Skip to main content
Skip to main content
Edit this page

2025 Changelog

Table of Contents​

ClickHouse release v25.2, 2025-02-27
ClickHouse release v25.1, 2025-01-28
Changelog for 2024
Changelog for 2023
Changelog for 2022
Changelog for 2021
Changelog for 2020
Changelog for 2019
Changelog for 2018
Changelog for 2017

ClickHouse release 25.2, 2025-02-27​

Backward Incompatible Change​

  • Completely enable async_load_databases by default (even for those installations that do not upgrade config.xml). #74772 (Azat Khuzhin).
  • Add JSONCompactEachRowWithProgress and JSONCompactStringsEachRowWithProgress formats. Continuation of #69989. The JSONCompactWithNames and JSONCompactWithNamesAndTypes no longer output "totals" - apparently, it was a mistake in the implementation. #75037 (Alexey Milovidov).
  • Change the format_alter_operations_with_parentheses default to true to disambiguate the ALTER commands list (see https://github.com/ClickHouse/ClickHouse/pull/59532). This breaks replication with clusters prior to 24.3. If you are upgrading a cluster using older releases turn off the setting in the server config or upgrade to 24.3 first. #75302 (Raúl Marín).
  • Remove the possibility of filtering log messages using regular expressions. The implementation introduced a data race, so it has to be removed. #75577 (János Benjamin Antal).
  • The setting min_chunk_bytes_for_parallel_parsing cannot be zero anymore. This fixes: #71110. #75239 (Nikita Mikhaylov).
  • Validate settings in the cache configuration. Non existing settings were ignored before, now they will throw an error and they should be removed. #75452 (Kseniia Sumarokova).

New Feature​

  • Support type Nullable(JSON). #73556 (Pavel Kruglov).
  • Support subcolumns in the DEFAULT and MATERIALIZED expressions. #74403 (Pavel Kruglov).
  • Support writing Parquet bloom filters using the output_format_parquet_write_bloom_filter setting (enabled by default). #71681 (Michael Kolupaev).
  • Web UI now has interactive database navigation. #75777 (Alexey Milovidov).
  • Allow the combination of read-only and read-write disks in the storage policy (as multiple volumes or multiple disks). This allows data to be read from the entire volume, while inserts will prefer the writable disk (i.e., Copy-on-Write storage policy). #75862 (Azat Khuzhin).
  • Add a new Database engine, DatabaseBackup, that allows to instantly attach table/database from backup. #75725 (Maksim Kita).
  • Support prepared statements in the Postgres wire protocol. #75035 (scanhex12).
  • Add the ability to ATTACH tables without a database layer, which is useful for MergeTree tables located on Web, S3, and similar external virtual filesystems. #75788 (Azat Khuzhin).
  • Add a new string comparison function, compareSubstrings to compare parts of two strings. Example: SELECT compareSubstrings('Saxony', 'Anglo-Saxon', 0, 6, 5) AS result means "compare 6 bytes of strings 'Saxon' and 'Anglo-Saxon' lexicographically, starting at offset 0 in the first string, offset 5 in the second string". #74070 (lgbo).
  • A new function, initialQueryStartTime is added. It returns the start time of the current query. The value is the same across all shards during a distributed query. #75087 (Roman Lomonosov).
  • Add support for SSL authentication with named collections for MySQL. Closes #59111. #59452 (Nikolay Degterinsky).

Experimental Features​

  • Added a new setting, enable_adaptive_memory_spill_scheduler that allows multiple Grace JOINs in the same query to monitor their combined memory footprint and trigger spilling into external storage adaptively to prevent MEMORY_LIMIT_EXCEEDED. #72728 (lgbo).
  • Make the new, experimental Kafka table engine fully respect Keeper feature flags. #76004 (János Benjamin Antal).
  • Restore the (Intel) QPL codec, which was removed in v24.10 due to licensing issues. #76021 (Konstantin Bogdanov).
  • For the integration with HDFS added support for the dfs.client.use.datanode.hostname configuration option. #74635 (Mikhail Tiukavkin).

Performance Improvement​

  • Improve performance of the whole JSON column reading in Wide parts from S3. It's done by adding prefetches for subcolumn prefixes deserialization, cache of deserialized prefixes, and parallel deserialization of subcolumn prefixes. It improves the reading of the JSON column from S3 4 times in queries like SELECT data FROM table and about 10 times in queries like SELECT data FROM table LIMIT 10. #74827 (Pavel Kruglov).
  • Fixed unnecessary contention in parallel_hash when max_rows_in_join = max_bytes_in_join = 0. #75155 (Nikita Taranov).
  • Fixed double preallocation in ConcurrentHashJoin in case join sides are swapped by the optimizer. #75149 (Nikita Taranov).
  • Slight improvement in some join scenarios: precalculate the number of output rows and reserve memory for them. #75376 (Alexander Gololobov).
  • For queries like WHERE a < b AND b < c AND c < 5, we can infer new comparing conditions (a < 5 AND b < 5) to have better filter ability. #73164 (Shichao Jin).
  • Keeper improvement: disable digest calculation when committing to in-memory storage for better performance. It can be enabled with keeper_server.digest_enabled_on_commit config. Digest is still calculated when preprocessing requests. #75490 (Antonio Andelic).
  • Push down filter expression from JOIN ON when possible. #75536 (Vladimir Cherkasov).
  • Calculate columns and indices sizes lazily in MergeTree. #75938 (Pavel Kruglov).
  • Reintroduce respect ttl_only_drop_parts on MATERIALIZE TTL; only read necessary columns to recalculate TTL and drop parts by replacing them with an empty ones. #72751 (Andrey Zvonov).
  • Reduce write buffer size for plain_rewritable metadata files #75758 (Julia Kartseva).
  • Reduce memory usage for some window functions. #65647 (lgbo).
  • Evaluate parquet bloom filters and min/max indexes together. Necessary to properly support: x = 3 or x > 5 where data = [1, 2, 4, 5]. #71383 (Arthur Passos).
  • Queries passed to the Executable storage are no longer limited to a single-threaded execution. #70084 (yawnt).
  • Fetch parts in parallel in ALTER TABLE FETCH PARTITION (thread pool size is controlled with max_fetch_partition_thread_pool_size). #74978 (Azat Khuzhin).
  • Allow to move predicates with the indexHint function to PREWHERE. #74987 (Anton Popov).

