diff --git a/.github/workflows/backport.yml b/.github/workflows/backport.yml index 620c698284e..66dddbee640 100644 --- a/.github/workflows/backport.yml +++ b/.github/workflows/backport.yml @@ -9,6 +9,8 @@ concurrency: on: # yamllint disable-line rule:truthy schedule: - cron: '0 */3 * * *' + workflow_dispatch: + jobs: CherryPick: runs-on: [self-hosted, style-checker] diff --git a/CHANGELOG.md b/CHANGELOG.md index a9c70b24548..317bb766336 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,4 +1,5 @@ ### Table of Contents +**[ClickHouse release v22.6, 2022-06-16](#226)**
**[ClickHouse release v22.5, 2022-05-19](#225)**
**[ClickHouse release v22.4, 2022-04-20](#224)**
**[ClickHouse release v22.3-lts, 2022-03-17](#223)**
@@ -6,6 +7,172 @@ **[ClickHouse release v22.1, 2022-01-18](#221)**
**[Changelog for 2021](https://clickhouse.com/docs/en/whats-new/changelog/2021/)**
+### ClickHouse release 22.6, 2022-06-16 + +#### Backward Incompatible Change +* Remove support for octal number literals in SQL. In previous versions they were parsed as Float64. [#37765](https://github.com/ClickHouse/ClickHouse/pull/37765) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* Changes how settings using `seconds` as type are parsed to support floating point values (for example: `max_execution_time=0.5`). Infinity or NaN values will throw an exception. [#37187](https://github.com/ClickHouse/ClickHouse/pull/37187) ([Raúl Marín](https://github.com/Algunenano)). +* Changed format of binary serialization of columns of experimental type `Object`. New format is more convenient to implement by third-party clients. [#37482](https://github.com/ClickHouse/ClickHouse/pull/37482) ([Anton Popov](https://github.com/CurtizJ)). +* Turn on setting `output_format_json_named_tuples_as_objects` by default. It allows to serialize named tuples as JSON objects in JSON formats. [#37756](https://github.com/ClickHouse/ClickHouse/pull/37756) ([Anton Popov](https://github.com/CurtizJ)). +* LIKE patterns with trailing escape symbol ('\\') are now disallowed (as mandated by the SQL standard). [#37764](https://github.com/ClickHouse/ClickHouse/pull/37764) ([Robert Schulze](https://github.com/rschu1ze)). +* If you run different ClickHouse versions on a cluster with AArch64 CPU or mix AArch64 and amd64 on a cluster, and use distributed queries with GROUP BY multiple keys of fixed-size type that fit in 256 bits but don't fit in 64 bits, and the size of the result is huge, the data will not be fully aggregated in the result of these queries during upgrade. Workaround: upgrade with downtime instead of a rolling upgrade. + +#### New Feature +* A new codec [FPC](https://userweb.cs.txstate.edu/~burtscher/papers/dcc07a.pdf) algorithm for floating point data compression. [#37553](https://github.com/ClickHouse/ClickHouse/pull/37553) ([Mikhail Guzov](https://github.com/koloshmet)). +* Add new columnar JSON formats: `JSONColumns`, `JSONCompactColumns`, `JSONColumnsWithMetadata`. Closes [#36338](https://github.com/ClickHouse/ClickHouse/issues/36338) Closes [#34509](https://github.com/ClickHouse/ClickHouse/issues/34509). [#36975](https://github.com/ClickHouse/ClickHouse/pull/36975) ([Kruglov Pavel](https://github.com/Avogar)). +* Added open telemetry traces visualizing tool based on d3js. [#37810](https://github.com/ClickHouse/ClickHouse/pull/37810) ([Sergei Trifonov](https://github.com/serxa)). +* Support INSERTs into `system.zookeeper` table. Closes [#22130](https://github.com/ClickHouse/ClickHouse/issues/22130). [#37596](https://github.com/ClickHouse/ClickHouse/pull/37596) ([Han Fei](https://github.com/hanfei1991)). +* Support non-constant pattern argument for `LIKE`, `ILIKE` and `match` functions. [#37251](https://github.com/ClickHouse/ClickHouse/pull/37251) ([Robert Schulze](https://github.com/rschu1ze)). +* Executable user defined functions now support parameters. Example: `SELECT test_function(parameters)(arguments)`. Closes [#37578](https://github.com/ClickHouse/ClickHouse/issues/37578). [#37720](https://github.com/ClickHouse/ClickHouse/pull/37720) ([Maksim Kita](https://github.com/kitaisreal)). +* Add `merge_reason` column to system.part_log table. [#36912](https://github.com/ClickHouse/ClickHouse/pull/36912) ([Sema Checherinda](https://github.com/CheSema)). +* Add support for Maps and Records in Avro format. Add new setting `input_format_avro_null_as_default ` that allow to insert null as default in Avro format. Closes [#18925](https://github.com/ClickHouse/ClickHouse/issues/18925) Closes [#37378](https://github.com/ClickHouse/ClickHouse/issues/37378) Closes [#32899](https://github.com/ClickHouse/ClickHouse/issues/32899). [#37525](https://github.com/ClickHouse/ClickHouse/pull/37525) ([Kruglov Pavel](https://github.com/Avogar)). +* Add `clickhouse-disks` tool to introspect and operate on virtual filesystems configured for ClickHouse. [#36060](https://github.com/ClickHouse/ClickHouse/pull/36060) ([Artyom Yurkov](https://github.com/Varinara)). +* Adds H3 unidirectional edge functions. [#36843](https://github.com/ClickHouse/ClickHouse/pull/36843) ([Bharat Nallan](https://github.com/bharatnc)). +* Add support for calculating [hashids](https://hashids.org/) from unsigned integers. [#37013](https://github.com/ClickHouse/ClickHouse/pull/37013) ([Michael Nutt](https://github.com/mnutt)). +* Explicit `SALT` specification is allowed for `CREATE USER IDENTIFIED WITH sha256_hash`. [#37377](https://github.com/ClickHouse/ClickHouse/pull/37377) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* Add two new settings `input_format_csv_skip_first_lines/input_format_tsv_skip_first_lines` to allow skipping specified number of lines in the beginning of the file in CSV/TSV formats. [#37537](https://github.com/ClickHouse/ClickHouse/pull/37537) ([Kruglov Pavel](https://github.com/Avogar)). +* `showCertificate` function shows current server's SSL certificate. [#37540](https://github.com/ClickHouse/ClickHouse/pull/37540) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* HTTP source for Data Dictionaries in Named Collections is supported. [#37581](https://github.com/ClickHouse/ClickHouse/pull/37581) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* Added a new window function `nonNegativeDerivative(metric_column, timestamp_column[, INTERVAL x SECOND])`. [#37628](https://github.com/ClickHouse/ClickHouse/pull/37628) ([Andrey Zvonov](https://github.com/zvonand)). +* Implemented changing the comment for `ReplicatedMergeTree` tables. [#37416](https://github.com/ClickHouse/ClickHouse/pull/37416) ([Vasily Nemkov](https://github.com/Enmk)). +* Added `SYSTEM UNFREEZE` query that deletes the whole backup regardless if the corresponding table is deleted or not. [#36424](https://github.com/ClickHouse/ClickHouse/pull/36424) ([Vadim Volodin](https://github.com/PolyProgrammist)). + +#### Experimental Feature +* Add `GROUPING` function. Closes [#19426](https://github.com/ClickHouse/ClickHouse/issues/19426). [#37163](https://github.com/ClickHouse/ClickHouse/pull/37163) ([Dmitry Novik](https://github.com/novikd)). The behavior of this function will be changed in subsequent releases. +* Enables `POPULATE` for `WINDOW VIEW`. [#36945](https://github.com/ClickHouse/ClickHouse/pull/36945) ([vxider](https://github.com/Vxider)). +* `ALTER TABLE ... MODIFY QUERY` support for `WINDOW VIEW`. [#37188](https://github.com/ClickHouse/ClickHouse/pull/37188) ([vxider](https://github.com/Vxider)). +* This PR changes the behavior of the `ENGINE` syntax in `WINDOW VIEW`, to make it like in `MATERIALIZED VIEW`. [#37214](https://github.com/ClickHouse/ClickHouse/pull/37214) ([vxider](https://github.com/Vxider)). + +#### Performance Improvement +* Added numerous optimizations for ARM NEON [#38093](https://github.com/ClickHouse/ClickHouse/pull/38093)([Daniel Kutenin](https://github.com/danlark1)), ([Alexandra Pilipyuk](https://github.com/chalice19)) Note: if you run different ClickHouse versions on a cluster with ARM CPU and use distributed queries with GROUP BY multiple keys of fixed-size type that fit in 256 bits but don't fit in 64 bits, the result of the aggregation query will be wrong during upgrade. Workaround: upgrade with downtime instead of a rolling upgrade. +* Improve performance and memory usage for select of subset of columns for formats Native, Protobuf, CapnProto, JSONEachRow, TSKV, all formats with suffixes WithNames/WithNamesAndTypes. Previously while selecting only subset of columns from files in these formats all columns were read and stored in memory. Now only required columns are read. This PR enables setting `input_format_skip_unknown_fields` by default, because otherwise in case of select of subset of columns exception will be thrown. [#37192](https://github.com/ClickHouse/ClickHouse/pull/37192) ([Kruglov Pavel](https://github.com/Avogar)). +* Now more filters can be pushed down for join. [#37472](https://github.com/ClickHouse/ClickHouse/pull/37472) ([Amos Bird](https://github.com/amosbird)). +* Load marks for only necessary columns when reading wide parts. [#36879](https://github.com/ClickHouse/ClickHouse/pull/36879) ([Anton Kozlov](https://github.com/tonickkozlov)). +* Improved performance of aggregation in case, when sparse columns (can be enabled by experimental setting `ratio_of_defaults_for_sparse_serialization` in `MergeTree` tables) are used as arguments in aggregate functions. [#37617](https://github.com/ClickHouse/ClickHouse/pull/37617) ([Anton Popov](https://github.com/CurtizJ)). +* Optimize function `COALESCE` with two arguments. [#37666](https://github.com/ClickHouse/ClickHouse/pull/37666) ([Anton Popov](https://github.com/CurtizJ)). +* Replace `multiIf` to `if` in case when `multiIf` has only one condition, because function `if` is more performant. [#37695](https://github.com/ClickHouse/ClickHouse/pull/37695) ([Anton Popov](https://github.com/CurtizJ)). +* Improve performance of `dictGetDescendants`, `dictGetChildren` functions, create temporary parent to children hierarchical index per query, not per function call during query. Allow to specify `BIDIRECTIONAL` for `HIERARHICAL` attributes, dictionary will maintain parent to children index in memory, that way functions `dictGetDescendants`, `dictGetChildren` will not create temporary index per query. Closes [#32481](https://github.com/ClickHouse/ClickHouse/issues/32481). [#37148](https://github.com/ClickHouse/ClickHouse/pull/37148) ([Maksim Kita](https://github.com/kitaisreal)). +* Aggregates state destruction now may be posted on a thread pool. For queries with LIMIT and big state it provides significant speedup, e.g. `select uniq(number) from numbers_mt(1e7) group by number limit 100` became around 2.5x faster. [#37855](https://github.com/ClickHouse/ClickHouse/pull/37855) ([Nikita Taranov](https://github.com/nickitat)). +* Improve sort performance by single column. [#37195](https://github.com/ClickHouse/ClickHouse/pull/37195) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of single column sorting using sorting queue specializations. [#37990](https://github.com/ClickHouse/ClickHouse/pull/37990) ([Maksim Kita](https://github.com/kitaisreal)). +* Improved performance on array norm and distance functions 2x-4x times. [#37394](https://github.com/ClickHouse/ClickHouse/pull/37394) ([Alexander Gololobov](https://github.com/davenger)). +* Improve performance of number comparison functions using dynamic dispatch. [#37399](https://github.com/ClickHouse/ClickHouse/pull/37399) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of ORDER BY with LIMIT. [#37481](https://github.com/ClickHouse/ClickHouse/pull/37481) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of `hasAll` function using dynamic dispatch infrastructure. [#37484](https://github.com/ClickHouse/ClickHouse/pull/37484) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of `greatCircleAngle`, `greatCircleDistance`, `geoDistance` functions. [#37524](https://github.com/ClickHouse/ClickHouse/pull/37524) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of insert into MergeTree if there are multiple columns in ORDER BY. [#35762](https://github.com/ClickHouse/ClickHouse/pull/35762) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix excessive CPU usage in background when there are a lot of tables. [#38028](https://github.com/ClickHouse/ClickHouse/pull/38028) ([Maksim Kita](https://github.com/kitaisreal)). +* Improve performance of `not` function using dynamic dispatch. [#38058](https://github.com/ClickHouse/ClickHouse/pull/38058) ([Maksim Kita](https://github.com/kitaisreal)). +* Optimized the internal caching of re2 patterns which occur e.g. in LIKE and MATCH functions. [#37544](https://github.com/ClickHouse/ClickHouse/pull/37544) ([Robert Schulze](https://github.com/rschu1ze)). +* Improve filter bitmask generator function all in one with AVX-512 instructions. [#37588](https://github.com/ClickHouse/ClickHouse/pull/37588) ([yaqi-zhao](https://github.com/yaqi-zhao)). +* Apply read method `threadpool` for Hive integration engine. This will significantly speed up reading. [#36328](https://github.com/ClickHouse/ClickHouse/pull/36328) ([李扬](https://github.com/taiyang-li)). +* When all the columns to read are partition keys, construct columns by the file's row number without real reading the Hive file. [#37103](https://github.com/ClickHouse/ClickHouse/pull/37103) ([lgbo](https://github.com/lgbo-ustc)). +* Support multi disks for caching hive files. [#37279](https://github.com/ClickHouse/ClickHouse/pull/37279) ([lgbo](https://github.com/lgbo-ustc)). +* Limiting the maximum cache usage per query can effectively prevent cache pool contamination. [Related Issues](https://github.com/ClickHouse/ClickHouse/issues/28961). [#37859](https://github.com/ClickHouse/ClickHouse/pull/37859) ([Han Shukai](https://github.com/KinderRiven)). +* Currently clickhouse directly downloads all remote files to the local cache (even if they are only read once), which will frequently cause IO of the local hard disk. In some scenarios, these IOs may not be necessary and may easily cause negative optimization. As shown in the figure below, when we run SSB Q1-Q4, the performance of the cache has caused negative optimization. [#37516](https://github.com/ClickHouse/ClickHouse/pull/37516) ([Han Shukai](https://github.com/KinderRiven)). +* Allow to prune the list of files via virtual columns such as `_file` and `_path` when reading from S3. This is for [#37174](https://github.com/ClickHouse/ClickHouse/issues/37174) , [#23494](https://github.com/ClickHouse/ClickHouse/issues/23494). [#37356](https://github.com/ClickHouse/ClickHouse/pull/37356) ([Amos Bird](https://github.com/amosbird)). +* In function: CompressedWriteBuffer::nextImpl(), there is an unnecessary write-copy step that would happen frequently during inserting data. Below shows the differentiation with this patch: - Before: 1. Compress "working_buffer" into "compressed_buffer" 2. write-copy into "out" - After: Directly Compress "working_buffer" into "out". [#37242](https://github.com/ClickHouse/ClickHouse/pull/37242) ([jasperzhu](https://github.com/jinjunzh)). + +#### Improvement +* Support types with non-standard defaults in ROLLUP, CUBE, GROUPING SETS. Closes [#37360](https://github.com/ClickHouse/ClickHouse/issues/37360). [#37667](https://github.com/ClickHouse/ClickHouse/pull/37667) ([Dmitry Novik](https://github.com/novikd)). +* Fix stack traces collection on ARM. Closes [#37044](https://github.com/ClickHouse/ClickHouse/issues/37044). Closes [#15638](https://github.com/ClickHouse/ClickHouse/issues/15638). [#37797](https://github.com/ClickHouse/ClickHouse/pull/37797) ([Maksim Kita](https://github.com/kitaisreal)). +* Client will try every IP address returned by DNS resolution until successful connection. [#37273](https://github.com/ClickHouse/ClickHouse/pull/37273) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* Allow to use String type instead of Binary in Arrow/Parquet/ORC formats. This PR introduces 3 new settings for it: `output_format_arrow_string_as_string`, `output_format_parquet_string_as_string`, `output_format_orc_string_as_string`. Default value for all settings is `false`. [#37327](https://github.com/ClickHouse/ClickHouse/pull/37327) ([Kruglov Pavel](https://github.com/Avogar)). +* Apply setting `input_format_max_rows_to_read_for_schema_inference` for all read rows in total from all files in globs. Previously setting `input_format_max_rows_to_read_for_schema_inference` was applied for each file in glob separately and in case of huge number of nulls we could read first `input_format_max_rows_to_read_for_schema_inference` rows from each file and get nothing. Also increase default value for this setting to 25000. [#37332](https://github.com/ClickHouse/ClickHouse/pull/37332) ([Kruglov Pavel](https://github.com/Avogar)). +* Add separate `CLUSTER` grant (and `access_control_improvements.on_cluster_queries_require_cluster_grant` configuration directive, for backward compatibility, default to `false`). [#35767](https://github.com/ClickHouse/ClickHouse/pull/35767) ([Azat Khuzhin](https://github.com/azat)). +* Added support for schema inference for `hdfsCluster`. [#35812](https://github.com/ClickHouse/ClickHouse/pull/35812) ([Nikita Mikhaylov](https://github.com/nikitamikhaylov)). +* Implement `least_used` load balancing algorithm for disks inside volume (multi disk configuration). [#36686](https://github.com/ClickHouse/ClickHouse/pull/36686) ([Azat Khuzhin](https://github.com/azat)). +* Modify the HTTP Endpoint to return the full stats under the `X-ClickHouse-Summary` header when `send_progress_in_http_headers=0` (before it would return all zeros). - Modify the HTTP Endpoint to return `X-ClickHouse-Exception-Code` header when progress has been sent before (`send_progress_in_http_headers=1`) - Modify the HTTP Endpoint to return `HTTP_REQUEST_TIMEOUT` (408) instead of `HTTP_INTERNAL_SERVER_ERROR` (500) on `TIMEOUT_EXCEEDED` errors. [#36884](https://github.com/ClickHouse/ClickHouse/pull/36884) ([Raúl Marín](https://github.com/Algunenano)). +* Allow a user to inspect grants from granted roles. [#36941](https://github.com/ClickHouse/ClickHouse/pull/36941) ([nvartolomei](https://github.com/nvartolomei)). +* Do not calculate an integral numerically but use CDF functions instead. This will speed up execution and will increase the precision. This fixes [#36714](https://github.com/ClickHouse/ClickHouse/issues/36714). [#36953](https://github.com/ClickHouse/ClickHouse/pull/36953) ([Nikita Mikhaylov](https://github.com/nikitamikhaylov)). +* Add default implementation for Nothing in functions. Now most of the functions will return column with type Nothing in case one of it's arguments is Nothing. It also solves problem with functions like arrayMap/arrayFilter and similar when they have empty array as an argument. Previously queries like `select arrayMap(x -> 2 * x, []);` failed because function inside lambda cannot work with type `Nothing`, now such queries return empty array with type `Array(Nothing)`. Also add support for arrays of nullable types in functions like arrayFilter/arrayFill. Previously, queries like `select arrayFilter(x -> x % 2, [1, NULL])` failed, now they work (if the result of lambda is NULL, then this value won't be included in the result). Closes [#37000](https://github.com/ClickHouse/ClickHouse/issues/37000). [#37048](https://github.com/ClickHouse/ClickHouse/pull/37048) ([Kruglov Pavel](https://github.com/Avogar)). +* Now if a shard has local replica we create a local plan and a plan to read from all remote replicas. They have shared initiator which coordinates reading. [#37204](https://github.com/ClickHouse/ClickHouse/pull/37204) ([Nikita Mikhaylov](https://github.com/nikitamikhaylov)). +* Do no longer abort server startup if configuration option "mark_cache_size" is not explicitly set. [#37326](https://github.com/ClickHouse/ClickHouse/pull/37326) ([Robert Schulze](https://github.com/rschu1ze)). +* Allows providing `NULL`/`NOT NULL` right after type in column declaration. [#37337](https://github.com/ClickHouse/ClickHouse/pull/37337) ([Igor Nikonov](https://github.com/devcrafter)). +* optimize file segment PARTIALLY_DOWNLOADED get read buffer. [#37338](https://github.com/ClickHouse/ClickHouse/pull/37338) ([xiedeyantu](https://github.com/xiedeyantu)). +* Try to improve short circuit functions processing to fix problems with stress tests. [#37384](https://github.com/ClickHouse/ClickHouse/pull/37384) ([Kruglov Pavel](https://github.com/Avogar)). +* Closes [#37395](https://github.com/ClickHouse/ClickHouse/issues/37395). [#37415](https://github.com/ClickHouse/ClickHouse/pull/37415) ([Memo](https://github.com/Joeywzr)). +* Fix extremely rare deadlock during part fetch in zero-copy replication. Fixes [#37423](https://github.com/ClickHouse/ClickHouse/issues/37423). [#37424](https://github.com/ClickHouse/ClickHouse/pull/37424) ([metahys](https://github.com/metahys)). +* Don't allow to create storage with unknown data format. [#37450](https://github.com/ClickHouse/ClickHouse/pull/37450) ([Kruglov Pavel](https://github.com/Avogar)). +* Set `global_memory_usage_overcommit_max_wait_microseconds` default value to 5 seconds. Add info about `OvercommitTracker` to OOM exception message. Add `MemoryOvercommitWaitTimeMicroseconds` profile event. [#37460](https://github.com/ClickHouse/ClickHouse/pull/37460) ([Dmitry Novik](https://github.com/novikd)). +* Do not display `-0.0` CPU time in clickhouse-client. It can appear due to rounding errors. This closes [#38003](https://github.com/ClickHouse/ClickHouse/issues/38003). This closes [#38038](https://github.com/ClickHouse/ClickHouse/issues/38038). [#38064](https://github.com/ClickHouse/ClickHouse/pull/38064) ([Alexey Milovidov](https://github.com/alexey-milovidov)). +* Play UI: Keep controls in place when the page is scrolled horizontally. This makes edits comfortable even if the table is wide and it was scrolled far to the right. The feature proposed by Maksym Tereshchenko from CaspianDB. [#37470](https://github.com/ClickHouse/ClickHouse/pull/37470) ([Alexey Milovidov](https://github.com/alexey-milovidov)). +* Modify query div in play.html to be extendable beyond 20% height. In case of very long queries it is helpful to extend the textarea element, only today, since the div is fixed height, the extended textarea hides the data div underneath. With this fix, extending the textarea element will push the data div down/up such the extended textarea won't hide it. Also, keeps query box width 100% even when the user adjusting the size of the query textarea. [#37488](https://github.com/ClickHouse/ClickHouse/pull/37488) ([guyco87](https://github.com/guyco87)). +* Added `ProfileEvents` for introspection of type of written (inserted or merged) parts (`Inserted{Wide/Compact/InMemory}Parts`, `MergedInto{Wide/Compact/InMemory}Parts`. Added column `part_type` to `system.part_log`. Resolves [#37495](https://github.com/ClickHouse/ClickHouse/issues/37495). [#37536](https://github.com/ClickHouse/ClickHouse/pull/37536) ([Anton Popov](https://github.com/CurtizJ)). +* clickhouse-keeper improvement: move broken logs to a timestamped folder. [#37565](https://github.com/ClickHouse/ClickHouse/pull/37565) ([Antonio Andelic](https://github.com/antonio2368)). +* Do not write expired columns by TTL after subsequent merges (before only first merge/optimize of the part will not write expired by TTL columns, all other will do). [#37570](https://github.com/ClickHouse/ClickHouse/pull/37570) ([Azat Khuzhin](https://github.com/azat)). +* More precise result of the `dumpColumnStructure` miscellaneous function in presence of LowCardinality or Sparse columns. In previous versions, these functions were converting the argument to a full column before returning the result. This is needed to provide an answer in [#6935](https://github.com/ClickHouse/ClickHouse/issues/6935). [#37633](https://github.com/ClickHouse/ClickHouse/pull/37633) ([Alexey Milovidov](https://github.com/alexey-milovidov)). +* clickhouse-keeper: store only unique session IDs for watches. [#37641](https://github.com/ClickHouse/ClickHouse/pull/37641) ([Azat Khuzhin](https://github.com/azat)). +* Fix possible "Cannot write to finalized buffer". [#37645](https://github.com/ClickHouse/ClickHouse/pull/37645) ([Azat Khuzhin](https://github.com/azat)). +* Add setting `support_batch_delete` for `DiskS3` to disable multiobject delete calls, which Google Cloud Storage doesn't support. [#37659](https://github.com/ClickHouse/ClickHouse/pull/37659) ([Fred Wulff](https://github.com/frew)). +* Add an option to disable connection pooling in ODBC bridge. [#37705](https://github.com/ClickHouse/ClickHouse/pull/37705) ([Anton Kozlov](https://github.com/tonickkozlov)). +* Functions `dictGetHierarchy`, `dictIsIn`, `dictGetChildren`, `dictGetDescendants` added support nullable `HIERARCHICAL` attribute in dictionaries. Closes [#35521](https://github.com/ClickHouse/ClickHouse/issues/35521). [#37805](https://github.com/ClickHouse/ClickHouse/pull/37805) ([Maksim Kita](https://github.com/kitaisreal)). +* Expose BoringSSL version related info in the `system.build_options` table. [#37850](https://github.com/ClickHouse/ClickHouse/pull/37850) ([Bharat Nallan](https://github.com/bharatnc)). +* Now clickhouse-server removes `delete_tmp` directories on server start. Fixes [#26503](https://github.com/ClickHouse/ClickHouse/issues/26503). [#37906](https://github.com/ClickHouse/ClickHouse/pull/37906) ([alesapin](https://github.com/alesapin)). +* Clean up broken detached parts after timeout. Closes [#25195](https://github.com/ClickHouse/ClickHouse/issues/25195). [#37975](https://github.com/ClickHouse/ClickHouse/pull/37975) ([Kseniia Sumarokova](https://github.com/kssenii)). +* Now in MergeTree table engines family failed-to-move parts will be removed instantly. [#37994](https://github.com/ClickHouse/ClickHouse/pull/37994) ([alesapin](https://github.com/alesapin)). +* Now if setting `always_fetch_merged_part` is enabled for ReplicatedMergeTree merges will try to find parts on other replicas rarely with smaller load for [Zoo]Keeper. [#37995](https://github.com/ClickHouse/ClickHouse/pull/37995) ([alesapin](https://github.com/alesapin)). +* Add implicit grants with grant option too. For example `GRANT CREATE TABLE ON test.* TO A WITH GRANT OPTION` now allows `A` to execute `GRANT CREATE VIEW ON test.* TO B`. [#38017](https://github.com/ClickHouse/ClickHouse/pull/38017) ([Vitaly Baranov](https://github.com/vitlibar)). + +#### Build/Testing/Packaging Improvement +* Use `clang-14` and LLVM infrastructure version 14 for builds. This closes [#34681](https://github.com/ClickHouse/ClickHouse/issues/34681). [#34754](https://github.com/ClickHouse/ClickHouse/pull/34754) ([Alexey Milovidov](https://github.com/alexey-milovidov)). Note: `clang-14` has [a bug](https://github.com/google/sanitizers/issues/1540) in ThreadSanitizer that makes our CI work worse. +* Allow to drop privileges at startup. This simplifies Docker images. Closes [#36293](https://github.com/ClickHouse/ClickHouse/issues/36293). [#36341](https://github.com/ClickHouse/ClickHouse/pull/36341) ([Alexey Milovidov](https://github.com/alexey-milovidov)). +* Add docs spellcheck to CI. [#37790](https://github.com/ClickHouse/ClickHouse/pull/37790) ([Vladimir C](https://github.com/vdimir)). +* Fix overly aggressive stripping which removed the embedded hash required for checking the consistency of the executable. [#37993](https://github.com/ClickHouse/ClickHouse/pull/37993) ([Robert Schulze](https://github.com/rschu1ze)). + +#### Bug Fix + +* Fix `SELECT ... INTERSECT` and `EXCEPT SELECT` statements with constant string types. [#37738](https://github.com/ClickHouse/ClickHouse/pull/37738) ([Antonio Andelic](https://github.com/antonio2368)). +* Fix `GROUP BY` `AggregateFunction` (i.e. you `GROUP BY` by the column that has `AggregateFunction` type). [#37093](https://github.com/ClickHouse/ClickHouse/pull/37093) ([Azat Khuzhin](https://github.com/azat)). +* (experimental WINDOW VIEW) Fix `addDependency` in WindowView. This bug can be reproduced like [#37237](https://github.com/ClickHouse/ClickHouse/issues/37237). [#37224](https://github.com/ClickHouse/ClickHouse/pull/37224) ([vxider](https://github.com/Vxider)). +* Fix inconsistency in ORDER BY ... WITH FILL feature. Query, containing ORDER BY ... WITH FILL, can generate extra rows when multiple WITH FILL columns are present. [#38074](https://github.com/ClickHouse/ClickHouse/pull/38074) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* This PR moving `addDependency` from constructor to `startup()` to avoid adding dependency to a *dropped* table, fix [#37237](https://github.com/ClickHouse/ClickHouse/issues/37237). [#37243](https://github.com/ClickHouse/ClickHouse/pull/37243) ([vxider](https://github.com/Vxider)). +* Fix inserting defaults for missing values in columnar formats. Previously missing columns were filled with defaults for types, not for columns. [#37253](https://github.com/ClickHouse/ClickHouse/pull/37253) ([Kruglov Pavel](https://github.com/Avogar)). +* (experimental Object type) Fix some cases of insertion nested arrays to columns of type `Object`. [#37305](https://github.com/ClickHouse/ClickHouse/pull/37305) ([Anton Popov](https://github.com/CurtizJ)). +* Fix unexpected errors with a clash of constant strings in aggregate function, prewhere and join. Close [#36891](https://github.com/ClickHouse/ClickHouse/issues/36891). [#37336](https://github.com/ClickHouse/ClickHouse/pull/37336) ([Vladimir C](https://github.com/vdimir)). +* Fix projections with GROUP/ORDER BY in query and optimize_aggregation_in_order (before the result was incorrect since only finish sorting was performed). [#37342](https://github.com/ClickHouse/ClickHouse/pull/37342) ([Azat Khuzhin](https://github.com/azat)). +* Fixed error with symbols in key name in S3. Fixes [#33009](https://github.com/ClickHouse/ClickHouse/issues/33009). [#37344](https://github.com/ClickHouse/ClickHouse/pull/37344) ([Vladimir Chebotarev](https://github.com/excitoon)). +* Throw an exception when GROUPING SETS used with ROLLUP or CUBE. [#37367](https://github.com/ClickHouse/ClickHouse/pull/37367) ([Dmitry Novik](https://github.com/novikd)). +* Fix LOGICAL_ERROR in getMaxSourcePartsSizeForMerge during merges (in case of non standard, greater, values of `background_pool_size`/`background_merges_mutations_concurrency_ratio` has been specified in `config.xml` (new way) not in `users.xml` (deprecated way)). [#37413](https://github.com/ClickHouse/ClickHouse/pull/37413) ([Azat Khuzhin](https://github.com/azat)). +* Stop removing UTF-8 BOM in RowBinary format. [#37428](https://github.com/ClickHouse/ClickHouse/pull/37428) ([Paul Loyd](https://github.com/loyd)). [#37428](https://github.com/ClickHouse/ClickHouse/pull/37428) ([Paul Loyd](https://github.com/loyd)). +* clickhouse-keeper bugfix: fix force recovery for single node cluster. [#37440](https://github.com/ClickHouse/ClickHouse/pull/37440) ([Antonio Andelic](https://github.com/antonio2368)). +* Fix logical error in normalizeUTF8 functions. Closes [#37298](https://github.com/ClickHouse/ClickHouse/issues/37298). [#37443](https://github.com/ClickHouse/ClickHouse/pull/37443) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix cast lowcard of nullable in JoinSwitcher, close [#37385](https://github.com/ClickHouse/ClickHouse/issues/37385). [#37453](https://github.com/ClickHouse/ClickHouse/pull/37453) ([Vladimir C](https://github.com/vdimir)). +* Fix named tuples output in ORC/Arrow/Parquet formats. [#37458](https://github.com/ClickHouse/ClickHouse/pull/37458) ([Kruglov Pavel](https://github.com/Avogar)). +* Fix optimization of monotonous functions in ORDER BY clause in presence of GROUPING SETS. Fixes [#37401](https://github.com/ClickHouse/ClickHouse/issues/37401). [#37493](https://github.com/ClickHouse/ClickHouse/pull/37493) ([Dmitry Novik](https://github.com/novikd)). +* Fix error on joining with dictionary on some conditions. Close [#37386](https://github.com/ClickHouse/ClickHouse/issues/37386). [#37530](https://github.com/ClickHouse/ClickHouse/pull/37530) ([Vladimir C](https://github.com/vdimir)). +* Prohibit `optimize_aggregation_in_order` with `GROUPING SETS` (fixes `LOGICAL_ERROR`). [#37542](https://github.com/ClickHouse/ClickHouse/pull/37542) ([Azat Khuzhin](https://github.com/azat)). +* Fix wrong dump information of ActionsDAG. [#37587](https://github.com/ClickHouse/ClickHouse/pull/37587) ([zhanglistar](https://github.com/zhanglistar)). +* Fix converting types for UNION queries (may produce LOGICAL_ERROR). [#37593](https://github.com/ClickHouse/ClickHouse/pull/37593) ([Azat Khuzhin](https://github.com/azat)). +* Fix `WITH FILL` modifier with negative intervals in `STEP` clause. Fixes [#37514](https://github.com/ClickHouse/ClickHouse/issues/37514). [#37600](https://github.com/ClickHouse/ClickHouse/pull/37600) ([Anton Popov](https://github.com/CurtizJ)). +* Fix illegal joinGet array usage when ` join_use_nulls = 1`. This fixes [#37562](https://github.com/ClickHouse/ClickHouse/issues/37562) . [#37650](https://github.com/ClickHouse/ClickHouse/pull/37650) ([Amos Bird](https://github.com/amosbird)). +* Fix columns number mismatch in cross join, close [#37561](https://github.com/ClickHouse/ClickHouse/issues/37561). [#37653](https://github.com/ClickHouse/ClickHouse/pull/37653) ([Vladimir C](https://github.com/vdimir)). +* Fix segmentation fault in `show create table` from mysql database when it is configured with named collections. Closes [#37683](https://github.com/ClickHouse/ClickHouse/issues/37683). [#37690](https://github.com/ClickHouse/ClickHouse/pull/37690) ([Kseniia Sumarokova](https://github.com/kssenii)). +* Fix RabbitMQ Storage not being able to startup on server restart if storage was create without SETTINGS clause. Closes [#37463](https://github.com/ClickHouse/ClickHouse/issues/37463). [#37691](https://github.com/ClickHouse/ClickHouse/pull/37691) ([Kseniia Sumarokova](https://github.com/kssenii)). +* SQL user defined functions disable CREATE/DROP in readonly mode. Closes [#37280](https://github.com/ClickHouse/ClickHouse/issues/37280). [#37699](https://github.com/ClickHouse/ClickHouse/pull/37699) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix formatting of Nullable arguments for executable user defined functions. Closes [#35897](https://github.com/ClickHouse/ClickHouse/issues/35897). [#37711](https://github.com/ClickHouse/ClickHouse/pull/37711) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix optimization enabled by setting `optimize_monotonous_functions_in_order_by` in distributed queries. Fixes [#36037](https://github.com/ClickHouse/ClickHouse/issues/36037). [#37724](https://github.com/ClickHouse/ClickHouse/pull/37724) ([Anton Popov](https://github.com/CurtizJ)). +* Fix possible logical error: `Invalid Field get from type UInt64 to type Float64` in `values` table function. Closes [#37602](https://github.com/ClickHouse/ClickHouse/issues/37602). [#37754](https://github.com/ClickHouse/ClickHouse/pull/37754) ([Kruglov Pavel](https://github.com/Avogar)). +* Fix possible segfault in schema inference in case of exception in SchemaReader constructor. Closes [#37680](https://github.com/ClickHouse/ClickHouse/issues/37680). [#37760](https://github.com/ClickHouse/ClickHouse/pull/37760) ([Kruglov Pavel](https://github.com/Avogar)). +* Fix setting cast_ipv4_ipv6_default_on_conversion_error for internal cast function. Closes [#35156](https://github.com/ClickHouse/ClickHouse/issues/35156). [#37761](https://github.com/ClickHouse/ClickHouse/pull/37761) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix toString error on DatatypeDate32. [#37775](https://github.com/ClickHouse/ClickHouse/pull/37775) ([LiuNeng](https://github.com/liuneng1994)). +* The clickhouse-keeper setting `dead_session_check_period_ms` was transformed into microseconds (multiplied by 1000), which lead to dead sessions only being cleaned up after several minutes (instead of 500ms). [#37824](https://github.com/ClickHouse/ClickHouse/pull/37824) ([Michael Lex](https://github.com/mlex)). +* Fix possible "No more packets are available" for distributed queries (in case of `async_socket_for_remote`/`use_hedged_requests` is disabled). [#37826](https://github.com/ClickHouse/ClickHouse/pull/37826) ([Azat Khuzhin](https://github.com/azat)). +* (experimental WINDOW VIEW) Do not drop the inner target table when executing `ALTER TABLE … MODIFY QUERY` in WindowView. [#37879](https://github.com/ClickHouse/ClickHouse/pull/37879) ([vxider](https://github.com/Vxider)). +* Fix directory ownership of coordination dir in clickhouse-keeper Docker image. Fixes [#37914](https://github.com/ClickHouse/ClickHouse/issues/37914). [#37915](https://github.com/ClickHouse/ClickHouse/pull/37915) ([James Maidment](https://github.com/jamesmaidment)). +* Dictionaries fix custom query with update field and `{condition}`. Closes [#33746](https://github.com/ClickHouse/ClickHouse/issues/33746). [#37947](https://github.com/ClickHouse/ClickHouse/pull/37947) ([Maksim Kita](https://github.com/kitaisreal)). +* Fix possible incorrect result of `SELECT ... WITH FILL` in the case when `ORDER BY` should be applied after `WITH FILL` result (e.g. for outer query). Incorrect result was caused by optimization for `ORDER BY` expressions ([#35623](https://github.com/ClickHouse/ClickHouse/issues/35623)). Closes [#37904](https://github.com/ClickHouse/ClickHouse/issues/37904). [#37959](https://github.com/ClickHouse/ClickHouse/pull/37959) ([Yakov Olkhovskiy](https://github.com/yakov-olkhovskiy)). +* (experimental WINDOW VIEW) Add missing default columns when pushing to the target table in WindowView, fix [#37815](https://github.com/ClickHouse/ClickHouse/issues/37815). [#37965](https://github.com/ClickHouse/ClickHouse/pull/37965) ([vxider](https://github.com/Vxider)). +* Fixed too large stack frame that would cause compilation to fail. [#37996](https://github.com/ClickHouse/ClickHouse/pull/37996) ([Han Shukai](https://github.com/KinderRiven)). +* When open enable_filesystem_query_cache_limit, throw Reserved cache size exceeds the remaining cache size. [#38004](https://github.com/ClickHouse/ClickHouse/pull/38004) ([xiedeyantu](https://github.com/xiedeyantu)). +* Fix converting types for UNION queries (may produce LOGICAL_ERROR). [#34775](https://github.com/ClickHouse/ClickHouse/pull/34775) ([Azat Khuzhin](https://github.com/azat)). +* TTL merge may not be scheduled again if BackgroundExecutor is busy. --merges_with_ttl_counter is increased in selectPartsToMerge() --merge task will be ignored if BackgroundExecutor is busy --merges_with_ttl_counter