Table of Contents
ClickHouse release v23.4, 2023-04-26
ClickHouse release v23.3 LTS, 2023-03-30
ClickHouse release v23.2, 2023-02-23
ClickHouse release v23.1, 2023-01-25
Changelog for 2022
2023 Changelog
ClickHouse release 23.4, 2023-04-26
Backward Incompatible Change
- Formatter '%M' in function formatDateTime() now prints the month name instead of the minutes. This makes the behavior consistent with MySQL. The previous behavior can be restored using setting "formatdatetime_parsedatetime_m_is_month_name = 0". #47246 (Robert Schulze).
- This change makes sense only if you are using the virtual filesystem cache. If
pathin the virtual filesystem cache configuration is not empty and is not an absolute path, then it will be put in<clickhouse server data directory>/caches/<path_from_cache_config>. #48784 (Kseniia Sumarokova). - Primary/secondary indices and sorting keys with identical expressions are now rejected. This behavior can be disabled using setting
allow_suspicious_indices. #48536 (凌涛).
New Feature
- Support new aggregate function
quantileGK/quantilesGK, like approx_percentile in spark. Greenwald-Khanna algorithm refer to http://infolab.stanford.edu/~datar/courses/cs361a/papers/quantiles.pdf. #46428 (李扬). - Add a statement
SHOW COLUMNSwhich shows distilled information from system.columns. #48017 (Robert Schulze). - Added
LIGHTWEIGHTandPULLmodifiers forSYSTEM SYNC REPLICAquery.LIGHTWEIGHTversion waits for fetches and drop-ranges only (merges and mutations are ignored).PULLversion pulls new entries from ZooKeeper and does not wait for them. Fixes #47794. #48085 (Alexander Tokmakov). - Add
kafkaMurmurHashfunction for compatibility with Kafka DefaultPartitioner. Closes #47834. #48185 (Nikolay Degterinsky). - Allow to easily create a user with the same grants as the current user by using
GRANT CURRENT GRANTS. #48262 (pufit). - Add statistical aggregate function
kolmogorovSmirnovTest. Close #48228. #48325 (FFFFFFFHHHHHHH). - Added a
lost_part_countcolumn to thesystem.replicastable. The column value shows the total number of lost parts in the corresponding table. Value is stored in zookeeper and can be used instead of not persistentReplicatedDataLossprofile event for monitoring. #48526 (Sergei Trifonov). - Add
soundexfunction for compatibility. Closes #39880. #48567 (FriendLey). - Support
Maptype for JSONExtract. #48629 (李扬). - Add
PrettyJSONEachRowformat to output pretty JSON with new line delimiters and 4 space indents. #48898 (Kruglov Pavel). - Add
ParquetMetadatainput format to read Parquet file metadata. #48911 (Kruglov Pavel). - Add
extractKeyValuePairsfunction to extract key value pairs from strings. Input strings might contain noise (i.e. log files / do not need to be 100% formatted in key-value-pair format), the algorithm will look for key value pairs matching the arguments passed to the function. As of now, function accepts the following arguments:data_column(mandatory),key_value_pair_delimiter(defaults to:),pair_delimiters(defaults to\space \, \;) andquoting_character(defaults to double quotes). #43606 (Arthur Passos). - Functions replaceOne(), replaceAll(), replaceRegexpOne() and replaceRegexpAll() can now be called with non-const pattern and replacement arguments. #46589 (Robert Schulze).
- Added functions to work with columns of type
Map:mapConcat,mapSort,mapExists. #48071 (Anton Popov).
Performance Improvement
- Reading files in
Parquetformat is now much faster. IO and decoding are parallelized (controlled bymax_threadssetting), and only required data ranges are read. #47964 (Michael Kolupaev). - If we run a mutation with IN (subquery) like this:
ALTER TABLE t UPDATE col='new value' WHERE id IN (SELECT id FROM huge_table)and the tablethas multiple parts than for each part a set for subquerySELECT id FROM huge_tableis built in memory. And if there are many parts then this might consume a lot of memory (and lead to an OOM) and CPU. The solution is to introduce a short-lived cache of sets that are currently being built by mutation tasks. If another task of the same mutation is executed concurrently it can look up the set in the cache, wait for it to be built and reuse it. #46835 (Alexander Gololobov). - Only check dependencies if necessary when applying
ALTER TABLEqueries. #48062 (Raúl Marín). - Optimize function
mapUpdate. #48118 (Anton Popov). - Now an internal query to local replica is sent explicitly and data from it received through loopback interface. Setting
prefer_localhost_replicais not respected for parallel replicas. This is needed for better scheduling and makes the code cleaner: the initiator is only responsible for coordinating of the reading process and merging results, continuously answering for requests while all the secondary queries read the data. Note: Using loopback interface is not so performant, otherwise some replicas could starve for tasks which could lead to even slower query execution and not utilizing all possible resources. The initialization of the coordinator is now even more lazy. All incoming requests contain the information about the reading algorithm we initialize the coordinator with it when first request comes. If any replica decides to read with a different algorithm–an exception will be thrown and a query will be aborted. #48246 (Nikita Mikhaylov). - Do not build set for the right side of
INclause with subquery when it is used only for analysis of skip indexes, and they are disabled by setting (use_skip_indexes=0). Previously it might affect the performance of queries. #48299 (Anton Popov). - Query processing is parallelized right after reading
FROM file(...). Related to #38755. #48525 (Igor Nikonov). Query processing is parallelized right after reading from any data source. Affected data sources are mostly simple or external storages like table functionsurl,file. #48727 (Igor Nikonov). This is controlled by the settingparallelize_output_from_storageswhich is not enabled by default. - Lowered contention of ThreadPool mutex (may increase performance for a huge amount of small jobs). #48750 (Sergei Trifonov).
- Reduce memory usage for multiple
ALTER DELETEmutations. #48522 (Nikolai Kochetov). - Remove the excessive connection attempts if the
skip_unavailable_shardssetting is enabled. #48771 (Azat Khuzhin).
Experimental Feature
- Entries in the query cache are now squashed to max_block_size and compressed. #45912 (Robert Schulze).
- It is now possible to define per-user quotas in the query cache. #48284 (Robert Schulze).
- Some fixes for parallel replicas #48433 (Nikita Mikhaylov).
- Implement zero-copy-replication (an experimental feature) on encrypted disks. #48741 (Vitaly Baranov).
Improvement
- Increase default value for
connect_timeout_with_failover_msto 1000 ms (because of adding async connections in #47229) . Closes #5188. #49009 (Kruglov Pavel). - Several improvements around data lakes: - Make
Icebergwork with non-partitioned data. - SupportIcebergformat version v2 (previously only v1 was supported) - Support reading partitioned data forDeltaLake/Hudi- Faster reading ofDeltaLakemetadata by using Delta's checkpoint files - Fixed incorrectHudireads: previously it incorrectly chose which data to read and therefore was able to read correctly only small size tables - Made these engines to pickup updates of changed data (previously the state was set on table creation) - Make proper testing forIceberg/DeltaLake/Hudiusing spark. #47307 (Kseniia Sumarokova). - Add async connection to socket and async writing to socket. Make creating connections and sending query/external tables async across shards. Refactor code with fibers. Closes #46931. We will be able to increase
connect_timeout_with_failover_msby default after this PR (#5188). #47229 (Kruglov Pavel). - Support config sections
keeper/keeper_serveras an alternative tozookeeper. Close #34766 , #34767. #35113 (李扬). - It is possible to set secure flag in named_collections for a dictionary with a ClickHouse table source. Addresses #38450 . #46323 (Ilya Golshtein).
bitCountfunction supportFixedStringandStringdata type. #49044 (flynn).- Added configurable retries for all operations with [Zoo]Keeper for Backup queries. #47224 (Nikita Mikhaylov).
- Enable
use_environment_credentialsfor S3 by default, so the entire provider chain is constructed by default. #47397 (Antonio Andelic). - Currently, the JSON_VALUE function is similar as spark's get_json_object function, which support to get value from JSON string by a path like '$.key'. But still has something different - 1. in spark's get_json_object will return null while the path is not exist, but in JSON_VALUE will return empty string; - 2. in spark's get_json_object will return a complex type value, such as a JSON object/array value, but in JSON_VALUE will return empty string. #47494 (KevinyhZou).
- For
use_structure_from_insertion_table_in_table_functionsmore flexible insert table structure propagation to table function. Fixed an issue with name mapping and using virtual columns. No more need for 'auto' setting. #47962 (Yakov Olkhovskiy). - Do not continue retrying to connect to Keeper if the query is killed or over limits. #47985 (Raúl Marín).
- Support Enum output/input in
BSONEachRow, allow all map key types and avoid extra calculations on output. #48122 (Kruglov Pavel). - Support more ClickHouse types in
ORC/Arrow/Parquetformats: Enum(8|16), (U)Int(128|256), Decimal256 (for ORC), allow reading IPv4 from Int32 values (ORC outputs IPv4 as Int32, and we couldn't read it back), fix reading Nullable(IPv6) from binary data forORC. #48126 (Kruglov Pavel). - Add columns
perform_ttl_move_on_insert,load_balancingfor tablesystem.storage_policies, modify columnvolume_typetype toEnum8. #48167 (lizhuoyu5). - Added support for
BACKUP ALLcommand which backups all tables and databases, including temporary and system ones. #48189 (Vitaly Baranov). - Function mapFromArrays supports
Maptype as an input. #48207 (李扬). - The output of some SHOW PROCESSLIST is now sorted. #48241 (Robert Schulze).
- Per-query/per-server throttling for remote IO/local IO/BACKUPs (server settings:
max_remote_read_network_bandwidth_for_server,max_remote_write_network_bandwidth_for_server,max_local_read_bandwidth_for_server,max_local_write_bandwidth_for_server,max_backup_bandwidth_for_server, settings:max_remote_read_network_bandwidth,max_remote_write_network_bandwidth,max_local_read_bandwidth,max_local_write_bandwidth,max_backup_bandwidth). #48242 (Azat Khuzhin). - Support more types in
CapnProtoformat: Map, (U)Int(128|256), Decimal(128|256). Allow integer conversions during input/output. #48257 (Kruglov Pavel). - Don't throw CURRENT_WRITE_BUFFER_IS_EXHAUSTED for normal behaviour. #48288 (Raúl Marín).
- Add new setting
keeper_map_strict_modewhich enforces extra guarantees on operations made on top ofKeeperMaptables. #48293 (Antonio Andelic). - Check primary key type for simple dictionary is native unsigned integer type Add setting
check_dictionary_primary_keyfor compatibility(setcheck_dictionary_primary_key =falseto disable checking). #48335 (lizhuoyu5). - Don't replicate mutations for
KeeperMapbecause it's unnecessary. #48354 (Antonio Andelic). - Allow to write/read unnamed tuple as nested Message in Protobuf format. Tuple elements and Message fields are matched by position. #48390 (Kruglov Pavel).
- Support
additional_table_filtersandadditional_result_filtersettings in the new planner. Also, add a documentation entry foradditional_result_filter. #48405 (Dmitry Novik). parseDateTimenow understands format string '%f' (fractional seconds). #48420 (Robert Schulze).- Format string "%f" in formatDateTime() now prints "000000" if the formatted value has no fractional seconds, the previous behavior (single zero) can be restored using setting "formatdatetime_f_prints_single_zero = 1". #48422 (Robert Schulze).
- Don't replicate DELETE and TRUNCATE for KeeperMap. #48434 (Antonio Andelic).
- Generate valid Decimals and Bools in generateRandom function. #48436 (Kruglov Pavel).