Improvement​

  • Fixed calculation of size in memory for LowCardinality columns. #74688 (Nikita Taranov).
  • processors_profile_log table now has a default configuration with a TTL of 30 days. #66139 (Ilya Yatsishin).
  • Allow naming shards in the cluster configuration. #72276 (MikhailBurdukov).
  • Change Prometheus remote write response success status from 200/OK to 204/NoContent. #74170 (Michael Dempsey).
  • Add ability to reload max_remote_read_network_bandwidth_for_serve and max_remote_write_network_bandwidth_for_server on the fly without restart server. #74206 (Kai Zhu).
  • Allow using blob paths to calculate checksums while making a backup. #74729 (Vitaly Baranov).
  • Added a query ID column to system.query_cache (closes #68205). #74982 (NamHoaiNguyen).
  • It is allowed to cancel ALTER TABLE ... FREEZE ... queries with KILL QUERY and automatically by a timeout (max_execution_time). #75016 (Kirill).
  • Add support for groupUniqArrayArrayMap as SimpleAggregateFunction. #75034 (Miel Donkers).
  • Hide catalog credential settings in database engine Iceberg. Closes #74559. #75080 (Kseniia Sumarokova).
  • intExp2 / intExp10: Define undefined behaviour: return 0 for too small argument, 18446744073709551615 for too big argument, throw exception if nan. #75312 (Vitaly Baranov).
  • Support s3.endpoint natively from catalog config in DatabaseIceberg. Closes #74558. #75375 (Kseniia Sumarokova).
  • Don't fail silently if a user executing SYSTEM DROP REPLICA doesn't have enough permissions. #75377 (Bharat Nallan).
  • Add a ProfileEvent about the number of times any of the system logs have failed to flush. #75466 (Alexey Milovidov).
  • Add a check and extra logging for decrypting and decompressing. #75471 (Vitaly Baranov).
  • Added support for the micro sign (U+00B5) in the parseTimeDelta function. Now both the micro sign (U+00B5) and the Greek letter mu (U+03BC) are recognized as valid representations for microseconds, aligning ClickHouse's behavior with Go’s implementation (see time.go and time/format.go). #75472 (Vitaly Orlov).
  • Replace server setting (send_settings_to_client) with client setting (apply_settings_from_server) that controls whether client-side code (e.g., parsing INSERT data and formatting query output) should use settings from server's users.xml and user profile. Otherwise, only settings from the client command line, session, and query are used. Note that this only applies to native client (not e.g. HTTP), and doesn't apply to most of query processing (which happens on the server). #75478 (Michael Kolupaev).
  • Better error messages for syntax errors. Previously, if the query was too large, and the token whose length exceeds the limit is a very large string literal, the message about the reason was lost in the middle of two examples of this very long token. Fix the issue when a query with UTF-8 was cut incorrectly in the error message. Fix excessive quoting of query fragments. This closes #75473. #75561 (Alexey Milovidov).
  • Add profile events in storage S3(Azure)Queue. #75618 (Kseniia Sumarokova).
  • Disable sending settings from server to client (send_settings_to_client=false) for compatibility (This feature will be re-implemented as a client setting later for better usability). #75648 (Michael Kolupaev).
  • Add a config memory_worker_correct_memory_tracker to enable correction of the internal memory tracker with information from different sources read in the background thread periodically. #75714 (Antonio Andelic).
  • Add column normalized_query_hash into system.processes. Note: while it can be easily calculated on the fly with the normalizedQueryHash function, this is needed to prepare for subsequent changes. #75756 (Alexey Milovidov).
  • Querying system.tables will not throw even if there is a Merge table created over a database that no longer exists. Remove the getTotalRows method from Hive tables because we don't allow it to do complex work. #75772 (Alexey Milovidov).
  • Store start_time/end_time for Backups with microseconds. #75929 (Aleksandr Musorin).
  • Add MemoryTrackingUncorrected metric showing the value of the internal global memory tracker which is not corrected by RSS. #75935 (Antonio Andelic).
  • Allow parsing endpoints like localhost:1234/handle in PostgreSQL or MySQL table functions. This fixes a regression introduced in https://github.com/ClickHouse/ClickHouse/pull/52503. #75944 (Nikita Mikhaylov).
  • Added a server setting throw_on_unknown_workload that allows to choose behavior on query with workload setting set to unknown value: either allow unlimited access (default) or throw a RESOURCE_ACCESS_DENIED error. It is useful to force all queries to use workload scheduling. #75999 (Sergei Trifonov).
  • Don't rewrite subcolumns to getSubcolumn in ARRAY JOIN if not necessary. #76018 (Pavel Kruglov).
  • Retry coordination errors when loading tables. #76020 (Alexander Tokmakov).
  • Support flushing individual logs in SYSTEM FLUSH LOGS. #76132 (Raúl Marín).
  • Improved the /binary server's page. Using the Hilbert curve instead of the Morton curve. Display 512 MB worth of addresses in the square, which fills the square better (in previous versions, addresses fill only half of the square). Color addresses closer to the library name rather than the function name. Allow scrolling a bit more outside of the area. #76192 (Alexey Milovidov).
  • Retry ON CLUSTER queries in case of TOO_MANY_SIMULTANEOUS_QUERIES. #76352 (Patrick Galbraith).
  • Add the CPUOverload asynchronous metric, which calculates the server's relative CPU deficit. #76404 (Alexey Milovidov).
  • Changed the default value of output_format_pretty_max_rows from 10000 to 1000. I think it is better for usability. #76407 (Alexey Milovidov).