will not be decrease. [#36387](https://github.com/ClickHouse/ClickHouse/pull/36387) ([lthaooo](https://github.com/lthaooo)). +* Fix overridden settings value of `normalize_function_names`. [#36937](https://github.com/ClickHouse/ClickHouse/pull/36937) ([李扬](https://github.com/taiyang-li)). +* Fix for exponential time decaying window functions. Now respecting boundaries of the window. [#36944](https://github.com/ClickHouse/ClickHouse/pull/36944) ([Vladimir Chebotarev](https://github.com/excitoon)). +* Fix possible heap-use-after-free error when reading system.projection_parts and system.projection_parts_columns . This fixes [#37184](https://github.com/ClickHouse/ClickHouse/issues/37184). [#37185](https://github.com/ClickHouse/ClickHouse/pull/37185) ([Amos Bird](https://github.com/amosbird)). +* Fixed `DateTime64` fractional seconds behavior prior to Unix epoch. [#37697](https://github.com/ClickHouse/ClickHouse/pull/37697) ([Andrey Zvonov](https://github.com/zvonand)). [#37039](https://github.com/ClickHouse/ClickHouse/pull/37039) ([李扬](https://github.com/taiyang-li)). + + ### ClickHouse release 22.5, 2022-05-19 #### Upgrade Notes diff --git a/CMakeLists.txt b/CMakeLists.txt index a6a09afc489..c8bb1a2d1ca 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -244,16 +244,18 @@ endif () # Add a section with the hash of the compiled machine code for integrity checks. # Only for official builds, because adding a section can be time consuming (rewrite of several GB). # And cross compiled binaries are not supported (since you cannot execute clickhouse hash-binary) -if (OBJCOPY_PATH AND CLICKHOUSE_OFFICIAL_BUILD AND (NOT CMAKE_TOOLCHAIN_FILE OR CMAKE_TOOLCHAIN_FILE MATCHES "linux/toolchain-x86_64.cmake$")) +if (CLICKHOUSE_OFFICIAL_BUILD AND (NOT CMAKE_TOOLCHAIN_FILE OR CMAKE_TOOLCHAIN_FILE MATCHES "linux/toolchain-x86_64.cmake$")) + message(STATUS "Official build: A checksum hash will be added to the clickhouse executable") set (USE_BINARY_HASH 1 CACHE STRING "Calculate binary hash and store it in the separate section") +else () + message(STATUS "No official build: A checksum hash will not be added to the clickhouse executable") endif () -# Allows to build stripped binary in a separate directory -if (OBJCOPY_PATH AND STRIP_PATH) - option(INSTALL_STRIPPED_BINARIES "Build stripped binaries with debug info in separate directory" OFF) - if (INSTALL_STRIPPED_BINARIES) - set(STRIPPED_BINARIES_OUTPUT "stripped" CACHE STRING "A separate directory for stripped information") - endif() +# Optionally split binaries and debug symbols. +option(INSTALL_STRIPPED_BINARIES "Split binaries and debug symbols" OFF) +if (INSTALL_STRIPPED_BINARIES) + message(STATUS "Will split binaries and debug symbols") + set(STRIPPED_BINARIES_OUTPUT "stripped" CACHE STRING "A separate directory for stripped information") endif() cmake_host_system_information(RESULT AVAILABLE_PHYSICAL_MEMORY QUERY AVAILABLE_PHYSICAL_MEMORY) # Not available under freebsd diff --git a/base/base/Decimal.h b/base/base/Decimal.h index aefe852b749..1efb8ba8d92 100644 --- a/base/base/Decimal.h +++ b/base/base/Decimal.h @@ -49,7 +49,7 @@ struct Decimal using NativeType = T; constexpr Decimal() = default; - constexpr Decimal(Decimal &&) = default; + constexpr Decimal(Decimal &&) noexcept = default; constexpr Decimal(const Decimal &) = default; constexpr Decimal(const T & value_): value(value_) {} @@ -57,7 +57,7 @@ struct Decimal template constexpr Decimal(const Decimal & x): value(x.value) {} - constexpr Decimal & operator = (Decimal &&) = default; + constexpr Decimal & operator=(Decimal &&) noexcept = default; constexpr Decimal & operator = (const Decimal &) = default; constexpr operator T () const { return value; } diff --git a/cmake/cpu_features.cmake b/cmake/cpu_features.cmake index fd4cc51b6f9..7b966e1acac 100644 --- a/cmake/cpu_features.cmake +++ b/cmake/cpu_features.cmake @@ -29,7 +29,7 @@ if (ARCH_NATIVE) set (COMPILER_FLAGS "${COMPILER_FLAGS} -march=native") elseif (ARCH_AARCH64) - set (COMPILER_FLAGS "${COMPILER_FLAGS} -march=armv8-a+crc") + set (COMPILER_FLAGS "${COMPILER_FLAGS} -march=armv8-a+crc+simd+crypto+dotprod+ssbs") elseif (ARCH_PPC64LE) # Note that gcc and clang have support for x86 SSE2 intrinsics when building for PowerPC diff --git a/cmake/strip_binary.cmake b/cmake/strip_binary.cmake index be23a4c1c30..6e38c86fc70 100644 --- a/cmake/strip_binary.cmake +++ b/cmake/strip_binary.cmake @@ -19,9 +19,12 @@ macro(clickhouse_strip_binary) COMMAND mkdir -p "${STRIP_DESTINATION_DIR}/lib/debug/bin" COMMAND mkdir -p "${STRIP_DESTINATION_DIR}/bin" COMMAND cp "${STRIP_BINARY_PATH}" "${STRIP_DESTINATION_DIR}/bin/${STRIP_TARGET}" + # Splits debug symbols into separate file, leaves the binary untouched: COMMAND "${OBJCOPY_PATH}" --only-keep-debug --compress-debug-sections "${STRIP_DESTINATION_DIR}/bin/${STRIP_TARGET}" "${STRIP_DESTINATION_DIR}/lib/debug/bin/${STRIP_TARGET}.debug" COMMAND chmod 0644 "${STRIP_DESTINATION_DIR}/lib/debug/bin/${STRIP_TARGET}.debug" - COMMAND "${STRIP_PATH}" --remove-section=.comment --remove-section=.note "${STRIP_DESTINATION_DIR}/bin/${STRIP_TARGET}" + # Strips binary, sections '.note' & '.comment' are removed in line with Debian's stripping policy: www.debian.org/doc/debian-policy/ch-files.html, section '.clickhouse.hash' is needed for integrity check: + COMMAND "${STRIP_PATH}" --remove-section=.comment --remove-section=.note --keep-section=.clickhouse.hash "${STRIP_DESTINATION_DIR}/bin/${STRIP_TARGET}" + # Associate stripped binary with debug symbols: COMMAND "${OBJCOPY_PATH}" --add-gnu-debuglink "${STRIP_DESTINATION_DIR}/lib/debug/bin/${STRIP_TARGET}.debug" "${STRIP_DESTINATION_DIR}/bin/${STRIP_TARGET}" COMMENT "Stripping clickhouse binary" VERBATIM ) diff --git a/contrib/grpc b/contrib/grpc index 7eac189a6ba..5e23e96c0c0 160000 --- a/contrib/grpc +++ b/contrib/grpc @@ -1 +1 @@ -Subproject commit 7eac189a6badddac593580ec2ad1478bd2656fc7 +Subproject commit 5e23e96c0c02e451dbb291cf9f66231d02b6cdb6 diff --git a/contrib/libunwind b/contrib/libunwind index c4ea9848a69..5022f30f3e0 160000 --- a/contrib/libunwind +++ b/contrib/libunwind @@ -1 +1 @@ -Subproject commit c4ea9848a697747dfa35325af9b3452f30841685 +Subproject commit 5022f30f3e092a54a7c101c335ce5e08769db366 diff --git a/contrib/simdjson b/contrib/simdjson index 8df32cea335..de196dd7a3a 160000 --- a/contrib/simdjson +++ b/contrib/simdjson @@ -1 +1 @@ -Subproject commit 8df32cea3359cb30120795da6020b3b73da01d38 +Subproject commit de196dd7a3a16e4056b0551ffa3b85c2f52581e1 diff --git a/contrib/zstd b/contrib/zstd index a488ba114ec..b944db0c451 160000 --- a/contrib/zstd +++ b/contrib/zstd @@ -1 +1 @@ -Subproject commit a488ba114ec17ea1054b9057c26a046fc122b3b6 +Subproject commit b944db0c451ba1bc6bbd8e201d5f88f9041bf1f9 diff --git a/contrib/zstd-cmake/CMakeLists.txt b/contrib/zstd-cmake/CMakeLists.txt index 4949c3f30d5..f44d5db12c4 100644 --- a/contrib/zstd-cmake/CMakeLists.txt +++ b/contrib/zstd-cmake/CMakeLists.txt @@ -50,7 +50,7 @@ GetLibraryVersion("${HEADER_CONTENT}" LIBVER_MAJOR LIBVER_MINOR LIBVER_RELEASE) MESSAGE(STATUS "ZSTD VERSION ${LIBVER_MAJOR}.${LIBVER_MINOR}.${LIBVER_RELEASE}") # cd contrib/zstd/lib -# find . -name '*.c' | grep -vP 'deprecated|legacy' | sort | sed 's/^\./ "${LIBRARY_DIR}/"' +# find . -name '*.c' -or -name '*.S' | grep -vP 'deprecated|legacy' | sort | sed 's/^\./ "${LIBRARY_DIR}/"' SET(Sources "${LIBRARY_DIR}/common/debug.c" "${LIBRARY_DIR}/common/entropy_common.c" @@ -73,6 +73,7 @@ SET(Sources "${LIBRARY_DIR}/compress/zstd_ldm.c" "${LIBRARY_DIR}/compress/zstdmt_compress.c" "${LIBRARY_DIR}/compress/zstd_opt.c" + "${LIBRARY_DIR}/decompress/huf_decompress_amd64.S" "${LIBRARY_DIR}/decompress/huf_decompress.c" "${LIBRARY_DIR}/decompress/zstd_ddict.c" "${LIBRARY_DIR}/decompress/zstd_decompress_block.c" @@ -85,6 +86,7 @@ SET(Sources # cd contrib/zstd/lib # find . -name '*.h' | grep -vP 'deprecated|legacy' | sort | sed 's/^\./ "${LIBRARY_DIR}/"' SET(Headers + "${LIBRARY_DIR}/common/bits.h" "${LIBRARY_DIR}/common/bitstream.h" "${LIBRARY_DIR}/common/compiler.h" "${LIBRARY_DIR}/common/cpu.h" @@ -94,11 +96,13 @@ SET(Headers "${LIBRARY_DIR}/common/huf.h" "${LIBRARY_DIR}/common/mem.h" "${LIBRARY_DIR}/common/pool.h" + "${LIBRARY_DIR}/common/portability_macros.h" "${LIBRARY_DIR}/common/threading.h" "${LIBRARY_DIR}/common/xxhash.h" "${LIBRARY_DIR}/common/zstd_deps.h" "${LIBRARY_DIR}/common/zstd_internal.h" "${LIBRARY_DIR}/common/zstd_trace.h" + "${LIBRARY_DIR}/compress/clevels.h" "${LIBRARY_DIR}/compress/hist.h" "${LIBRARY_DIR}/compress/zstd_compress_internal.h" "${LIBRARY_DIR}/compress/zstd_compress_literals.h" diff --git a/docker/test/util/process_functional_tests_result.py b/docker/test/util/process_functional_tests_result.py index 687c03697d7..fdd193e9c7f 100755 --- a/docker/test/util/process_functional_tests_result.py +++ b/docker/test/util/process_functional_tests_result.py @@ -11,6 +11,7 @@ TIMEOUT_SIGN = "[ Timeout! " UNKNOWN_SIGN = "[ UNKNOWN " SKIPPED_SIGN = "[ SKIPPED " HUNG_SIGN = "Found hung queries in processlist" +DATABASE_SIGN = "Database: " SUCCESS_FINISH_SIGNS = ["All tests have finished", "No tests were run"] @@ -27,14 +28,19 @@ def process_test_log(log_path): retries = False success_finish = False test_results = [] + test_end = True with open(log_path, "r") as test_file: for line in test_file: original_line = line line = line.strip() + if any(s in line for s in SUCCESS_FINISH_SIGNS): success_finish = True + # Ignore hung check report, since it may be quite large. + # (and may break python parser which has limit of 128KiB for each row). if HUNG_SIGN in line: hung = True + break if RETRIES_SIGN in line: retries = True if any( @@ -67,8 +73,17 @@ def process_test_log(log_path): else: success += int(OK_SIGN in line) test_results.append((test_name, "OK", test_time, [])) - elif len(test_results) > 0 and test_results[-1][1] == "FAIL": + test_end = False + elif ( + len(test_results) > 0 and test_results[-1][1] == "FAIL" and not test_end + ): test_results[-1][3].append(original_line) + # Database printed after everything else in case of failures, + # so this is a stop marker for capturing test output. + # + # And it is handled after everything else to include line with database into the report. + if DATABASE_SIGN in line: + test_end = True test_results = [ (test[0], test[1], test[2], "".join(test[3])) for test in test_results @@ -113,7 +128,7 @@ def process_result(result_path): test_results, ) = process_test_log(result_path) is_flacky_check = 1 < int(os.environ.get("NUM_TRIES", 1)) - logging.info("Is flacky check: %s", is_flacky_check) + logging.info("Is flaky check: %s", is_flacky_check) # If no tests were run (success == 0) it indicates an error (e.g. server did not start or crashed immediately) # But it's Ok for "flaky checks" - they can contain just one test for check which is marked as skipped. if failed != 0 or unknown != 0 or (success == 0 and (not is_flacky_check)): diff --git a/docs/en/operations/settings/merge-tree-settings.md b/docs/en/operations/settings/merge-tree-settings.md index cf6b3459fe8..e72314d84a5 100644 --- a/docs/en/operations/settings/merge-tree-settings.md +++ b/docs/en/operations/settings/merge-tree-settings.md @@ -2,6 +2,18 @@ The values of `merge_tree` settings (for all MergeTree tables) can be viewed in the table `system.merge_tree_settings`, they can be overridden in `config.xml` in the `merge_tree` section, or set in the `SETTINGS` section of each table. +These are example overrides for `max_suspicious_broken_parts`: + +## max_suspicious_broken_parts + +If the number of broken parts in a single partition exceeds the `max_suspicious_broken_parts` value, automatic deletion is denied. + +Possible values: + +- Any positive integer. + +Default value: 10. + Override example in `config.xml`: ``` text @@ -152,6 +164,187 @@ Default value: 604800 (1 week). Similar to [replicated_deduplication_window](#replicated-deduplication-window), `replicated_deduplication_window_seconds` specifies how long to store hash sums of blocks for insert deduplication. Hash sums older than `replicated_deduplication_window_seconds` are removed from ClickHouse Keeper, even if they are less than ` replicated_deduplication_window`. +## max_replicated_logs_to_keep + +How many records may be in the ClickHouse Keeper log if there is inactive replica. An inactive replica becomes lost when when this number exceed. + +Possible values: + +- Any positive integer. + +Default value: 1000 + +## min_replicated_logs_to_keep + +Keep about this number of last records in ZooKeeper log, even if they are obsolete. It doesn't affect work of tables: used only to diagnose ZooKeeper log before cleaning. + +Possible values: + +- Any positive integer. + +Default value: 10 + +## prefer_fetch_merged_part_time_threshold + +If the time passed since a replication log (ClickHouse Keeper or ZooKeeper) entry creation exceeds this threshold, and the sum of the size of parts is greater than `prefer_fetch_merged_part_size_threshold`, then prefer fetching merged part from a replica instead of doing merge locally. This is to speed up very long merges. + +Possible values: + +- Any positive integer. + +Default value: 3600 + +## prefer_fetch_merged_part_size_threshold + +If the sum of the size of parts exceeds this threshold and the time since a replication log entry creation is greater than `prefer_fetch_merged_part_time_threshold`, then prefer fetching merged part from a replica instead of doing merge locally. This is to speed up very long merges. + +Possible values: + +- Any positive integer. + +Default value: 10,737,418,240 + +## execute_merges_on_single_replica_time_threshold + +When this setting has a value greater than zero, only a single replica starts the merge immediately, and other replicas wait up to that amount of time to download the result instead of doing merges locally. If the chosen replica doesn't finish the merge during that amount of time, fallback to standard behavior happens. + +Possible values: + +- Any positive integer. + +Default value: 0 (seconds) + +## remote_fs_execute_merges_on_single_replica_time_threshold + +When this setting has a value greater than than zero only a single replica starts the merge immediately if merged part on shared storage and `allow_remote_fs_zero_copy_replication` is enabled. + +Possible values: + +- Any positive integer. + +Default value: 1800 + +## try_fetch_recompressed_part_timeout + +Recompression works slow in most cases, so we don't start merge with recompression until this timeout and trying to fetch recompressed part from replica which assigned this merge with recompression. + +Possible values: + +- Any positive integer. + +Default value: 7200 + +## always_fetch_merged_part + +If true, this replica never merges parts and always downloads merged parts from other replicas. + +Possible values: + +- true, false + +Default value: false + +## max_suspicious_broken_parts + +Max broken parts, if more - deny automatic deletion. + +Possible values: + +- Any positive integer. + +Default value: 10 + +## max_suspicious_broken_parts_bytes + + +Max size of all broken parts, if more - deny automatic deletion. + +Possible values: + +- Any positive integer. + +Default value: 1,073,741,824 + +## max_files_to_modify_in_alter_columns + +Do not apply ALTER if number of files for modification(deletion, addition) is greater than this setting. + +Possible values: + +- Any positive integer. + +Default value: 75 + +## max_files_to_remove_in_alter_columns + +Do not apply ALTER, if the number of files for deletion is greater than this setting. + +Possible values: + +- Any positive integer. + +Default value: 50 + +## replicated_max_ratio_of_wrong_parts + +If the ratio of wrong parts to total number of parts is less than this - allow to start. + +Possible values: + +- Float, 0.0 - 1.0 + +Default value: 0.5 + +## replicated_max_parallel_fetches_for_host + +Limit parallel fetches from endpoint (actually pool size). + +Possible values: + +- Any positive integer. + +Default value: 15 + +## replicated_fetches_http_connection_timeout + +HTTP connection timeout for part fetch requests. Inherited from default profile `http_connection_timeout` if not set explicitly. + +Possible values: + +- Any positive integer. + +Default value: Inherited from default profile `http_connection_timeout` if not set explicitly. + +## replicated_can_become_leader + +If true, replicated tables replicas on this node will try to acquire leadership. + +Possible values: + +- true, false + +Default value: true + +## zookeeper_session_expiration_check_period + +ZooKeeper session expiration check period, in seconds. + +Possible values: + +- Any positive integer. + +Default value: 60 + +## detach_old_local_parts_when_cloning_replica + +Do not remove old local parts when repairing lost replica. + +Possible values: + +- true, false + +Default value: true + ## replicated_fetches_http_connection_timeout {#replicated_fetches_http_connection_timeout} HTTP connection timeout (in seconds) for part fetch requests. Inherited from default profile [http_connection_timeout](./settings.md#http_connection_timeout) if not set explicitly. diff --git a/docs/en/sql-reference/statements/create/table.md b/docs/en/sql-reference/statements/create/table.md index d8bd741c2cf..bdf6c02c737 100644 --- a/docs/en/sql-reference/statements/create/table.md +++ b/docs/en/sql-reference/statements/create/table.md @@ -248,6 +248,7 @@ Specialized codecs: - `Delta(delta_bytes)` — Compression approach in which raw values are replaced by the difference of two neighboring values, except for the first value that stays unchanged. Up to `delta_bytes` are used for storing delta values, so `delta_bytes` is the maximum size of raw values. Possible `delta_bytes` values: 1, 2, 4, 8. The default value for `delta_bytes` is `sizeof(type)` if equal to 1, 2, 4, or 8. In all other cases, it’s 1. - `DoubleDelta` — Calculates delta of deltas and writes it in compact binary form. Optimal compression rates are achieved for monotonic sequences with a constant stride, such as time series data. Can be used with any fixed-width type. Implements the algorithm used in Gorilla TSDB, extending it to support 64-bit types. Uses 1 extra bit for 32-byte deltas: 5-bit prefixes instead of 4-bit prefixes. For additional information, see Compressing Time Stamps in [Gorilla: A Fast, Scalable, In-Memory Time Series Database](http://www.vldb.org/pvldb/vol8/p1816-teller.pdf). - `Gorilla` — Calculates XOR between current and previous value and writes it in compact binary form. Efficient when storing a series of floating point values that change slowly, because the best compression rate is achieved when neighboring values are binary equal. Implements the algorithm used in Gorilla TSDB, extending it to support 64-bit types. For additional information, see Compressing Values in [Gorilla: A Fast, Scalable, In-Memory Time Series Database](http://www.vldb.org/pvldb/vol8/p1816-teller.pdf). +- `FPC` - Repeatedly predicts the next floating point value in the sequence using the better of two predictors, then XORs the actual with the predicted value, and leading-zero compresses the result. Similar to Gorilla, this is efficient when storing a series of floating point values that change slowly. For 64-bit values (double), FPC is faster than Gorilla, for 32-bit