- Allow trailing commas in expression list of SELECT query, for example
SELECT a, b, c, FROM table. Closes #37802. #48438 (Nikolay Degterinsky). - Override
CLICKHOUSE_USERandCLICKHOUSE_PASSWORDenvironment variables with--userand--passwordclient parameters. Closes #38909. #48440 (Nikolay Degterinsky). - Added retries to loading of data parts in
MergeTreetables in case of retryable errors. #48442 (Anton Popov). - Add support for
Date,Date32,DateTime,DateTime64data types toarrayMin,arrayMax,arrayDifferencefunctions. Closes #21645. #48445 (Nikolay Degterinsky). - Add support for
{server_uuid}macro. It is useful for identifying replicas in autoscaled clusters when new replicas are constantly added and removed in runtime. This closes #48554. #48563 (Alexey Milovidov). - The installation script will create a hard link instead of copying if it is possible. #48578 (Alexey Milovidov).
- Support
SHOW TABLEsyntax meaning the same asSHOW CREATE TABLE. Closes #48580. #48591 (flynn). - HTTP temporary buffers now support working by evicting data from the virtual filesystem cache. #48664 (Vladimir C).
- Make Schema inference works for
CREATE AS SELECT. Closes #47599. #48679 (flynn). - Added a
replicated_max_mutations_in_one_entrysetting forReplicatedMergeTreethat allows limiting the number of mutation commands per oneMUTATE_PARTentry (default is 10000). #48731 (Alexander Tokmakov). - In AggregateFunction types, don't count unused arena bytes as
read_bytes. #48745 (Raúl Marín). - Fix some MySQL-related settings not being handled with the MySQL dictionary source + named collection. Closes #48402. #48759 (Kseniia Sumarokova).
- If a user set
max_single_part_upload_sizeto a very large value, it can lead to a crash due to a bug in the AWS S3 SDK. This fixes #47679. #48816 (Alexey Milovidov). - Fix data race in
RabbitMQ(report), refactor the code. #48845 (Kseniia Sumarokova). - Add aliases
nameandpart_nameformsystem.partsandsystem.part_log. Closes #48718. #48850 (sichenzhao). - Functions "arrayDifferenceSupport()", "arrayCumSum()" and "arrayCumSumNonNegative()" now support input arrays of wide integer types (U)Int128/256. #48866 (cluster).
- Multi-line history in clickhouse-client is now no longer padded. This makes pasting more natural. #48870 (Joanna Hulboj).
- Implement a slight improvement for the rare case when ClickHouse is run inside LXC and LXCFS is used. The LXCFS has an issue: sometimes it returns an error "Transport endpoint is not connected" on reading from the file inside
/proc. This error was correctly logged into ClickHouse's server log. We have additionally workaround this issue by reopening a file. This is a minuscule change. #48922 (Real). - Improve memory accounting for prefetches. Randomise prefetch settings In CI. #48973 (Kseniia Sumarokova).
- Correctly set headers for native copy operations on GCS. #48981 (Antonio Andelic).
- Add support for specifying setting names in the command line with dashes instead of underscores, for example,
--max-threadsinstead of--max_threads. Additionally, support Unicode dash characters like—instead of--- this is useful when you communicate with a team in another company, and a manager from that team copy-pasted code from MS Word. #48985 (alekseygolub). - Add fallback to password authentication when authentication with SSL user certificate has failed. Closes #48974. #48989 (Nikolay Degterinsky).
- Improve the embedded dashboard. Close #46671. #49036 (Kevin Zhang).
- Add profile events for log messages, so you can easily see the count of log messages by severity. #49042 (Alexey Milovidov).
- In previous versions, the
LineAsStringformat worked inconsistently when the parallel parsing was enabled or not, in presence of DOS or macOS Classic line breaks. This closes #49039. #49052 (Alexey Milovidov). - The exception message about the unparsed query parameter will also tell about the name of the parameter. Reimplement #48878. Close #48772. #49061 (Alexey Milovidov).
Build/Testing/Packaging Improvement
- Update time zones. The following were updated: Africa/Cairo, Africa/Casablanca, Africa/El_Aaiun, America/Bogota, America/Cambridge_Bay, America/Ciudad_Juarez, America/Godthab, America/Inuvik, America/Iqaluit, America/Nuuk, America/Ojinaga, America/Pangnirtung, America/Rankin_Inlet, America/Resolute, America/Whitehorse, America/Yellowknife, Asia/Gaza, Asia/Hebron, Asia/Kuala_Lumpur, Asia/Singapore, Canada/Yukon, Egypt, Europe/Kirov, Europe/Volgograd, Singapore. #48572 (Alexey Milovidov).
- Reduce the number of dependencies in the header files to speed up the build. #47984 (Dmitry Novik).
- Randomize compression of marks and indices in tests. #48286 (Alexey Milovidov).
- Bump internal ZSTD from 1.5.4 to 1.5.5. #46797 (Robert Schulze).
- Randomize vertical merges from compact to wide parts in tests. #48287 (Raúl Marín).
- Support for CRC32 checksum in HDFS. Fix performance issues. #48614 (Alexey Milovidov).
- Remove remainders of GCC support. #48671 (Robert Schulze).
- Add CI run with new analyzer infrastructure enabled. #48719 (Dmitry Novik).
Bug Fix (user-visible misbehavior in an official stable release)
- Fix system.query_views_log for MVs that are pushed from background threads #46668 (Azat Khuzhin).
- Fix several
RENAME COLUMNbugs #46946 (alesapin). - Fix minor hiliting issues in clickhouse-format #47610 (Natasha Murashkina).
- Fix a bug in LLVM's libc++ leading to a crash for uploading parts to S3 which size is greater than INT_MAX #47693 (Azat Khuzhin).
- Fix overflow in the
sparkbarfunction #48121 (Vladimir C). - Fix race in S3 #48190 (Anton Popov).
- Disable JIT for aggregate functions due to inconsistent behavior #48195 (Alexey Milovidov).
- Fix alter formatting (minor) #48289 (Natasha Murashkina).
- Fix CPU usage in RabbitMQ (was worsened in 23.2 after #44404) #48311 (Kseniia Sumarokova).
- Fix crash in EXPLAIN PIPELINE for Merge over Distributed #48320 (Azat Khuzhin).
- Fix serializing LowCardinality as Arrow dictionary #48361 (Kruglov Pavel).
- Reset downloader for cache file segment in TemporaryFileStream #48386 (Vladimir C).
- Fix possible SYSTEM SYNC REPLICA stuck in case of DROP/REPLACE PARTITION #48391 (Azat Khuzhin).
- Fix a startup error when loading a distributed table that depends on a dictionary #48419 (MikhailBurdukov).
- Don't check dependencies when renaming system tables automatically #48431 (Raúl Marín).
- Update only affected rows in KeeperMap storage #48435 (Antonio Andelic).
- Fix possible segfault in the VFS cache #48469 (Kseniia Sumarokova).
toTimeZonefunction throws an error when no constant string is provided #48471 (Jordi Villar).- Fix logical error with IPv4 in Protobuf, add support for Date32 #48486 (Kruglov Pavel).
- "changed" flag in system.settings was calculated incorrectly for settings with multiple values #48516 (MikhailBurdukov).
- Fix storage
Memorywith enabled compression #48517 (Anton Popov). - Fix bracketed-paste mode messing up password input in the event of client reconnection #48528 (Michael Kolupaev).
- Fix nested map for keys of IP and UUID types #48556 (Yakov Olkhovskiy).
- Fix an uncaught exception in case of parallel loader for hashed dictionaries #48571 (Azat Khuzhin).
- The
groupArrayaggregate function correctly works for empty result over nullable types #48593 (lgbo). - Fix bug in Keeper when a node is not created with scheme
authin ACL sometimes. #48595 (Aleksei Filatov). - Allow IPv4 comparison operators with UInt #48611 (Yakov Olkhovskiy).
- Fix possible error from cache #48636 (Kseniia Sumarokova).
- Async inserts with empty data will no longer throw exception. #48663 (Anton Popov).
- Fix table dependencies in case of failed RENAME TABLE #48683 (Azat Khuzhin).
- If the primary key has duplicate columns (which is only possible for projections), in previous versions it might lead to a bug #48838 (Amos Bird).
- Fix for a race condition in ZooKeeper when joining send_thread/receive_thread #48849 (Alexander Gololobov).
- Fix unexpected part name error when trying to drop a ignored detached part with zero copy replication #48862 (Michael Lex).
- Fix reading
Date32Parquet/Arrow column into not aDate32column #48864 (Kruglov Pavel). - Fix
UNKNOWN_IDENTIFIERerror while selecting from table with row policy and column with dots #48976 (Kruglov Pavel). - Fix aggregation by empty nullable strings #48999 (LiuNeng).
ClickHouse release 23.3 LTS, 2023-03-30
Upgrade Notes
- Lightweight DELETEs are production ready and enabled by default. The
DELETEquery for MergeTree tables is now available by default. - The behavior of
*domain*RFCandnetlocfunctions is slightly changed: relaxed the set of symbols that are allowed in the URL authority for better conformance. #46841 (Azat Khuzhin). - Prohibited creating tables based on KafkaEngine with DEFAULT/EPHEMERAL/ALIAS/MATERIALIZED statements for columns. #47138 (Aleksandr Musorin).
- An "asynchronous connection drain" feature is removed. Related settings and metrics are removed as well. It was an internal feature, so the removal should not affect users who had never heard about that feature. #47486 (Alexander Tokmakov).
- Support 256-bit Decimal data type (more than 38 digits) in
arraySum/Min/Max/Avg/Product,arrayCumSum/CumSumNonNegative,arrayDifference, array construction, IN operator, query parameters,groupArrayMovingSum, statistical functions,min/max/any/argMin/argMax, PostgreSQL wire protocol, MySQL table engine and function,sumMap,mapAdd,mapSubtract,arrayIntersect. Add support for big integers inarrayIntersect. Statistical aggregate functions involving moments (such ascorror variousTTests) will useFloat64as their internal representation (they were usingDecimal128before this change, but it was pointless), and these functions can returnnaninstead ofinfin case of infinite variance. Some functions were allowed onDecimal256data types but returnedDecimal128in previous versions - now it is fixed. This closes #47569. This closes #44864. This closes #28335. #47594 (Alexey Milovidov). - Make backup_threads/restore_threads server settings (instead of user settings). #47881 (Azat Khuzhin).
- Do not allow const and non-deterministic secondary indices #46839 (Anton Popov).
New Feature
- Add a new mode for splitting the work on replicas using settings
parallel_replicas_custom_keyandparallel_replicas_custom_key_filter_type. If the cluster consists of a single shard with multiple replicas, up tomax_parallel_replicaswill be randomly picked and turned into shards. For each shard, a corresponding filter is added to the query on the initiator before being sent to the shard. If the cluster consists of multiple shards, it will behave the same assample_keybut with the possibility to define an arbitrary key. #45108 (Antonio Andelic). - An option to display partial result on cancel: Added query setting
partial_result_on_first_cancelallowing the canceled query (e.g. due to Ctrl-C) to return a partial result. #45689 (Alexey Perevyshin). - Added support of arbitrary tables engines for temporary tables (except for Replicated and KeeperMap engines). Close #31497. #46071 (Roman Vasin).
- Add support for replication of user-defined SQL functions using centralized storage in Keeper. #46085 (Aleksei Filatov).
- Implement
system.server_settings(similar tosystem.settings), which will contain server configurations. #46550 (pufit). - Support for
UNDROP TABLEquery. Closes #46811. #47241 (chen). - Allow separate grants for named collections (e.g. to be able to give
SHOW/CREATE/ALTER/DROP named collectionaccess only to certain collections, instead of all at once). Closes #40894. Add new access typeNAMED_COLLECTION_CONTROLwhich is not given to user default unless explicitly added to the user config (is required to be able to doGRANT ALL), alsoshow_named_collectionsis no longer obligatory to be manually specified for user default to be able to have full access rights as was in 23.2. #46241 (Kseniia Sumarokova). - Allow nested custom disks. Previously custom disks supported only flat disk structure. #47106 (Kseniia Sumarokova).