Bug Fix (user-visible misbehavior in an official stable release)​

  • Fix formatting of exceptions using a custom format if they appear during query interpretation. In previous versions, exceptions were formatted using the default format rather than the format specified in the query. This closes #55422. #74994 (Alexey Milovidov).
  • Fix type mapping for SQLite (integer types into int64, floating points into float64). #73853 (Joanna Hulboj).
  • Fix identifier resolution from parent scopes. Allow the use of aliases to expressions in the WITH clause. Fixes #58994. Fixes #62946. Fixes #63239. Fixes #65233. Fixes #71659. Fixes #71828. Fixes #68749. #66143 (Dmitry Novik).
  • Fix negate function monotonicity. In previous versions, the query select * from a where -x = -42; where x is the primary key, can return a wrong result. #71440 (Michael Kolupaev).
  • Fix empty tuple handling in arrayIntersect. This fixes #72578. #72581 (Amos Bird).
  • Fix reading JSON sub-object subcolumns with incorrect prefix. #73182 (Pavel Kruglov).
  • Propagate Native format settings properly for client-server communication. #73924 (Pavel Kruglov).
  • Check for not supported types for some storages. #74218 (Pavel Kruglov).
  • Fix crash with query INSERT INTO SELECT over PostgreSQL interface on macOS (issue #72938). #74231 (Artem Yurov).
  • Fixed uninitialized max_log_ptr in the replicated database. #74336 (Konstantin Morozov).
  • Fix crash when inserting interval (issue #74299). #74478 (NamHoaiNguyen).
  • Fix formatting constant JSON literals. Previously it could lead to syntax errors during sending the query to another server. #74533 (Pavel Kruglov).
  • Fix broken create query when using constant partition expressions with implicit projections enabled. This fixes #74596 . #74634 (Amos Bird).
  • Avoid leaving connection in broken state after INSERT finishes with exception. #74740 (Azat Khuzhin).
  • Avoid reusing connections that had been left in the intermediate state. #74749 (Azat Khuzhin).
  • Fix crash during JSON type declaration parsing when type name is not uppercase. #74784 (Pavel Kruglov).
  • Keeper: fix logical_error when the connection had been terminated before establishing. #74844 (Michael Kolupaev).
  • Fix a behavior when the server couldn't startup when there's a table using AzureBlobStorage. Tables are loaded without any requests to Azure. #74880 (Alexey Katsman).
  • Fix missing used_privileges and missing_privileges fields in query_log for BACKUP and RESTORE operations. #74887 (Alexey Katsman).
  • HDFS refresh krb ticket if sasl error during hdfs select request. #74930 (inv2004).
  • Fix queries to Replicated database in startup_scripts. #74942 (Azat Khuzhin).
  • Fix issues with expressions type aliased in the JOIN ON clause when a null-safe comparison is used. #74970 (Vladimir Cherkasov).
  • Revert part's state from deleting back to outdated when remove operation has failed. #74985 (Sema Checherinda).
  • In previous versions, when there was a scalar subquery, we started writing the progress (accumulated from processing the subquery) during the initialization of the data format, which was before HTTP headers were written. This led to the loss of HTTP headers, such as X-ClickHouse-QueryId and X-ClickHouse-Format, as well as Content-Type. #74991 (Alexey Milovidov).
  • Fix CREATE TABLE AS... queries for database_replicated_allow_replicated_engine_arguments=0. #75000 (Bharat Nallan).
  • Fix leaving connection in a bad state in client after INSERT exceptions. #75030 (Azat Khuzhin).
  • Fix crash due to uncaught exception in PSQL replication. #75062 (Azat Khuzhin).
  • Sasl can fail any rpc call, the fix helps to repeat the call in case if krb5 ticker is expired. #75063 (inv2004).
  • Fixed usage of indexes (primary and secondary) for Array, Map and Nullable(..) columns with enabled setting optimize_function_to_subcolumns. Previously, indexes for these columns could have been ignored. #75081 (Anton Popov).
  • Disable flatten_nested when creating materialized views with inner tables since it will not be possible to use such flattened columns. #75085 (Christoph Wurm).
  • Fix for some of IPv6 addresses (such as ::ffff:1.1.1.1) in forwarded_for field is wrongly interpreted resulting in client disconnect with exception. #75133 (Yakov Olkhovskiy).
  • Fix nullsafe JOIN handling for LowCardinality nullable data type. Previously JOIN ON with nullsafe comparison, such as IS NOT DISTINCT FROM, <=> , a IS NULL AND b IS NULL OR a == b didn't work correctly with LowCardinality columns. #75143 (Vladimir Cherkasov).
  • Checks that we don't specify key_condition when counting total_number_of_rows for NumRowsCache. #75164 (Daniil Ivanik).
  • Fix queries with unused interpolation with the new analyzer. #75173 (János Benjamin Antal).
  • Fix the crash bug of CTE with Insert. #75188 (Shichao Jin).
  • Keeper fix: avoid writing to broken changelogs when rolling back logs. #75197 (Antonio Andelic).
  • Use BFloat16 as a supertype where appropriate. This closes: #74404. #75236 (Nikita Mikhaylov).
  • Fix unexpected defaults in join result with any_join_distinct_right_table_keys and OR in JOIN ON. #75262 (Vladimir Cherkasov).
  • Mask azureblobstorage table engine credentials. #75319 (Garrett Thomas).
  • Fixed behavior when ClickHouse may erroneously do a filter pushdown to an external database like PostgreSQL, MySQL, or SQLite. This closes: #71423. #75320 (Nikita Mikhaylov).
  • Fix crash in protobuf schema cache that can happen during output in Protobuf format and parallel query SYSTEM DROP FORMAT SCHEMA CACHE. #75357 (Pavel Kruglov).
  • Fix a possible logical error or uninitialized memory issue when a filter from HAVING is pushed down with parallel replicas. #75363 (Vladimir Cherkasov).
  • Hide sensitive info for icebergS3, icebergAzure table functions and table engines. #75378 (Kseniia Sumarokova).
  • Function TRIM with computed empty trim characters are now correctly handled. Example: SELECT TRIM(LEADING concat('') FROM 'foo') (Issue #69922). #75399 (Manish Gill).
  • Fix data race in IOutputFormat. #75448 (Pavel Kruglov).
  • Fix possible error Elements ... and ... of Nested data structure ... (Array columns) have different array sizes when JSON subcolumns with Array type are used in JOIN over distributed tables. #75512 (Pavel Kruglov).
  • Fix data corruption with CODEC(ZSTD, DoubleDelta). Closes #70031. #75548 (Konstantin Bogdanov).
  • Fix interaction between allow_feature_tier and compatibility mergetree setting. #75635 (Raúl Marín).
  • Fix incorrect processed_rows value in system.s3queue_log in case file was retried. #75666 (Kseniia Sumarokova).
  • Respect materialized_views_ignore_errors when a materialized view writes to a URL engine and there is a connectivity issue. #75679 (Christoph Wurm).
  • Fixed rare crashes while reading from MergeTree table after multiple asynchronous RENAME queries (with alter_sync = 0) between columns with different types. #75693 (Anton Popov).
  • Fix Block structure mismatch in QueryPipeline stream error for some queries with UNION ALL. #75715 (Nikolai Kochetov).
  • Rebuild projection on alter modify of its PK column. Previously it could lead to CANNOT_READ_ALL_DATA errors during selects after alter modify of the column used in projection PK. #75720 (Pavel Kruglov).
  • Fix incorrect result of ARRAY JOIN for scalar subqueries (with analyzer). #75732 (Nikolai Kochetov).
  • Fixed null pointer dereference in DistinctSortedStreamTransform. #75734 (Nikita Taranov).
  • Fix allow_suspicious_ttl_expressions behaviour. #75771 (Aleksei Filatov).
  • Fix uninitialized memory read in function translate. This closes #75592. #75794 (Alexey Milovidov).
  • Propagate format settings to JSON as string formatting in Native format. #75832 (Pavel Kruglov).
  • Recorded the default enablement of parallel hash as join algorithm in v24.12 in the settings change history. This means that ClickHouse will continue to join using non-parallel hash if an older compatibility level than v24.12 is configured. #75870 (Robert Schulze).
  • Fixed a bug that tables with implicitly added min-max indices could not be copied into a new table (issue #75677). #75877 (Smita Kulkarni).
  • clickhouse-library-bridge allows opening arbitrary libraries from the filesystem, which makes it safe to run only inside an isolated environment. To prevent a vulnerability when it is run near the clickhouse-server, we will limit the paths of libraries to a location, provided in the configuration. This vulnerability was found with the ClickHouse Bug Bounty Program by Arseniy Dugin. #75954 (Alexey Milovidov).
  • We happened to use JSON serialization for some metadata, which was a mistake, because JSON does not support binary data inside string literals, including zero bytes. SQL queries can contain binary data and invalid UTF-8, so we have to support this in our metadata files as well. At the same time, ClickHouse's JSONEachRow and similar formats work around that by deviating from the JSON standard in favor of a perfect roundtrip for the binary data. See the motivation here: https://github.com/ClickHouse/ClickHouse/pull/73668#issuecomment-2560501790. The solution is to make Poco::JSON library consistent with the JSON format serialization in ClickHouse. This closes #73668. #75963 (Alexey Milovidov).
  • Fix check for commit limits in storage S3Queue. #76104 (Kseniia Sumarokova).
  • Fix attaching MergeTree tables with auto indexes (add_minmax_index_for_numeric_columns/add_minmax_index_for_string_columns). #76139 (Azat Khuzhin).
  • Fixed issue of stack traces from parent threads of a job (enable_job_stack_trace setting) are not printed out. Fixed issue enable_job_stack_trace setting is not properly propagated to the threads resulting stack trace content not always respects this setting. #76191 (Yakov Olkhovskiy).
  • Fix incorrect permission check where ALTER RENAME required CREATE USER grant. Closes #74372. #76241 (pufit).
  • Fix reinterpretAs with FixedString on big-endian architecture. #76253 (Azat Khuzhin).
  • Fix logical error in S3Queue "Expected current processor to be equal to for bucket ". #76358 (Kseniia Sumarokova).
  • Fix deadlock for ALTER with Memory database. #76359 (Azat Khuzhin).
  • Fix logical error in index analysis if condition in WHERE has pointInPolygon function. #76360 (Anton Popov).
  • Fix potentially unsafe call in signal handler. #76549 (Yakov Olkhovskiy).
  • Fix reverse key support in PartsSplitter. This fixes #73400. #73418 (Amos Bird).

