beyond1920 commented on code in PR #11559: URL: https://github.com/apache/hudi/pull/11559#discussion_r1698757780
########## rfc/rfc-80/rfc-80.md: ########## @@ -0,0 +1,180 @@ +<!-- + Licensed to the Apache Software Foundation (ASF) under one or more + contributor license agreements. See the NOTICE file distributed with + this work for additional information regarding copyright ownership. + The ASF licenses this file to You under the Apache License, Version 2.0 + (the "License"); you may not use this file except in compliance with + the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. +--> +# RFC-80: Support column families for wide tables + +## Proposers +- @xiarixiaoyao +- @wombatu-kun + +## Approvers + - @vinothchandar + - @danny0405 + - @beyond1920 + +## Status + +JIRA: https://issues.apache.org/jira/browse/HUDI-7947 + +## Abstract + +In streaming processing, there are often scenarios where the table is widened. The current mainstream real-time wide table concatenation is completed through Flink's multi-layer join; +Flink's join will cache a large amount of data in the state backend. As the data set increases, the pressure on the Flink task state backend will gradually increase, and may even become unavailable. +In multi-layer join scenarios, this problem is more obvious. +1.x also supports partial updates being encoded in logfiles. That should be able to handle this scenario. But even with partial-update, the column families will reduce write amplification on compaction. + +So, main gains of clustering columns for wide tables are: +Write performance: +- Writing is similar to ordinary bucket writing, but it involves splitting column clusters and sorting. Therefore, the writing performance of full data is 10% lower than that of native bucket writing. +- However, if only some columns are updated among a large number of columns, the writing efficiency is much faster than that of non-column clustered tables. (Note: 1.0 introduces a partial update functionality, that can also avoid such writing costs, by writing only the changed columns) + +Read performance: +Since the data is already sorted when it is written, the SortMerge method can be used directly to merge the data; compared with the native bucket data reading performance is improved a lot, and the memory consumption is reduced significantly. + +Compaction performance: +The logic of compaction and reading is the same. Compaction costs across column families is where there real savings are. + +The log merge we can make it pluggable to decide between hash or sort merge - we need to introduce new log headers or standard mechanism for merging to determine if base file or log files are sorted. + +## Background +Currently, Hudi organizes data according to fileGroup granularity. The fileGroup is further divided into column clusters to introduce the columnFamily concept. +The organizational form of Hudi files is divided according to the following rules: +The data in the partition is divided into buckets according to hash (each bucket maps to a file group); the files in each bucket are divided according to columnFamily; multiple colFamily files in the bucket form a completed fileGroup; when there is only one columnFamily, it degenerates into the native Hudi bucket table. + + + +## Implementation +This feature should be implemented for both Spark and Flink. So, a table written by Flink this way, also can be read by Spark. + +### Constraints and Restrictions +1. The overall design relies on the non-blocking concurrent writing feature of Hudi 1.0. +2. Lower version Hudi cannot read and write column family tables. +3. Only MOR bucketed tables support setting column families. +4. Column families do not support repartitioning and renaming. +5. Schema evolution does not take effect on the current column family table. +6. Like native bucket tables, clustering operations are not supported. + +### Model change +After the column family is introduced, the storage structure of the entire Hudi bucket table changes: + + + +The bucket is divided into multiple columnFamilies by column cluster. When columnFamily is 1, it will automatically degenerate into the native bucket table. + + + +### Proposed Storage Format Changes +After splitting the fileGroup by columnFamily, the naming rules for base files and log files change. We add the cfName suffix at the end of all file names to facilitate Hudi itself to distinguish column families. If it's not present, we assume default column family. +So, new file name templates will be as follows: +- Base file: [file_id]\_[write_token]\_[begin_time]**[#cfName]**.[extension] +- Log file: [file_id]\_[begin_instant_time]**[#cfName]**.log.