- Introduce a function
widthBucket(with aWIDTH_BUCKETalias for compatibility). #42974. #46790 (avoiderboi). - Add new function
parseDateTime/parseDateTimeInJodaSyntaxaccording to the specified format string. parseDateTime parses String to DateTime in MySQL syntax, parseDateTimeInJodaSyntax parses in Joda syntax. #46815 (李扬). - Use
dummy UInt8for the default structure of table functionnull. Closes #46930. #47006 (flynn). - Support for date format with a comma, like
Dec 15, 2021in theparseDateTimeBestEffortfunction. Closes #46816. #47071 (chen). - Add settings
http_wait_end_of_queryandhttp_response_buffer_sizethat corresponds to URL paramswait_end_of_queryandbuffer_sizefor the HTTP interface. This allows changing these settings in the profiles. #47108 (Vladimir C). - Add
system.dropped_tablestable that shows tables that were dropped fromAtomicdatabases but were not completely removed yet. #47364 (chen). - Add
INSTRas alias ofpositionCaseInsensitivefor MySQL compatibility. Closes #47529. #47535 (flynn). - Added
toDecimalStringfunction allowing to convert numbers to string with fixed precision. #47838 (Andrey Zvonov). - Add a merge tree setting
max_number_of_mutations_for_replica. It limits the number of part mutations per replica to the specified amount. Zero means no limit on the number of mutations per replica (the execution can still be constrained by other settings). #48047 (Vladimir C). - Add the Map-related function
mapFromArrays, which allows the creation of a map from a pair of arrays. #31125 (李扬). - Allow control of compression in Parquet/ORC/Arrow output formats, adds support for more compression input formats. This closes #13541. #47114 (Kruglov Pavel).
- Add SSL User Certificate authentication to the native protocol. Closes #47077. #47596 (Nikolay Degterinsky).
- Add *OrNull() and *OrZero() variants for
parseDateTime, add aliasstr_to_datefor MySQL parity. #48000 (Robert Schulze). - Added operator
REGEXP(similar to operators "LIKE", "IN", "MOD" etc.) for better compatibility with MySQL #47869 (Robert Schulze).
Performance Improvement
- Marks in memory are now compressed, using 3-6x less memory. #47290 (Michael Kolupaev).
- Backups for large numbers of files were unbelievably slow in previous versions. Not anymore. Now they are unbelievably fast. #47251 (Alexey Milovidov). Introduced a separate thread pool for backup's IO operations. This will allow scaling it independently of other pools and increase performance. #47174 (Nikita Mikhaylov). Use MultiRead request and retries for collecting metadata at the final stage of backup processing. #47243 (Nikita Mikhaylov). If a backup and restoring data are both in S3 then server-side copy should be used from now on. #47546 (Vitaly Baranov).
- Fixed excessive reading in queries with
FINAL. #47801 (Nikita Taranov). - Setting
max_final_threadswould be set to the number of cores at server startup (by the same algorithm as used formax_threads). This improves the concurrency offinalexecution on servers with high number of CPUs. #47915 (Nikita Taranov). - Allow executing reading pipeline for DIRECT dictionary with CLICKHOUSE source in multiple threads. To enable set
dictionary_use_async_executor=1inSETTINGSsection for source inCREATE DICTIONARYstatement. #47986 (Vladimir C). - Optimize one nullable key aggregate performance. #45772 (LiuNeng).
- Implemented lowercase
tokenbf_v1index utilization forhasTokenOrNull,hasTokenCaseInsensitiveandhasTokenCaseInsensitiveOrNull. #46252 (ltrk2). - Optimize functions
positionandLIKEby searching the first two chars using SIMD. #46289 (Jiebin Sun). - Optimize queries from the
system.detached_parts, which could be significantly large. Added several sources with respect to the block size limitation; in each block, an IO thread pool is used to calculate the part size, i.e. to make syscalls in parallel. #46624 (Sema Checherinda). - Increase the default value of
max_replicated_merges_in_queuefor ReplicatedMergeTree tables from 16 to 1000. It allows faster background merge operation on clusters with a very large number of replicas, such as clusters with shared storage in ClickHouse Cloud. #47050 (Alexey Milovidov). - Updated
clickhouse-copierto useGROUP BYinstead ofDISTINCTto get the list of partitions. For large tables, this reduced the select time from over 500s to under 1s. #47386 (Clayton McClure). - Fix performance degradation in
ASOF JOIN. #47544 (Ongkong). - Even more batching in Keeper. Improve performance by avoiding breaking batches on read requests. #47978 (Antonio Andelic).
- Allow PREWHERE for Merge with different DEFAULT expressions for columns. #46831 (Azat Khuzhin).
Experimental Feature
- Parallel replicas: Improved the overall performance by better utilizing the local replica, and forbid the reading with parallel replicas from non-replicated MergeTree by default. #47858 (Nikita Mikhaylov).
- Support filter push down to left table for JOIN with
Join,DictionaryandEmbeddedRocksDBtables if the experimental Analyzer is enabled. #47280 (Maksim Kita). - Now ReplicatedMergeTree with zero copy replication has less load to Keeper. #47676 (alesapin).
- Fix create materialized view with MaterializedPostgreSQL #40807 (Maksim Buren).
Improvement
- Enable
input_format_json_ignore_unknown_keys_in_named_tupleby default. #46742 (Kruglov Pavel). - Allow errors to be ignored while pushing to MATERIALIZED VIEW (add new setting
materialized_views_ignore_errors, by default tofalse, but it is set totruefor flushing logs tosystem.*_logtables unconditionally). #46658 (Azat Khuzhin). - Track the file queue of distributed sends in memory. #45491 (Azat Khuzhin).
- Now
X-ClickHouse-Query-IdandX-ClickHouse-Timezoneheaders are added to responses in all queries via HTTP protocol. Previously it was done only forSELECTqueries. #46364 (Anton Popov). - External tables from
MongoDB: support for connection to a replica set via a URI with a host:port enum and support for the readPreference option in MongoDB dictionaries. Example URI: mongodb://db0.example.com:27017,db1.example.com:27017,db2.example.com:27017/?replicaSet=myRepl&readPreference=primary. #46524 (artem-yadr). - This improvement should be invisible for users. Re-implement projection analysis on top of query plan. Added setting
query_plan_optimize_projection=1to switch between old and new version. Fixes #44963. #46537 (Nikolai Kochetov). - Use Parquet format v2 instead of v1 in output format by default. Add setting
output_format_parquet_versionto control parquet version, possible values1.0,2.4,2.6,2.latest(default). #46617 (Kruglov Pavel). - It is now possible to use the new configuration syntax to configure Kafka topics with periods (
.) in their name. #46752 (Robert Schulze). - Fix heuristics that check hyperscan patterns for problematic repeats. #46819 (Robert Schulze).
- Don't report ZK node exists to system.errors when a block was created concurrently by a different replica. #46820 (Raúl Marín).
- Increase the limit for opened files in
clickhouse-local. It will be able to read fromwebtables on servers with a huge number of CPU cores. Do not back off reading from the URL table engine in case of too many opened files. This closes #46852. #46853 (Alexey Milovidov). - Exceptions thrown when numbers cannot be parsed now have an easier-to-read exception message. #46917 (Robert Schulze).
- Added update
system.backupsafter every processed task to track the progress of backups. #46989 (Aleksandr Musorin). - Allow types conversion in Native input format. Add settings
input_format_native_allow_types_conversionthat controls it (enabled by default). #46990 (Kruglov Pavel). - Allow IPv4 in the
rangefunction to generate IP ranges. #46995 (Yakov Olkhovskiy). - Improve exception message when it's impossible to move a part from one volume/disk to another. #47032 (alesapin).
- Support
Booltype inJSONTypefunction. PreviouslyNulltype was mistakenly returned for bool values. #47046 (Anton Popov). - Use
_request_bodyparameter to configure predefined HTTP queries. #47086 (Constantine Peresypkin). - Automatic indentation in the built-in UI SQL editor when Enter is pressed. #47113 (Alexey Korepanov).
- Self-extraction with 'sudo' will attempt to set uid and gid of extracted files to running user. #47116 (Yakov Olkhovskiy).
- Previously, the
repeatfunction's second argument only accepted an unsigned integer type, which meant it could not accept values such as -1. This behavior differed from that of the Spark function. In this update, the repeat function has been modified to match the behavior of the Spark function. It now accepts the same types of inputs, including negative integers. Extensive testing has been performed to verify the correctness of the updated implementation. #47134 (KevinyhZou). Note: the changelog entry was rewritten by ChatGPT. - Remove
::__1part from stacktraces. Displaystd::basic_string<char, ...asStringin stacktraces. #47171 (Mike Kot). - Reimplement interserver mode to avoid replay attacks (note, that change is backward compatible with older servers). #47213 (Azat Khuzhin).
- Improve recognition of regular expression groups and refine the regexp_tree dictionary. #47218 (Han Fei).
- Keeper improvement: Add new 4LW
clrsto clean resources used by Keeper (e.g. release unused memory). #47256 (Antonio Andelic). - Add optional arguments to codecs
DoubleDelta(bytes_size),Gorilla(bytes_size),FPC(level, float_size), this allows using these codecs without column type inclickhouse-compressor. Fix possible aborts and arithmetic errors inclickhouse-compressorwith these codecs. Fixes: #47262. #47271 (Kruglov Pavel). - Add support for big int types to the
runningDifferencefunction. Closes #47194. #47322 (Nikolay Degterinsky). - Add an expiration window for S3 credentials that have an expiration time to avoid
ExpiredTokenerrors in some edge cases. It can be controlled withexpiration_window_secondsconfig, the default is 120 seconds. #47423 (Antonio Andelic). - Support Decimals and Date32 in
Avroformat. #47434 (Kruglov Pavel). - Do not start the server if an interrupted conversion from
OrdinarytoAtomicwas detected, print a better error message with troubleshooting instructions. #47487 (Alexander Tokmakov). - Add a new column
kindto thesystem.opentelemetry_span_log. This column holds the value of SpanKind defined in OpenTelemtry. #47499 (Frank Chen). - Allow reading/writing nested arrays in
Protobufformat with only the root field name as column name. Previously column name should've contained all nested field names (likea.b.c Array(Array(Array(UInt32))), now you can use justa Array(Array(Array(UInt32))). #47650 (Kruglov Pavel). - Added an optional
STRICTmodifier forSYSTEM SYNC REPLICAwhich makes the query wait for the replication queue to become empty (just like it worked before #45648). #47659 (Alexander Tokmakov). - Improve the naming of some OpenTelemetry span logs. #47667 (Frank Chen).
- Prevent using too long chains of aggregate function combinators (they can lead to slow queries in the analysis stage). This closes #47715. #47716 (Alexey Milovidov).
- Support for subquery in parameterized views; resolves #46741 #47725 (SmitaRKulkarni).
- Fix memory leak in MySQL integration (reproduces with
connection_auto_close=1). #47732 (Kseniia Sumarokova). - Improved error handling in the code related to Decimal parameters, resulting in more informative error messages. Previously, when incorrect Decimal parameters were supplied, the error message generated was unclear or unhelpful. With this update, the error message printed has been fixed to provide more detailed and useful information, making it easier to identify and correct issues related to Decimal parameters. #47812 (Yu Feng). Note: this changelog entry is rewritten by ChatGPT.
- The parameter
exact_rows_before_limitis used to makerows_before_limit_at_leastis designed to accurately reflect the number of rows returned before the limit is reached. This pull request addresses issues encountered when the query involves distributed processing across multiple shards or sorting operations. Prior to this update, these scenarios were not functioning as intended. #47874 (Amos Bird). - ThreadPools metrics introspection. #47880 (Azat Khuzhin).
- Add
WriteBufferFromS3MicrosecondsandWriteBufferFromS3RequestsErrorsprofile events. #47885 (Antonio Andelic). - Add
--linkand--noninteractive(-y) options to ClickHouse install. Closes #47750. #47887 (Nikolay Degterinsky). - Fixed
UNKNOWN_TABLEexception when attaching to a materialized view that has dependent tables that are not available. This might be useful when trying to restore state from a backup. #47975 (MikhailBurdukov). - Fix case when the (optional) path is not added to an encrypted disk configuration. #47981 (Kseniia Sumarokova).