Build/Testing/Packaging Improvement​

ClickHouse release 25.1, 2025-01-28​

Backward Incompatible Change​

  • JSONEachRowWithProgress will write the progress whenever the progress happens. In previous versions, the progress was shown only after each block of the result, which made it useless. Change the way how the progress is displayed: it will not show zero values. This closes #70800. #73834 (Alexey Milovidov).
  • Merge tables will unify the structure of underlying tables by using a union of their columns and deriving common types. This closes #64864. In certain cases, this change could be backward incompatible. One example is when there is no common type between tables, but conversion to the type of the first table is still possible, such as in the case of UInt64 and Int64 or any numeric type and String. If you want to return to the old behavior, set merge_table_max_tables_to_look_for_schema_inference to 1 or set compatibility to 24.12 or earlier. #73956 (Alexey Milovidov).
  • Parquet output format converts Date and DateTime columns to date/time types supported by Parquet, instead of writing them as raw numbers. DateTime becomes DateTime64(3) (was: UInt32); setting output_format_parquet_datetime_as_uint32 brings back the old behavior. Date becomes Date32 (was: UInt16). #70950 (Michael Kolupaev).
  • Don't allow not comparable types (like JSON/Object/AggregateFunction) in ORDER BY and comparison functions less/greater/equal/etc by default. #73276 (Pavel Kruglov).
  • The obsolete MaterializedMySQL database engine has been removed and is no longer available. #73879 (Alexey Milovidov).
  • The mysql dictionary source no longer does SHOW TABLE STATUS query, because it does not provide any value for InnoDB tables, as long as for any recent MySQL versions. This closes #72636. This change is backward compatible, but put in this category so you have a chance to notice it. #73914 (Alexey Milovidov).
  • CHECK TABLE queries now require a separate, CHECK grant. In previous versions, it was enough to have SHOW TABLES grant to run these queries. But a CHECK TABLE query can be heavy, and usual query complexity limits for SELECT queries don't apply to it. It led to the potential of DoS. #74471 (Alexey Milovidov).
  • Function h3ToGeo() now returns the results in the order (lat, lon) (which is the standard order for geometric functions). Users who wish to retain the legacy result order (lon, lat) can set setting h3togeo_lon_lat_result_order = true. #74719 (Manish Gill).
  • A new MongoDB driver is now the default. Users who like to continue using the legacy driver can set server setting use_legacy_mongodb_integration to true. #73359 (Robert Schulze).

New Feature​

  • Added an ability to apply non-finished (not materialized by background process) mutations during the execution of SELECT queries immediately after submitting. It can be enabled by setting apply_mutations_on_fly. #74877 (Anton Popov).
  • Implement Iceberg tables partition pruning for time-related transform partition operations in Iceberg. #72044 (Daniil Ivanik).
  • Support subcolumns in MergeTree sorting key and skip indexes. #72644 (Pavel Kruglov).
  • Support reading HALF_FLOAT values from Apache Arrow/Parquet/ORC (they are read into Float32). This closes #72960. Keep in mind that IEEE-754 half float is not the same as BFloat16. Closes #73835. #73836 (Alexey Milovidov).
  • The system.trace_log table will contain two new columns, symbols and lines containing symbolized stack trace. It allows for easy collection and export of profile information. This is controlled by the server configuration value symbolize inside trace_log and is enabled by default. #73896 (Alexey Milovidov).
  • Add a new function, generateSerialID, which can be used to generate auto-incremental numbers in tables. Continuation of #64310 by kazalika. This closes #62485. #73950 (Alexey Milovidov).
  • Add syntax query1 PARALLEL WITH query2 PARALLEL WITH query3 ... PARALLEL WITH queryN for DDL queries. That means subqueries {query1, query2, ... queryN} are allowed to run in parallel with each other (and it's preferable). #73983 (Vitaly Baranov).
  • Added an in-memory cache for deserialized skipping index granules. This should make repeated queries that use skipping indexes faster. The size of the new cache is controlled by server settings skipping_index_cache_size and skipping_index_cache_max_entries. The original motivation for the cache were vector similarity indexes which became a lot faster now. #70102 (Robert Schulze).
  • Now, the embedded Web UI has a progress bar during query runtime. It allows cancelling queries. It displays the total number of records and the extended information about the speed. The table can be rendered incrementally as soon as data arrives. Enable HTTP compression. Rendering of the table became faster. The table header became sticky. It allows selecting cells and navigating them by arrow keys. Fix the issue when the outline of the selected cell makes it smaller. Cells no longer expand on mouse hover but only on selection. The moment to stop rendering the incoming data is decided on the client rather than on the server side. Highlight digit groups for numbers. The overall design was refreshed and became bolder. It checks if the server is reachable and the correctness of credentials and displays the server version and uptime. The cloud icon is contoured in every font, even in Safari. Big integers inside nested data types will be rendered better. It will display inf/nan correctly. It will display data types when the mouse is over a column header. #74204 (Alexey Milovidov).
  • Add the ability to create min-max (skipping) indices by default for columns managed by MergeTree using settings add_minmax_index_for_numeric_columns (for numeric columns) and add_minmax_index_for_string_columns (for string columns). For now, both settings are disabled, so there is no behavior change yet. #74266 (Smita Kulkarni).
  • Add script_query_number and script_line_number fields to system.query_log, to the ClientInfo in the native protocol, and to server logs. This closes #67542. Credits to pinsvin00 for kicking off this feature earlier in #68133. #74477 (Alexey Milovidov).
  • Added aggregation function sequenceMatchEvents which return timestamps of matched events for longest chain of events in pattern. #72349 (UnamedRus).
  • Added function arrayNormalizedGini. #72823 (flynn).
  • Add minus operator support for DateTime64, to allow subtraction between DateTime64 values, as well as DateTime. #74482 (Li Yin).

