This release contains bug fixes for the previous release 1.1.54318:
- Fixed bug with possible race condition in replication that could lead to data loss. This issue affects versions 1.1.54310 and 1.1.54318. If you use one of these versions with Replicated tables, the update is strongly recommended. This issue shows in logs in Warning messages like
Part ... from own log doesn't exist.The issue is relevant even if you don’t see these messages in logs.
This release contains bug fixes for the previous release 1.1.54310:
- Fixed incorrect row deletions during merges in the SummingMergeTree engine
- Fixed a memory leak in unreplicated MergeTree engines
- Fixed performance degradation with frequent inserts in MergeTree engines
- Fixed an issue that was causing the replication queue to stop running
- Fixed rotation and archiving of server logs
- Custom partitioning key for the MergeTree family of table engines.
- Kafka table engine.
- Added support for loading CatBoost models and applying them to data stored in ClickHouse.
- Added support for time zones with non-integer offsets from UTC.
- Added support for arithmetic operations with time intervals.
- The range of values for the Date and DateTime types is extended to the year 2105.
- Added the
CREATE MATERIALIZED VIEW x TO yquery (specifies an existing table for storing the data of a materialized view).
- Added the
ATTACH TABLEquery without arguments.
- The processing logic for Nested columns with names ending in -Map in a SummingMergeTree table was extracted to the sumMap aggregate function. You can now specify such columns explicitly.
- Max size of the IP trie dictionary is increased to 128M entries.
- Added the getSizeOfEnumType function.
- Added the sumWithOverflow aggregate function.
- Added support for the Cap’n Proto input format.
- You can now customize compression level when using the zstd algorithm.
- Creation of temporary tables with an engine other than Memory is not allowed.
- Explicit creation of tables with the View or MaterializedView engine is not allowed.
- During table creation, a new check verifies that the sampling key expression is included in the primary key.
- Fixed hangups when synchronously inserting into a Distributed table.
- Fixed nonatomic adding and removing of parts in Replicated tables.
- Data inserted into a materialized view is not subjected to unnecessary deduplication.
- Executing a query to a Distributed table for which the local replica is lagging and remote replicas are unavailable does not result in an error anymore.
- Users don’t need access permissions to the
defaultdatabase to create temporary tables anymore.
- Fixed crashing when specifying the Array type without arguments.
- Fixed hangups when the disk volume containing server logs is full.
- Fixed an overflow in the toRelativeWeekNum function for the first week of the Unix epoch.
- Several third-party libraries (notably Poco) were updated and converted to git submodules.
- TLS support in the native protocol (to enable, set
ALTERfor replicated tables now tries to start running as soon as possible.
- Fixed crashing when reading data with the setting
- Fixed crashes of
- Correct interpretation of certain complex queries with
FREEZE PARTITIONalways works atomically now.
- Empty POST requests now return a response with code 411.
- Fixed interpretation errors for expressions like
CAST(1 AS Nullable(UInt8)).
- Fixed an error when reading
- Fixed crashing when parsing queries like
SELECT dummy AS dummy, dummy AS b
- Users are updated correctly with invalid
- Correct handling when an executable dictionary returns a non-zero response code.
- Added the
pointInPolygonfunction for working with coordinates on a coordinate plane.
- Added the
sumMapaggregate function for calculating the sum of arrays, similar to
- Added the
truncfunction. Improved performance of the rounding functions (
roundToExp2) and corrected the logic of how they work. Changed the logic of the
roundToExp2function for fractions and negative numbers.
- The ClickHouse executable file is now less dependent on the libc version. The same ClickHouse executable file can run on a wide variety of Linux systems. There is still a dependency when using compiled queries (with the setting
compile = 1, which is not used by default).
- Reduced the time needed for dynamic compilation of queries.
- Fixed an error that sometimes produced
part ... intersects previous partmessages and weakened replica consistency.
- Fixed an error that caused the server to lock up if ZooKeeper was unavailable during shutdown.
- Removed excessive logging when restoring replicas.
- Fixed an error in the UNION ALL implementation.
- Fixed an error in the concat function that occurred if the first column in a block has the Array type.
- Progress is now displayed correctly in the system.merges table.
SYSTEMqueries for server administration:
SYSTEM RELOAD DICTIONARY,
SYSTEM RELOAD DICTIONARIES,
SYSTEM DROP DNS CACHE,
- Added functions for working with arrays:
identityparameters for the ZooKeeper configuration. This allows you to isolate individual users on the same ZooKeeper cluster.
- Added aggregate functions
groupBitXor(for compatibility, they are also available under the names
- External dictionaries can be loaded from MySQL by specifying a socket in the filesystem.