- Support for CTE in parameterized views Implementation: Updated to allow query parameters while evaluating scalar subqueries. #48065 (SmitaRKulkarni).
- Support big integers
(U)Int128/(U)Int256,Mapwith any key type andDateTime64with any precision (not only 3 and 6). #48119 (Kruglov Pavel). - Allow skipping errors related to unknown enum values in row input formats. #48133 (Alexey Milovidov).
Build/Testing/Packaging Improvement
- ClickHouse now builds with
C++23. #47424 (Robert Schulze). - Fuzz
EXPLAINqueries in the AST Fuzzer. #47803 #47852 (flynn). - Split stress test and the automated backward compatibility check (now Upgrade check). #44879 (Kruglov Pavel).
- Updated the Ubuntu Image for Docker to calm down some bogus security reports. #46784 (Julio Jimenez). Please note that ClickHouse has no dependencies and does not require Docker.
- Adds a prompt to allow the removal of an existing
clickhousedownload when using "curl | sh" download of ClickHouse. Prompt is "ClickHouse binary clickhouse already exists. Overwrite? [y/N]". #46859 (Dan Roscigno). - Fix error during server startup on old distros (e.g. Amazon Linux 2) and on ARM that glibc 2.28 symbols are not found. #47008 (Robert Schulze).
- Prepare for clang 16. #47027 (Amos Bird).
- Added a CI check which ensures ClickHouse can run with an old glibc on ARM. #47063 (Robert Schulze).
- Add a style check to prevent incorrect usage of the
NDEBUGmacro. #47699 (Alexey Milovidov). - Speed up the build a little. #47714 (Alexey Milovidov).
- Bump
vectorscanto 5.4.9. #47955 (Robert Schulze). - Add a unit test to assert Apache Arrow's fatal logging does not abort. It covers the changes in ClickHouse/arrow#16. #47958 (Arthur Passos).
- Restore the ability of native macOS debug server build to start. #48050 (Robert Schulze). Note: this change is only relevant for development, as the ClickHouse official builds are done with cross-compilation.
Bug Fix (user-visible misbehavior in an official stable release)
- Fix formats parser resetting, test processing bad messages in
Kafka#45693 (Kruglov Pavel). - Fix data size calculation in Keeper #46086 (Antonio Andelic).
- Fixed a bug in automatic retries of
DROP TABLEquery withReplicatedMergeTreetables andAtomicdatabases. In rare cases it could lead toCan't get data for node /zk_path/log_pointerandThe specified key does not existerrors if the ZooKeeper session expired during DROP and a new replicated table with the same path in ZooKeeper was created in parallel. #46384 (Alexander Tokmakov). - Fix incorrect alias recursion while normalizing queries that prevented some queries to run. #46609 (Raúl Marín).
- Fix IPv4/IPv6 serialization/deserialization in binary formats #46616 (Kruglov Pavel).
- ActionsDAG: do not change result of
andduring optimization #46653 (Salvatore Mesoraca). - Improve query cancellation when a client dies #46681 (Alexander Tokmakov).
- Fix arithmetic operations in aggregate optimization #46705 (Duc Canh Le).
- Fix possible
clickhouse-local's abort on JSONEachRow schema inference #46731 (Kruglov Pavel). - Fix changing an expired role #46772 (Vitaly Baranov).
- Fix combined PREWHERE column accumulation from multiple steps #46785 (Alexander Gololobov).
- Use initial range for fetching file size in HTTP read buffer. Without this change, some remote files couldn't be processed. #46824 (Antonio Andelic).
- Fix the incorrect progress bar while using the URL tables #46830 (Antonio Andelic).
- Fix MSan report in
maxIntersectionsfunction #46847 (Alexey Milovidov). - Fix a bug in
Mapdata type #46856 (Alexey Milovidov). - Fix wrong results of some LIKE searches when the LIKE pattern contains quoted non-quotable characters #46875 (Robert Schulze).
- Fix - WITH FILL would produce abort when the Filling Transform processing an empty block #46897 (Yakov Olkhovskiy).
- Fix date and int inference from string in JSON #46972 (Kruglov Pavel).
- Fix bug in zero-copy replication disk choice during fetch #47010 (alesapin).
- Fix a typo in systemd service definition #47051 (Palash Goel).
- Fix the NOT_IMPLEMENTED error with CROSS JOIN and algorithm = auto #47068 (Vladimir C).
- Fix the problem that the 'ReplicatedMergeTree' table failed to insert two similar data when the 'part_type' is configured as 'InMemory' mode (experimental feature). #47121 (liding1992).
- External dictionaries / library-bridge: Fix error "unknown library method 'extDict_libClone'" #47136 (alex filatov).
- Fix race condition in a grace hash join with limit #47153 (Vladimir C).
- Fix concrete columns PREWHERE support #47154 (Azat Khuzhin).
- Fix possible deadlock in Query Status #47161 (Kruglov Pavel).
- Forbid insert select for the same
Jointable, as it leads to a deadlock #47260 (Vladimir C). - Skip merged partitions for
min_age_to_force_merge_secondsmerges #47303 (Antonio Andelic). - Modify find_first_symbols, so it works as expected for find_first_not_symbols #47304 (Arthur Passos).
- Fix big numbers inference in CSV #47410 (Kruglov Pavel).
- Disable logical expression optimizer for expression with aliases. #47451 (Nikolai Kochetov).
- Fix error in
decodeURLComponent#47457 (Alexey Milovidov). - Fix explain graph with projection #47473 (flynn).
- Fix query parameters #47488 (Alexey Milovidov).
- Parameterized view: a bug fix. #47495 (SmitaRKulkarni).
- Fuzzer of data formats, and the corresponding fixes. #47519 (Alexey Milovidov).
- Fix monotonicity check for
DateTime64#47526 (Antonio Andelic). - Fix "block structure mismatch" for a Nullable LowCardinality column #47537 (Nikolai Kochetov).
- Proper fix for a bug in Apache Parquet #45878 #47538 (Kruglov Pavel).
- Fix
BSONEachRowparallel parsing when document size is invalid #47540 (Kruglov Pavel). - Preserve error in
system.distribution_queueonSYSTEM FLUSH DISTRIBUTED#47541 (Azat Khuzhin). - Check for duplicate column in
BSONEachRowformat #47609 (Kruglov Pavel). - Fix wait for zero copy lock during move #47631 (alesapin).
- Fix aggregation by partitions #47634 (Nikita Taranov).
- Fix bug in tuple as array serialization in
BSONEachRowformat #47690 (Kruglov Pavel). - Fix crash in
polygonsSymDifferenceCartesian#47702 (pufit). - Fix reading from storage
Filecompressed files withzlibandgzipcompression #47796 (Anton Popov). - Improve empty query detection for PostgreSQL (for pgx golang driver) #47854 (Azat Khuzhin).
- Fix DateTime monotonicity check for LowCardinality types #47860 (Antonio Andelic).
- Use restore_threads (not backup_threads) for RESTORE ASYNC #47861 (Azat Khuzhin).
- Fix DROP COLUMN with ReplicatedMergeTree containing projections #47883 (Antonio Andelic).
- Fix for Replicated database recovery #47901 (Alexander Tokmakov).
- Hotfix for too verbose warnings in HTTP #47903 (Alexander Tokmakov).
- Fix "Field value too long" in
catboostEvaluate#47970 (Robert Schulze). - Fix #36971: Watchdog: exit with non-zero code if child process exits #47973 (Коренберг Марк).
- Fix for "index file
cidxis unexpectedly long" #48010 (SmitaRKulkarni). - Fix MaterializedPostgreSQL query to get attributes (replica-identity) #48015 (Solomatov Sergei).
- parseDateTime(): Fix UB (signed integer overflow) #48019 (Robert Schulze).
- Use unique names for Records in Avro to avoid reusing its schema #48057 (Kruglov Pavel).
- Correctly set TCP/HTTP socket timeouts in Keeper #48108 (Antonio Andelic).
- Fix possible member call on null pointer in
Avroformat #48184 (Kruglov Pavel).
ClickHouse release 23.2, 2023-02-23
Backward Incompatible Change
- Extend function "toDayOfWeek()" (alias: "DAYOFWEEK") with a mode argument that encodes whether the week starts on Monday or Sunday and whether counting starts at 0 or 1. For consistency with other date time functions, the mode argument was inserted between the time and the time zone arguments. This breaks existing usage of the (previously undocumented) 2-argument syntax "toDayOfWeek(time, time_zone)". A fix is to rewrite the function into "toDayOfWeek(time, 0, time_zone)". #45233 (Robert Schulze).
- Rename setting
max_query_cache_sizetofilesystem_cache_max_download_size. #45614 (Kseniia Sumarokova). - The
defaultuser will not have permissions for access typeSHOW NAMED COLLECTIONby default (e.g.defaultuser will no longer be able to grant ALL to other users as it was before, therefore this PR is backward incompatible). #46010 (Kseniia Sumarokova). - If the SETTINGS clause is specified before the FORMAT clause, the settings will be applied to formatting as well. #46003 (Azat Khuzhin).
- Remove support for setting
materialized_postgresql_allow_automatic_update(which was by default turned off). #46106 (Kseniia Sumarokova). - Slightly improve performance of
countDigitson realistic datasets. This closed #44518. In previous versions,countDigits(0)returned0; now it returns1, which is more correct, and follows the existing documentation. #46187 (Alexey Milovidov). - Disallow creation of new columns compressed by a combination of codecs "Delta" or "DoubleDelta" followed by codecs "Gorilla" or "FPC". This can be bypassed using setting "allow_suspicious_codecs = true". #45652 (Robert Schulze).
New Feature
- Add
StorageIcebergand table functionicebergto access iceberg table store on S3. #45384 (flynn). - Allow configuring storage as
SETTINGS disk = '<disk_name>'(instead ofstorage_policy) and with explicit disk creationSETTINGS disk = disk(type=s3, ...). #41976 (Kseniia Sumarokova). - Expose
ProfileEventscounters insystem.part_log. #38614 (Bharat Nallan). - Enrichment of the existing
ReplacingMergeTreeengine to allow duplicate the insertion. It leverages the power of bothReplacingMergeTreeandCollapsingMergeTreein one MergeTree engine. Deleted data are not returned when queried, but not removed from disk neither. #41005 (youennL-cs). - Add
generateULIDfunction. Closes #36536. #44662 (Nikolay Degterinsky). - Add
corrMatrixaggregate function, calculating each two columns. In addition, since AggregatefunctionscovarSampandcovarPopare similar tocorr, I addcovarSampMatrix,covarPopMatrixby the way. @alexey-milovidov closes #44587. #44680 (FFFFFFFHHHHHHH). - Introduce arrayShuffle function for random array permutations. #45271 (Joanna Hulboj).
- Support types
FIXED_SIZE_BINARYtype in Arrow,FIXED_LENGTH_BYTE_ARRAYinParquetand match them toFixedString. Add settingsoutput_format_parquet_fixed_string_as_fixed_byte_array/output_format_arrow_fixed_string_as_fixed_byte_arrayto control default output type for FixedString. Closes #45326. #45340 (Kruglov Pavel). - Add a new column
last_exception_timeto system.replication_queue. #45457 (Frank Chen). - Add two new functions which allow for user-defined keys/seeds with SipHash{64,128}. #45513 (Salvatore Mesoraca).