values your mileage may vary. For a detailed description of the algorithm see [High Throughput Compression of Double-Precision Floating-Point Data](https://userweb.cs.txstate.edu/~burtscher/papers/dcc07a.pdf). - `T64` — Compression approach that crops unused high bits of values in integer data types (including `Enum`, `Date` and `DateTime`). At each step of its algorithm, codec takes a block of 64 values, puts them into 64x64 bit matrix, transposes it, crops the unused bits of values and returns the rest as a sequence. Unused bits are the bits, that do not differ between maximum and minimum values in the whole data part for which the compression is used. `DoubleDelta` and `Gorilla` codecs are used in Gorilla TSDB as the components of its compressing algorithm. Gorilla approach is effective in scenarios when there is a sequence of slowly changing values with their timestamps. Timestamps are effectively compressed by the `DoubleDelta` codec, and values are effectively compressed by the `Gorilla` codec. For example, to get an effectively stored table, you can create it in the following configuration: diff --git a/programs/CMakeLists.txt b/programs/CMakeLists.txt index afde5bd9a47..a2c6eb1a27e 100644 --- a/programs/CMakeLists.txt +++ b/programs/CMakeLists.txt @@ -508,7 +508,7 @@ else () endif() if (USE_BINARY_HASH) - add_custom_command(TARGET clickhouse POST_BUILD COMMAND ./clickhouse hash-binary > hash && ${OBJCOPY_PATH} --add-section .note.ClickHouse.hash=hash clickhouse COMMENT "Adding .note.ClickHouse.hash to clickhouse" VERBATIM) + add_custom_command(TARGET clickhouse POST_BUILD COMMAND ./clickhouse hash-binary > hash && ${OBJCOPY_PATH} --add-section .clickhouse.hash=hash clickhouse COMMENT "Adding section '.clickhouse.hash' to clickhouse binary" VERBATIM) endif() if (INSTALL_STRIPPED_BINARIES) diff --git a/programs/main.cpp b/programs/main.cpp index 4148cbfc4cd..175504a85fa 100644 --- a/programs/main.cpp +++ b/programs/main.cpp @@ -82,7 +82,7 @@ int mainEntryClickHouseDisks(int argc, char ** argv); int mainEntryClickHouseHashBinary(int, char **) { /// Intentionally without newline. So you can run: - /// objcopy --add-section .note.ClickHouse.hash=<(./clickhouse hash-binary) clickhouse + /// objcopy --add-section .clickhouse.hash=<(./clickhouse hash-binary) clickhouse std::cout << getHashOfLoadedBinaryHex(); return 0; } diff --git a/programs/server/Server.cpp b/programs/server/Server.cpp index bc5a959c88b..b013ba9ee05 100644 --- a/programs/server/Server.cpp +++ b/programs/server/Server.cpp @@ -744,16 +744,18 @@ int Server::main(const std::vector & /*args*/) /// But there are other sections of the binary (e.g. exception handling tables) /// that are interpreted (not executed) but can alter the behaviour of the program as well. + /// Please keep the below log messages in-sync with the ones in daemon/BaseDaemon.cpp + String calculated_binary_hash = getHashOfLoadedBinaryHex(); if (stored_binary_hash.empty()) { - LOG_WARNING(log, "Calculated checksum of the binary: {}." - " There is no information about the reference checksum.", calculated_binary_hash); + LOG_WARNING(log, "Integrity check of the executable skipped because the reference checksum could not be read." + " (calculated checksum: {})", calculated_binary_hash); } else if (calculated_binary_hash == stored_binary_hash) { - LOG_INFO(log, "Calculated checksum of the binary: {}, integrity check passed.", calculated_binary_hash); + LOG_INFO(log, "Integrity check of the executable successfully passed (checksum: {})", calculated_binary_hash); } else { @@ -769,14 +771,14 @@ int Server::main(const std::vector & /*args*/) else { throw Exception(ErrorCodes::CORRUPTED_DATA, - "Calculated checksum of the ClickHouse binary ({0}) does not correspond" - " to the reference checksum stored in the binary ({1})." - " It may indicate one of the following:" - " - the file {2} was changed just after startup;" - " - the file {2} is damaged on disk due to faulty hardware;" - " - the loaded executable is damaged in memory due to faulty hardware;" + "Calculated checksum of the executable ({0}) does not correspond" + " to the reference checksum stored in the executable ({1})." + " This may indicate one of the following:" + " - the executable {2} was changed just after startup;" + " - the executable {2} was corrupted on disk due to faulty hardware;" + " - the loaded executable was corrupted in memory due to faulty hardware;" " - the file {2} was intentionally modified;" - " - logical error in code." + " - a logical error in the code." , calculated_binary_hash, stored_binary_hash, executable_path); } } diff --git a/src/Columns/ColumnVector.cpp b/src/Columns/ColumnVector.cpp index 391baa188d6..f9fc14874aa 100644 --- a/src/Columns/ColumnVector.cpp +++ b/src/Columns/ColumnVector.cpp @@ -337,7 +337,7 @@ void ColumnVector::updatePermutation(IColumn::PermutationSortDirection direct PaddedPODArray> pairs(size); size_t index = 0; - for (auto it = begin; it != end; ++it) + for (auto * it = begin; it != end; ++it) { pairs[index] = {data[*it], static_cast(*it)}; ++index; diff --git a/src/Columns/ColumnsCommon.h b/src/Columns/ColumnsCommon.h index 99489e7e4cc..1e5849e2b88 100644 --- a/src/Columns/ColumnsCommon.h +++ b/src/Columns/ColumnsCommon.h @@ -8,6 +8,9 @@ #if defined(__AVX512F__) || defined(__AVX512BW__) || defined(__AVX__) || defined(__AVX2__) #include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +#endif /// Common helper methods for implementation of different columns. @@ -44,6 +47,22 @@ inline UInt64 bytes64MaskToBits64Mask(const UInt8 * bytes64) _mm_loadu_si128(reinterpret_cast(bytes64 + 32)), zero16))) << 32) & 0xffff00000000) | ((static_cast(_mm_movemask_epi8(_mm_cmpeq_epi8( _mm_loadu_si128(reinterpret_cast(bytes64 + 48)), zero16))) << 48) & 0xffff000000000000); +#elif defined(__aarch64__) && defined(__ARM_NEON) + const uint8x16_t bitmask = {0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80, 0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80}; + const auto * src = reinterpret_cast(bytes64); + const uint8x16_t p0 = vceqzq_u8(vld1q_u8(src)); + const uint8x16_t p1 = vceqzq_u8(vld1q_u8(src + 16)); + const uint8x16_t p2 = vceqzq_u8(vld1q_u8(src + 32)); + const uint8x16_t p3 = vceqzq_u8(vld1q_u8(src + 48)); + uint8x16_t t0 = vandq_u8(p0, bitmask); + uint8x16_t t1 = vandq_u8(p1, bitmask); + uint8x16_t t2 = vandq_u8(p2, bitmask); + uint8x16_t t3 = vandq_u8(p3, bitmask); + uint8x16_t sum0 = vpaddq_u8(t0, t1); + uint8x16_t sum1 = vpaddq_u8(t2, t3); + sum0 = vpaddq_u8(sum0, sum1); + sum0 = vpaddq_u8(sum0, sum0); + UInt64 res = vgetq_lane_u64(vreinterpretq_u64_u8(sum0), 0); #else UInt64 res = 0; for (size_t i = 0; i < 64; ++i) diff --git a/src/Common/Elf.cpp b/src/Common/Elf.cpp index 27ff3bad310..b735367b179 100644 --- a/src/Common/Elf.cpp +++ b/src/Common/Elf.cpp @@ -176,9 +176,9 @@ String Elf::getBuildID(const char * nhdr_pos, size_t size) #endif // OS_SUNOS -String Elf::getBinaryHash() const +String Elf::getStoredBinaryHash() const { - if (auto section = findSectionByName(".note.ClickHouse.hash")) + if (auto section = findSectionByName(".clickhouse.hash")) return {section->begin(), section->end()}; else return {}; diff --git a/src/Common/Elf.h b/src/Common/Elf.h index f23458cfc2e..5a6bd9e302d 100644 --- a/src/Common/Elf.h +++ b/src/Common/Elf.h @@ -61,7 +61,7 @@ public: static String getBuildID(const char * nhdr_pos, size_t size); /// Hash of the binary for integrity checks. - String getBinaryHash() const; + String getStoredBinaryHash() const; private: MMapReadBufferFromFile in; diff --git a/src/Common/HashTable/Hash.h b/src/Common/HashTable/Hash.h index 3cf8978f418..bf691458d27 100644 --- a/src/Common/HashTable/Hash.h +++ b/src/Common/HashTable/Hash.h @@ -296,6 +296,19 @@ struct UInt128HashCRC32 } }; +#elif defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) + +struct UInt128HashCRC32 +{ + size_t operator()(UInt128 x) const + { + UInt64 crc = -1ULL; + crc = __crc32cd(crc, x.items[0]); + crc = __crc32cd(crc, x.items[1]); + return crc; + } +}; + #else /// On other platforms we do not use CRC32. NOTE This can be confusing. @@ -339,6 +352,21 @@ struct UInt256HashCRC32 } }; +#elif defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) + +struct UInt256HashCRC32 +{ + size_t operator()(UInt256 x) const + { + UInt64 crc = -1ULL; + crc = __crc32cd(crc, x.items[0]); + crc = __crc32cd(crc, x.items[1]); + crc = __crc32cd(crc, x.items[2]); + crc = __crc32cd(crc, x.items[3]); + return crc; + } +}; + #else /// We do not need to use CRC32 on other platforms. NOTE This can be confusing. diff --git a/src/Common/UTF8Helpers.h b/src/Common/UTF8Helpers.h index a940ddc0234..72bdb965789 100644 --- a/src/Common/UTF8Helpers.h +++ b/src/Common/UTF8Helpers.h @@ -9,6 +9,13 @@ #include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif + namespace DB { @@ -66,6 +73,18 @@ inline size_t countCodePoints(const UInt8 * data, size_t size) for (; data < src_end_sse; data += bytes_sse) res += __builtin_popcount(_mm_movemask_epi8( _mm_cmpgt_epi8(_mm_loadu_si128(reinterpret_cast(data)), threshold))); +#elif defined(__aarch64__) && defined(__ARM_NEON) + /// Returns a 64 bit mask of nibbles (4 bits for each byte). + auto get_nibble_mask + = [](uint8x16_t input) -> uint64_t { return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(input), 4)), 0); }; + constexpr auto bytes_sse = 16; + const auto * src_end_sse = data + size / bytes_sse * bytes_sse; + + const auto threshold = vdupq_n_s8(0xBF); + + for (; data < src_end_sse; data += bytes_sse) + res += __builtin_popcountll(get_nibble_mask(vcgtq_s8(vld1q_s8(reinterpret_cast(data)), threshold))); + res >>= 2; #endif for (; data < end; ++data) /// Skip UTF-8 continuation bytes. diff --git a/src/Common/memcmpSmall.h b/src/Common/memcmpSmall.h index 57b9c731897..7b977a4a23c 100644 --- a/src/Common/memcmpSmall.h +++ b/src/Common/memcmpSmall.h @@ -1,7 +1,7 @@ #pragma once -#include #include +#include #include @@ -26,7 +26,7 @@ inline int cmp(T a, T b) /// Results don't depend on the values inside uninitialized memory but Memory Sanitizer cannot see it. /// Disable optimized functions if compile with Memory Sanitizer. #if defined(__AVX512BW__) && defined(__AVX512VL__) && !defined(MEMORY_SANITIZER) -#include +# include /** All functions works under the following assumptions: @@ -45,7 +45,8 @@ inline int memcmpSmallAllowOverflow15(const Char * a, size_t a_size, const Char { uint16_t mask = _mm_cmp_epi8_mask( _mm_loadu_si128(reinterpret_cast(a + offset)), - _mm_loadu_si128(reinterpret_cast(b + offset)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(b + offset)), + _MM_CMPINT_NE); if (mask) { @@ -76,7 +77,8 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz { uint16_t mask = _mm_cmp_epi8_mask( _mm_loadu_si128(reinterpret_cast(a + offset)), - _mm_loadu_si128(reinterpret_cast(b + offset)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(b + offset)), + _MM_CMPINT_NE); if (mask) { @@ -117,9 +119,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz for (size_t offset = min_size; offset < max_size; offset += 16) { - uint16_t mask = _mm_cmpneq_epi8_mask( - _mm_loadu_si128(reinterpret_cast(longest + offset)), - zero16); + uint16_t mask = _mm_cmpneq_epi8_mask(_mm_loadu_si128(reinterpret_cast(longest + offset)), zero16); if (mask) { @@ -145,7 +145,8 @@ inline int memcmpSmallAllowOverflow15(const Char * a, const Char * b, size_t siz { uint16_t mask = _mm_cmp_epi8_mask( _mm_loadu_si128(reinterpret_cast(a + offset)), - _mm_loadu_si128(reinterpret_cast(b + offset)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(b + offset)), + _MM_CMPINT_NE); if (mask) { @@ -174,7 +175,8 @@ inline bool memequalSmallAllowOverflow15(const Char * a, size_t a_size, const Ch { uint16_t mask = _mm_cmp_epi8_mask( _mm_loadu_si128(reinterpret_cast(a + offset)), - _mm_loadu_si128(reinterpret_cast(b + offset)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(b + offset)), + _MM_CMPINT_NE); if (mask) { @@ -196,7 +198,8 @@ inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size) { uint16_t mask = _mm_cmp_epi8_mask( _mm_loadu_si128(reinterpret_cast(a + offset)), - _mm_loadu_si128(reinterpret_cast(b + offset)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(b + offset)), + _MM_CMPINT_NE); if (mask) { @@ -215,8 +218,7 @@ template inline int memcmp16(const Char * a, const Char * b) { uint16_t mask = _mm_cmp_epi8_mask( - _mm_loadu_si128(reinterpret_cast(a)), - _mm_loadu_si128(reinterpret_cast(b)), _MM_CMPINT_NE); + _mm_loadu_si128(reinterpret_cast(a)), _mm_loadu_si128(reinterpret_cast(b)), _MM_CMPINT_NE); if (mask) { @@ -232,9 +234,9 @@ inline int memcmp16(const Char * a, const Char * b) */ inline bool memequal16(const void * a, const void * b) { - return 0xFFFF == _mm_cmp_epi8_mask( - _mm_loadu_si128(reinterpret_cast(a)), - _mm_loadu_si128(reinterpret_cast(b)), _MM_CMPINT_EQ); + return 0xFFFF + == _mm_cmp_epi8_mask( + _mm_loadu_si128(reinterpret_cast(a)), _mm_loadu_si128(reinterpret_cast(b)), _MM_CMPINT_EQ); } @@ -245,8 +247,8 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size) for (size_t offset = 0; offset < size; offset += 16) { - uint16_t mask = _mm_cmp_epi8_mask(zero16, - _mm_loadu_si128(reinterpret_cast(reinterpret_cast(data) + offset)), _MM_CMPINT_NE); + uint16_t mask = _mm_cmp_epi8_mask( + zero16, _mm_loadu_si128(reinterpret_cast(reinterpret_cast(data) + offset)), _MM_CMPINT_NE); if (mask) { @@ -259,7 +261,7 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size) } #elif defined(__SSE2__) && !defined(MEMORY_SANITIZER) -#include +# include /** All functions works under the following assumptions: @@ -352,9 +354,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz for (size_t offset = min_size; offset < max_size; offset += 16) { - uint16_t mask = _mm_movemask_epi8(_mm_cmpeq_epi8( - _mm_loadu_si128(reinterpret_cast(longest + offset)), - zero16)); + uint16_t mask = _mm_movemask_epi8(_mm_cmpeq_epi8(_mm_loadu_si128(reinterpret_cast(longest + offset)), zero16)); mask = ~mask; if (mask) @@ -453,9 +453,8 @@ inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size) template inline int memcmp16(const Char * a, const Char * b) { - uint16_t mask = _mm_movemask_epi8(_mm_cmpeq_epi8( - _mm_loadu_si128(reinterpret_cast(a)), - _mm_loadu_si128(reinterpret_cast(b)))); + uint16_t mask = _mm_movemask_epi8( + _mm_cmpeq_epi8(_mm_loadu_si128(reinterpret_cast(a)), _mm_loadu_si128(reinterpret_cast(b)))); mask = ~mask; if (mask) @@ -472,9 +471,9 @@ inline int memcmp16(const Char * a, const Char * b) */ inline bool memequal16(const void * a, const void * b) { - return 0xFFFF == _mm_movemask_epi8(_mm_cmpeq_epi8( - _mm_loadu_si128(reinterpret_cast(a)), - _mm_loadu_si128(reinterpret_cast(b)))); + return 0xFFFF + == _mm_movemask_epi8(_mm_cmpeq_epi8( + _mm_loadu_si128(reinterpret_cast(a)), _mm_loadu_si128(reinterpret_cast(b)))); } @@ -485,8 +484,8 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size) for (size_t offset = 0; offset < size; offset += 16) { - uint16_t mask = _mm_movemask_epi8(_mm_cmpeq_epi8(zero16, - _mm_loadu_si128(reinterpret_cast(reinterpret_cast(data) + offset)))); + uint16_t mask = _mm_movemask_epi8( + _mm_cmpeq_epi8(zero16, _mm_loadu_si128(reinterpret_cast(reinterpret_cast(data) + offset)))); mask = ~mask; if (mask) @@ -499,6 +498,209 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size) return true; } +#elif defined(__aarch64__) && defined(__ARM_NEON) + +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif + +inline uint64_t getNibbleMask(uint8x16_t res) +{ + return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(res), 4)), 0); +} + +template +inline int memcmpSmallAllowOverflow15(const Char * a, size_t a_size, const Char * b, size_t b_size) +{ + size_t min_size = std::min(a_size, b_size); + + for (size_t offset = 0; offset < min_size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqq_u8( + vld1q_u8(reinterpret_cast(a + offset)), vld1q_u8(reinterpret_cast(b + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + + if (offset >= min_size) + break; + + return detail::cmp(a[offset], b[offset]); + } + } + + return detail::cmp(a_size, b_size); +} + +template +inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_size, const Char * b, size_t b_size) +{ + size_t min_size = std::min(a_size, b_size); + + for (size_t offset = 0; offset < min_size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqq_u8( + vld1q_u8(reinterpret_cast(a + offset)), vld1q_u8(reinterpret_cast(b + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + + if (offset >= min_size) + break; + + return detail::cmp(a[offset], b[offset]); + } + } + + /// The strings are equal up to min_size. + /// If the rest of the larger string is zero bytes then the strings are + /// considered equal. + + size_t max_size; + const Char * longest; + int cmp; + + if (a_size == b_size) + { + return 0; + } + else if (a_size > b_size) + { + max_size = a_size; + longest = a; + cmp = 1; + } + else + { + max_size = b_size; + longest = b; + cmp = -1; + } + + for (size_t offset = min_size; offset < max_size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqzq_u8(vld1q_u8(reinterpret_cast(longest + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + + if (offset >= max_size) + return 0; + return cmp; + } + } + + return 0; +} + +template +inline int memcmpSmallAllowOverflow15(const Char * a, const Char * b, size_t size) +{ + for (size_t offset = 0; offset < size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqq_u8( + vld1q_u8(reinterpret_cast(a + offset)), vld1q_u8(reinterpret_cast(b + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + + if (offset >= size) + return 0; + + return detail::cmp(a[offset], b[offset]); + } + } + + return 0; +} + +template +inline bool memequalSmallAllowOverflow15(const Char * a, size_t a_size, const Char * b, size_t b_size) +{ + if (a_size != b_size) + return false; + + for (size_t offset = 0; offset < a_size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqq_u8( + vld1q_u8(reinterpret_cast(a + offset)), vld1q_u8(reinterpret_cast(b + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + return offset >= a_size; + } + } + + return true; +} + +template +inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size) +{ + for (size_t offset = 0; offset < size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqq_u8( + vld1q_u8(reinterpret_cast(a + offset)), vld1q_u8(reinterpret_cast(b + offset)))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + return detail::cmp(a[offset], b[offset]); + } + } + + return 0; +} + +template +inline int memcmp16(const Char * a, const Char * b) +{ + uint64_t mask = getNibbleMask( + vceqq_u8(vld1q_u8(reinterpret_cast(a)), vld1q_u8(reinterpret_cast(b)))); + mask = ~mask; + if (mask) + { + auto offset = __builtin_ctzll(mask) >> 2; + return detail::cmp(a[offset], b[offset]); + } + return 0; +} + +inline bool memequal16(const void * a, const void * b) +{ + return 0xFFFFFFFFFFFFFFFFull + == getNibbleMask( + vceqq_u8(vld1q_u8(reinterpret_cast(a)), vld1q_u8(reinterpret_cast(b)))); +} + +inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size) +{ + for (size_t offset = 0; offset < size; offset += 16) + { + uint64_t mask = getNibbleMask(vceqzq_u8(vld1q_u8(reinterpret_cast(data) + offset))); + mask = ~mask; + + if (mask) + { + offset += __builtin_ctzll(mask) >> 2; + return offset >= size; + } + } + + return