- External dictionaries can be loaded from MySQL over SSL (
- Added the
max_network_bandwidth_for_usersetting to restrict the overall bandwidth use for queries per user.
- Support for
DROP TABLEfor temporary tables.
- Support for reading
DateTimevalues in Unix timestamp format from the
- Lagging replicas in distributed queries are now excluded by default (the default threshold is 5 minutes).
- FIFO locking is used during ALTER: an ALTER query isn’t blocked indefinitely for continuously running queries.
- Option to set
umaskin the config file.
- Improved performance for queries with
- Improved the process for deleting old nodes in ZooKeeper. Previously, old nodes sometimes didn’t get deleted if there were very frequent inserts, which caused the server to be slow to shut down, among other things.
- Fixed randomization when choosing hosts for the connection to ZooKeeper.
- Fixed the exclusion of lagging replicas in distributed queries if the replica is localhost.
- Fixed an error where a data part in a
ReplicatedMergeTreetable could be broken after running
ALTER MODIFYon an element in a
- Fixed an error that could cause SELECT queries to “hang”.
- Improvements to distributed DDL queries.
- Fixed the query
CREATE TABLE ... AS <materialized view>.
- Resolved the deadlock in the
ALTER ... CLEAR COLUMN IN PARTITIONquery for
- Fixed the invalid default value for
Enums (0 instead of the minimum) when using the
- Resolved the appearance of zombie processes when using a dictionary with an
- Fixed segfault for the HEAD query.
- You can use
pbuilderto build ClickHouse.
- You can use
libstdc++for builds on Linux.
- Added instructions for using static code analysis tools:
- There is now a higher default value for the MergeTree setting
max_bytes_to_merge_at_max_space_in_pool(the maximum total size of data parts to merge, in bytes): it has increased from 100 GiB to 150 GiB. This might result in large merges running after the server upgrade, which could cause an increased load on the disk subsystem. If the free space available on the server is less than twice the total amount of the merges that are running, this will cause all other merges to stop running, including merges of small data parts. As a result, INSERT queries will fail with the message “Merges are processing significantly slower than inserts.” Use the
SELECT * FROM system.mergesquery to monitor the situation. You can also check the
DiskSpaceReservedForMergemetric in the
system.metricstable, or in Graphite. You don’t need to do anything to fix this, since the issue will resolve itself once the large merges finish. If you find this unacceptable, you can restore the previous value for the
max_bytes_to_merge_at_max_space_in_poolsetting. To do this, go to the
<merge_tree>section in config.xml, set
<merge_tree>``<max_bytes_to_merge_at_max_space_in_pool>107374182400</max_bytes_to_merge_at_max_space_in_pool>and restart the server.
- This is a bugfix release for the previous 1.1.54282 release. It fixes leaks in the parts directory in ZooKeeper.
This release contains bug fixes for the previous release 1.1.54276:
DB::Exception: Assertion violation: !_path.empty()when inserting into a Distributed table.
- Fixed parsing when inserting in RowBinary format if input data starts with’;’.
- Errors during runtime compilation of certain aggregate functions (e.g.
- Added an optional WITH section for a SELECT query. Example query:
WITH 1+1 AS a SELECT a, a*a
- INSERT can be performed synchronously in a Distributed table: OK is returned only after all the data is saved on all the shards. This is activated by the setting insert_distributed_sync=1.
- Added the UUID data type for working with 16-byte identifiers.
- Added aliases of CHAR, FLOAT and other types for compatibility with the Tableau.
- Added the functions toYYYYMM, toYYYYMMDD, and toYYYYMMDDhhmmss for converting time into numbers.
- You can use IP addresses (together with the hostname) to identify servers for clustered DDL queries.
- Added support for non-constant arguments and negative offsets in the function
substring(str, pos, len).
- Added the max_size parameter for the
groupArray(max_size)(column)aggregate function, and optimized its performance.
- Security improvements: all server files are created with 0640 permissions (can be changed via
- Improved error messages for queries with invalid syntax.
- Significantly reduced memory consumption and improved performance when merging large sections of MergeTree data.
- Significantly increased the performance of data merges for the ReplacingMergeTree engine.
- Improved performance for asynchronous inserts from a Distributed table by combining multiple source inserts. To enable this functionality, use the setting distributed_directory_monitor_batch_inserts=1.
- Changed the binary format of aggregate states of
groupArray(array_column)functions for arrays.
- Added the
output_format_json_quote_denormalssetting, which enables outputting nan and inf values in JSON format.
- Optimized stream allocation when reading from a Distributed table.
- Settings can be configured in readonly mode if the value doesn’t change.
- Added the ability to retrieve non-integer granules of the MergeTree engine in order to meet restrictions on the block size specified in the preferred_block_size_bytes setting. The purpose is to reduce the consumption of RAM and increase cache locality when processing queries from tables with large columns.