- Allow a three-argument version for table function
format. close #45808. #45873 (FFFFFFFHHHHHHH). - Add
JodaTimeformat support for 'x','w','S'. Refer to https://joda-time.sourceforge.net/apidocs/org/joda/time/format/DateTimeFormat.html. #46073 (zk_kiger). - Support window function
ntile. (lgbo). - Add setting
finalto implicitly apply theFINALmodifier to every table. #40945 (Arthur Passos). - Added
arrayPartialSortandarrayPartialReverseSortfunctions. #46296 (Joanna Hulboj). - The new http parameter
client_protocol_versionallows setting a client protocol version for HTTP responses using the Native format. #40397. #46360 (Geoff Genz). - Add new function
regexpExtract, like spark functionREGEXP_EXTRACTfor compatibility. It is similar to the existing functionextract. #46469 (李扬). - Add new function
JSONArrayLength, which returns the number of elements in the outermost JSON array. The function returns NULL if the input JSON string is invalid. #46631 (李扬).
Performance Improvement
- The introduced logic works if PREWHERE condition is a conjunction of multiple conditions (cond1 AND cond2 AND ... ). It groups those conditions that require reading the same columns into steps. After each step the corresponding part of the full condition is computed and the result rows might be filtered. This allows to read fewer rows in the next steps thus saving IO bandwidth and doing less computation. This logic is disabled by default for now. It will be enabled by default in one of the future releases once it is known to not have any regressions, so it is highly encouraged to be used for testing. It can be controlled by 2 settings: "enable_multiple_prewhere_read_steps" and "move_all_conditions_to_prewhere". #46140 (Alexander Gololobov).
- An option added to aggregate partitions independently if table partition key and group by key are compatible. Controlled by the setting
allow_aggregate_partitions_independently. Disabled by default because of limited applicability (please refer to the docs). #45364 (Nikita Taranov). - Allow using Vertical merge algorithm with parts in Compact format. This will allow ClickHouse server to use much less memory for background operations. This closes #46084. #45681 #46282 (Anton Popov).
- Optimize
Parquetreader by using batch reader. #45878 (LiuNeng). - Add new
local_filesystem_read_methodmethodio_uringbased on the asynchronous Linux io_uring subsystem, improving read performance almost universally compared to the defaultpreadmethod. #38456 (Saulius Valatka). - Rewrite aggregate functions with
ifexpression as argument when logically equivalent. For example,avg(if(cond, col, null))can be rewritten to avgIf(cond, col). It is helpful in performance. #44730 (李扬). - Improve lower/upper function performance with avx512 instructions. #37894 (yaqi-zhao).
- Remove the limitation that on systems with >=32 cores and SMT disabled ClickHouse uses only half of the cores (the case when you disable Hyper Threading in BIOS). #44973 (Robert Schulze).
- Improve performance of function
multiIfby columnar executing, speed up by 2.3x. #45296 (李扬). - Add fast path for function
positionwhen the needle is empty. #45382 (李扬). - Enable
query_plan_remove_redundant_sortingoptimization by default. Optimization implemented in #45420. #45567 (Igor Nikonov). - Increased HTTP Transfer Encoding chunk size to improve performance of large queries using the HTTP interface. #45593 (Geoff Genz).
- Fixed performance of short
SELECTqueries that read from tables with large number ofArray/Map/Nestedcolumns. #45630 (Anton Popov). - Improve performance of filtering for big integers and decimal types. #45949 (李扬).
- This change could effectively reduce the overhead of obtaining the filter from ColumnNullable(UInt8) and improve the overall query performance. To evaluate the impact of this change, we adopted TPC-H benchmark but revised the column types from non-nullable to nullable, and we measured the QPS of its queries as the performance indicator. #45962 (Zhiguo Zhou).
- Make the
_partand_partition_idvirtual column beLowCardinality(String)type. Closes #45964. #45975 (flynn). - Improve the performance of Decimal conversion when the scale does not change. #46095 (Alexey Milovidov).
- Allow to increase prefetching for read data. #46168 (Kseniia Sumarokova).
- Rewrite
arrayExists(x -> x = 1, arr)->has(arr, 1), which improve performance by 1.34x. #46188 (李扬). - Fix too big memory usage for vertical merges on non-remote disk. Respect
max_insert_delayed_streams_for_parallel_writefor the remote disk. #46275 (Nikolai Kochetov). - Update zstd to v1.5.4. It has some minor improvements in performance and compression ratio. If you run replicas with different versions of ClickHouse you may see reasonable error messages
Data after merge/mutation is not byte-identical to data on another replicas.with explanation. These messages are Ok and you should not worry. #46280 (Raúl Marín). - Fix performance degradation caused by #39737. #46309 (Alexey Milovidov).
- The
replicas_statushandle will answer quickly even in case of a large replication queue. #46310 (Alexey Milovidov). - Add avx512 support for aggregate function
sum, function unary arithmetic, function comparison. #37870 (zhao zhou). - Rewrote the code around marks distribution and the overall coordination of the reading in order to achieve the maximum performance improvement. This closes #34527. #43772 (Nikita Mikhaylov).
- Remove redundant DISTINCT clauses in query (subqueries). Implemented on top of query plan. It does similar optimization as
optimize_duplicate_order_by_and_distinctregarding DISTINCT clauses. Can be enabled viaquery_plan_remove_redundant_distinctsetting. Related to #42648. #44176 (Igor Nikonov). - A few query rewrite optimizations:
sumIf(123, cond) -> 123 * countIf(1, cond),sum(if(cond, 123, 0)) -> 123 * countIf(cond),sum(if(cond, 0, 123)) -> 123 * countIf(not(cond))#44728 (李扬). - Improved how memory bound merging and aggregation in order on top query plan interact. Previously we fell back to explicit sorting for AIO in some cases when it wasn't actually needed. #45892 (Nikita Taranov).
- Concurrent merges are scheduled using round-robin by default to ensure fair and starvation-free operation. Previously in heavily overloaded shards, big merges could possibly be starved by smaller merges due to the use of strict priority scheduling. Added
background_merges_mutations_scheduling_policyserver config option to select scheduling algorithm (round_robinorshortest_task_first). #46247 (Sergei Trifonov).
Improvement
- Enable retries for INSERT by default in case of ZooKeeper session loss. We already use it in production. #46308 (Alexey Milovidov).
- Add ability to ignore unknown keys in JSON object for named tuples (
input_format_json_ignore_unknown_keys_in_named_tuple). #45678 (Azat Khuzhin). - Support optimizing the
whereclause with sorting key expression move toprewherefor query withfinal. #38893. #38950 (hexiaoting). - Add new metrics for backups: num_processed_files and processed_files_size described actual number of processed files. #42244 (Aleksandr).
- Added retries on interserver DNS errors. #43179 (Anton Kozlov).
- Keeper improvement: try preallocating space on the disk to avoid undefined out-of-space issues. Introduce setting
max_log_file_sizefor the maximum size of Keeper's Raft log files. #44370 (Antonio Andelic). - Optimize behavior for a replica delay api logic in case the replica is read-only. #45148 (mateng915).
- Ask for the password in clickhouse-client interactively in a case when the empty password is wrong. Closes #46702. #46730 (Nikolay Degterinsky).
- Mark
Gorillacompression on columns of non-Float* type as suspicious. #45376 (Robert Schulze). - Show replica name that is executing a merge in the
postpone_reasoncolumn. #45458 (Frank Chen). - Save exception stack trace in part_log. #45459 (Frank Chen).
- The
regexp_treedictionary is polished and now it is compatible with https://github.com/ua-parser/uap-core. #45631 (Han Fei). - Updated checking of
SYSTEM SYNC REPLICA, resolves #45508 #45648 (SmitaRKulkarni). - Rename setting
replication_alter_partitions_synctoalter_sync. #45659 (Antonio Andelic). - The
generateRandomtable function and the engine now supportLowCardinalitydata types. This is useful for testing, for example you can writeINSERT INTO table SELECT * FROM generateRandom() LIMIT 1000. This is needed to debug #45590. #45661 (Alexey Milovidov). - The experimental query result cache now provides more modular configuration settings. #45679 (Robert Schulze).
- Renamed "query result cache" to "query cache". #45682 (Robert Schulze).
- add
SYSTEM SYNC FILE CACHEcommand. It will do thesyncsyscall. #8921. #45685 (DR). - Add a new S3 setting
allow_head_object_request. This PR makes usage ofGetObjectAttributesrequest instead ofHeadObjectintroduced in #45288 optional (and disabled by default). #45701 (Vitaly Baranov). - Add ability to override connection settings based on connection names (that said that now you can forget about storing password for each connection, you can simply put everything into
~/.clickhouse-client/config.xmland even use different history files for them, which can be also useful). #45715 (Azat Khuzhin). - Arrow format: support the duration type. Closes #45669. #45750 (flynn).
- Extend the logging in the Query Cache to improve investigations of the caching behavior. #45751 (Robert Schulze).
- The query cache's server-level settings are now reconfigurable at runtime. #45758 (Robert Schulze).
- Hide password in logs when a table function's arguments are specified with a named collection. #45774 (Vitaly Baranov).
- Improve internal S3 client to correctly deduce regions and redirections for different types of URLs. #45783 (Antonio Andelic).
- Add support for Map, IPv4 and IPv6 types in generateRandom. Mostly useful for testing. #45785 (Raúl Marín).
- Support empty/notEmpty for IP types. #45799 (Yakov Olkhovskiy).
- The column
num_processed_fileswas split into two columns:num_files(for BACKUP) andfiles_read(for RESTORE). The columnprocessed_files_sizewas split into two columns:total_size(for BACKUP) andbytes_read(for RESTORE). #45800 (Vitaly Baranov). - Add support for
SHOW ENGINESquery for MySQL compatibility. #45859 (Filatenkov Artur). - Improved how the obfuscator deals with queries. #45867 (Raúl Marín).
- Improve behaviour of conversion into Date for boundary value 65535 (2149-06-06). #46042 #45914 (Joanna Hulboj).
- Add setting
check_referential_table_dependenciesto check referential dependencies onDROP TABLE. This PR solves #38326. #45936 (Vitaly Baranov). - Fix
tupleElementto returnNullwhen havingNullargument. Closes #45894. #45952 (flynn). - Throw an error on no files satisfying the S3 wildcard. Closes #45587. #45957 (chen).
- Use cluster state data to check concurrent backup/restore. #45982 (SmitaRKulkarni).
- ClickHouse Client: Use "exact" matching for fuzzy search, which has correct case ignorance and more appropriate algorithm for matching SQL queries. #46000 (Azat Khuzhin).
- Forbid wrong create View syntax
CREATE View X TO Y AS SELECT. Closes #4331. #46043 (flynn). - Storage
Logfamily support setting thestorage_policy. Closes #43421. #46044 (flynn). - Improve
JSONColumnsformat when the result is empty. Closes #46024. #46053 (flynn). - Add reference implementation for SipHash128. #46065 (Salvatore Mesoraca).
- Add a new metric to record allocations times and bytes using mmap. #46068 (李扬).
- Currently for functions like
leftPad,rightPad,leftPadUTF8,rightPadUTF8, the second argumentlengthmust be UInt8|16|32|64|128|256. Which is too strict for clickhouse users, besides, it is not consistent with other similar functions likearrayResize,substringand so on. #46103 (李扬). - Fix assertion in the
welchTTestfunction in debug build when the resulting statistics is NaN. Unified the behavior with other similar functions. Change the behavior ofstudentTTestto return NaN instead of throwing an exception because the previous behavior was inconvenient. This closes #41176 This closes #42162. #46141 (Alexey Milovidov). - More convenient usage of big integers and ORDER BY WITH FILL. Allow using plain integers for start and end points in WITH FILL when ORDER BY big (128-bit and 256-bit) integers. Fix the wrong result for big integers with negative start or end points. This closes #16733. #46152 (Alexey Milovidov).
- Add
parts,active_partsandtotal_markscolumns tosystem.tableson issue. #46161 (attack204). - Functions "multi[Fuzzy]Match(Any|AnyIndex|AllIndices}" now reject regexes which will likely evaluate very slowly in vectorscan. #46167 (Robert Schulze).