Experimental Features​

Performance Improvement​

  • Optimized function indexHint. Now, columns that are only used as arguments of function indexHint are not read from the table. #74314 (Anton Popov). If the indexHint function is a central piece of your enterprise data architecture, this optimization will save your life.
  • More accurate accounting for max_joined_block_size_rows setting for parallel_hash JOIN algorithm. Helps to avoid increased memory consumption compared to hash algorithm. #74630 (Nikita Taranov).
  • Support predicate push down optimization on the query plan level for the MergingAggregated step. It improves performance for some queries with the analyzer. #74073 (Nikolai Kochetov).
  • Splitting of left table blocks by hash was removed from the probe phase of the parallel_hash JOIN algorithm. #73089 (Nikita Taranov).
  • Optimize RowBinary input format. Closes #63805. #65059 (Pavel Kruglov).
  • Write parts with level 1 if optimize_on_insert is enabled. It allows to use several optimizations of queries with FINAL for freshly written parts. #73132 (Anton Popov).
  • Speedup string deserialization by some low-level optimisation. #65948 (Nikita Taranov).
  • When running an equality comparison between records, such as during merges, start to compare rows from most likely unequal columns first. #63780 (UnamedRus).
  • Improve grace hash join performance by re-ranking the right join table by keys. #72237 (kevinyhzou).
  • Allow arrayROCAUC and arrayAUCPR to compute partial area of the whole curve, so that its calculation can be parallelized over huge datasets. #72904 (Emmanuel).
  • Avoid spawn too many idle threads. #72920 (Guo Wangyang).
  • Don't list blob storage keys if we only have curly brackets expansion in table function. Closes #73333. #73518 (Konstantin Bogdanov).
  • Short circuit optimization for functions executed over Nullable arguments. #73820 (李扬).
  • Do not apply maskedExecute on non-function columns, improve the performance of short circuit execution. #73965 (lgbo).
  • Disable the autodetection of headers in input formats for Kafka/NATS/RabbitMQ/FileLog to improve performance. #74006 (Azat Khuzhin).
  • Execute pipeline with a higher degree of parallelism after aggregation with grouping sets. #74082 (Nikita Taranov).
  • Reduce critical section in MergeTreeReadPool. #74202 (Guo Wangyang).
  • Parallel replicas performance improvement. Packets deserialization on query initiator, for packets not related to parallel replicas protocol, now always happens in pipeline thread. Before, it could happen in a thread responsible for pipeline scheduling, which could make initiator less responsive and delay pipeline execution. #74398 (Igor Nikonov).
  • Improve performance of larger multi requests in Keeper. #74849 (Antonio Andelic).
  • Use log wrappers by value and don't allocate them in a heap. #74034 (Mikhail Artemenko).
  • Reestablish connection to MySQL and Postgres dictionary replicas in the background, so it wouldn't delay requests to corresponding dictionaries. #71101 (Yakov Olkhovskiy).
  • Parallel replicas used historical information about replica availability to improve replica selection but did not update the replica's error count when the connection was unavailable. This PR updates the replica's error count when unavailable. #72666 (zoomxi).
  • Added a merge tree setting materialize_skip_indexes_on_merge which suppresses the creation of skip indexes during merge. This allows users to control explicitly (via ALTER TABLE [..] MATERIALIZE INDEX [...]) when skip indexes are created. This can be useful if skip indexes are expensive to build (e.g. vector similarity indexes). #74401 (Robert Schulze).
  • Optimize keeper requests in Storage(S3/Azure)Queue. #74410 (Kseniia Sumarokova). #74538 (Kseniia Sumarokova).
  • Use up to 1000 parallel replicas by default. #74504 (Konstantin Bogdanov).
  • Improve HTTP session reuse when reading from s3 disk (#72401). #74548 (Julian Maicher).