[version]_[write_token] + +Also, we should evolve the metadata table files schema to additionally track a column family name. + +### Specifying column families when creating a table +In the table creation statement, column family division is specified in the options/tblproperties attribute; +Column family attributes are specified in key-value mode: +* Key is the column family name. Format: hoodie.colFamily. Column family name naming rules specified. +* Value is the specific content of the column family: it consists of all the columns included in the column family plus the precombine field. Format: " col1,col2...colN; precombineCol", the column family list and the preCombine field are separated by ";"; in the column family list the columns are split by ",". + +Constraints: The column family list must contain the primary key, and columns contained in different column families cannot overlap except for the primary key. The preCombie field does not need to be specified. If not specified, the primary key will be taken by default. + +After the table is created, the column family attributes will be persisted to hoodie's metadata for subsequent use. + +### Adding and deleting column families in existing table +Use the SQL alter command to modify the column family attributes and persist it: +* Execute ALTER TABLE table_name SET TBLPROPERTIES ('hoodie.columnFamily.k'='a,b,c;a'); to add a new column family. +* Execute ALTER TABLE table_name UNSET TBLPROPERTIES('hoodie.columnFamily.k'); to delete the column family. + +Specific steps are as follows: +1. Execute the ALTER command to modify the column family +2. Verify whether the column family modified by alter is legal. Column family modification must meet the following conditions, otherwise the verification will not pass: + * The column family name of an existing column family cannot be modified. + * Columns in other column families cannot be divided into new column families. + * When creating a new column family, it must meet the format requirements from previous chapter. +3. Save the modified column family to the .hoodie directory. + +### Writing data +The Hudi kernel divides the input data according to column families; the data belonging to a certain column family is sorted and directly written to the corresponding column family log file. + + + +Specific steps: +1. The engine divides the written data into buckets according to hash and shuffles the data (the writing engine completes it by itself and is consistent with the current writing of the native bucket). +2. The Hudi kernel sorts the data to be written to each bucket by primary key (both Spark and Flink has its own ExternalSorter, we can refer those ExternalSorter to finish sort). +3. After sorting, split the data into column families. +4. Write the segmented data into the log file of the corresponding column family. + +#### Common API interface +After the table columns are clustered, the writing process includes the process of sorting and splitting the data compared to the original bucket bucketing. A new append interface needs to be introduced to support column families. +Introduce ColumnFamilyAppendHandle extend AppendHandle to implement column family writing. + + + +### Reading data +#### ColumnFamilyReader and RowReader + + +Hudi internal row reader reading steps: +1. Hudi organizes files by column families to be read. +2. Introduce familyReader to merge and read each column family's own baseFile and logfile to achieve column family-level data reading. + * Since log files are written after being sorted by primary key, familyReader merges its own baseFile and logFile by primary key using sortMerge. + * familyReader supports upstream incoming column pruning to reduce IO overhead when reading data. + * During the merging process, if the user specifies the precombie field for the column family, the merging strategy will be selected based on the precombie field. This logic reuses Hudi's own precombine logic and does not need to be modified. +3. Row reader merges the data read by multiple familyReaders according to the primary key. + +Since the data read by each familyReader is sorted by the primary key, the row reader merges the data read by each familyReader in the form of sortMergeJoin and returns the complete data. + +The entire reading process involves a large amount of data merging, but because the data itself is sorted, the memory consumption of the entire merging process is very low and the merging is fast. Compared with Hudi's native merging method, the memory pressure and the merging time are significantly reduced. + +#### Engine reads pseudo process + + +1) The engine itself delivers the data files that need to be scanned to executor/woker/taskmanger. +2) executor/worker/taskmanger calls Hudi’s rowReader interface and passes in column clipping and filter conditions to rowReader. +3) The Hudi kernel completes the data reading of rowReader and returns complete data. The data format is Avro. +4) The engine gets the Avro format data and needs to convert it into the data format it needs. For example, spark needs to be converted into unsaferow, hetu into block, flink into row, and hive into arrayWritable. + +### Column family level compaction +Extend Hudi's compaction schedule module to merge each column family's own base file and log file: + + + +### Full compaction +Extend Hudi's compaction schedule module to merge and update all column families in the entire table. +After merging at the column family level, multiple column families are finally merged into a complete row and saved. + + + +ColumnFamily can be used not only in ML scenarios, but also can be used to simulate multi-stream join and concatenation of wide tables. 。 +In simulation of multi-stream join scenario, Hudi should produce complete rows, so full compaction is needed. Review Comment: Just curiously, do we plan to let log compaction support wide table with column family as well? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