- When
insert_null_as_defaultis enabled and column doesn't have defined default value, the default of column type will be used. Also this PR fixes using default values on nulls in case of LowCardinality columns. #46171 (Kruglov Pavel). - Prefer explicitly defined access keys for S3 clients. If
use_environment_credentialsis set totrue, and the user has provided the access key through query or config, they will be used instead of the ones from the environment variable. #46191 (Antonio Andelic). - Add an alias "DATE_FORMAT()" for function "formatDateTime()" to improve compatibility with MySQL's SQL dialect, extend function
formatDateTimewith substitutions "a", "b", "c", "h", "i", "k", "l" "r", "s", "W". ### Documentation entry for user-facing changes User-readable short description:DATE_FORMATis an alias offormatDateTime. Formats a Time according to the given Format string. Format is a constant expression, so you cannot have multiple formats for a single result column. (Provide link to formatDateTime). #46302 (Jake Bamrah). - Add
ProfileEventsandCurrentMetricsabout the callback tasks for parallel replicas (s3ClusterandMergeTreetables). #46313 (Alexey Milovidov). - Add support for
DELETEandUPDATEfor tables usingKeeperMapstorage engine. #46330 (Antonio Andelic). - Allow writing RENAME queries with query parameters. Resolves #45778. #46407 (Nikolay Degterinsky).
- Fix parameterized SELECT queries with REPLACE transformer. Resolves #33002. #46420 (Nikolay Degterinsky).
- Exclude the internal database used for temporary/external tables from the calculation of asynchronous metric "NumberOfDatabases". This makes the behavior consistent with system table "system.databases". #46435 (Robert Schulze).
- Added
last_exception_timecolumn into distribution_queue table. #46564 (Aleksandr). - Support for IN clause with parameter in parameterized views. #46583 (SmitaRKulkarni).
- Do not load named collections on server startup (load them on first access instead). #46607 (Kseniia Sumarokova).
Build/Testing/Packaging Improvement
- Introduce GWP-ASan implemented by the LLVM runtime. This closes #27039. #45226 (Han Fei).
- We want to make our tests less stable and more flaky: add randomization for merge tree settings in tests. #38983 (Anton Popov).
- Enable the HDFS support in PowerPC and which helps to fixes the following functional tests 02113_hdfs_assert.sh, 02244_hdfs_cluster.sql and 02368_cancel_write_into_hdfs.sh. #44949 (MeenaRenganathan22).
- Add systemd.service file for clickhouse-keeper. Fixes #44293. #45568 (Mikhail f. Shiryaev).
- ClickHouse's fork of poco was moved from "contrib/" to "base/poco/". #46075 (Robert Schulze).
- Add an option for
clickhouse-watchdogto restart the child process. This does not make a lot of use. #46312 (Alexey Milovidov). - If the environment variable
CLICKHOUSE_DOCKER_RESTART_ON_EXITis set to 1, the Docker container will runclickhouse-serveras a child instead of the first process, and restart it when it exited. #46391 (Alexey Milovidov). - Fix Systemd service file. #46461 (SuperDJY).
- Raised the minimum Clang version needed to build ClickHouse from 12 to 15. #46710 (Robert Schulze).
- Upgrade Intel QPL from v0.3.0 to v1.0.0 2. Build libaccel-config and link it statically to QPL library instead of dynamically. #45809 (jasperzhu).
Bug Fix (user-visible misbehavior in official stable release)
- Flush data exactly by
rabbitmq_flush_interval_msor byrabbitmq_max_block_sizeinStorageRabbitMQ. Closes #42389. Closes #45160. #44404 (Kseniia Sumarokova). - Use PODArray to render in sparkBar function, so we can control the memory usage. Close #44467. #44489 (Duc Canh Le).
- Fix functions (quantilesExactExclusive, quantilesExactInclusive) return unsorted array element. #45379 (wujunfu).
- Fix uncaught exception in HTTPHandler when open telemetry is enabled. #45456 (Frank Chen).
- Don't infer Dates from 8 digit numbers. It could lead to wrong data to be read. #45581 (Kruglov Pavel).
- Fixes to correctly use
odbc_bridge_use_connection_poolingsetting. #45591 (Bharat Nallan). - When the callback in the cache is called, it is possible that this cache is destructed. To keep it safe, we capture members by value. It's also safe for task schedule because it will be deactivated before storage is destroyed. Resolve #45548. #45601 (Han Fei).
- Fix data corruption when codecs Delta or DoubleDelta are combined with codec Gorilla. #45615 (Robert Schulze).
- Correctly check types when using N-gram bloom filter index to avoid invalid reads. #45617 (Antonio Andelic).
- A couple of segfaults have been reported around
c-ares. They were introduced in my previous pull requests. I have fixed them with the help of Alexander Tokmakov. #45629 (Arthur Passos). - Fix key description when encountering duplicate primary keys. This can happen in projections. See #45590 for details. #45686 (Amos Bird).
- Set compression method and level for backup Closes #45690. #45737 (Pradeep Chhetri).
- Should use
select_query_typed.limitByOffset()instead ofselect_query_typed.limitOffset(). #45817 (刘陶峰). - When use experimental analyzer, queries like
SELECT number FROM numbers(100) LIMIT 10 OFFSET 10;get wrong results (empty result for this sql). That is caused by an unnecessary offset step added by planner. #45822 (刘陶峰). - Backward compatibility - allow implicit narrowing conversion from UInt64 to IPv4 - required for "INSERT ... VALUES ..." expression. #45865 (Yakov Olkhovskiy).
- Bugfix IPv6 parser for mixed ip4 address with missed first octet (like
::.1.2.3). #45871 (Yakov Olkhovskiy). - Add the
query_kindcolumn to thesystem.processestable and theSHOW PROCESSLISTquery. Remove duplicate code. It fixes a bug: the global configuration parametermax_concurrent_select_querieswas not respected to queries withINTERSECTorEXCEPTchains. #45872 (Alexey Milovidov). - Fix crash in a function
stochasticLinearRegression. Found by WingFuzz. #45985 (Nikolai Kochetov). - Fix crash in
SELECTqueries withINTERSECTandEXCEPTmodifiers that read data from tables with enabled sparse columns (controlled by settingratio_of_defaults_for_sparse_serialization). #45987 (Anton Popov). - Fix read in order optimization for DESC sorting with FINAL, close #45815. #46009 (Vladimir C).
- Fix reading of non existing nested columns with multiple level in compact parts. #46045 (Azat Khuzhin).
- Fix elapsed column in system.processes (10x error). #46047 (Azat Khuzhin).
- Follow-up fix for Replace domain IP types (IPv4, IPv6) with native #43221. #46087 (Yakov Olkhovskiy).
- Fix environment variable substitution in the configuration when a parameter already has a value. This closes #46131. This closes #9547. #46144 (pufit).
- Fix incorrect predicate push down with grouping sets. Closes #45947. #46151 (flynn).
- Fix possible pipeline stuck error on
fulls_sorting_joinwith constant keys. #46175 (Vladimir C). - Never rewrite tuple functions as literals during formatting to avoid incorrect results. #46232 (Salvatore Mesoraca).
- Fix possible out of bounds error while reading LowCardinality(Nullable) in Arrow format. #46270 (Kruglov Pavel).
- Fix
SYSTEM UNFREEZEqueries failing with the exceptionCANNOT_PARSE_INPUT_ASSERTION_FAILED. #46325 (Aleksei Filatov). - Fix possible crash which can be caused by an integer overflow while deserializing aggregating state of a function that stores HashTable. #46349 (Nikolai Kochetov).
- Fix possible
LOGICAL_ERRORin asynchronous inserts with invalid data sent in formatVALUES. #46350 (Anton Popov). - Fixed a LOGICAL_ERROR on an attempt to execute
ALTER ... MOVE PART ... TO TABLE. This type of query was never actually supported. #46359 (Alexander Tokmakov). - Fix s3Cluster schema inference in parallel distributed insert select when
parallel_distributed_insert_selectis enabled. #46381 (Kruglov Pavel). - Fix queries like
ALTER TABLE ... UPDATE nested.arr1 = nested.arr2 ..., wherearr1andarr2are fields of the sameNestedcolumn. #46387 (Anton Popov). - Scheduler may fail to schedule a task. If it happens, the whole MulityPartUpload should be aborted and
UploadHelpermust wait for already scheduled tasks. #46451 (Dmitry Novik). - Fix PREWHERE for Merge with different default types (fixes some
NOT_FOUND_COLUMN_IN_BLOCKwhen the default type for the column differs, also allowPREWHEREwhen the type of column is the same across tables, and prohibit it, only if it differs). #46454 (Azat Khuzhin). - Fix a crash that could happen when constant values are used in
ORDER BY. Fixes #46466. #46493 (Nikolai Kochetov). - Do not throw exception if
disksetting was specified on query level, butstorage_policywas specified in config merge tree settings section.diskwill override setting from config. #46533 (Kseniia Sumarokova). - Fix an invalid processing of constant
LowCardinalityargument in functionarrayMap. This bug could lead to a segfault in release, and logical errorBad castin debug build. #46569 (Alexey Milovidov). - fixes #46557. #46611 (Alexander Gololobov).
- Fix endless restarts of clickhouse-server systemd unit if server cannot start within 1m30sec (Disable timeout logic for starting clickhouse-server from systemd service). #46613 (Azat Khuzhin).
- Allocated during asynchronous inserts memory buffers were deallocated in the global context and MemoryTracker counters for corresponding user and query were not updated correctly. That led to false positive OOM exceptions. #46622 (Dmitry Novik).
- Updated to not clear on_expression from table_join as its used by future analyze runs resolves #45185. #46487 (SmitaRKulkarni).
ClickHouse release 23.1, 2023-01-26
ClickHouse release 23.1
Upgrade Notes
- The
SYSTEM RESTART DISKquery becomes a no-op. #44647 (alesapin). - The
PREALLOCATEoption forHASHED/SPARSE_HASHEDdictionaries becomes a no-op. #45388 (Azat Khuzhin). It does not give significant advantages anymore. - Disallow
Gorillacodec on columns of non-Float32 or non-Float64 type. #45252 (Robert Schulze). It was pointless and led to inconsistencies. - Parallel quorum inserts might work incorrectly with
*MergeTreetables created with the deprecated syntax. Therefore, parallel quorum inserts support is completely disabled for such tables. It does not affect tables created with a new syntax. #45430 (Alexander Tokmakov). - Use the
GetObjectAttributesrequest instead of theHeadObjectrequest to get the size of an object in AWS S3. This change fixes handling endpoints without explicit regions after updating the AWS SDK, for example. #45288 (Vitaly Baranov). AWS S3 and Minio are tested, but keep in mind that various S3-compatible services (GCS, R2, B2) may have subtle incompatibilities. This change also may require you to adjust the ACL to allow theGetObjectAttributesrequest. - Forbid paths in timezone names. For example, a timezone name like
/usr/share/zoneinfo/Asia/Adenis not allowed; the IANA timezone database name likeAsia/Adenshould be used. #44225 (Kruglov Pavel). - Queries combining equijoin and constant expressions (e.g.,
JOIN ON t1.x = t2.x AND 1 = 1) are forbidden due to incorrect results. #44016 (Vladimir C).
New Feature
- Dictionary source for extracting keys by traversing regular expressions tree. It can be used for User-Agent parsing. #40878 (Vage Ogannisian). #43858 (Han Fei).
- Added parametrized view functionality, now it's possible to specify query parameters for the View table engine. resolves #40907. #41687 (SmitaRKulkarni).
- Add
quantileInterpolatedWeighted/quantilesInterpolatedWeightedfunctions. #38252 (Bharat Nallan). - Array join support for the
Maptype, like the function "explode" in Spark. #43239 (李扬). - Support SQL standard binary and hex string literals. #43785 (Mo Xuan).