Improvement​

  • Support SETTINGS in a CREATE TABLE query with an implicit ENGINE and support mixing engine and query settings. #73120 (Raúl Marín).
  • Enable use_hive_partitioning by default. #71636 (Yarik Briukhovetskyi).
  • Support CAST and ALTER between JSON types with different parameters. #72303 (Pavel Kruglov).
  • Support equal comparison for values of JSON column. #72991 (Pavel Kruglov).
  • Improve formatting of identifiers with JSON subcolumns to avoid unnecessary back quotes. #73085 (Pavel Kruglov).
  • Interactive metrics improvements. Fix metrics from parallel replicas not being fully displayed. Display the metrics in order of the most recent update, then lexicographically by name. Do not display stale metrics. #71631 (Julia Kartseva).
  • Make JSON output format pretty by default. Add new setting output_format_json_pretty_print to control it and enable it by default. #72148 (Pavel Kruglov).
  • Allow LowCardinality(UUID) by default. This has proven practical among ClickHouse Cloud customers. #73826 (Alexey Milovidov).
  • Better message during installation. #73827 (Alexey Milovidov).
  • Better message about password reset for ClickHouse Cloud. #73831 (Alexey Milovidov).
  • Improve the error message with a File table that cannot perform appends into a file. #73832 (Alexey Milovidov).
  • Ask when a user accidentally requests to output binary format (such as Native, Parquet, Avro) in the terminal. This closes #59524. #73833 (Alexey Milovidov).
  • Highlight trailing spaces in Pretty and Vertical formats in the terminal for better clarity. This is controlled with the output_format_pretty_highlight_trailing_spaces setting. Initial implementation by Braden Burns from #72996. Closes #71590. #73847 (Alexey Milovidov).
  • clickhouse-client and clickhouse-local will autodetect compression of stdin when it is redirected from a file. This closes #70865. #73848 (Alexey Milovidov).
  • Cut too long column names in pretty formats by default. This is controlled by the output_format_pretty_max_column_name_width_cut_to and output_format_pretty_max_column_name_width_min_chars_to_cut settings. This is the continuation of the work of tanmaydatta in #66502. This closes #65968. #73851 (Alexey Milovidov).
  • Make Pretty formats prettier: squash blocks if not much time passed since the output of the previous block. This is controlled by new settings output_format_pretty_squash_consecutive_ms (50 ms by default) and output_format_pretty_squash_max_wait_ms (1000 ms by default). Continuation of #49537. This closes #49153. #73852 (Alexey Milovidov).
  • Add a metric on the number of currently merging source parts. This closes #70809. #73868 (Alexey Milovidov).
  • Highlight columns in the Vertical format if the output is to a terminal. This can be disabled with the output_format_pretty_color setting. #73898 (Alexey Milovidov).
  • Enhanced the MySQL compatibility to a level that now, mysqlsh (a rich MySQL CLI from Oracle) can connect to ClickHouse. This is needed to facilitate testing. #73912 (Alexey Milovidov).
  • Pretty formats can render multi-line fields inside a table cell, which improves readability. This is enabled by default and can be controlled by the setting output_format_pretty_multiline_fields. Continuation of the work by Volodyachan in #64094. This closes #56912. #74032 (Alexey Milovidov).
  • Expose X-ClickHouse HTTP headers to JavaScript in the browser. It makes writing applications more convenient. #74180 (Alexey Milovidov).
  • The JSONEachRowWithProgress format will include events with metadata, as well as totals and extremes. It also includes rows_before_limit_at_least and rows_before_aggregation. The format prints the exception properly if it arrives after partial results. The progress now includes elapsed nanoseconds. One final progress event is emitted at the end. The progress during query runtime will be printed no more frequently than the value of the interactive_delay setting. #74181 (Alexey Milovidov).
  • Hourglass will rotate smoothly in Play UI. #74182 (Alexey Milovidov).
  • Even if the HTTP response is compressed, send packets as soon as they arrive. This allows the browser to receive progress packets and compressed data. #74201 (Alexey Milovidov).
  • If the number of output records is larger than N = output_format_pretty_max_rows, instead of displaying only the first N rows, we will cut the output table in the middle, displaying N/2 first rows and N/2 last rows. Continuation of #64200. This closes #59502. #73929 (Alexey Milovidov).
  • Allow more general join planning algorithm when hash join algorithm is enabled. #71926 (János Benjamin Antal).
  • Allow to create bloom_filter index on columns with data type DateTime64. #66416 (Yutong Xiao).
  • When min_age_to_force_merge_seconds and min_age_to_force_merge_on_partition_only are both enabled, the part merging will ignore the max bytes limit. #73656 (Kai Zhu).
  • Added HTTP headers to OpenTelemetry span logs table for enhanced traceability. #70516 (jonymohajanGmail).
  • Support writing orc file by custom time zone, not always by the GMT time zone. #70615 (kevinyhzou).
  • Respect IO scheduling settings when writing backups across clouds. #71093 (János Benjamin Antal).
  • Add metric column alias name to system.asynchronous_metrics. #71164 (megao).
  • Historically for some reason, the query ALTER TABLE MOVE PARTITION TO TABLE checked SELECT and ALTER DELETE rights instead of dedicated ALTER_MOVE_PARTITION. This PR makes use of this access type. For compatibility, this permission is also will be granted implicitly if SELECT and ALTER DELETE are granted, but this behavior will be removed in future releases. Closes #16403. #71632 (pufit).
  • Throw an exception when trying to materialize a column in the sort key instead of allowing it to break the sort order. #71891 (Peter Nguyen).
  • Hide secrets in EXPLAIN QUERY TREE. #72025 (Yakov Olkhovskiy).
  • Support parquet integer logical types in the "native" reader. #72105 (Arthur Passos).
  • Interactively request credentials in the browser if the default user requires a password. In previous versions, the server returned HTTP 403; now, it returns HTTP 401. #72198 (Alexey Milovidov).
  • Convert access types CREATE_USER, ALTER_USER, DROP_USER, CREATE_ROLE, ALTER_ROLE, DROP_ROLE from global to parameterized. That means users can now grant access management grants more precise:. #72246 (pufit).
  • Add the latest_fail_error_code_name column to system.mutations. We need this column to introduce a new metric on stuck mutations and use it to build graphs of the errors encountered in the cloud as well as, optionally, adding a new less-noisy alert. #72398 (Miсhael Stetsyuk).
  • Reduce amount of allocation in the ATTACH PARTITION query. #72583 (Konstantin Morozov).
  • Make max_bytes_before_external_sort limit depends on total query memory consumption (previously it was number of bytes in the sorting block for one sorting thread, now it has the same meaning as max_bytes_before_external_group_by - it is total limit for the whole query memory for all threads). Also one more setting added to control on disk block size - min_external_sort_block_bytes. #72598 (Azat Khuzhin).
