codope commented on code in PR #9790:
URL: https://github.com/apache/hudi/pull/9790#discussion_r1345180240


##########
website/releases/release-0.14.0.md:
##########
@@ -0,0 +1,339 @@
+---
+title: "Release 0.14.0"
+sidebar_position: 1
+layout: releases
+toc: true
+---
+import Tabs from '@theme/Tabs';
+import TabItem from '@theme/TabItem';
+
+## [Release 
0.14.0](https://github.com/apache/hudi/releases/tag/release-0.14.0) 
([docs](/docs/quick-start-guide))
+Apache Hudi 0.14.0 marks a significant milestone with a range of new 
functionalities and enhancements. 
+These include the introduction of Record Level Index, automatic generation of 
record keys, the `hudi_table_changes` 
+function for incremental reads, and more. Notably, this release also 
incorporates support for Spark 3.4. On the Flink 
+front, version 0.14.0 brings several exciting features such as consistent 
hashing index support, Flink 1.17 support, and U
+pdate and Delete statement support. Additionally, this release upgrades the 
Hudi table version, prompting users to consult
+the Migration Guide provided below. We encourage users to review the [release 
highlights](#release-highlights),
+[breaking changes](#breaking-changes), and [behavior 
changes](#behavior-changes) before 
+adopting the 0.14.0 release.
+
+
+
+## Migration Guide
+In version 0.14.0, we've made changes such as the removal of compaction plans 
from the ".aux" folder and the introduction
+of a new log block version. As part of this release, the table version is 
updated to version `6`. When running a Hudi job 
+with version 0.14.0 on a table with an older table version, an automatic 
upgrade process is triggered to bring the table 
+up to version `6`. This upgrade is a one-time occurrence for each Hudi table, 
as the `hoodie.table.version` is updated in
+the property file upon completion of the upgrade. Additionally, a command-line 
tool for downgrading has been included, 
+allowing users to move from table version `6` to `5`, or revert from Hudi 
0.14.0 to a version prior to 0.14.0. To use this 
+tool, execute it from a 0.14.0 environment. For more details, refer to the 
+[hudi-cli](/docs/cli/#upgrade-and-downgrade-table).
+
+:::caution
+If migrating from an older release (pre 0.14.0), please also check the upgrade 
instructions from each older release in
+sequence.
+:::
+
+### Bundle Updates
+
+#### New Spark Bundles
+In this release, we've expanded our support to include bundles for both Spark 
3.4 
+([hudi-spark3.4-bundle_2.12](https://mvnrepository.com/artifact/org.apache.hudi/hudi-spark3.4-bundle_2.12))
 
+and Spark 3.0 
([hudi-spark3.0-bundle_2.12](https://mvnrepository.com/artifact/org.apache.hudi/hudi-spark3.0-bundle_2.12)).
+Please note that, the support for Spark 3.0 had been discontinued after Hudi 
version 0.10.1, but due to strong community 
+interest, it has been reinstated in this release.
+
+### Breaking Changes
+
+#### INSERT INTO behavior with Spark SQL
+Before version 0.14.0, data ingested through `INSERT INTO` in Spark SQL 
followed the upsert flow, where multiple versions 
+of records would be merged into one version. However, starting from 0.14.0, 
we've altered the default behavior of 
+`INSERT INTO` to utilize the `insert` flow internally. This change 
significantly enhances write performance as it 
+bypasses index lookups.
+
+If a table is created with a *preCombine* key, the default operation for 
`INSERT INTO` remains as `upsert`. Conversely, 
+if no *preCombine* key is set, the underlying write operation for `INSERT 
INTO` defaults to `insert`. Users have the 
+flexibility to override this behavior by explicitly setting values for the 
config 
+[`hoodie.spark.sql.insert.into.operation`](https://hudi.apache.org/docs/configurations#hoodiesparksqlinsertintooperation)
 
+as per their requirements. Possible values for this config include `insert`, 
`bulk_insert`, and `upsert`.
+
+Additionally, in version 0.14.0, we have **deprecated** two related older 
configs:
+- `hoodie.sql.insert.mode`
+- `hoodie.sql.bulk.insert.enable`.
+
+### Behavior changes
+
+#### Simplified duplicates handling with Inserts in Spark SQL
+In cases where the operation type is configured as `insert` for the Spark SQL 
`INSERT INTO` flow, users now have the 
+option to enforce a duplicate policy using the configuration setting 
+[`hoodie.datasource.insert.dup.policy`](https://hudi.apache.org/docs/configurations#hoodiedatasourceinsertduppolicy).
 