- Allow formatting
DateTimein Joda-Time style. Refer to the Joda-Time docs. #43818 (李扬). - Implemented a fractional second formatter (
%f) forformatDateTime. #44060 (ltrk2). #44497 (Alexander Gololobov). - Added
agefunction to calculate the difference between two dates or dates with time values expressed as the number of full units. Closes #41115. #44421 (Robert Schulze). - Add
Nullsource for dictionaries. Closes #44240. #44502 (mayamika). - Allow configuring the S3 storage class with the
s3_storage_classconfiguration option. Such as<s3_storage_class>STANDARD/INTELLIGENT_TIERING</s3_storage_class>Closes #44443. #44707 (chen). - Insert default values in case of missing elements in JSON object while parsing named tuple. Add setting
input_format_json_defaults_for_missing_elements_in_named_tuplethat controls this behaviour. Closes #45142#issuecomment-1380153217. #45231 (Kruglov Pavel). - Record server startup time in ProfileEvents (
ServerStartupMilliseconds). Resolves #43188. #45250 (SmitaRKulkarni). - Refactor and Improve streaming engines Kafka/RabbitMQ/NATS and add support for all formats, also refactor formats a bit: - Fix producing messages in row-based formats with suffixes/prefixes. Now every message is formatted completely with all delimiters and can be parsed back using input format. - Support block-based formats like Native, Parquet, ORC, etc. Every block is formatted as a separate message. The number of rows in one message depends on the block size, so you can control it via the setting
max_block_size. - Add new engine settingskafka_max_rows_per_message/rabbitmq_max_rows_per_message/nats_max_rows_per_message. They control the number of rows formatted in one message in row-based formats. Default value: 1. - Fix high memory consumption in the NATS table engine. - Support arbitrary binary data in NATS producer (previously it worked only with strings contained \0 at the end) - Add missing Kafka/RabbitMQ/NATS engine settings in the documentation. - Refactor producing and consuming in Kafka/RabbitMQ/NATS, separate it from WriteBuffers/ReadBuffers semantic. - Refactor output formats: remove callbacks on each row used in Kafka/RabbitMQ/NATS (now we don't use callbacks there), allow to use IRowOutputFormat directly, clarify row end and row between delimiters, make it possible to reset output format to start formatting again - Add proper implementation in formatRow function (bonus after formats refactoring). #42777 (Kruglov Pavel). - Support reading/writing
Nestedtables asListofStructinCapnProtoformat. Read/writeDecimal32/64asInt32/64. Closes #43319. #43379 (Kruglov Pavel). - Added a
message_format_stringcolumn tosystem.text_log. The column contains a pattern that was used to format the message. #44543 (Alexander Tokmakov). This allows various analytics over the ClickHouse logs. - Try to autodetect headers with column names (and maybe types) for CSV/TSV/CustomSeparated input formats. Add settings input_format_tsv/csv/custom_detect_header that enable this behaviour (enabled by default). Closes #44640. #44953 (Kruglov Pavel).
Experimental Feature
- Add an experimental inverted index as a new secondary index type for efficient text search. #38667 (larryluogit).
- Add experimental query result cache. #43797 (Robert Schulze).
- Added extendable and configurable scheduling subsystem for IO requests (not yet integrated with IO code itself). #41840 (Sergei Trifonov). This feature does nothing at all, enjoy.
- Added
SYSTEM DROP DATABASE REPLICAthat removes metadata of a dead replica of aReplicateddatabase. Resolves #41794. #42807 (Alexander Tokmakov).
Performance Improvement
- Do not load inactive parts at startup of
MergeTreetables. #42181 (Anton Popov). - Improved latency of reading from storage
S3and table functions3with large numbers of small files. Now settingsremote_filesystem_read_methodandremote_filesystem_read_prefetchtake effect while reading from storageS3. #43726 (Anton Popov). - Optimization for reading struct fields in Parquet/ORC files. Only the required fields are loaded. #44484 (lgbo).
- Two-level aggregation algorithm was mistakenly disabled for queries over the HTTP interface. It was enabled back, and it leads to a major performance improvement. #45450 (Nikolai Kochetov).
- Added mmap support for StorageFile, which should improve the performance of clickhouse-local. #43927 (pufit).
- Added sharding support in HashedDictionary to allow parallel load (almost linear scaling based on number of shards). #40003 (Azat Khuzhin).
- Speed up query parsing. #42284 (Raúl Marín).
- Always replace OR chain
expr = x1 OR ... OR expr = xNtoexpr IN (x1, ..., xN)in the case whereexpris aLowCardinalitycolumn. Settingoptimize_min_equality_disjunction_chain_lengthis ignored in this case. #42889 (Guo Wangyang). - Slightly improve performance by optimizing the code around ThreadStatus. #43586 (Zhiguo Zhou).
- Optimize the column-wise ternary logic evaluation by achieving auto-vectorization. In the performance test of this microbenchmark, we've observed a peak performance gain of 21x on the ICX device (Intel Xeon Platinum 8380 CPU). #43669 (Zhiguo Zhou).
- Avoid acquiring read locks in the
system.tablestable if possible. #43840 (Raúl Marín). - Optimize ThreadPool. The performance experiments of SSB (Star Schema Benchmark) on the ICX device (Intel Xeon Platinum 8380 CPU, 80 cores, 160 threads) shows that this change could effectively decrease the lock contention for ThreadPoolImpl::mutex by 75%, increasing the CPU utilization and improving the overall performance by 2.4%. #44308 (Zhiguo Zhou).
- Now the optimisation for predicting the hash table size is applied only if the cached hash table size is sufficiently large (thresholds were determined empirically and hardcoded). #44455 (Nikita Taranov).
- Small performance improvement for asynchronous reading from remote filesystems. #44868 (Kseniia Sumarokova).
- Add fast path for: -
col like '%%'; -col like '%'; -col not like '%'; -col not like '%'; -match(col, '.*'). #45244 (李扬). - Slightly improve happy path optimisation in filtering (WHERE clause). #45289 (Nikita Taranov).
- Provide monotonicity info for
toUnixTimestamp64*to enable more algebraic optimizations for index analysis. #44116 (Nikita Taranov). - Allow the configuration of temporary data for query processing (spilling to disk) to cooperate with the filesystem cache (taking up the space from the cache disk) #43972 (Vladimir C). This mainly improves ClickHouse Cloud, but can be used for self-managed setups as well, if you know what to do.
- Make
system.replicastable do parallel fetches of replicas statuses. Closes #43918. #43998 (Nikolay Degterinsky). - Optimize memory consumption during backup to S3: files to S3 now will be copied directly without using
WriteBufferFromS3(which could use a lot of memory). #45188 (Vitaly Baranov). - Add a cache for async block ids. This will reduce the number of requests of ZooKeeper when we enable async inserts deduplication. #45106 (Han Fei).
Improvement
- Use structure from insertion table in generateRandom without arguments. #45239 (Kruglov Pavel).
- Allow to implicitly convert floats stored in string fields of JSON to integers in
JSONExtractfunctions. E.g.JSONExtract('{"a": "1000.111"}', 'a', 'UInt64')->1000, previously it returned 0. #45432 (Anton Popov). - Added fields
supports_parallel_parsingandsupports_parallel_formattingto tablesystem.formatsfor better introspection. #45499 (Anton Popov). - Improve reading CSV field in CustomSeparated/Template format. Closes #42352 Closes #39620. #43332 (Kruglov Pavel).
- Unify query elapsed time measurements. #43455 (Raúl Marín).
- Improve automatic usage of structure from insertion table in table functions file/hdfs/s3 when virtual columns are present in a select query, it fixes the possible error
Block structure mismatchornumber of columns mismatch. #43695 (Kruglov Pavel). - Add support for signed arguments in the function
range. Fixes #43333. #43733 (sanyu). - Remove redundant sorting, for example, sorting related ORDER BY clauses in subqueries. Implemented on top of query plan. It does similar optimization as
optimize_duplicate_order_by_and_distinctregardingORDER BYclauses, but more generic, since it's applied to any redundant sorting steps (not only caused by ORDER BY clause) and applied to subqueries of any depth. Related to #42648. #43905 (Igor Nikonov). - Add the ability to disable deduplication of files for BACKUP (for backups without deduplication ATTACH can be used instead of full RESTORE). For example
BACKUP foo TO S3(...) SETTINGS deduplicate_files=0(defaultdeduplicate_files=1). #43947 (Azat Khuzhin). - Refactor and improve schema inference for text formats. Add new setting
schema_inference_make_columns_nullablethat controls making result typesNullable(enabled by default);. #44019 (Kruglov Pavel). - Better support for
PROXYv1protocol. #44135 (Yakov Olkhovskiy). - Add information about the latest part check by cleanup threads into
system.partstable. #44244 (Dmitry Novik). - Disable table functions in readonly mode for inserts. #44290 (SmitaRKulkarni).
- Add a setting
simultaneous_parts_removal_limitto allow limiting the number of parts being processed by one iteration of CleanupThread. #44461 (Dmitry Novik). - Do not initialize ReadBufferFromS3 when only virtual columns are needed in a query. This may be helpful to #44246. #44493 (chen).
- Prevent duplicate column names hints. Closes #44130. #44519 (Joanna Hulboj).
- Allow macro substitution in endpoint of disks. Resolve #40951. #44533 (SmitaRKulkarni).
- Improve schema inference when
input_format_json_read_object_as_stringis enabled. #44546 (Kruglov Pavel). - Add a user-level setting
database_replicated_allow_replicated_engine_argumentswhich allows banning the creation ofReplicatedMergeTreetables with arguments inDatabaseReplicated. #44566 (alesapin). - Prevent users from mistakenly specifying zero (invalid) value for
index_granularity. This closes #44536. #44578 (Alexey Milovidov). - Added possibility to set path to service keytab file in
keytabparameter inkerberossection of config.xml. #44594 (Roman Vasin). - Use already written part of the query for fuzzy search (pass to the
skimlibrary, which is written in Rust and linked statically to ClickHouse). #44600 (Azat Khuzhin). - Enable
input_format_json_read_objects_as_stringsby default to be able to read nested JSON objects while JSON Object type is experimental. #44657 (Kruglov Pavel). - Improvement for deduplication of async inserts: when users do duplicate async inserts, we should deduplicate inside the memory before we query Keeper. #44682 (Han Fei).
- Input/ouptut
Avroformat will parse bool type as ClickHouse bool type. #44684 (Kruglov Pavel). - Support Bool type in Arrow/Parquet/ORC. Closes #43970. #44698 (Kruglov Pavel).
- Don't greedily parse beyond the quotes when reading UUIDs - it may lead to mistakenly successful parsing of incorrect data. #44686 (Raúl Marín).
- Infer UInt64 in case of Int64 overflow and fix some transforms in schema inference. #44696 (Kruglov Pavel).
- Previously dependency resolving inside
Replicateddatabase was done in a hacky way, and now it's done right using an explicit graph. #44697 (Nikita Mikhaylov). - Fix
output_format_pretty_row_numbersdoes not preserve the counter across the blocks. Closes #44815. #44832 (flynn). - Don't report errors in
system.errorsdue to parts being merged concurrently with the background cleanup process. #44874 (Raúl Marín). - Optimize and fix metrics for Distributed async INSERT. #44922 (Azat Khuzhin).
- Added settings to disallow concurrent backups and restores resolves #43891 Implementation: * Added server-level settings to disallow concurrent backups and restores, which are read and set when BackupWorker is created in Context. * Settings are set to true by default. * Before starting backup or restores, added a check to see if any other backups/restores are running. For internal requests, it checks if it is from the self node using backup_uuid. #45072 (SmitaRKulkarni).
- Add
<storage_policy>config parameter for system logs. #45320 (Stig Bakken).
Build/Testing/Packaging Improvement
- Statically link with the
skimlibrary (it is written in Rust) for fuzzy search in clickhouse client/local history. #44239 (Azat Khuzhin). - We removed support for shared linking because of Rust. Actually, Rust is only an excuse for this removal, and we wanted to remove it nevertheless. #44828 (Alexey Milovidov).