  • Ignore memory restrictions by trace collector. #72606 (Azat Khuzhin).
  • Add server settings dictionaries_lazy_load and wait_dictionaries_load_at_startup to system.server_settings. #72664 (Christoph Wurm).
  • Adds setting max_backup_bandwidth to the list of settings that can be specified as part of BACKUP/RESTORE queries. #72665 (Christoph Wurm).
  • Reducing the log level for appearing replicated parts in the ReplicatedMergeTree engine to help minimize the volume of logs generated in a replicated cluster. #72876 (mor-akamai).
  • Improve extraction of common expression in disjunctions. Allow simplifying the resulting filter expression even if there's no common subexpression for all the disjuncts. Continuation of #71537. #73271 (Dmitry Novik).
  • In Storage S3Queue/AzureQueue allow to add settings where table was created without settings. #73283 (Kseniia Sumarokova).
  • Introduce a setting least_greatest_legacy_null_behavior (default: false) which controls if functions least and greatest handle NULL arguments by unconditionally returning NULL (if true) or by ignoring them (if false). #73344 (Robert Schulze).
  • Use Keeper multi requests in the cleanup thread of ObjectStorageQueueMetadata. #73357 (Antonio Andelic).
  • When ClickHouse runs under a cgroup we will still collect system-wide asynchronous metrics related to system load, process scheduling, memory etc. They might provide useful signals when ClickHouse is the only process on the host with high resource consumption. #73369 (Nikita Taranov).
  • In storage S3Queue allow to transfer old ordered tables created before 24.6 to new structure with buckets. #73467 (Kseniia Sumarokova).
  • Add system.azure_queue similar to existing system.s3queue. #73477 (Kseniia Sumarokova).
  • Function parseDateTime64 (and its variants) now produces correct results for input dates before 1970 / after 2106. Example: SELECT parseDateTime64InJodaSyntax('2200-01-01 00:00:00.000', 'yyyy-MM-dd HH:mm:ss.SSS'). #73594 (zhanglistar).
  • Address some clickhouse-disks usability issues addressed by users. Closes #67136. #73616 (Daniil Ivanik).
  • Allow to alter commit settings in storage S3(Azure)Queue. (Commit settings are: max_processed_files_before_commit, max_processed_rows_before_commit, max_processed_bytes_before_commit, max_processing_time_sec_before_commit). #73635 (Kseniia Sumarokova).
  • In storage S3(Azure)Queue aggregate progress between sources to compare with commit limit settings. #73641 (Kseniia Sumarokova).
  • Support core settings in BACKUP/RESTORE query. #73650 (Vitaly Baranov).
  • Take into account the output_format_compression_level on Parquet output. #73651 (Arthur Passos).
  • Adds reading Apache Arrow's fixed_size_list as an Array instead of treating it as an unsupported type. #73654 (Julian Meyers).
  • Add two backup engines: Memory (keeps backups inside the current user session), and Null (don't keep backups anywhere), which is for testing. #73690 (Vitaly Baranov).
  • concurrent_threads_soft_limit_num and concurrent_threads_soft_limit_num_ratio_to_cores could be changed w/o restart of a server. #73713 (Sergei Trifonov).
  • Add support for extended numeric types (Decimal, big integers) in formatReadable functions. #73765 (Raúl Marín).
  • Support TLS for Postgres wire protocol compatibility. #73812 (scanhex12).
  • The function isIPv4String returned true if the correct IPv4 address was followed by a zero byte, while it should return false in this case. Continuation of #65387. #73946 (Alexey Milovidov).
  • Make the error code in the MySQL wire protocol compatible with MySQL. Continuation of #56831. Closes #50957. #73948 (Alexey Milovidov).
  • Add setting validate_enum_literals_in_opearators to validate enum literals in operators like IN, NOT IN against the enum type and throw an exception if the literal is not a valid enum value. #73985 (Vladimir Cherkasov).
  • In Storage S3(Azure)Queue commit all files (in a single butch defined by commit settings) in a single keeper transaction. #73991 (Kseniia Sumarokova).
  • Disable header detection for executable UDFs and dictionaries (could lead to Function 'X': wrong result, expected Y row(s), actual Y-1). #73992 (Azat Khuzhin).
  • Add the distributed option for EXPLAIN PLAN. Now, EXPLAIN distributed=1 ... appends remote plan to ReadFromParallelRemote* steps. #73994 (Nikolai Kochetov).
  • Use correct return type for not/xor with Dynamic arguments. #74013 (Pavel Kruglov).
  • Allow changing add_implicit_sign_column_constraint_for_collapsing_engine after table creation. #74014 (Christoph Wurm).
  • Support subcolumns in materialized view select query. #74030 (Pavel Kruglov).
  • There are now three simple ways to set a custom prompt in clickhouse-client: 1. via command-line parameter --prompt, 2. in the configuration file, via settings <prompt>[...]</prompt>, and 3. also in the configuration file, via per-connection settings <connections_credentials><prompt>[...]</prompt></connection_credentials>. #74168 (Christoph Wurm).
  • Autodetect secure connection based on connecting to port 9440 in ClickHouse Client. #74212 (Christoph Wurm).
  • Authenticate users with username only for http_handlers (previously it requires user to put the password as well). #74221 (Azat Khuzhin).
  • Support for the alternative query languages PRQL and KQL was marked experimental. To use them, specify settings allow_experimental_prql_dialect = 1 and allow_experimental_kusto_dialect = 1. #74224 (Robert Schulze).
  • Support returning the default Enum type in more aggregate functions. #74272 (Raúl Marín).
  • In OPTIMIZE TABLE, it is now possible to specify keyword FORCE as an alternative to existing keyword FINAL. #74342 (Robert Schulze).
  • Add the IsServerShuttingDown metric, which is needed to trigger an alert when the server shutdown takes too much time. #74429 (Miсhael Stetsyuk).
  • Added Iceberg tables names to EXPLAIN. #74485 (alekseev-maksim).
  • Provide a better error message when using RECURSIVE CTE with the old analyzer. #74523 (Raúl Marín).
  • Show extended error messages in system.errors. #74574 (Vitaly Baranov).
  • Allow to use password for client communication with clickhouse-keeper. This feature is not very useful if you specify proper SSL configuration for server and client, but still can be useful for some cases. Password cannot be longer than 16 characters. It's not connected with Keeper Auth model. #74673 (alesapin).
  • Add error code for config reloader. #74746 (Garrett Thomas).
  • Added support for IPv6 addresses in MySQL and PostgreSQL table functions and engines. #74796 (Mikhail Koviazin).
  • Implement short circuit optimization for divideDecimal. Fixes #74280. #74843 (Kevin Mingtarja).
  • Now users can be specified inside the startup scripts. #74894 (pufit).
  • Add support for Azure SAS Tokens. #72959 (Azat Khuzhin).