true; +} #else diff --git a/src/Common/memcpySmall.h b/src/Common/memcpySmall.h index aaedfb81fe5..6cadc19262f 100644 --- a/src/Common/memcpySmall.h +++ b/src/Common/memcpySmall.h @@ -3,8 +3,15 @@ #include #ifdef __SSE2__ -#include +# include +#endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif /** memcpy function could work suboptimal if all the following conditions are met: * 1. Size of memory region is relatively small (approximately, under 50 bytes). @@ -27,6 +34,7 @@ * Use with caution. */ +#ifdef __SSE2__ /// Implementation for x86 platform namespace detail { inline void memcpySmallAllowReadWriteOverflow15Impl(char * __restrict dst, const char * __restrict src, ssize_t n) @@ -51,11 +59,33 @@ inline void memcpySmallAllowReadWriteOverflow15(void * __restrict dst, const voi detail::memcpySmallAllowReadWriteOverflow15Impl(reinterpret_cast(dst), reinterpret_cast(src), n); } +#elif defined(__aarch64__) && defined(__ARM_NEON) /// Implementation for arm platform, similar to x86 + +namespace detail +{ +inline void memcpySmallAllowReadWriteOverflow15Impl(char * __restrict dst, const char * __restrict src, ssize_t n) +{ + while (n > 0) + { + vst1q_s8(reinterpret_cast(dst), vld1q_s8(reinterpret_cast(src))); + + dst += 16; + src += 16; + n -= 16; + } +} +} + +inline void memcpySmallAllowReadWriteOverflow15(void * __restrict dst, const void * __restrict src, size_t n) +{ + detail::memcpySmallAllowReadWriteOverflow15Impl(reinterpret_cast(dst), reinterpret_cast(src), n); +} + /** NOTE There was also a function, that assumes, that you could read any bytes inside same memory page of src. * This function was unused, and also it requires special handling for Valgrind and ASan. */ -#else /// Implementation for other platforms. +#else /// Implementation for other platforms. inline void memcpySmallAllowReadWriteOverflow15(void * __restrict dst, const void * __restrict src, size_t n) { diff --git a/src/Compression/CompressionCodecFPC.cpp b/src/Compression/CompressionCodecFPC.cpp new file mode 100644 index 00000000000..247eb73b65b --- /dev/null +++ b/src/Compression/CompressionCodecFPC.cpp @@ -0,0 +1,491 @@ +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include + + +namespace DB +{ + +/// An implementation of the FPC codec for floating-point values described in the paper +/// M. Burtscher, P. Ratanaworabhan: "FPC: A high-speed compressor for double-precision floating-point data" (2008). +/// Note: The paper only describes compression of 64-bit doubles and leaves 32-bit floats to future work. The code +/// implements them anyways. Your mileage with respect to performance and compression may vary. +class CompressionCodecFPC : public ICompressionCodec +{ +public: + CompressionCodecFPC(UInt8 float_size, UInt8 compression_level); + + uint8_t getMethodByte() const override; + + void updateHash(SipHash & hash) const override; + + static constexpr UInt8 MAX_COMPRESSION_LEVEL = 28; + static constexpr UInt8 DEFAULT_COMPRESSION_LEVEL = 12; + +protected: + UInt32 doCompressData(const char * source, UInt32 source_size, char * dest) const override; + + void doDecompressData(const char * source, UInt32 source_size, char * dest, UInt32 uncompressed_size) const override; + + UInt32 getMaxCompressedDataSize(UInt32 uncompressed_size) const override; + + bool isCompression() const override { return true; } + bool isGenericCompression() const override { return false; } + +private: + static constexpr UInt32 HEADER_SIZE = 2; + + // below members are used by compression, decompression ignores them: + const UInt8 float_width; // size of uncompressed float in bytes + const UInt8 level; // compression level, 2^level * float_width is the size of predictors table in bytes +}; + + +namespace ErrorCodes +{ + extern const int CANNOT_COMPRESS; + extern const int CANNOT_DECOMPRESS; + extern const int ILLEGAL_CODEC_PARAMETER; + extern const int ILLEGAL_SYNTAX_FOR_CODEC_TYPE; + extern const int BAD_ARGUMENTS; +} + +uint8_t CompressionCodecFPC::getMethodByte() const +{ + return static_cast(CompressionMethodByte::FPC); +} + +void CompressionCodecFPC::updateHash(SipHash & hash) const +{ + getCodecDesc()->updateTreeHash(hash); +} + +CompressionCodecFPC::CompressionCodecFPC(UInt8 float_size, UInt8 compression_level) + : float_width{float_size}, level{compression_level} +{ + setCodecDescription("FPC", {std::make_shared(static_cast(level))}); +} + +UInt32 CompressionCodecFPC::getMaxCompressedDataSize(UInt32 uncompressed_size) const +{ + auto float_count = (uncompressed_size + float_width - 1) / float_width; + if (float_count % 2 != 0) + ++float_count; + return HEADER_SIZE + float_count * float_width + float_count / 2; +} + +namespace +{ + +UInt8 getFloatBytesSize(const IDataType & column_type) +{ + if (!WhichDataType(column_type).isFloat()) + { + throw Exception(ErrorCodes::BAD_ARGUMENTS, "FPC codec is not applicable for {} because the data type is not float", + column_type.getName()); + } + + if (auto float_size = column_type.getSizeOfValueInMemory(); float_size >= 4) + { + return static_cast(float_size); + } + throw Exception(ErrorCodes::BAD_ARGUMENTS, "FPC codec is not applicable for floats of size less than 4 bytes. Given type {}", + column_type.getName()); +} + +} + +void registerCodecFPC(CompressionCodecFactory & factory) +{ + auto method_code = static_cast(CompressionMethodByte::FPC); + auto codec_builder = [&](const ASTPtr & arguments, const IDataType * column_type) -> CompressionCodecPtr + { + UInt8 float_width = 0; + if (column_type != nullptr) + float_width = getFloatBytesSize(*column_type); + + UInt8 level = CompressionCodecFPC::DEFAULT_COMPRESSION_LEVEL; + if (arguments && !arguments->children.empty()) + { + if (arguments->children.size() > 1) + { + throw Exception(ErrorCodes::ILLEGAL_SYNTAX_FOR_CODEC_TYPE, + "FPC codec must have 1 parameter, given {}", arguments->children.size()); + } + + const auto * literal = arguments->children.front()->as(); + if (!literal) + throw Exception("FPC codec argument must be integer", ErrorCodes::ILLEGAL_CODEC_PARAMETER); + + level = literal->value.safeGet(); + if (level < 1 || level > CompressionCodecFPC::MAX_COMPRESSION_LEVEL) + throw Exception(ErrorCodes::ILLEGAL_CODEC_PARAMETER, "FPC codec level must be between {} and {}", 1, static_cast(CompressionCodecFPC::MAX_COMPRESSION_LEVEL)); + } + return std::make_shared(float_width, level); + }; + factory.registerCompressionCodecWithType("FPC", method_code, codec_builder); +} + +namespace +{ + +template +requires (sizeof(TUint) >= 4) +class DfcmPredictor +{ +public: + explicit DfcmPredictor(std::size_t table_size) + : table(table_size, 0), prev_value{0}, hash{0} + { + } + + [[nodiscard]] + TUint predict() const noexcept + { + return table[hash] + prev_value; + } + + void add(TUint value) noexcept + { + table[hash] = value - prev_value; + recalculateHash(); + prev_value = value; + } + +private: + void recalculateHash() noexcept + { + auto value = table[hash]; + if constexpr (sizeof(TUint) >= 8) + { + hash = ((hash << 2) ^ static_cast(value >> 40)) & (table.size() - 1); + } + else + { + hash = ((hash << 4) ^ static_cast(value >> 23)) & (table.size() - 1); + } + } + + std::vector table; + TUint prev_value; + std::size_t hash; +}; + +template +requires (sizeof(TUint) >= 4) +class FcmPredictor +{ +public: + explicit FcmPredictor(std::size_t table_size) + : table(table_size, 0), hash{0} + { + } + + [[nodiscard]] + TUint predict() const noexcept + { + return table[hash]; + } + + void add(TUint value) noexcept + { + table[hash] = value; + recalculateHash(); + } + +private: + void recalculateHash() noexcept + { + auto value = table[hash]; + if constexpr (sizeof(TUint) >= 8) + { + hash = ((hash << 6) ^ static_cast(value >> 48)) & (table.size() - 1); + } + else + { + hash = ((hash << 1) ^ static_cast(value >> 22)) & (table.size() - 1); + } + } + + std::vector table; + std::size_t hash; +}; + +template +class FPCOperation +{ + static constexpr auto VALUE_SIZE = sizeof(TUint); + static constexpr std::byte FCM_BIT{0}; + static constexpr std::byte DFCM_BIT{1u << 3}; + static constexpr auto DFCM_BIT_1 = DFCM_BIT << 4; + static constexpr auto DFCM_BIT_2 = DFCM_BIT; + static constexpr unsigned MAX_ZERO_BYTE_COUNT = 0b111u; + static constexpr std::endian ENDIAN = std::endian::little; + static constexpr std::size_t CHUNK_SIZE = 64; + +public: + FPCOperation(std::span destination, UInt8 compression_level) + : dfcm_predictor(1u << compression_level), fcm_predictor(1u << compression_level), chunk{}, result{destination} + { + } + + std::size_t encode(std::span data) && + { + auto initial_size = result.size(); + + std::span chunk_view(chunk); + for (std::size_t i = 0; i < data.size(); i += chunk_view.size_bytes()) + { + auto written_values_count = importChunk(data.subspan(i), chunk_view); + encodeChunk(chunk_view.subspan(0, written_values_count)); + } + + return initial_size - result.size(); + } + + void decode(std::span values, std::size_t decoded_size) && + { + std::size_t read_bytes = 0; + + std::span chunk_view(chunk); + for (std::size_t i = 0; i < decoded_size; i += chunk_view.size_bytes()) + { + if (i + chunk_view.size_bytes() > decoded_size) + chunk_view = chunk_view.first(ceilBytesToEvenValues(decoded_size - i)); + read_bytes += decodeChunk(values.subspan(read_bytes), chunk_view); + exportChunk(chunk_view); + } + } + +private: + static std::size_t ceilBytesToEvenValues(std::size_t bytes_count) + { + auto values_count = (bytes_count + VALUE_SIZE - 1) / VALUE_SIZE; + return values_count % 2 == 0 ? values_count : values_count + 1; + } + + std::size_t importChunk(std::span values, std::span chnk) + { + if (auto chunk_view = std::as_writable_bytes(chnk); chunk_view.size() <= values.size()) + { + std::memcpy(chunk_view.data(), values.data(), chunk_view.size()); + return chunk_view.size() / VALUE_SIZE; + } + else + { + std::memset(chunk_view.data(), 0, chunk_view.size()); + std::memcpy(chunk_view.data(), values.data(), values.size()); + return ceilBytesToEvenValues(values.size()); + } + } + + void exportChunk(std::span chnk) + { + auto chunk_view = std::as_bytes(chnk).first(std::min(result.size(), chnk.size_bytes())); + std::memcpy(result.data(), chunk_view.data(), chunk_view.size()); + result = result.subspan(chunk_view.size()); + } + + void encodeChunk(std::span seq) + { + for (std::size_t i = 0; i < seq.size(); i += 2) + { + encodePair(seq[i], seq[i + 1]); + } + } + + struct CompressedValue + { + TUint value; + unsigned compressed_size; + std::byte predictor; + }; + + unsigned encodeCompressedZeroByteCount(unsigned compressed) + { + if constexpr (VALUE_SIZE == MAX_ZERO_BYTE_COUNT + 1) + { + if (compressed >= 4) + --compressed; + } + return std::min(compressed, MAX_ZERO_BYTE_COUNT); + } + + unsigned decodeCompressedZeroByteCount(unsigned encoded_size) + { + if constexpr (VALUE_SIZE == MAX_ZERO_BYTE_COUNT + 1) + { + if (encoded_size > 3) + ++encoded_size; + } + return encoded_size; + } + + CompressedValue compressValue(TUint value) noexcept + { + static constexpr auto BITS_PER_BYTE = std::numeric_limits::digits; + + TUint compressed_dfcm = dfcm_predictor.predict() ^ value; + TUint compressed_fcm = fcm_predictor.predict() ^ value; + dfcm_predictor.add(value); + fcm_predictor.add(value); + auto zeroes_dfcm = std::countl_zero(compressed_dfcm); + auto zeroes_fcm = std::countl_zero(compressed_fcm); + if (zeroes_dfcm > zeroes_fcm) + return {compressed_dfcm, encodeCompressedZeroByteCount(static_cast(zeroes_dfcm) / BITS_PER_BYTE), DFCM_BIT}; + return {compressed_fcm, encodeCompressedZeroByteCount(static_cast(zeroes_fcm) / BITS_PER_BYTE), FCM_BIT}; + } + + void encodePair(TUint first, TUint second) + { + auto [compressed_value1, zero_byte_count1, predictor1] = compressValue(first); + auto [compressed_value2, zero_byte_count2, predictor2] = compressValue(second); + std::byte header{0x0}; + header |= (predictor1 << 4) | predictor2; + header |= static_cast((zero_byte_count1 << 4) | zero_byte_count2); + result.front() = header; + + zero_byte_count1 = decodeCompressedZeroByteCount(zero_byte_count1); + zero_byte_count2 = decodeCompressedZeroByteCount(zero_byte_count2); + auto tail_size1 = VALUE_SIZE - zero_byte_count1; + auto tail_size2 = VALUE_SIZE - zero_byte_count2; + + std::memcpy(result.data() + 1, valueTail(compressed_value1, zero_byte_count1), tail_size1); + std::memcpy(result.data() + 1 + tail_size1, valueTail(compressed_value2, zero_byte_count2), tail_size2); + result = result.subspan(1 + tail_size1 + tail_size2); + } + + std::size_t decodeChunk(std::span values, std::span seq) + { + std::size_t read_bytes = 0; + for (std::size_t i = 0; i < seq.size(); i += 2) + { + read_bytes += decodePair(values.subspan(read_bytes), seq[i], seq[i + 1]); + } + return read_bytes; + } + + TUint decompressValue(TUint value, bool isDfcmPredictor) + { + TUint decompressed; + if (isDfcmPredictor) + { + decompressed = dfcm_predictor.predict() ^ value; + } + else + { + decompressed = fcm_predictor.predict() ^ value; + } + dfcm_predictor.add(decompressed); + fcm_predictor.add(decompressed); + return decompressed; + } + + std::size_t decodePair(std::span bytes, TUint& first, TUint& second) + { + if (bytes.empty()) [[unlikely]] + throw Exception(ErrorCodes::CANNOT_DECOMPRESS, "Unexpected end of encoded sequence"); + + auto zero_byte_count1 = decodeCompressedZeroByteCount( + std::to_integer(bytes.front() >> 4) & MAX_ZERO_BYTE_COUNT); + auto zero_byte_count2 = decodeCompressedZeroByteCount( + std::to_integer(bytes.front()) & MAX_ZERO_BYTE_COUNT); + + auto tail_size1 = VALUE_SIZE - zero_byte_count1; + auto tail_size2 = VALUE_SIZE - zero_byte_count2; + + if (bytes.size() < 1 + tail_size1 + tail_size2) [[unlikely]] + throw Exception(ErrorCodes::CANNOT_DECOMPRESS, "Unexpected end of encoded sequence"); + + TUint value1 = 0; + TUint value2 = 0; + + std::memcpy(valueTail(value1, zero_byte_count1), bytes.data() + 1, tail_size1); + std::memcpy(valueTail(value2, zero_byte_count2), bytes.data() + 1 + tail_size1, tail_size2); + + auto is_dfcm_predictor1 = std::to_integer(bytes.front() & DFCM_BIT_1) != 0; + auto is_dfcm_predictor2 = std::to_integer(bytes.front() & DFCM_BIT_2) != 0; + first = decompressValue(value1, is_dfcm_predictor1); + second = decompressValue(value2, is_dfcm_predictor2); + + return 1 + tail_size1 + tail_size2; + } + + static void* valueTail(TUint& value, unsigned compressed_size) + { + if constexpr (ENDIAN == std::endian::little) + { + return &value; + } + else + { + return reinterpret_cast(&value) + compressed_size; + } + } + + DfcmPredictor dfcm_predictor; + FcmPredictor fcm_predictor; + + // memcpy the input into this buffer to align reads, this improves performance compared to unaligned reads (bit_cast) by ~10% + std::array chunk{}; + + std::span result{}; +}; + +} + +UInt32 CompressionCodecFPC::doCompressData(const char * source, UInt32 source_size, char * dest) const +{ + dest[0] = static_cast(float_width); + dest[1] = static_cast(level); + + auto dest_size = getMaxCompressedDataSize(source_size); + auto destination = std::as_writable_bytes(std::span(dest, dest_size).subspan(HEADER_SIZE)); + auto src = std::as_bytes(std::span(source, source_size)); + switch (float_width) + { + case sizeof(Float64): + return HEADER_SIZE + FPCOperation(destination, level).encode(src); + case sizeof(Float32): + return HEADER_SIZE + FPCOperation(destination, level).encode(src); + default: + break; + } + throw Exception("Cannot compress. File has incorrect float width", ErrorCodes::CANNOT_COMPRESS); +} + +void CompressionCodecFPC::doDecompressData(const char * source, UInt32 source_size, char * dest, UInt32 uncompressed_size) const +{ + if (source_size < HEADER_SIZE) + throw Exception("Cannot decompress. File has wrong header", ErrorCodes::CANNOT_DECOMPRESS); + + auto compressed_data = std::as_bytes(std::span(source, source_size)); + auto compressed_float_width = std::to_integer(compressed_data[0]); + auto compressed_level = std::to_integer(compressed_data[1]); + if (compressed_level == 0 || compressed_level > MAX_COMPRESSION_LEVEL) + throw Exception("Cannot decompress. File has incorrect level", ErrorCodes::CANNOT_DECOMPRESS); + + auto destination = std::as_writable_bytes(std::span(dest, uncompressed_size)); + auto src = compressed_data.subspan(HEADER_SIZE); + switch (compressed_float_width) + { + case sizeof(Float64): + FPCOperation(destination, compressed_level).decode(src, uncompressed_size); + break; + case sizeof(Float32): + FPCOperation(destination, compressed_level).decode(src, uncompressed_size); + break; + default: + throw Exception("Cannot decompress. File has incorrect float width", ErrorCodes::CANNOT_DECOMPRESS); + } +} + +} diff --git a/src/Compression/CompressionFactory.cpp b/src/Compression/CompressionFactory.cpp index abf5e38a8c3..b8a1c5877a4 100644 --- a/src/Compression/CompressionFactory.cpp +++ b/src/Compression/CompressionFactory.cpp @@ -177,6 +177,7 @@ void registerCodecT64(CompressionCodecFactory & factory); void registerCodecDoubleDelta(CompressionCodecFactory & factory); void registerCodecGorilla(CompressionCodecFactory & factory); void registerCodecEncrypted(CompressionCodecFactory & factory); +void registerCodecFPC(CompressionCodecFactory & factory); #endif @@ -194,6 +195,7 @@ CompressionCodecFactory::CompressionCodecFactory() registerCodecDoubleDelta(*this); registerCodecGorilla(*this); registerCodecEncrypted(*this); + registerCodecFPC(*this); #endif default_codec = get("LZ4", {}); diff --git a/src/Compression/CompressionInfo.h b/src/Compression/CompressionInfo.h index bbe8315f3ea..839fb68e8c3 100644 --- a/src/Compression/CompressionInfo.h +++ b/src/Compression/CompressionInfo.h @@ -44,7 +44,8 @@ enum class CompressionMethodByte : uint8_t DoubleDelta = 0x94, Gorilla = 0x95, AES_128_GCM_SIV = 0x96, - AES_256_GCM_SIV = 0x97 + AES_256_GCM_SIV = 0x97, + FPC = 0x98 }; } diff --git a/src/Core/BackgroundSchedulePool.cpp b/src/Core/BackgroundSchedulePool.cpp index 344860ca5dd..6f5a5a1fa54 100644 --- a/src/Core/BackgroundSchedulePool.cpp +++ b/src/Core/BackgroundSchedulePool.cpp @@ -10,18 +10,6 @@ namespace DB { - -class TaskNotification final : public Poco::Notification -{ -public: - explicit TaskNotification(const BackgroundSchedulePoolTaskInfoPtr & task_) : task(task_) {} - void execute() { task->execute(); } - -private: - BackgroundSchedulePoolTaskInfoPtr task; -}; - - BackgroundSchedulePoolTaskInfo::BackgroundSchedulePoolTaskInfo( BackgroundSchedulePool & pool_, const std::string & log_name_, const BackgroundSchedulePool::TaskFunc & function_) : pool(pool_), log_name(log_name_), function(function_) @@ -39,7 +27,7 @@ bool BackgroundSchedulePoolTaskInfo::schedule() return true; } -bool BackgroundSchedulePoolTaskInfo::scheduleAfter(size_t ms, bool overwrite) +bool BackgroundSchedulePoolTaskInfo::scheduleAfter(size_t milliseconds, bool overwrite) { std::lock_guard lock(schedule_mutex); @@ -48,7 +36,7 @@ bool BackgroundSchedulePoolTaskInfo::scheduleAfter(size_t ms, bool overwrite) if (delayed && !overwrite) return false; - pool.scheduleDelayedTask(shared_from_this(), ms, lock); + pool.scheduleDelayedTask(shared_from_this(), milliseconds, lock); return true; } @@ -106,7 +94,7 @@ void BackgroundSchedulePoolTaskInfo::execute() UInt64 milliseconds = watch.elapsedMilliseconds(); /// If the task is executed longer than specified time, it will be logged. - static