+This policy determines the action taken when incoming records being ingested 
already exist in storage. The available 
+values for this configuration are as follows:
+
+- `none`: No specific action is taken, allowing duplicates to exist in the 
Hudi table if the incoming records contain duplicates.
+- `drop`: Matching records from the incoming writes will be dropped, and the 
remaining ones will be ingested.
+- `fail`: The write operation will fail if the same records are re-ingested. 
In essence, a given record, as determined 
+by the key generation policy, can only be ingested once into the target table.
+
+With this addition, an older related configuration setting, 
+[`hoodie.datasource.write.insert.drop.duplicates`](https://hudi.apache.org/docs/configurations#hoodiedatasourcewriteinsertdropduplicates),
 
+is now deprecated. The newer configuration will take precedence over the old 
one when both are specified. If no specific 
+configurations are provided, the default value for the newer configuration 
will be assumed. Users are strongly encouraged 
+to migrate to the use of these newer configurations. 
+
+
+#### Compaction with MOR table
+For Spark batch writers (both the Spark datasource and Spark SQL), compaction 
is automatically enabled by default for 
+MOR (Merge On Read) tables, unless users explicitly override this behavior. 
Users have the option to disable compaction 
+explicitly by setting 
[`hoodie.compact.inline`](https://hudi.apache.org/docs/configurations#hoodiecompactinline)
 to false. 
+In case users do not override this configuration, compaction may be triggered 
for MOR tables approximately once every 
+5 delta commits (the default value for 
+[`hoodie.compact.inline.max.delta.commits`](https://hudi.apache.org/docs/configurations#hoodiecompactinlinemaxdeltacommits)).
+
+
+#### HoodieDeltaStreamer renamed to HoodieStreamer
+Starting from version 0.14.0, we have renamed 
[HoodieDeltaStreamer](https://github.com/apache/hudi/blob/84a80e21b5f0cdc1f4a33957293272431b221aa9/hudi-utilities/src/main/java/org/apache/hudi/utilities/deltastreamer/HoodieDeltaStreamer.java)
+to 
[HoodieStreamer](https://github.com/apache/hudi/blob/84a80e21b5f0cdc1f4a33957293272431b221aa9/hudi-utilities/src/main/java/org/apache/hudi/utilities/streamer/HoodieStreamer.java).
 
+We have ensured backward compatibility so that existing user jobs remain 
unaffected. However, in upcoming 
+releases, support for Deltastreamer might be discontinued. Hence, we strongly 
advise users to transition to using 
+HoodieStreamer instead.
+
+
+#### MERGE INTO JOIN condition 
+Starting from version 0.14.0, Hudi has the capability to automatically 
generate primary record keys when users do not 
+provide explicit specifications. This enhancement enables the `MERGE INTO 
JOIN` clause to reference any data column for 
+the join condition in Hudi tables where the primary keys are generated by Hudi 
itself. However, in cases where users 
+configure the primary record key, the join condition still expects the primary 
key fields as specified by the user.
+
+
+## Release Highlights
+
+### Record Level Index
+Hudi version 0.14.0, introduces a new index implementation -  
+[Record Level 
Index](https://github.com/apache/hudi/blob/master/rfc/rfc-8/rfc-8.md#rfc-8-metadata-based-record-index).
 
+The Record level Index significantly enhances write performance for large 
tables by efficiently storing per-record 
+locations and enabling swift retrieval during index lookup operations. It can 
effectively replace other 
+[Global 
indices](https://hudi.apache.org/docs/next/indexing#global-and-non-global-indexes)
 like Global_bloom, 
+Global_Simple, or Hbase, commonly used in Hudi.
+
+Bloom and Simple Indexes exhibit slower performance for large datasets due to 
the high costs associated with gathering 
+index data from various data files during lookup. Moreover, these indexes do 
not preserve a one-to-one record-key to 
+record file path mapping; instead, they deduce the mapping through an 
optimized search at lookup time. The per-file 
+overhead required by these indexes makes them less effective for datasets with 
a larger number of files or records.
+
+On the other hand, the Hbase Index saves a one-to-one mapping for each record 
key, resulting in fast performance that 
+scales with the dataset size. However, it necessitates a separate HBase 
cluster for maintenance, which is operationally 
+challenging and resource-intensive, requiring specialized expertise.
+
+The Record Index combines the speed and scalability of the HBase Index without 
its limitations and overhead. Being a 
+part of the HUDI Metadata Table, any future performance enhancements in writes 
and queries will automatically translate 
+into improved performance for the Record Index. Adopting the Record Level 
Index has the potential to boost index lookup 
+performance by 4 to 10 times, depending on the workload, even for extremely 
large-scale datasets (e.g., 1TB).
+

Review Comment:
   I see it's added here.



-- 
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]

Reply via email to