Bug Fix (user-visible misbehavior in an official stable release)​

  • Set parquet compression level only if compression codec supports it. #74659 (Arthur Passos).
  • Fixed a regression that using collation locales with modifiers throws an error. As an example, SELECT arrayJoin(['kk 50', 'KK 01', ' KK 2', ' KK 3', 'kk 1', 'x9y99', 'x9y100']) item ORDER BY item ASC COLLATE 'tr-u-kn-true-ka-shifted now works. #73544 (Robert Schulze).
  • Fix cannot create SEQUENTIAL node with keeper-client. #64177 (Duc Canh Le).
  • Fix incorrect character counting in the position functions. #71003 (思维).
  • RESTORE operations for access entities required more permission than necessary because of unhandled partial revokes. This PR fixes the issue. Closes #71853. #71958 (pufit).
  • Avoid pause after ALTER TABLE REPLACE/MOVE PARTITION FROM/TO TABLE. Retrieve correct settings for background task scheduling. #72024 (Aleksei Filatov).
  • Fix handling of empty tuples in some input and output formats (e.g. Parquet, Arrow). #72616 (Michael Kolupaev).
  • Column-level GRANT SELECT/INSERT statements on wildcard databases/tables now throw an error. #72646 (Johann Gan).
  • Fix the situation when a user can't run REVOKE ALL ON *.* because of implicit grants in the target access entity. #72872 (pufit).
  • Fix positive timezone formatting of formatDateTime scalar function. #73091 (ollidraese).
  • Fix to correctly reflect source port when connection made through PROXYv1 and auth_use_forwarded_address is set - previously proxy port was incorrectly used. Add currentQueryID() function. #73095 (Yakov Olkhovskiy).
  • Propagate format settings to NativeWriter in TCPHandler, so settings like output_format_native_write_json_as_string are applied correctly. #73179 (Pavel Kruglov).
  • Fix a crash in StorageObjectStorageQueue. #73274 (Kseniia Sumarokova).
  • Fix rare crash in refreshable materialized view during server shutdown. #73323 (Michael Kolupaev).
  • The %f placeholder of function formatDateTime now unconditionally generates six (sub-second) digits. This makes the behavior compatible with MySQL DATE_FORMAT function. The previous behavior can be restored using setting formatdatetime_f_prints_scale_number_of_digits = 1. #73324 (ollidraese).
  • Fixed filtering by _etag column while reading from s3 storage and table function. #73353 (Anton Popov).
  • Fix Not-ready Set is passed as the second argument for function 'in' error when IN (subquery) is used in JOIN ON expression, with the old analyzer. #73382 (Nikolai Kochetov).
  • Fix preparing for squashin for Dynamic and JSON columns. Previously in some cases new types could be inserted into shared variant/shared data even when the limit on types/paths is not reached. #73388 (Pavel Kruglov).
  • Check for corrupted sizes during types binary decoding to avoid too big allocations. #73390 (Pavel Kruglov).
  • Fixed a logical error when reading from single-replica cluster with parallel replicas enabled. #73403 (Michael Kolupaev).
  • Fix ObjectStorageQueue with ZooKeeper and older Keeper. #73420 (Antonio Andelic).
  • Implements fix, needed to enable hive partitioning by default. #73479 (Yarik Briukhovetskyi).
  • Fix data race when creating vector similarity index. #73517 (Antonio Andelic).
  • Fixes segfault when the source of the dictionary contains a function with wrong data. #73535 (Yarik Briukhovetskyi).
  • Fix retries on failed insert in storage S3(Azure)Queue. Closes #70951. #73546 (Kseniia Sumarokova).
  • Fixed error in function tupleElement which may appear in some cases for tuples with LowCardinality elements and enabled setting optimize_functions_to_subcolumns. #73548 (Anton Popov).
  • Fix parsing enum glob followed by range one. Fixes #73473. #73569 (Konstantin Bogdanov).
  • Fixed parallel_replicas_for_non_replicated_merge_tree being ignored in subqueries for non-replicated tables. #73584 (Igor Nikonov).
  • Fix for std::logical_error thrown when task cannot be scheduled. Found in stress tests. #73629 (Alexander Gololobov).
  • Do not interpret queries in EXPLAIN SYNTAX to avoid logical errors with incorrect processing stage for distributed queries. Fixes #65205. #73634 (Dmitry Novik).
  • Fix possible data inconsistency in Dynamic column. Fixes possible logical error Nested columns sizes are inconsistent with local_discriminators column size. #73644 (Pavel Kruglov).
  • Fixed NOT_FOUND_COLUMN_IN_BLOCK in queries with FINAL and SAMPLE. Fixed incorrect result in selects with FINAL from CollapsingMergeTree and enabled optimizations of FINAL . #73682 (Anton Popov).
  • Fix crash in LIMIT BY COLUMNS. #73686 (Raúl Marín).
  • Fix the bug when the normal projection is forced to use, and query is exactly the same as the projection defined, but the projection is not selected and thus error is prompted. #73700 (Shichao Jin).
  • Fix deserialization of Dynamic/Object structure. It could lead to CANNOT_READ_ALL_DATA exceptions. #73767 (Pavel Kruglov).
  • Skip metadata_version.txt in while restoring parts from a backup. #73768 (Vitaly Baranov).
  • Fix segmentation fault when Casting to Enum with LIKE. #73775 (zhanglistar).
  • Fix for S3 Express bucket not working as disk. #73777 (Sameer Tamsekar).
  • Allow merging of rows with invalid sign column values in CollapsingMergeTree tables. #73864 (Christoph Wurm).
  • Fix getting error when querying ddl with offline replica. #73876 (Tuan Pham Anh).
  • Fixes occasional failure to compare map() types due to possibility to create Map lacking explicit naming ('keys','values') of its nested tuple. #73878 (Yakov Olkhovskiy).
  • Ignore window functions during GROUP BY ALL clause resolution. Fix #73501. #73916 (Dmitry Novik).
  • Fix implicit privileges (worked as wildcard before). #73932 (Azat Khuzhin).
  • Fix high memory usage during nested Maps creation. #73982 (Pavel Kruglov).
  • Fix parsing nested JSON with empty keys. #73993 (Pavel Kruglov).
  • Fix: alias can be not added to the projection if it is referenced by another alias and selected in inverse order. #74033 (Yakov Olkhovskiy).
  • Ignore object not found errors for Azure during plain_rewritable disk initialization. #74059 (Julia Kartseva).
  • Fix behaviour of any and anyLast with enum types and empty table. #74061 (Joanna Hulboj).
  • Fixes case when the user specifies keyword arguments in the kafka table engine. #74064 (Yarik Briukhovetskyi).
  • Fix altering Storage S3Queue settings with "s3queue_" prefix to without and vice versa. #74075 (Kseniia Sumarokova).
  • Add a setting allow_push_predicate_ast_for_distributed_subqueries. This adds AST-based predicate push-down for distributed queries with the analyzer. This is a temporary solution that we use until distributed queries with query plan serialization are supported. Closes #66878 #69472 #65638 #68030 #73718. #74085 (Nikolai Kochetov).
  • Fixes issue when after #73095 port can be present in the forwarded_for field, which leads to inability to resolve host name with port included. #74116 (Yakov Olkhovskiy).
  • Fixed incorrect formatting of ALTER TABLE (DROP STATISTICS ...) (DROP STATISTICS ...). #74126 (Han Fei).
  • Fix for issue #66112. #74128 (Anton Ivashkin).
  • It is no longer possible to use Loop as a table engine in CREATE TABLE. This combination was previously causing segfaults. #74137 (Yarik Briukhovetskyi).
  • Fix security issue to prevent SQL injection in postgresql and sqlite table functions. #74144 (Pablo Marcos).
  • Fix crash when reading a subcolumn from the compressed Memory engine table. Fixes #74009. #74161 (Nikita Taranov).
  • Fixed an infinite loop occurring with queries to the system.detached_tables. #74190 (Konstantin Morozov).
  • Fix logical error in s3queue during setting file as failed. #74216 (Kseniia Sumarokova).
  • Fix native copy settings (allow_s3_native_copy/allow_azure_native_copy) for RESTORE from base backup. #74286 (Azat Khuzhin).
  • Fixed the issue when the number of detached tables in the database is a multiple of max_block_size. #74289 (Konstantin Morozov).
  • Fix copying via ObjectStorage (i.e. S3) when source and destination credentials differs. #74331 (Azat Khuzhin).
  • Fix detection of "use the Rewrite method in the JSON API" for native copy on GCS. #74338 (Azat Khuzhin).
  • Fix incorrect calculation of BackgroundMergesAndMutationsPoolSize (it was x2 from real value). #74509 (alesapin).
  • Fix the bug of leaking keeper watches when enable Cluster Discovery. #74521 (RinChanNOW).
  • Fix mem alignment issue reported by UBSan #74512. #74534 (Arthur Passos).
  • Fix KeeperMap concurrent cleanup during table creation. #74568 (Antonio Andelic).
  • Do not remove unused projection columns in subqueries in the presence of EXCEPT or INTERSECT to preserve the correct query result. Fixes #73930. Fixes #66465. #74577 (Dmitry Novik).
  • Fixed INSERT SELECT queries between tables with Tuple columns and enabled sparse serialization. #74698 (Anton Popov).
  • Function right works incorrectly for const negative offset. #74701 (Daniil Ivanik).
  • Fix insertion of gzip-ed data sometimes fails due to flawed decompression on client side. #74707 (siyuan).
  • Partial revokes with wildcard grants could remove more privileges than expected. Closes #74263. #74751 (pufit).
  • Keeper fix: fix reading log entries from disk. #74785 (Antonio Andelic).
  • Fixed checking grants for SYSTEM REFRESH/START/STOP VIEW, now it's not required to have this grant on *.* to execute a query for a specific view, only grant for this view are required. #74789 (Alexander Tokmakov).
  • The hasColumnInTable function doesn't account for alias columns. Fix it to also work for alias columns. #74841 (Bharat Nallan).
  • Fix FILE_DOESNT_EXIST error occurring during data parts merge for a table with an empty column in Azure Blob Storage. #74892 (Julia Kartseva).
  • Fix projection column name when joining temporary tables, close #68872. #74897 (Vladimir Cherkasov).

Build/Testing/Packaging Improvement​

  • Table of Contents
  • ClickHouse release 25.2, 2025-02-27
  • ClickHouse release 25.1, 2025-01-28