const int32_t slow_execution_threshold_ms = 200; + static constexpr UInt64 slow_execution_threshold_ms = 200; if (milliseconds >= slow_execution_threshold_ms) LOG_TRACE(&Poco::Logger::get(log_name), "Execution took {} ms.", milliseconds); @@ -121,7 +109,7 @@ void BackgroundSchedulePoolTaskInfo::execute() /// will have their chance to execute if (scheduled) - pool.queue.enqueueNotification(new TaskNotification(shared_from_this())); + pool.scheduleTask(shared_from_this()); } } @@ -136,14 +124,14 @@ void BackgroundSchedulePoolTaskInfo::scheduleImpl(std::lock_guard & /// But if it is currently executing, do nothing because it will be enqueued /// at the end of the execute() method. if (!executing) - pool.queue.enqueueNotification(new TaskNotification(shared_from_this())); + pool.scheduleTask(shared_from_this()); } Coordination::WatchCallback BackgroundSchedulePoolTaskInfo::getWatchCallback() { - return [t = shared_from_this()](const Coordination::WatchResponse &) + return [task = shared_from_this()](const Coordination::WatchResponse &) { - t->schedule(); + task->schedule(); }; } @@ -184,15 +172,18 @@ BackgroundSchedulePool::~BackgroundSchedulePool() try { { - std::unique_lock lock(delayed_tasks_mutex); + std::lock_guard lock_tasks(tasks_mutex); + std::lock_guard lock_delayed_tasks(delayed_tasks_mutex); + shutdown = true; - wakeup_cond.notify_all(); } - queue.wakeUpAll(); - delayed_thread.join(); + tasks_cond_var.notify_all(); + delayed_tasks_cond_var.notify_all(); LOG_TRACE(&Poco::Logger::get("BackgroundSchedulePool/" + thread_name), "Waiting for threads to finish."); + delayed_thread.join(); + for (auto & thread : threads) thread.join(); } @@ -208,6 +199,15 @@ BackgroundSchedulePool::TaskHolder BackgroundSchedulePool::createTask(const std: return TaskHolder(std::make_shared(*this, name, function)); } +void BackgroundSchedulePool::scheduleTask(TaskInfoPtr task_info) +{ + { + std::lock_guard tasks_lock(tasks_mutex); + tasks.push_back(std::move(task_info)); + } + + tasks_cond_var.notify_one(); +} void BackgroundSchedulePool::scheduleDelayedTask(const TaskInfoPtr & task, size_t ms, std::lock_guard & /* task_schedule_mutex_lock */) { @@ -223,7 +223,7 @@ void BackgroundSchedulePool::scheduleDelayedTask(const TaskInfoPtr & task, size_ task->delayed = true; } - wakeup_cond.notify_all(); + delayed_tasks_cond_var.notify_all(); } @@ -235,7 +235,7 @@ void BackgroundSchedulePool::cancelDelayedTask(const TaskInfoPtr & task, std::lo task->delayed = false; } - wakeup_cond.notify_all(); + delayed_tasks_cond_var.notify_all(); } @@ -264,20 +264,25 @@ void BackgroundSchedulePool::threadFunction() while (!shutdown) { - /// We have to wait with timeout to prevent very rare deadlock, caused by the following race condition: - /// 1. Background thread N: threadFunction(): checks for shutdown (it's false) - /// 2. Main thread: ~BackgroundSchedulePool(): sets shutdown to true, calls queue.wakeUpAll(), it triggers - /// all existing Poco::Events inside Poco::NotificationQueue which background threads are waiting on. - /// 3. Background thread N: threadFunction(): calls queue.waitDequeueNotification(), it creates - /// new Poco::Event inside Poco::NotificationQueue and starts to wait on it - /// Background thread N will never be woken up. - /// TODO Do we really need Poco::NotificationQueue? Why not to use std::queue + mutex + condvar or maybe even DB::ThreadPool? - constexpr size_t wait_timeout_ms = 500; - if (Poco::AutoPtr notification = queue.waitDequeueNotification(wait_timeout_ms)) + TaskInfoPtr task; + { - TaskNotification & task_notification = static_cast(*notification); - task_notification.execute(); + std::unique_lock tasks_lock(tasks_mutex); + + tasks_cond_var.wait(tasks_lock, [&]() + { + return shutdown || !tasks.empty(); + }); + + if (!tasks.empty()) + { + task = tasks.front(); + tasks.pop_front(); + } } + + if (task) + task->execute(); } } @@ -309,7 +314,7 @@ void BackgroundSchedulePool::delayExecutionThreadFunction() if (!task) { - wakeup_cond.wait(lock); + delayed_tasks_cond_var.wait(lock); continue; } @@ -317,7 +322,7 @@ void BackgroundSchedulePool::delayExecutionThreadFunction() if (min_time > current_time) { - wakeup_cond.wait_for(lock, std::chrono::microseconds(min_time - current_time)); + delayed_tasks_cond_var.wait_for(lock, std::chrono::microseconds(min_time - current_time)); continue; } else diff --git a/src/Core/BackgroundSchedulePool.h b/src/Core/BackgroundSchedulePool.h index fbd7e3f749a..ebd0d52ee20 100644 --- a/src/Core/BackgroundSchedulePool.h +++ b/src/Core/BackgroundSchedulePool.h @@ -62,6 +62,8 @@ private: void threadFunction(); void delayExecutionThreadFunction(); + void scheduleTask(TaskInfoPtr task_info); + /// Schedule task for execution after specified delay from now. void scheduleDelayedTask(const TaskInfoPtr & task_info, size_t ms, std::lock_guard & task_schedule_mutex_lock); @@ -69,12 +71,16 @@ private: void cancelDelayedTask(const TaskInfoPtr & task_info, std::lock_guard & task_schedule_mutex_lock); std::atomic shutdown {false}; + + /// Tasks. + std::condition_variable tasks_cond_var; + std::mutex tasks_mutex; + std::deque tasks; Threads threads; - Poco::NotificationQueue queue; - /// Delayed notifications. + /// Delayed tasks. - std::condition_variable wakeup_cond; + std::condition_variable delayed_tasks_cond_var; std::mutex delayed_tasks_mutex; /// Thread waiting for next delayed task. ThreadFromGlobalPool delayed_thread; @@ -102,7 +108,7 @@ public: /// Schedule for execution after specified delay. /// If overwrite is set then the task will be re-scheduled (if it was already scheduled, i.e. delayed == true). - bool scheduleAfter(size_t ms, bool overwrite = true); + bool scheduleAfter(size_t milliseconds, bool overwrite = true); /// Further attempts to schedule become no-op. Will wait till the end of the current execution of the task. void deactivate(); diff --git a/src/Daemon/BaseDaemon.cpp b/src/Daemon/BaseDaemon.cpp index c203a96ff11..62fcebb10bb 100644 --- a/src/Daemon/BaseDaemon.cpp +++ b/src/Daemon/BaseDaemon.cpp @@ -352,26 +352,27 @@ private: #if defined(OS_LINUX) /// Write information about binary checksum. It can be difficult to calculate, so do it only after printing stack trace. + /// Please keep the below log messages in-sync with the ones in programs/server/Server.cpp String calculated_binary_hash = getHashOfLoadedBinaryHex(); if (daemon.stored_binary_hash.empty()) { - LOG_FATAL(log, "Calculated checksum of the binary: {}." - " There is no information about the reference checksum.", calculated_binary_hash); + LOG_FATAL(log, "Integrity check of the executable skipped because the reference checksum could not be read." + " (calculated checksum: {})", calculated_binary_hash); } else if (calculated_binary_hash == daemon.stored_binary_hash) { - LOG_FATAL(log, "Checksum of the binary: {}, integrity check passed.", calculated_binary_hash); + LOG_FATAL(log, "Integrity check of the executable successfully passed (checksum: {})", calculated_binary_hash); } else { - LOG_FATAL(log, "Calculated checksum of the ClickHouse binary ({0}) does not correspond" - " to the reference checksum stored in the binary ({1})." - " It may indicate one of the following:" - " - the file was changed just after startup;" - " - the file is damaged on disk due to faulty hardware;" - " - the loaded executable is damaged in memory due to faulty hardware;" + LOG_FATAL(log, "Calculated checksum of the executable ({0}) does not correspond" + " to the reference checksum stored in the executable ({1})." + " This may indicate one of the following:" + " - the executable was changed just after startup;" + " - the executable was corrupted on disk due to faulty hardware;" + " - the loaded executable was corrupted in memory due to faulty hardware;" " - the file was intentionally modified;" - " - logical error in code." + " - a logical error in the code." , calculated_binary_hash, daemon.stored_binary_hash); } #endif @@ -872,7 +873,7 @@ void BaseDaemon::initializeTerminationAndSignalProcessing() std::string executable_path = getExecutablePath(); if (!executable_path.empty()) - stored_binary_hash = DB::Elf(executable_path).getBinaryHash(); + stored_binary_hash = DB::Elf(executable_path).getStoredBinaryHash(); #endif } diff --git a/src/Functions/FunctionsStringSimilarity.cpp b/src/Functions/FunctionsStringSimilarity.cpp index 6b2d34a182c..ee3bce4c0db 100644 --- a/src/Functions/FunctionsStringSimilarity.cpp +++ b/src/Functions/FunctionsStringSimilarity.cpp @@ -20,6 +20,10 @@ # include #endif +#if defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) +# include +#endif + namespace DB { /** Distance function implementation. @@ -64,6 +68,8 @@ struct NgramDistanceImpl UInt64 combined = (static_cast(code_points[0]) << 32) | code_points[1]; #ifdef __SSE4_2__ return _mm_crc32_u64(code_points[2], combined) & 0xFFFFu; +#elif defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) + return __crc32cd(code_points[2], combined) & 0xFFFFu; #else return (intHashCRC32(combined) ^ intHashCRC32(code_points[2])) & 0xFFFFu; #endif diff --git a/src/Functions/modulo.cpp b/src/Functions/modulo.cpp index d494dfca74d..175f74bab8f 100644 --- a/src/Functions/modulo.cpp +++ b/src/Functions/modulo.cpp @@ -2,7 +2,13 @@ #include #if defined(__SSE2__) -# define LIBDIVIDE_SSE2 1 +# define LIBDIVIDE_SSE2 +#elif defined(__AVX512F__) || defined(__AVX512BW__) || defined(__AVX512VL__) +# define LIBDIVIDE_AVX512 +#elif defined(__AVX2__) +# define LIBDIVIDE_AVX2 +#elif defined(__aarch64__) && defined(__ARM_NEON) +# define LIBDIVIDE_NEON #endif #include diff --git a/src/Functions/toValidUTF8.cpp b/src/Functions/toValidUTF8.cpp index 189556f48ea..0ee62bd4961 100644 --- a/src/Functions/toValidUTF8.cpp +++ b/src/Functions/toValidUTF8.cpp @@ -11,8 +11,16 @@ # include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif + namespace DB { + namespace ErrorCodes { extern const int ILLEGAL_COLUMN; @@ -61,6 +69,24 @@ struct ToValidUTF8Impl while (p < simd_end && !_mm_movemask_epi8(_mm_loadu_si128(reinterpret_cast(p)))) p += SIMD_BYTES; + if (!(p < end)) + break; +#elif defined(__aarch64__) && defined(__ARM_NEON) + /// Fast skip of ASCII for aarch64. + static constexpr size_t SIMD_BYTES = 16; + const char * simd_end = p + (end - p) / SIMD_BYTES * SIMD_BYTES; + /// Returns a 64 bit mask of nibbles (4 bits for each byte). + auto get_nibble_mask = [](uint8x16_t input) -> uint64_t + { return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(input), 4)), 0); }; + /// Other options include + /// vmaxvq_u8(input) < 0b10000000; + /// Used by SIMDJSON, has latency 3 for M1, 6 for everything else + /// SIMDJSON uses it for 64 byte masks, so it's a little different. + /// vmaxvq_u32(vandq_u32(input, vdupq_n_u32(0x80808080))) // u32 version has latency 3 + /// shrn version has universally <=3 cycles, on servers 2 cycles. + while (p < simd_end && get_nibble_mask(vcgeq_u8(vld1q_u8(reinterpret_cast(p)), vdupq_n_u8(0x80))) == 0) + p += SIMD_BYTES; + if (!(p < end)) break; #endif diff --git a/src/IO/ReadHelpers.cpp b/src/IO/ReadHelpers.cpp index 21e943b36ef..f09292cd349 100644 --- a/src/IO/ReadHelpers.cpp +++ b/src/IO/ReadHelpers.cpp @@ -15,6 +15,13 @@ #include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif + namespace DB { @@ -695,6 +702,24 @@ void readCSVStringInto(Vector & s, ReadBuffer & buf, const FormatSettings::CSV & return; } } +#elif defined(__aarch64__) && defined(__ARM_NEON) + auto rc = vdupq_n_u8('\r'); + auto nc = vdupq_n_u8('\n'); + auto dc = vdupq_n_u8(delimiter); + /// Returns a 64 bit mask of nibbles (4 bits for each byte). + auto get_nibble_mask = [](uint8x16_t input) -> uint64_t + { return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(input), 4)), 0); }; + for (; next_pos + 15 < buf.buffer().end(); next_pos += 16) + { + uint8x16_t bytes = vld1q_u8(reinterpret_cast(next_pos)); + auto eq = vorrq_u8(vorrq_u8(vceqq_u8(bytes, rc), vceqq_u8(bytes, nc)), vceqq_u8(bytes, dc)); + uint64_t bit_mask = get_nibble_mask(eq); + if (bit_mask) + { + next_pos += __builtin_ctzll(bit_mask) >> 2; + return; + } + } #endif while (next_pos < buf.buffer().end() && *next_pos != delimiter && *next_pos != '\r' && *next_pos != '\n') diff --git a/src/IO/ReadHelpers.h b/src/IO/ReadHelpers.h index feacbd085c2..8fe8e2aa23e 100644 --- a/src/IO/ReadHelpers.h +++ b/src/IO/ReadHelpers.h @@ -908,6 +908,8 @@ inline ReturnType readDateTimeTextImpl(DateTime64 & datetime64, UInt32 scale, Re return ReturnType(false); } + int negative_multiplier = 1; + DB::DecimalUtils::DecimalComponents components{static_cast(whole), 0}; if (!buf.eof() && *buf.position() == '.') @@ -934,29 +936,18 @@ inline ReturnType readDateTimeTextImpl(DateTime64 & datetime64, UInt32 scale, Re while (!buf.eof() && isNumericASCII(*buf.position())) ++buf.position(); - /// Keep sign of fractional part the same with whole part if datetime64 is negative - /// Case1: - /// 1965-12-12 12:12:12.123 - /// => whole = -127914468, fractional = 123(coefficient>0) - /// => new whole = -127914467, new fractional = 877(coefficient<0) - /// - /// Case2: - /// 1969-12-31 23:59:59.123 - /// => whole = -1, fractional = 123(coefficient>0) - /// => new whole = 0, new fractional = -877(coefficient>0) + /// Fractional part (subseconds) is treated as positive by users + /// (as DateTime64 itself is a positive, although underlying decimal is negative) + /// setting fractional part to be negative when whole is 0 results in wrong value, + /// so we multiply result by -1. if (components.whole < 0 && components.fractional != 0) { const auto scale_multiplier = DecimalUtils::scaleMultiplier(scale); ++components.whole; - if (components.whole) + components.fractional = scale_multiplier - components.fractional; + if (!components.whole) { - /// whole keep the sign, fractional should be non-negative - components.fractional = scale_multiplier - components.fractional; - } - else - { - /// when whole is zero, fractional should keep the sign - components.fractional = components.fractional - scale_multiplier; + negative_multiplier = -1; } } } @@ -969,7 +960,7 @@ inline ReturnType readDateTimeTextImpl(DateTime64 & datetime64, UInt32 scale, Re components.whole = components.whole / common::exp10_i32(scale); } - datetime64 = DecimalUtils::decimalFromComponents(components, scale); + datetime64 = negative_multiplier * DecimalUtils::decimalFromComponents(components, scale); return ReturnType(true); } diff --git a/src/IO/WriteBufferValidUTF8.cpp b/src/IO/WriteBufferValidUTF8.cpp index f711db0f1e7..10e86f01343 100644 --- a/src/IO/WriteBufferValidUTF8.cpp +++ b/src/IO/WriteBufferValidUTF8.cpp @@ -6,6 +6,12 @@ #include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif namespace DB { @@ -67,13 +73,31 @@ void WriteBufferValidUTF8::nextImpl() while (p < pos) { #ifdef __SSE2__ - /// Fast skip of ASCII + /// Fast skip of ASCII for x86. static constexpr size_t SIMD_BYTES = 16; const char * simd_end = p + (pos - p) / SIMD_BYTES * SIMD_BYTES; while (p < simd_end && !_mm_movemask_epi8(_mm_loadu_si128(reinterpret_cast(p)))) p += SIMD_BYTES; + if (!(p < pos)) + break; +#elif defined(__aarch64__) && defined(__ARM_NEON) + /// Fast skip of ASCII for aarch64. + static constexpr size_t SIMD_BYTES = 16; + const char * simd_end = p + (pos - p) / SIMD_BYTES * SIMD_BYTES; + /// Returns a 64 bit mask of nibbles (4 bits for each byte). + auto get_nibble_mask = [](uint8x16_t input) -> uint64_t + { return vget_lane_u64(vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(input), 4)), 0); }; + /// Other options include + /// vmaxvq_u8(input) < 0b10000000; + /// Used by SIMDJSON, has latency 3 for M1, 6 for everything else + /// SIMDJSON uses it for 64 byte masks, so it's a little different. + /// vmaxvq_u32(vandq_u32(input, vdupq_n_u32(0x80808080))) // u32 version has latency 3 + /// shrn version has universally <=3 cycles, on servers 2 cycles. + while (p < simd_end && get_nibble_mask(vcgeq_u8(vld1q_u8(reinterpret_cast(p)), vdupq_n_u8(0x80))) == 0) + p += SIMD_BYTES; + if (!(p < pos)) break; #endif diff --git a/src/Interpreters/ClusterProxy/SelectStreamFactory.cpp b/src/Interpreters/ClusterProxy/SelectStreamFactory.cpp index 498e296a3af..fce2e9b2f08 100644 --- a/src/Interpreters/ClusterProxy/SelectStreamFactory.cpp +++ b/src/Interpreters/ClusterProxy/SelectStreamFactory.cpp @@ -257,7 +257,6 @@ SelectStreamFactory::ShardPlans SelectStreamFactory::createForShardWithParallelR std::move(scalars), std::move(external_tables), &Poco::Logger::get("ReadFromParallelRemoteReplicasStep"), - shard_count, storage_limits); remote_plan->addStep(std::move(read_from_remote)); diff --git a/src/Interpreters/FillingRow.cpp b/src/Interpreters/FillingRow.cpp index bb8661d0ef9..561ac255326 100644 --- a/src/Interpreters/FillingRow.cpp +++ b/src/Interpreters/FillingRow.cpp @@ -76,7 +76,7 @@ bool FillingRow::next(const FillingRow & to_row) auto next_value = row[pos]; getFillDescription(pos).step_func(next_value); - if (less(to_row.row[pos], next_value, getDirection(pos))) + if (less(to_row.row[pos], next_value, getDirection(pos)) || equals(next_value, getFillDescription(pos).fill_to)) return false; row[pos] = next_value; diff --git a/src/Interpreters/createBlockSelector.cpp b/src/Interpreters/createBlockSelector.cpp index d7a09743b46..b1a9a4e9e35 100644 --- a/src/Interpreters/createBlockSelector.cpp +++ b/src/Interpreters/createBlockSelector.cpp @@ -6,7 +6,13 @@ #include #if defined(__SSE2__) -# define LIBDIVIDE_SSE2 1 +# define LIBDIVIDE_SSE2 +#elif defined(__AVX512F__) || defined(__AVX512BW__) || defined(__AVX512VL__) +# define LIBDIVIDE_AVX512 +#elif defined(__AVX2__) +# define LIBDIVIDE_AVX2 +#elif defined(__aarch64__) && defined(__ARM_NEON) +# define LIBDIVIDE_NEON #endif #include diff --git a/src/Processors/QueryPlan/ReadFromRemote.cpp b/src/Processors/QueryPlan/ReadFromRemote.cpp index 6e2d776e1e4..65b902230f4 100644 --- a/src/Processors/QueryPlan/ReadFromRemote.cpp +++ b/src/Processors/QueryPlan/ReadFromRemote.cpp @@ -239,7 +239,6 @@ ReadFromParallelRemoteReplicasStep::ReadFromParallelRemoteReplicasStep( Scalars scalars_, Tables external_tables_, Poco::Logger * log_, - UInt32 shard_count_, std::shared_ptr storage_limits_) : ISourceStep(DataStream{.header = std::move(header_)}) , coordinator(std::move(coordinator_)) @@ -253,7 +252,6 @@ ReadFromParallelRemoteReplicasStep::ReadFromParallelRemoteReplicasStep( , external_tables{external_tables_} , storage_limits(std::move(storage_limits_)) , log(log_) - , shard_count(shard_count_) { std::vector description; diff --git a/src/Processors/QueryPlan/ReadFromRemote.h b/src/Processors/QueryPlan/ReadFromRemote.h index 0a21f240f5a..4d37a637250 100644 --- a/src/Processors/QueryPlan/ReadFromRemote.h +++ b/src/Processors/QueryPlan/ReadFromRemote.h @@ -83,7 +83,6 @@ public: Scalars scalars_, Tables external_tables_, Poco::Logger * log_, - UInt32 shard_count_, std::shared_ptr storage_limits_); String getName() const override { return "ReadFromRemoteParallelReplicas"; } @@ -110,8 +109,6 @@ private: std::shared_ptr storage_limits; Poco::Logger * log; - - UInt32 shard_count{0}; }; } diff --git a/src/Processors/Transforms/PartialSortingTransform.cpp b/src/Processors/Transforms/PartialSortingTransform.cpp index 3687fa770f0..131bf4f8e7c 100644 --- a/src/Processors/Transforms/PartialSortingTransform.cpp +++ b/src/Processors/Transforms/PartialSortingTransform.cpp @@ -1,50 +1,26 @@ #include #include +#include #include namespace DB { -PartialSortingTransform::PartialSortingTransform( - const Block & header_, SortDescription & description_, UInt64 limit_) - : ISimpleTransform(header_, header_, false) - , description(description_), limit(limit_) +namespace { - // Sorting by no columns doesn't make sense. - assert(!description.empty()); -} -static ColumnRawPtrs extractColumns(const Block & block, const SortDescription & description) +ColumnRawPtrs extractRawColumns(const Block & block, const SortDescriptionWithPositions & description) { size_t size = description.size(); - ColumnRawPtrs res; - res.reserve(size); + ColumnRawPtrs result(size); for (size_t i = 0; i < size; ++i) - { - const IColumn * column = block.getByName(description[i].column_name).column.get(); - res.emplace_back(column); - } + result[i] = block.safeGetByPosition(description[i].column_number).column.get(); - return res; + return result; } -bool less(const ColumnRawPtrs & lhs, UInt64 lhs_row_num, - const ColumnRawPtrs & rhs, UInt64 rhs_row_num, const SortDescription & description) -{ - size_t size = description.size(); - for (size_t i = 0; i < size; ++i) - { - int res = description[i].direction * lhs[i]->compareAt(lhs_row_num, rhs_row_num, *rhs[i], description[i].nulls_direction); - if (res < 0) - return true; - else if (res > 0) - return false; - } - return false; -} - -size_t getFilterMask(const ColumnRawPtrs & lhs, const ColumnRawPtrs & rhs, size_t rhs_row_num, +size_t getFilterMask(const ColumnRawPtrs & raw_block_columns, const Columns & threshold_columns, const SortDescription & description, size_t num_rows, IColumn::Filter & filter, PaddedPODArray & rows_to_compare, PaddedPODArray & compare_results) { @@ -54,7 +30,7 @@ size_t getFilterMask(const ColumnRawPtrs & lhs, const ColumnRawPtrs & rhs, size_ if (description.size() == 1) { /// Fast path for single column - lhs[0]->compareColumn(*rhs[0], rhs_row_num, nullptr, compare_results, + raw_block_columns[0]->compareColumn(*threshold_columns[0], 0, nullptr, compare_results, description[0].direction, description[0].nulls_direction); } else @@ -67,7 +43,7 @@ size_t getFilterMask(const ColumnRawPtrs & lhs, const ColumnRawPtrs & rhs, size_ size_t size = description.size(); for (size_t i = 0; i < size; ++i) { - lhs[i]->compareColumn(*rhs[i], rhs_row_num, &rows_to_compare, compare_results, + raw_block_columns[i]->compareColumn(*threshold_columns[i], 0, &rows_to_compare, compare_results, description[i].direction, description[i].nulls_direction); if (rows_to_compare.empty()) @@ -87,6 +63,41 @@ size_t getFilterMask(const ColumnRawPtrs & lhs, const ColumnRawPtrs & rhs, size_ return result_size_hint; } +bool compareWithThreshold(const ColumnRawPtrs & raw_block_columns, size_t min_block_index, const Columns & threshold_columns, const SortDescription & sort_description) +{ + assert(raw_block_columns.size() == threshold_columns.size()); + assert(raw_block_columns.size() == sort_description.size()); + + size_t raw_block_columns_size = raw_block_columns.size(); + for (size_t i = 0; i < raw_block_columns_size; ++i) + { + int res = sort_description[i].direction * raw_block_columns[i]->compareAt(min_block_index, 0, *threshold_columns[0], sort_description[i].nulls_direction); + + if (res < 0) + return true; + else if (res > 0) + return false; + } + + return false; +} + +} + +PartialSortingTransform::PartialSortingTransform( + const Block & header_, SortDescription & description_, UInt64 limit_) + : ISimpleTransform(header_, header_, false) + , description(description_) + , limit(limit_) +{ + // Sorting by no columns doesn't make sense. + assert(!description_.empty()); + + size_t description_size = description.size(); + for (size_t i = 0; i < description_size; ++i) + description_with_positions.emplace_back(description[i], header_.getPositionByName(description[i].column_name)); +} + void PartialSortingTransform::transform(Chunk & chunk) { if (chunk.getNumRows()) @@ -101,17 +112,18 @@ void PartialSortingTransform::transform(Chunk & chunk) read_rows->add(chunk.getNumRows()); auto block = getInputPort().getHeader().cloneWithColumns(chunk.detachColumns()); + size_t block_rows_before_filter = block.rows(); /** If we've saved columns from previously blocks we could filter all rows from current block * which are unnecessary for sortBlock(...) because they obviously won't be in the top LIMIT rows. */ - if (!threshold_block_columns.empty()) + if (!sort_description_threshold_columns.empty()) { UInt64 rows_num = block.rows(); - auto block_columns = extractColumns(block, description); + auto block_columns = extractRawColumns(block, description_with_positions); size_t result_size_hint = getFilterMask( - block_columns, threshold_block_columns, limit - 1, + block_columns, sort_description_threshold_columns, description, rows_num, filter, rows_to_compare, compare_results); /// Everything was filtered. Skip whole chunk. @@ -127,16 +139,34 @@ void PartialSortingTransform::transform(Chunk & chunk) sortBlock(block, description, limit); - /// Check if we can use this block for optimization. - if (min_limit_for_partial_sort_optimization <= limit && limit <= block.rows()) - { - auto block_columns = extractColumns(block, description); + size_t block_rows_after_filter = block.rows(); - if (threshold_block_columns.empty() || - less(block_columns, limit - 1, threshold_block_columns, limit - 1, description)) + /// Check if we can use this block for optimization. + if (min_limit_for_partial_sort_optimization <= limit && block_rows_after_filter > 0 && limit <= block_rows_before_filter) + { + /** If we filtered more than limit rows from block take block last row. + * Otherwise take last limit row. + * + * If current threshold value is empty, update current threshold value. + * If min block value is less than current threshold value, update current threshold value. + */ + size_t min_row_to_compare = limit <= block_rows_after_filter ? (limit - 1) : (block_rows_after_filter - 1); + auto raw_block_columns = extractRawColumns(block, description_with_positions); + + if (sort_description_threshold_columns.empty() || + compareWithThreshold(raw_block_columns, min_row_to_compare, sort_description_threshold_columns, description)) { - threshold_block = block; - threshold_block_columns.swap(block_columns); + size_t raw_block_columns_size = raw_block_columns.size(); + Columns sort_description_threshold_columns_updated(raw_block_columns_size); + + for (size_t i = 0; i < raw_block_columns_size; ++i) + { + MutableColumnPtr sort_description_threshold_column_updated = raw_block_columns[i]->cloneEmpty(); + sort_description_threshold_column_updated->insertFrom(*raw_block_columns[i], min_row_to_compare); + sort_description_threshold_columns_updated[i] = std::move(sort_description_threshold_column_updated); + } + + sort_description_threshold_columns = std::move(sort_description_threshold_columns_updated); } } diff --git a/src/Processors/Transforms/PartialSortingTransform.h b/src/Processors/Transforms/PartialSortingTransform.h index bd2465f11a1..78ce80bdeeb 100644 --- a/src/Processors/Transforms/PartialSortingTransform.h +++ b/src/Processors/Transforms/PartialSortingTransform.h @@ -8,7 +8,6 @@ namespace DB { /** Sorts each block individually by the values of the specified columns. - * At the moment, not very optimal algorithm is used. */ class PartialSortingTransform : public ISimpleTransform { @@ -28,15 +27,11 @@ protected: private: SortDescription description; + SortDescriptionWithPositions description_with_positions; UInt64 limit; RowsBeforeLimitCounterPtr read_rows; - /** threshold_block is using for saving columns from previously processed block. - * threshold_block_columns contains pointers to columns from threshold_block which used for comparison. - * That's all for PartialSort optimization - */ - Block threshold_block; - ColumnRawPtrs threshold_block_columns; + Columns sort_description_threshold_columns; /// This are just buffers which reserve memory to reduce the number of allocations. PaddedPODArray rows_to_compare; diff --git a/src/Storages/MergeTree/MergeTreeData.cpp b/src/Storages/MergeTree/MergeTreeData.cpp index dc24e0b9806..dea6c8b0066 100644 --- a/src/Storages/MergeTree/MergeTreeData.cpp +++ b/src/Storages/MergeTree/MergeTreeData.cpp @@ -4438,10 +4438,7 @@ std::set MergeTreeData::getPartitionIdsAffectedByCommands( MergeTreeData::DataPartsVector MergeTreeData::getDataPartsVectorForInternalUsage( - const DataPartStates & affordable_states, - const DataPartsLock & /*lock*/, - DataPartStateVector * out_states, - bool require_projection_parts) const + const DataPartStates & affordable_states, const DataPartsLock & /*lock*/, DataPartStateVector * out_states) const { DataPartsVector res; DataPartsVector buf; @@ -4449,86 +4446,89 @@ MergeTreeData::DataPartsVector MergeTreeData::getDataPartsVectorForInternalUsage for (auto state : affordable_states) { auto range = getDataPartsStateRange(state); - - if (require_projection_parts) - { - for (const auto & part : range) - { - for (const auto & [_, projection_part] : part->getProjectionParts()) - res.push_back(projection_part); - } - } - else - { - std::swap(buf, res); - res.clear(); - std::merge(range.begin(), range.end(), buf.begin(), buf.end(), std::back_inserter(res), LessDataPart()); //-V783 - } + std::swap(buf, res); + res.clear(); + std::merge(range.begin(), range.end(), buf.begin(), buf.end(), std::back_inserter(res), LessDataPart()); //-V783 } if (out_states != nullptr) { out_states->resize(res.size()); - if (require_projection_parts) - { - for (size_t i = 0; i < res.size(); ++i) - (*out_states)[i] = res[i]->getParentPart()->getState(); - } - else - { - for (size_t i = 0; i < res.size(); ++i) - (*out_states)[i] = res[i]->getState(); - } + for (size_t i = 0; i < res.size(); ++i) + (*out_states)[i] = res[i]->getState(); } return res; } -MergeTreeData::DataPartsVector MergeTreeData::getDataPartsVectorForInternalUsage( - const DataPartStates & affordable_states, - DataPartStateVector * out_states, - bool require_projection_parts) const +MergeTreeData::DataPartsVector +MergeTreeData::getDataPartsVectorForInternalUsage(const DataPartStates & affordable_states, DataPartStateVector * out_states) const { auto lock = lockParts(); - return getDataPartsVectorForInternalUsage(affordable_states, lock, out_states, require_projection_parts); + return getDataPartsVectorForInternalUsage(affordable_states, lock, out_states); } -MergeTreeData::DataPartsVector -MergeTreeData::getAllDataPartsVector(MergeTreeData::DataPartStateVector * out_states, bool require_projection_parts) const +MergeTreeData::ProjectionPartsVector +MergeTreeData::getProjectionPartsVectorForInternalUsage(const DataPartStates & affordable_states, DataPartStateVector * out_states) const +{ + auto lock = lockParts(); + ProjectionPartsVector res; + for (auto state : affordable_states) + { + auto range = getDataPartsStateRange(state); + for (const auto & part : range) + { + res.data_parts.push_back(part); + for (const auto & [_, projection_part] : part->getProjectionParts()) + res.projection_parts.push_back(projection_part); + } + } + + if (out_states != nullptr) + { + out_states->resize(res.projection_parts.size()); + for (size_t i = 0; i < res.projection_parts.size(); ++i) + (*out_states)[i] = res.projection_parts[i]->getParentPart()->getState(); + } + + return res; +} + +MergeTreeData::DataPartsVector MergeTreeData::getAllDataPartsVector(MergeTreeData::DataPartStateVector * out_states) const { DataPartsVector res; - if (require_projection_parts) + auto lock = lockParts(); + res.assign(data_parts_by_info.begin(), data_parts_by_info.end()); + if (out_states != nullptr) { - auto lock = lockParts(); - for (const auto & part : data_parts_by_info) - { - for (const auto & [p_name, projection_part] : part->getProjectionParts()) - res.push_back(projection_part); - } - - if (out_states != nullptr) - { - out_states->resize(res.size()); - for (size_t i = 0; i < res.size(); ++i) - (*out_states)[i] = res[i]->getParentPart()->getState(); - } - } - else - { - auto lock = lockParts(); - res.assign(data_parts_by_info.begin(), data_parts_by_info.end()); - - if (out_states != nullptr) - { - out_states->resize(res.size()); - for (size_t i = 0; i < res.size(); ++i) - (*out_states)[i] = res[i]->getState(); - } + out_states->resize(res.size()); + for (size_t i = 0; i < res.size(); ++i) + (*out_states)[i] = res[i]->getState(); } return res; } +MergeTreeData::ProjectionPartsVector MergeTreeData::getAllProjectionPartsVector(MergeTreeData::DataPartStateVector * out_states) const +{ + ProjectionPartsVector res; + auto lock = lockParts(); + for (const auto & part : data_parts_by_info) + { + res.data_parts.push_back(part); + for (const auto & [p_name, projection_part] : part->getProjectionParts()) + res.projection_parts.push_back(projection_part); + } + + if (out_states != nullptr) + { + out_states->resize(res.projection_parts.size()); + for (size_t i = 0; i < res.projection_parts.size(); ++i) + (*out_states)[i] = res.projection_parts[i]->getParentPart()->getState(); + } + return res; +} + DetachedPartsInfo MergeTreeData::getDetachedParts() const { DetachedPartsInfo res; diff --git a/src/Storages/MergeTree/MergeTreeData.h b/src/Storages/MergeTree/MergeTreeData.h index d3dba49f3a5..9ebdeca416b 100644 --- a/src/Storages/MergeTree/MergeTreeData.h +++ b/src/Storages/MergeTree/MergeTreeData.h @@ -449,26 +449,31 @@ public: Int64 getMaxBlockNumber() const; + struct ProjectionPartsVector + { + DataPartsVector projection_parts; + DataPartsVector data_parts; + }; /// Returns a copy of the list so that the caller shouldn't worry about locks. DataParts getDataParts(const DataPartStates & affordable_states) const; DataPartsVector getDataPartsVectorForInternalUsage( - const DataPartStates & affordable_states, - const DataPartsLock & lock, - DataPartStateVector * out_states = nullptr, - bool require_projection_parts = false) const; + const DataPartStates & affordable_states, const DataPartsLock & lock, DataPartStateVector * out_states = nullptr) const; /// Returns sorted list of the parts with specified states /// out_states will contain snapshot of each part state DataPartsVector getDataPartsVectorForInternalUsage( - const DataPartStates & affordable_states, DataPartStateVector * out_states = nullptr, bool require_projection_parts = false) const; + const DataPartStates & affordable_states, DataPartStateVector * out_states = nullptr) const; + /// Same as above but only returns projection parts + ProjectionPartsVector getProjectionPartsVectorForInternalUsage( + const DataPartStates & affordable_states, DataPartStateVector * out_states = nullptr) const; /// Returns absolutely all parts (and snapshot of their states) - DataPartsVector getAllDataPartsVector( - DataPartStateVector * out_states = nullptr, - bool require_projection_parts = false) const; + DataPartsVector getAllDataPartsVector(DataPartStateVector * out_states = nullptr) const; + /// Same as above but only returns projection parts + ProjectionPartsVector getAllProjectionPartsVector(MergeTreeData::DataPartStateVector * out_states = nullptr) const; /// Returns parts in Active state DataParts getDataPartsForInternalUsage() const; diff --git a/src/Storages/MergeTree/MergeTreeRangeReader.cpp b/src/Storages/MergeTree/MergeTreeRangeReader.cpp index c142530de58..d71dddc938c 100644 --- a/src/Storages/MergeTree/MergeTreeRangeReader.cpp +++ b/src/Storages/MergeTree/MergeTreeRangeReader.cpp @@ -11,6 +11,12 @@ #include #endif +#if defined(__aarch64__) && defined(__ARM_NEON) +# include +# ifdef HAS_RESERVED_IDENTIFIER +# pragma clang diagnostic ignored "-Wreserved-identifier" +# endif +#endif namespace DB { @@ -551,6 +557,34 @@ size_t MergeTreeRangeReader::ReadResult::numZerosInTail(const UInt8 * begin, con return count; } } +#elif defined(__aarch64__) && defined(__ARM_NEON) + const uint8x16_t bitmask = {0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80, 0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80}; + while (end - begin >= 64) + { + end -= 64; + const auto * src = reinterpret_cast(end); + const uint8x16_t p0 = vceqzq_u8(vld1q_u8(src)); + const uint8x16_t p1 = vceqzq_u8(vld1q_u8(src + 16)); + const uint8x16_t p2 = vceqzq_u8(vld1q_u8(src + 32)); + const uint8x16_t p3 = vceqzq_u8(vld1q_u8(src + 48)); + uint8x16_t t0 = vandq_u8(p0, bitmask); + uint8x16_t t1 = vandq_u8(p1, bitmask); + uint8x16_t t2 = vandq_u8(p2, bitmask); + uint8x16_t t3 = vandq_u8(p3, bitmask); + uint8x16_t sum0 = vpaddq_u8(t0, t1); + uint8x16_t sum1 = vpaddq_u8(t2, t3); + sum0 = vpaddq_u8(sum0, sum1); + sum0 = vpaddq_u8(sum0, sum0); + UInt64 val = vgetq_lane_u64(vreinterpretq_u64_u8(sum0), 0); + val = ~val; + if (val == 0) + count += 64; + else + { + count += __builtin_clzll(val); + return count; + } + } #endif while (end > begin && *(--end) == 0) diff --git a/src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp b/src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp index ebd34474066..9f679f121b8 100644 --- a/src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp +++ b/src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp @@ -1256,27 +1256,6 @@ bool ReplicatedMergeTreeQueue::shouldExecuteLogEntry( { ignore_max_size = max_source_parts_size == data_settings->max_bytes_to_merge_at_max_space_in_pool; - if (data_settings->always_fetch_merged_part && entry.num_tries > 0) - { - static constexpr auto MAX_SECONDS_TO_WAIT = 300L; - static constexpr auto BACKOFF_SECONDS = 3; - - auto time_to_wait_seconds = std::min(MAX_SECONDS_TO_WAIT, entry.num_tries * BACKOFF_SECONDS); - auto time_since_last_try_seconds = std::time(nullptr) - entry.last_attempt_time; - /// Otherwise we will constantly look for part on other replicas - /// and load zookeeper too much. - if (time_to_wait_seconds > time_since_last_try_seconds) - { - out_postpone_reason = fmt::format( - "Not executing log entry ({}) to merge parts for part {} because `always_fetch_merged_part` enabled and " - " not enough time had been passed since last try, have to wait {} seconds", - entry.znode_name, entry.new_part_name, time_to_wait_seconds - time_since_last_try_seconds); - - LOG_DEBUG(log, fmt::runtime(out_postpone_reason)); - return false; - } - } - if (isTTLMergeType(entry.merge_type)) { if (merger_mutator.ttl_merges_blocker.isCancelled()) diff --git a/src/Storages/System/StorageSystemPartsBase.cpp b/src/Storages/System/StorageSystemPartsBase.cpp index 0b976680d00..b272c080262 100644 --- a/src/Storages/System/StorageSystemPartsBase.cpp +++ b/src/Storages/System/StorageSystemPartsBase.cpp @@ -48,9 +48,25 @@ bool StorageSystemPartsBase::hasStateColumn(const Names & column_names, const St } MergeTreeData::DataPartsVector -StoragesInfo::getParts(MergeTreeData::DataPartStateVector & state, bool has_state_column, bool require_projection_parts) const +StoragesInfo::getParts(MergeTreeData::DataPartStateVector & state, bool has_state_column) const { - if (require_projection_parts && data->getInMemoryMetadataPtr()->projections.empty()) + using State = MergeTreeData::DataPartState; + if (need_inactive_parts) + { + /// If has_state_column is requested, return all states. + if (!has_state_column) + return data->getDataPartsVectorForInternalUsage({State::Active, State::Outdated}, &state); + + return data->getAllDataPartsVector(&state); + } + + return