- Remove the dependency on the
addusertool from the packages, because we don't use it. This fixes #44934. #45011 (Alexey Milovidov). - The
SQLitelibrary is updated to the latest. It is used for the SQLite database and table integration engines. Also, fixed a false-positive TSan report. This closes #45027. #45031 (Alexey Milovidov). - CRC-32 changes to address the WeakHash collision issue in PowerPC. #45144 (MeenaRenganathan22).
- Update aws-c* submodules #43020 (Vitaly Baranov).
- Automatically merge green backport PRs and green approved PRs #41110 (Mikhail f. Shiryaev).
- Introduce a website for the status of ClickHouse CI. Source.
Bug Fix
- Replace domain IP types (IPv4, IPv6) with native. #43221 (Yakov Olkhovskiy). It automatically fixes some missing implementations in the code.
- Fix the backup process if mutations get killed during the backup process. #45351 (Vitaly Baranov).
- Fix the
Invalid number of rows in Chunkexception message. #41404. #42126 (Alexander Gololobov). - Fix possible use of an uninitialized value after executing expressions after sorting. Closes #43386 #43635 (Kruglov Pavel).
- Better handling of NULL in aggregate combinators, fix possible segfault/logical error while using an obscure optimization
optimize_rewrite_sum_if_to_count_if. Closes #43758. #43813 (Kruglov Pavel). - Fix CREATE USER/ROLE query settings constraints. #43993 (Nikolay Degterinsky).
- Fixed bug with non-parsable default value for
EPHEMERALcolumn in table metadata. #44026 (Yakov Olkhovskiy). - Fix parsing of bad version from compatibility setting. #44224 (Kruglov Pavel).
- Bring interval subtraction from datetime in line with addition. #44241 (ltrk2).
- Remove limits on the maximum size of the result for view. #44261 (lizhuoyu5).
- Fix possible logical error in cache if
do_not_evict_index_and_mrk_files=1. Closes #42142. #44268 (Kseniia Sumarokova). - Fix possible too early cache write interruption in write-through cache (caching could be stopped due to false assumption when it shouldn't have). #44289 (Kseniia Sumarokova).
- Fix possible crash in the case function
INwith constant arguments was used as a constant argument together withLowCardinality. Fixes #44221. #44346 (Nikolai Kochetov). - Fix support for complex parameters (like arrays) of parametric aggregate functions. This closes #30975. The aggregate function
sumMapFilteredwas unusable in distributed queries before this change. #44358 (Alexey Milovidov). - Fix reading ObjectId in BSON schema inference. #44382 (Kruglov Pavel).
- Fix race which can lead to premature temp parts removal before merge finishes in ReplicatedMergeTree. This issue could lead to errors like
No such file or directory: xxx. Fixes #43983. #44383 (alesapin). - Some invalid
SYSTEM ... ON CLUSTERqueries worked in an unexpected way if a cluster name was not specified. It's fixed, now invalid queries throwSYNTAX_ERRORas they should. Fixes #44264. #44387 (Alexander Tokmakov). - Fix reading Map type in ORC format. #44400 (Kruglov Pavel).
- Fix reading columns that are not presented in input data in Parquet/ORC formats. Previously it could lead to error
INCORRECT_NUMBER_OF_COLUMNS. Closes #44333. #44405 (Kruglov Pavel). - Previously the
barfunction used the same '▋' (U+258B "Left five eighths block") character to display both 5/8 and 6/8 bars. This change corrects this behavior by using '▊' (U+258A "Left three quarters block") for displaying 6/8 bar. #44410 (Alexander Gololobov). - Placing profile settings after profile settings constraints in the configuration file made constraints ineffective. #44411 (Konstantin Bogdanov).
- Fix
SYNTAX_ERRORwhile runningEXPLAIN AST INSERTqueries with data. Closes #44207. #44413 (save-my-heart). - Fix reading bool value with CRLF in CSV format. Closes #44401. #44442 (Kruglov Pavel).
- Don't execute and/or/if/multiIf on a LowCardinality dictionary, so the result type cannot be LowCardinality. It could lead to the error
Illegal column ColumnLowCardinalityin some cases. Fixes #43603. #44469 (Kruglov Pavel). - Fix mutations with the setting
max_streams_for_merge_tree_reading. #44472 (Anton Popov). - Fix potential null pointer dereference with GROUPING SETS in ASTSelectQuery::formatImpl (#43049). #44479 (Robert Schulze).
- Validate types in table function arguments, CAST function arguments, JSONAsObject schema inference according to settings. #44501 (Kruglov Pavel).
- Fix IN function with LowCardinality and const column, close #44503. #44506 (Duc Canh Le).
- Fixed a bug in the normalization of a
DEFAULTexpression inCREATE TABLEstatement. The second argument of the functionin(or the right argument of operatorIN) might be replaced with the result of its evaluation during CREATE query execution. Fixes #44496. #44547 (Alexander Tokmakov). - Projections do not work in presence of WITH ROLLUP, WITH CUBE and WITH TOTALS. In previous versions, a query produced an exception instead of skipping the usage of projections. This closes #44614. This closes #42772. #44615 (Alexey Milovidov).
- Async blocks were not cleaned because the function
get all blocks sorted by timedidn't get async blocks. #44651 (Han Fei). - Fix
LOGICAL_ERRORThe top step of the right pipeline should be ExpressionStepfor JOIN with subquery, UNION, and TOTALS. Fixes #43687. #44673 (Nikolai Kochetov). - Avoid
std::out_of_rangeexception in the Executable table engine. #44681 (Kruglov Pavel). - Do not apply
optimize_syntax_fuse_functionsto quantiles on AST, close #44712. #44713 (Vladimir C). - Fix bug with wrong type in Merge table and PREWHERE, close #43324. #44716 (Vladimir C).
- Fix a possible crash during shutdown (while destroying TraceCollector). Fixes #44757. #44758 (Nikolai Kochetov).
- Fix a possible crash in distributed query processing. The crash could happen if a query with totals or extremes returned an empty result and there are mismatched types in the Distributed and the local tables. Fixes #44738. #44760 (Nikolai Kochetov).
- Fix fsync for fetches (
min_compressed_bytes_to_fsync_after_fetch)/small files (ttl.txt, columns.txt) in mutations (min_rows_to_fsync_after_merge/min_compressed_bytes_to_fsync_after_merge). #44781 (Azat Khuzhin). - A rare race condition was possible when querying the
system.partsorsystem.parts_columnstables in the presence of parts being moved between disks. Introduced in #41145. #44809 (Alexey Milovidov). - Fix the error
Context has expiredwhich could appear with enabled projections optimization. Can be reproduced for queries with specific functions, likedictHas/dictGetwhich use context in runtime. Fixes #44844. #44850 (Nikolai Kochetov). - A fix for
Cannot read all dataerror which could happen while readingLowCardinalitydictionary from remote fs. Fixes #44709. #44875 (Nikolai Kochetov). - Ignore cases when hardware monitor sensors cannot be read instead of showing a full exception message in logs. #44895 (Raúl Marín).
- Use
max_delay_to_insertvalue in case the calculated time to delay INSERT exceeds the setting value. Related to #44902. #44916 (Igor Nikonov). - Fix error
Different order of columns in UNION subqueryfor queries withUNION. Fixes #44866. #44920 (Nikolai Kochetov). - Delay for INSERT can be calculated incorrectly, which can lead to always using
max_delay_to_insertsetting as delay instead of a correct value. Using simple formulamax_delay_to_insert * (parts_over_threshold/max_allowed_parts_over_threshold)i.e. delay grows proportionally to parts over threshold. Closes #44902. #44954 (Igor Nikonov). - Fix alter table TTL error when a wide part has the lightweight delete mask. #44959 (Mingliang Pan).
- Follow-up fix for Replace domain IP types (IPv4, IPv6) with native #43221. #45024 (Yakov Olkhovskiy).
- Follow-up fix for Replace domain IP types (IPv4, IPv6) with native #43221. #45043 (Yakov Olkhovskiy).
- A buffer overflow was possible in the parser. Found by fuzzer. #45047 (Alexey Milovidov).
- Fix possible cannot-read-all-data error in storage FileLog. Closes #45051, #38257. #45057 (Kseniia Sumarokova).
- Memory efficient aggregation (setting
distributed_aggregation_memory_efficient) is disabled when grouping sets are present in the query. #45058 (Nikita Taranov). - Fix
RANGE_HASHEDdictionary to count range columns as part of the primary key during updates whenupdate_fieldis specified. Closes #44588. #45061 (Maksim Kita). - Fix error
Cannot capture columnforLowCardinalitycaptured argument of nested lambda. Fixes #45028. #45065 (Nikolai Kochetov). - Fix the wrong query result of
additional_table_filters(additional filter was not applied) in case the minmax/count projection is used. #45133 (Nikolai Kochetov). - Fixed bug in
histogramfunction accepting negative values. #45147 (simpleton). - Fix wrong column nullability in StoreageJoin, close #44940. #45184 (Vladimir C).
- Fix
background_fetches_pool_sizesettings reload (increase at runtime). #45189 (Raúl Marín). - Correctly process
SELECTqueries on KV engines (e.g. KeeperMap, EmbeddedRocksDB) usingINon the key with subquery producing different type. #45215 (Antonio Andelic). - Fix logical error in SEMI JOIN & join_use_nulls in some cases, close #45163, close #45209. #45230 (Vladimir C).
- Fix heap-use-after-free in reading from s3. #45253 (Kruglov Pavel).
- Fix bug when the Avro Union type is ['null', Nested type], closes #45275. Fix bug that incorrectly infers
bytestype toFloat. #45276 (flynn). - Throw a correct exception when explicit PREWHERE cannot be used with a table using the storage engine
Merge. #45319 (Antonio Andelic). - Under WSL1 Ubuntu self-extracting ClickHouse fails to decompress due to inconsistency - /proc/self/maps reporting 32bit file's inode, while stat reporting 64bit inode. #45339 (Yakov Olkhovskiy).
- Fix race in Distributed table startup (that could lead to processing file of async INSERT multiple times). #45360 (Azat Khuzhin).
- Fix a possible crash while reading from storage
S3and table functions3in the case whenListObjectrequest has failed. #45371 (Anton Popov). - Fix
SELECT ... FROM system.dictionariesexception when there is a dictionary with a bad structure (e.g. incorrect type in XML config). #45399 (Aleksei Filatov). - Fix s3Cluster schema inference when structure from insertion table is used in
INSERT INTO ... SELECT * FROM s3Clusterqueries. #45422 (Kruglov Pavel). - Fix bug in JSON/BSONEachRow parsing with HTTP that could lead to using default values for some columns instead of values from data. #45424 (Kruglov Pavel).
- Fixed bug (Code: 632. DB::Exception: Unexpected data ... after parsed IPv6 value ...) with typed parsing of IP types from text source. #45425 (Yakov Olkhovskiy).
- close #45297 Add check for empty regular expressions. #45428 (Han Fei).
- Fix possible (likely distributed) query hung. #45448 (Azat Khuzhin).
- Fix possible deadlock with
allow_asynchronous_read_from_io_pool_for_merge_treeenabled in case of exception fromThreadPool::schedule. #45481 (Nikolai Kochetov). - Fix possible in-use table after DETACH. #45493 (Azat Khuzhin).
- Fix rare abort in the case when a query is canceled and parallel parsing was used during its execution. #45498 (Anton Popov).
- Fix a race between Distributed table creation and INSERT into it (could lead to CANNOT_LINK during INSERT into the table). #45502 (Azat Khuzhin).
- Add proper default (SLRU) to cache policy getter. Closes #45514. #45524 (Kseniia Sumarokova).
- Disallow array join in mutations closes #42637 #44447 (SmitaRKulkarni).
- Fix for qualified asterisks with alias table name and column transformer. Resolves #44736. #44755 (SmitaRKulkarni).