- Efficient use of indexes that contain expressions like
toStartOfHour(x)for conditions like
toStartOfHour(x) op сonstexpr.
- Added new settings for MergeTree engines (the merge_tree section in config.xml):
- replicated_deduplication_window_seconds sets the number of seconds allowed for deduplicating inserts in Replicated tables.
- cleanup_delay_period sets how often to start cleanup to remove outdated data.
- replicated_can_become_leader can prevent a replica from becoming the leader (and assigning merges).
- Accelerated cleanup to remove outdated data from ZooKeeper.
- Multiple improvements and fixes for clustered DDL queries. Of particular interest is the new setting distributed_ddl_task_timeout, which limits the time to wait for a response from the servers in the cluster. If a ddl request has not been performed on all hosts, a response will contain a timeout error and a request will be executed in an async mode.
- Improved display of stack traces in the server logs.
- Added the “none” value for the compression method.
- You can use multiple dictionaries_config sections in config.xml.
- It is possible to connect to MySQL through a socket in the file system.
- The system.parts table has a new column with information about the size of marks, in bytes.
- Distributed tables using a Merge table now work correctly for a SELECT query with a condition on the
- Fixed a rare race condition in ReplicatedMergeTree when checking data parts.
- Fixed possible freezing on “leader election” when starting a server.
- The max_replica_delay_for_distributed_queries setting was ignored when using a local replica of the data source. This has been fixed.
- Fixed incorrect behavior of
ALTER TABLE CLEAR COLUMN IN PARTITIONwhen attempting to clean a non-existing column.
- Fixed an exception in the multiIf function when using empty arrays or strings.
- Fixed excessive memory allocations when deserializing Native format.
- Fixed incorrect auto-update of Trie dictionaries.
- Fixed an exception when running queries with a GROUP BY clause from a Merge table when using SAMPLE.
- Fixed a crash of GROUP BY when using distributed_aggregation_memory_efficient=1.
- Now you can specify the database.table in the right side of IN and JOIN.
- Too many threads were used for parallel aggregation. This has been fixed.
- Fixed how the “if” function works with FixedString arguments.
- SELECT worked incorrectly from a Distributed table for shards with a weight of 0. This has been fixed.
CREATE VIEW IF EXISTS no longer causes crashes.
- Fixed incorrect behavior when input_format_skip_unknown_fields=1 is set and there are negative numbers.
- Fixed an infinite loop in the
dictGetHierarchy()function if there is some invalid data in the dictionary.
Syntax error: unexpected (...)errors when running distributed queries with subqueries in an IN or JOIN clause and Merge tables.
- Fixed an incorrect interpretation of a SELECT query from Dictionary tables.
- Fixed the “Cannot mremap” error when using arrays in IN and JOIN clauses with more than 2 billion elements.
- Fixed the failover for dictionaries with MySQL as the source.
- Builds can be assembled in Arcadia.
- You can use gcc 7 to compile ClickHouse.
- Parallel builds using ccache+distcc are faster now.
- Distributed DDL (for example,
CREATE TABLE ON CLUSTER)
- The replicated query
ALTER TABLE CLEAR COLUMN IN PARTITION.
- The engine for Dictionary tables (access to dictionary data in the form of a table).
- Dictionary database engine (this type of database automatically has Dictionary tables available for all the connected external dictionaries).
- You can check for updates to the dictionary by sending a request to the source.
- Qualified column names
- Quoting identifiers using double quotation marks.
- Sessions in the HTTP interface.
- The OPTIMIZE query for a Replicated table can can run not only on the leader.
- Removed SET GLOBAL.
- Now after an alert is triggered, the log prints the full stack trace.
- Relaxed the verification of the number of damaged/extra data parts at startup (there were too many false positives).
- Fixed a bad connection “sticking” when inserting into a Distributed table.
- GLOBAL IN now works for a query from a Merge table that looks at a Distributed table.
- The incorrect number of cores was detected on a Google Compute Engine virtual machine. This has been fixed.
- Changes in how an executable source of cached external dictionaries works.
- Fixed the comparison of strings containing null characters.
- Fixed the comparison of Float32 primary key fields with constants.
- Previously, an incorrect estimate of the size of a field could lead to overly large allocations.
- Fixed a crash when querying a Nullable column added to a table using ALTER.
- Fixed a crash when sorting by a Nullable column, if the number of rows is less than LIMIT.
- Fixed an ORDER BY subquery consisting of only constant values.
- Previously, a Replicated table could remain in the invalid state after a failed DROP TABLE.
- Aliases for scalar subqueries with empty results are no longer lost.
- Now a query that used compilation does not fail with an error if the .so file gets damaged.