data->getDataPartsVectorForInternalUsage({State::Active}, &state); +} + +MergeTreeData::ProjectionPartsVector +StoragesInfo::getProjectionParts(MergeTreeData::DataPartStateVector & state, bool has_state_column) const +{ + if (data->getInMemoryMetadataPtr()->projections.empty()) return {}; using State = MergeTreeData::DataPartState; @@ -58,12 +74,12 @@ StoragesInfo::getParts(MergeTreeData::DataPartStateVector & state, bool has_stat { /// If has_state_column is requested, return all states. if (!has_state_column) - return data->getDataPartsVectorForInternalUsage({State::Active, State::Outdated}, &state, require_projection_parts); + return data->getProjectionPartsVectorForInternalUsage({State::Active, State::Outdated}, &state); - return data->getAllDataPartsVector(&state, require_projection_parts); + return data->getAllProjectionPartsVector(&state); } - return data->getDataPartsVectorForInternalUsage({State::Active}, &state, require_projection_parts); + return data->getProjectionPartsVectorForInternalUsage({State::Active}, &state); } StoragesInfoStream::StoragesInfoStream(const SelectQueryInfo & query_info, ContextPtr context) diff --git a/src/Storages/System/StorageSystemPartsBase.h b/src/Storages/System/StorageSystemPartsBase.h index 36c0fd551df..8db96700e1a 100644 --- a/src/Storages/System/StorageSystemPartsBase.h +++ b/src/Storages/System/StorageSystemPartsBase.h @@ -23,8 +23,9 @@ struct StoragesInfo MergeTreeData * data = nullptr; explicit operator bool() const { return storage != nullptr; } - MergeTreeData::DataPartsVector - getParts(MergeTreeData::DataPartStateVector & state, bool has_state_column, bool require_projection_parts = false) const; + + MergeTreeData::DataPartsVector getParts(MergeTreeData::DataPartStateVector & state, bool has_state_column) const; + MergeTreeData::ProjectionPartsVector getProjectionParts(MergeTreeData::DataPartStateVector & state, bool has_state_column) const; }; /** A helper class that enumerates the storages that match given query. */ diff --git a/src/Storages/System/StorageSystemProjectionParts.cpp b/src/Storages/System/StorageSystemProjectionParts.cpp index 591277c1a66..21ca1f57703 100644 --- a/src/Storages/System/StorageSystemProjectionParts.cpp +++ b/src/Storages/System/StorageSystemProjectionParts.cpp @@ -94,14 +94,13 @@ void StorageSystemProjectionParts::processNextStorage( { using State = IMergeTreeDataPart::State; MergeTreeData::DataPartStateVector all_parts_state; - MergeTreeData::DataPartsVector all_parts; - - all_parts = info.getParts(all_parts_state, has_state_column, true /* require_projection_parts */); - - for (size_t part_number = 0; part_number < all_parts.size(); ++part_number) + MergeTreeData::ProjectionPartsVector all_parts = info.getProjectionParts(all_parts_state, has_state_column); + for (size_t part_number = 0; part_number < all_parts.projection_parts.size(); ++part_number) { - const auto & part = all_parts[part_number]; + const auto & part = all_parts.projection_parts[part_number]; const auto * parent_part = part->getParentPart(); + chassert(parent_part); + auto part_state = all_parts_state[part_number]; ColumnSize columns_size = part->getTotalColumnsSize(); diff --git a/src/Storages/System/StorageSystemProjectionPartsColumns.cpp b/src/Storages/System/StorageSystemProjectionPartsColumns.cpp index 8f6db9fcbe8..1eec6825d5a 100644 --- a/src/Storages/System/StorageSystemProjectionPartsColumns.cpp +++ b/src/Storages/System/StorageSystemProjectionPartsColumns.cpp @@ -100,14 +100,15 @@ void StorageSystemProjectionPartsColumns::processNextStorage( } } - /// Go through the list of parts. + /// Go through the list of projection parts. MergeTreeData::DataPartStateVector all_parts_state; - MergeTreeData::DataPartsVector all_parts; - all_parts = info.getParts(all_parts_state, has_state_column, true /* require_projection_parts */); - for (size_t part_number = 0; part_number < all_parts.size(); ++part_number) + MergeTreeData::ProjectionPartsVector all_parts = info.getProjectionParts(all_parts_state, has_state_column); + for (size_t part_number = 0; part_number < all_parts.projection_parts.size(); ++part_number) { - const auto & part = all_parts[part_number]; + const auto & part = all_parts.projection_parts[part_number]; const auto * parent_part = part->getParentPart(); + chassert(parent_part); + auto part_state = all_parts_state[part_number]; auto columns_size = part->getTotalColumnsSize(); auto parent_columns_size = parent_part->getTotalColumnsSize(); diff --git a/tests/performance/codecs_float_insert.xml b/tests/performance/codecs_float_insert.xml index b31e0eafdd7..64325d30189 100644 --- a/tests/performance/codecs_float_insert.xml +++ b/tests/performance/codecs_float_insert.xml @@ -12,6 +12,7 @@ ZSTD DoubleDelta Gorilla + FPC diff --git a/tests/performance/codecs_float_select.xml b/tests/performance/codecs_float_select.xml index 82489daf524..325c140d9a0 100644 --- a/tests/performance/codecs_float_select.xml +++ b/tests/performance/codecs_float_select.xml @@ -12,6 +12,7 @@ ZSTD DoubleDelta Gorilla + FPC @@ -36,7 +37,11 @@ - CREATE TABLE IF NOT EXISTS codec_{seq_type}_{type}_{codec} (n {type} CODEC({codec})) ENGINE = MergeTree PARTITION BY tuple() ORDER BY tuple(); + + CREATE TABLE IF NOT EXISTS codec_{seq_type}_{type}_{codec} (n {type} CODEC({codec})) + ENGINE = MergeTree PARTITION BY tuple() + ORDER BY tuple(); + INSERT INTO codec_seq_{type}_{codec} (n) SELECT number/pi() FROM system.numbers LIMIT {num_rows} SETTINGS max_threads=1 INSERT INTO codec_mon_{type}_{codec} (n) SELECT number+sin(number) FROM system.numbers LIMIT {num_rows} SETTINGS max_threads=1 diff --git a/tests/queries/0_stateless/00995_order_by_with_fill.reference b/tests/queries/0_stateless/00995_order_by_with_fill.reference index adb0e1aa2c3..0036aabda40 100644 --- a/tests/queries/0_stateless/00995_order_by_with_fill.reference +++ b/tests/queries/0_stateless/00995_order_by_with_fill.reference @@ -376,11 +376,6 @@ 2019-05-03 4 2019-05-03 1 2019-05-03 -2 -2019-05-01 10 -2019-05-01 7 -2019-05-01 4 -2019-05-01 1 -2019-05-01 -2 *** date WITH FILL TO 2019-06-23 STEP 3, val WITH FILL FROM -10 STEP 2 2019-05-07 -10 2019-05-07 -8 diff --git a/tests/queries/0_stateless/01375_compact_parts_codecs.reference b/tests/queries/0_stateless/01375_compact_parts_codecs.reference index b48892597b6..0c9e9c8ac47 100644 --- a/tests/queries/0_stateless/01375_compact_parts_codecs.reference +++ b/tests/queries/0_stateless/01375_compact_parts_codecs.reference @@ -4,6 +4,6 @@ 11965 11890 499500 499500 999 499500 499500 999 -5858 11890 +5857 11890 499500 499500 999 499500 499500 999 diff --git a/tests/queries/0_stateless/01417_freeze_partition_verbose.sh b/tests/queries/0_stateless/01417_freeze_partition_verbose.sh index 38f84c2fa15..1f67100a4b6 100755 --- a/tests/queries/0_stateless/01417_freeze_partition_verbose.sh +++ b/tests/queries/0_stateless/01417_freeze_partition_verbose.sh @@ -1,5 +1,5 @@ #!/usr/bin/env bash -# Tags: no-replicated-database, no-parallel +# Tags: no-replicated-database, no-parallel, no-ordinary-database # Tag no-replicated-database: Unsupported type of ALTER query CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd) diff --git a/tests/queries/0_stateless/02019_multiple_weird_with_fill.reference b/tests/queries/0_stateless/02019_multiple_weird_with_fill.reference index 822d290564a..703dd6e6aac 100644 --- a/tests/queries/0_stateless/02019_multiple_weird_with_fill.reference +++ b/tests/queries/0_stateless/02019_multiple_weird_with_fill.reference @@ -34,12 +34,3 @@ 6 -4 6 -3 6 -2 -7 -10 -7 -9 -7 -8 -7 -7 -7 -6 -7 -5 -7 -4 -7 -3 -7 -2 diff --git a/tests/queries/0_stateless/02112_with_fill_interval.reference b/tests/queries/0_stateless/02112_with_fill_interval.reference index 4bb99803eb1..a1b1d3b0d4d 100644 --- a/tests/queries/0_stateless/02112_with_fill_interval.reference +++ b/tests/queries/0_stateless/02112_with_fill_interval.reference @@ -103,10 +103,6 @@ 2020-04-01 2 0 2020-04-01 3 0 2020-04-01 4 0 -2020-05-01 1 0 -2020-05-01 2 0 -2020-05-01 3 0 -2020-05-01 4 0 1970-01-04 1970-01-03 1970-01-02 diff --git a/tests/queries/0_stateless/02313_negative_datetime64.reference b/tests/queries/0_stateless/02313_negative_datetime64.reference new file mode 100644 index 00000000000..17474c96202 --- /dev/null +++ b/tests/queries/0_stateless/02313_negative_datetime64.reference @@ -0,0 +1,4 @@ +1 1 1 1 +1 1 1 1 +1 1 1 1 +1 1 1 1 diff --git a/tests/queries/0_stateless/02313_negative_datetime64.sql b/tests/queries/0_stateless/02313_negative_datetime64.sql new file mode 100644 index 00000000000..a5728074cf5 --- /dev/null +++ b/tests/queries/0_stateless/02313_negative_datetime64.sql @@ -0,0 +1,39 @@ +-- Before UNIX epoch +WITH + toDateTime64('1959-09-16 19:20:12.999999998', 9, 'UTC') AS dt1, + toDateTime64('1959-09-16 19:20:12.999999999', 9, 'UTC') AS dt2 +SELECT + dt1 < dt2, + (dt1 + INTERVAL 1 NANOSECOND) = dt2, + (dt1 + INTERVAL 2 NANOSECOND) > dt2, + (dt1 + INTERVAL 3 NANOSECOND) > dt2; + +-- At UNIX epoch border +WITH + toDateTime64('1969-12-31 23:59:59.999999998', 9, 'UTC') AS dt1, + toDateTime64('1969-12-31 23:59:59.999999999', 9, 'UTC') AS dt2 +SELECT + dt1 < dt2, + (dt1 + INTERVAL 1 NANOSECOND) = dt2, + (dt1 + INTERVAL 2 NANOSECOND) > dt2, + (dt1 + INTERVAL 3 NANOSECOND) > dt2; + +-- After UNIX epoch +WITH + toDateTime64('2001-12-31 23:59:59.999999998', 9, 'UTC') AS dt1, + toDateTime64('2001-12-31 23:59:59.999999999', 9, 'UTC') AS dt2 +SELECT + dt1 < dt2, + (dt1 + INTERVAL 1 NANOSECOND) = dt2, + (dt1 + INTERVAL 2 NANOSECOND) > dt2, + (dt1 + INTERVAL 3 NANOSECOND) > dt2; + +-- At upper DT64 bound (DT64 precision is lower here by design) +WITH + toDateTime64('2282-12-31 23:59:59.999998', 6, 'UTC') AS dt1, + toDateTime64('2282-12-31 23:59:59.999999', 6, 'UTC') AS dt2 +SELECT + dt1 < dt2, + (dt1 + INTERVAL 1 MICROSECOND) = dt2, + (dt1 + INTERVAL 2 MICROSECOND) > dt2, + (dt1 + INTERVAL 3 MICROSECOND) > dt2; diff --git a/tests/queries/0_stateless/02313_test_fpc_codec.reference b/tests/queries/0_stateless/02313_test_fpc_codec.reference new file mode 100644 index 00000000000..23c75ed1ac0 --- /dev/null +++ b/tests/queries/0_stateless/02313_test_fpc_codec.reference @@ -0,0 +1,4 @@ +F64 +F32 +F64 +F32 diff --git a/tests/queries/0_stateless/02313_test_fpc_codec.sql b/tests/queries/0_stateless/02313_test_fpc_codec.sql new file mode 100644 index 00000000000..3b1127350f0 --- /dev/null +++ b/tests/queries/0_stateless/02313_test_fpc_codec.sql @@ -0,0 +1,121 @@ +DROP TABLE IF EXISTS codecTest; + +CREATE TABLE codecTest ( + key UInt64, + name String, + ref_valueF64 Float64, + ref_valueF32 Float32, + valueF64 Float64 CODEC(FPC), + valueF32 Float32 CODEC(FPC) +) Engine = MergeTree ORDER BY key; + +-- best case - same value +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'e()', e() AS v, v, v, v FROM system.numbers LIMIT 1, 100; + +-- good case - values that grow insignificantly +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'log2(n)', log2(n) AS v, v, v, v FROM system.numbers LIMIT 101, 100; + +-- bad case - values differ significantly +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'n*sqrt(n)', n*sqrt(n) AS v, v, v, v FROM system.numbers LIMIT 201, 100; + +-- worst case - almost like a random values +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'sin(n*n*n)*n', sin(n * n * n * n* n) AS v, v, v, v FROM system.numbers LIMIT 301, 100; + + +-- These floating-point values are expected to be BINARY equal, so comparing by-value is Ok here. + +-- referencing previous row key, value, and case name to simplify debugging. +SELECT 'F64'; +SELECT + c1.key, c1.name, + c1.ref_valueF64, c1.valueF64, c1.ref_valueF64 - c1.valueF64 AS dF64, + 'prev:', + c2.key, c2.ref_valueF64 +FROM + codecTest as c1, codecTest as c2 +WHERE + dF64 != 0 +AND + c2.key = c1.key - 1 +LIMIT 10; + + +SELECT 'F32'; +SELECT + c1.key, c1.name, + c1.ref_valueF32, c1.valueF32, c1.ref_valueF32 - c1.valueF32 AS dF32, + 'prev:', + c2.key, c2.ref_valueF32 +FROM + codecTest as c1, codecTest as c2 +WHERE + dF32 != 0 +AND + c2.key = c1.key - 1 +LIMIT 10; + +DROP TABLE IF EXISTS codecTest; + +CREATE TABLE codecTest ( + key UInt64, + name String, + ref_valueF64 Float64, + ref_valueF32 Float32, + valueF64 Float64 CODEC(FPC(4)), + valueF32 Float32 CODEC(FPC(4)) +) Engine = MergeTree ORDER BY key; + +-- best case - same value +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'e()', e() AS v, v, v, v FROM system.numbers LIMIT 1, 100; + +-- good case - values that grow insignificantly +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'log2(n)', log2(n) AS v, v, v, v FROM system.numbers LIMIT 101, 100; + +-- bad case - values differ significantly +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'n*sqrt(n)', n*sqrt(n) AS v, v, v, v FROM system.numbers LIMIT 201, 100; + +-- worst case - almost like a random values +INSERT INTO codecTest (key, name, ref_valueF64, valueF64, ref_valueF32, valueF32) + SELECT number AS n, 'sin(n*n*n)*n', sin(n * n * n * n* n) AS v, v, v, v FROM system.numbers LIMIT 301, 100; + + +-- These floating-point values are expected to be BINARY equal, so comparing by-value is Ok here. + +-- referencing previous row key, value, and case name to simplify debugging. +SELECT 'F64'; +SELECT + c1.key, c1.name, + c1.ref_valueF64, c1.valueF64, c1.ref_valueF64 - c1.valueF64 AS dF64, + 'prev:', + c2.key, c2.ref_valueF64 +FROM + codecTest as c1, codecTest as c2 +WHERE + dF64 != 0 +AND + c2.key = c1.key - 1 +LIMIT 10; + + +SELECT 'F32'; +SELECT + c1.key, c1.name, + c1.ref_valueF32, c1.valueF32, c1.ref_valueF32 - c1.valueF32 AS dF32, + 'prev:', + c2.key, c2.ref_valueF32 +FROM + codecTest as c1, codecTest as c2 +WHERE + dF32 != 0 +AND + c2.key = c1.key - 1 +LIMIT 10; + +DROP TABLE IF EXISTS codecTest; diff --git a/utils/self-extracting-executable/compressor.cpp b/utils/self-extracting-executable/compressor.cpp index 5478dfcb419..7108e34e220 100644 --- a/utils/self-extracting-executable/compressor.cpp +++ b/utils/self-extracting-executable/compressor.cpp @@ -8,6 +8,7 @@ #include #include #include +#include #include "types.h" @@ -54,7 +55,7 @@ int doCompress(char * input, char * output, off_t & in_offset, off_t & out_offse size_t compressed_size = ZSTD_compress2(cctx, output + out_offset, output_size, input + in_offset, input_size); if (ZSTD_isError(compressed_size)) { - fprintf(stderr, "Error (ZSTD): %zu %s\n", compressed_size, ZSTD_getErrorName(compressed_size)); + std::cerr << "Error (ZSTD): " << compressed_size << " " << ZSTD_getErrorName(compressed_size) << std::endl; return 1; } in_offset += input_size; @@ -79,7 +80,7 @@ int compress(int in_fd, int out_fd, int level, off_t & pointer, const struct sta ZSTD_CCtx * cctx = ZSTD_createCCtx(); if (cctx == nullptr) { - fprintf(stderr, "Error (ZSTD): failed to create compression context\n"); + std::cerr << "Error (ZSTD): failed to create compression context" << std::endl; return 1; } @@ -89,13 +90,13 @@ int compress(int in_fd, int out_fd, int level, off_t & pointer, const struct sta check_result = ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, level); if (ZSTD_isError(check_result)) { - fprintf(stderr, "Error (ZSTD): %zu %s\n", check_result, ZSTD_getErrorName(check_result)); + std::cerr << "Error (ZSTD): " << check_result << " " << ZSTD_getErrorName(check_result) << std::endl; return 1; } check_result = ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1); if (ZSTD_isError(check_result)) { - fprintf(stderr, "Error (ZSTD): %zu %s\n", check_result, ZSTD_getErrorName(check_result)); + std::cerr << "Error (ZSTD): " << check_result << " " << ZSTD_getErrorName(check_result) << std::endl; return 1; } @@ -254,7 +255,7 @@ int compressFiles(char* filenames[], int count, int output_fd, int level, const continue; } - printf("Size: %td\n", info_in.st_size); + std::cout << "Size: " << info_in.st_size << std::endl; /// Save umask files_data[i].umask = info_in.st_mode; @@ -317,7 +318,7 @@ int copy_decompressor(const char *self, int output_fd) if (sz < 0) perror(nullptr); else - fprintf(stderr, "Error: unable to extract decompressor.\n"); + std::cerr << "Error: unable to extract decompressor" << std::endl; close(input_fd); return 1; } @@ -398,7 +399,7 @@ int main(int argc, char* argv[]) struct stat info_out; if (stat(argv[start_of_files], &info_out) != -1 || errno != ENOENT) { - fprintf(stderr, "Error: output file [%s] already exists.\n", argv[start_of_files]); + std::cerr << "Error: output file [" << argv[start_of_files] << "] already exists" << std::endl; return 1; } @@ -419,7 +420,7 @@ int main(int argc, char* argv[]) return 1; } - printf("Compression with level %d\n", level); + std::cout << "Compression with level: " << level << std::endl; if (0 != compressFiles(&argv[start_of_files], argc - start_of_files, output_fd, level, info_out)) { printf("Compression failed.\n"); diff --git a/utils/self-extracting-executable/decompressor.cpp b/utils/self-extracting-executable/decompressor.cpp index e10d1413cd0..d5c9cf278de 100644 --- a/utils/self-extracting-executable/decompressor.cpp +++ b/utils/self-extracting-executable/decompressor.cpp @@ -11,6 +11,7 @@ #include #include #include +#include #include "types.h" @@ -21,7 +22,7 @@ int doDecompress(char * input, char * output, off_t & in_offset, off_t & out_off size_t decompressed_size = ZSTD_decompressDCtx(dctx, output + out_offset, output_size, input + in_offset, input_size); if (ZSTD_isError(decompressed_size)) { - fprintf(stderr, "Error (ZSTD): %zu %s\n", decompressed_size, ZSTD_getErrorName(decompressed_size)); + std::cerr << "Error (ZSTD):" << decompressed_size << " " << ZSTD_getErrorName(decompressed_size) << std::endl; return 1; } return 0; @@ -40,7 +41,7 @@ int decompress(char * input, char * output, off_t start, off_t end, size_t max_n ZSTD_DCtx * dctx = ZSTD_createDCtx(); if (dctx == nullptr) { - fprintf(stderr, "Error (ZSTD): failed to create decompression context\n"); + std::cerr << "Error (ZSTD): failed to create decompression context" << std::endl; return 1; } pid_t pid; @@ -52,7 +53,7 @@ int decompress(char * input, char * output, off_t start, off_t end, size_t max_n size = ZSTD_findFrameCompressedSize(input + in_pointer, max_block_size); if (ZSTD_isError(size)) { - fprintf(stderr, "Error (ZSTD): %td %s\n", size, ZSTD_getErrorName(size)); + std::cerr << "Error (ZSTD): " << size << " " << ZSTD_getErrorName(size) << std::endl; error_happened = true; break; } @@ -60,7 +61,7 @@ int decompress(char * input, char * output, off_t start, off_t end, size_t max_n decompressed_size = ZSTD_getFrameContentSize(input + in_pointer, max_block_size); if (ZSTD_isError(decompressed_size)) { - fprintf(stderr, "Error (ZSTD): %td %s\n", decompressed_size, ZSTD_getErrorName(decompressed_size)); + std::cerr << "Error (ZSTD): " << decompressed_size << " " << ZSTD_getErrorName(decompressed_size) << std::endl; error_happened = true; break; } @@ -171,8 +172,7 @@ int decompressFiles(int input_fd, char * path, char * name, bool & have_compress } if (fs_info.f_blocks * info_in.st_blksize < decompressed_full_size) { - fprintf(stderr, "Not enough space for decompression. Have %tu, need %zu.", - fs_info.f_blocks * info_in.st_blksize, decompressed_full_size); + std::cerr << "Not enough space for decompression. Have " << fs_info.f_blocks * info_in.st_blksize << ", need " << decompressed_full_size << std::endl; return 1; }