 (Alexey Milovidov).
- Backported in #16326: Fix a possible memory leak during
GROUP BY
with string keys, caused by an error inTwoLevelStringHashTable
implementation. #16264 (Amos Bird). - Backported in #16377: Fix async Distributed INSERT w/ prefer_localhost_replica=0 and internal_replication. #16358 (Azat Khuzhin).
- Backported in #16432: Fix group by with totals/rollup/cube modifers and min/max functions over group by keys. Fixes #16393. #16397 (Anton Popov).
- Backported in #16449: Fix double free in case of exception in function
dictGet
. It could have happened if dictionary was loaded with error. #16429 (Nikolai Kochetov).
Other
- Unfold
{database}
,{table}
and{uuid}
macros inReplicatedMergeTree
arguments on table creation. #16160 (Alexander Tokmakov).