This is an automated email from the ASF dual-hosted git repository.
lzljs3620320 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/paimon-website.git
The following commit(s) were added to refs/heads/master by this push:
new 5f22f9b3f Add release note for 0.9
5f22f9b3f is described below
commit 5f22f9b3ff0a29c3cd2d66fac16e7d6857fad5a8
Author: Jingsong <jingsongl...@gmail.com>
AuthorDate: Fri Sep 13 19:17:27 2024 +0800
Add release note for 0.9
---
main/template/nav.html | 2 +-
pages/content/releases/release-0.9.md | 331 ++++++++++++++++++++++++++++++++++
2 files changed, 332 insertions(+), 1 deletion(-)
diff --git a/main/template/nav.html b/main/template/nav.html
index fdade3090..08948081a 100644
--- a/main/template/nav.html
+++ b/main/template/nav.html
@@ -19,7 +19,7 @@
<a class="nav-link"
href="https://paimon.apache.org/downloads.html";>Downloads</a>
</li>
<li class="nav-item active px-3">
- <a class="nav-link"
href="https://paimon.apache.org/release-0.8.2.html";>Releases</a>
+ <a class="nav-link"
href="https://paimon.apache.org/release-0.9.html";>Releases</a>
</li>
<li class="nav-item active px-3">
<a class="nav-link" href="https://github.com/apache/paimon/";>Github</a>
diff --git a/pages/content/releases/release-0.9.md
b/pages/content/releases/release-0.9.md
new file mode 100644
index 000000000..728a48bcb
--- /dev/null
+++ b/pages/content/releases/release-0.9.md
@@ -0,0 +1,331 @@
+---
+title: "Release 0.9"
+weight: 992
+type: docs
+aliases:
+- /release-0.9.html
+---
+<!--
+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.
+-->
+
+# Apache Paimon 0.9 Available
+
+Sep 13, 2024 - Jingsong Lee (jingsongl...@gmail.com)
+
+The Apache Paimon PMC officially announces the release of Apache Paimon 0.9.0.
This version was developed over the
+course of four months, with the participation of 80 contributors, resulting in
more than 600 commits.
+
+Thank you to all contributors for your support!
+
+The community has decided that the next version will be 1.0, marking that most
functionalities of Apache Paimon have
+become relatively mature and stable.
+
+## Version Overview
+
+Paimon's long-term plan is to become a unified lake storage format that meets
the main requirements for minute level
+big data: offline batch computing, real-time stream computing, and OLAP
computing.
+
+Notable changes in this version include:
+
+1. Paimon Branch: The branch functionality is now officially production-ready,
and the introduction of the 'scan.fallback-branch'
+ feature helps to better unify stream and batch storage for businesses.
+2. Universal Format: This version introduces native Iceberg compatibility. You
can enable Iceberg compatibility mode, and
+ Paimon will generate Iceberg-compatible snapshots in real-time, allowing
you to use the Iceberg ecosystem to read this Paimon table.
+3. Caching Catalog: This version introduces the implementation of Caching
Catalog by default. Table metadata and manifest
+ files will be cached in the catalog, which can accelerate OLAP query
performance.
+4. Improvements in Bucketed Append Table Availability: The small file issue
has been significantly alleviated, and it
+ can be applied to bucketed joins in Spark, reducing shuffles during joins.
+5. Support for DELETE & UPDATE & MERGEINTO in Append Tables: This version
introduces support for DELETE, UPDATE, and
+ MERGE INTO in append tables. You can modify and delete records in append
tables using Spark SQL, and it also supports
+ Deletion Vectors mode.
+
+## Compatibility Changes
+
+The following changes may impact the compatibility of your usage.
+
+### Bucketed Append Tables
+
+When defining a table without a primary key, if the number of 'bucket' is
defined, the table is considered a Bucketed
+Append table, previously referred to as an Append Queue table, as it is more
commonly used in ordered stream writes and
+reads. The issue of small files has been significantly alleviated, and it can
be applied in Spark for Bucketed Joins,
+reducing shuffles during joins.
+
+Here are some changes to its default values:
+
+1. Bucketed Append tables are prohibited from being defined without a
bucket-key. The behavior in previous versions was
+ to hash the entire row to determine the corresponding bucket, which is an
unintuitive behavior. We recommend using
+ an older version of Paimon to refresh the data (write into a new valid
table).
+2. The default value of the 'compaction.max.file-num' option for Bucketed
Append tables has been adjusted to 5, meaning
+ there will be fewer small files within a single bucket to avoid excessive
small files affecting production usability.
+
+Despite this, we still recommend that you avoid defining Bucketed Append
tables unless necessary; the default bucket -1
+mode is more user-friendly.
+
+### File Format and Compression
+
+The Paimon community is focused on improving overall performance under default
options. The following option default
+values have been modified in version 0.9:
+
+1. File Format 'file.format': The default has changed from ORC to Parquet.
There is no essential difference between these
+ formats, but Parquet generally performs better, and the community has
completed all capabilities for Parquet, including
+ support for nested types, Filter PushDown, and more.
+2. File Size 'target-file-size': The size for primary key tables remains at
128MB, while the default for non-primary key
+ tables (Append tables) has been adjusted to 256MB.
+3. Compression Default 'file.compression': The default has changed from LZ4 to
ZSTD, with the default ZSTD compression
+ level set to 1. You can adjust the compression level using
'file.compression.zstd-level', consuming more CPU for
+ greater compression rates.
+4. Local Spill Compression Level 'spill-compression.zstd-level': Likewise,
local spill can also achieve greater
+ compression rates by adjusting the level.
+
+### CDC Ingestion
+
+The dependency for Flink CDC has been upgraded to version 3.1. Since Flink CDC
has become a sub-project of Flink in this
+version, the package names have been modified, rendering older versions of CDC
unsupported. MySQL CDC, MongoDB CDC, and
+Postgres CDC will be affected.
+
+## Paimon Branch
+
+Branching is an interesting feature that allows us to manipulate Paimon tables
in a manner similar to Git. It has
+reached a production-ready state in Paimon 0.9, and Alibaba's internal teams
are already using it in production
+environments for tasks such as data correction and stream-batch integration.
+
+For example, you can use branches for data correction:
+
+```sql
+-- create branch named 'branch1' from tag 'tag1'
+CALL sys.create_branch('default.T', 'branch1', 'tag1');
+
+-- write to branch 'branch1'
+INSERT INTO `t$branch_branch1` SELECT ...
+
+-- read from branch 'branch1'
+SELECT * FROM `t$branch_branch1`;
+
+-- replace master branch with 'branch1'
+CALL sys.fast_forward('default.T', 'branch1');
+```
+
+You can also use branches for unified stream-batch storage. You can set up a
separate stream branch and then configure
+'scan.fallback-branch'. This way, when a batch processing job reads from the
current branch and a partition is missing,
+it will attempt to read that partition from the fallback branch.
+
+Suppose you create a Paimon table partitioned by date. You have a long-running
streaming job that inserts records into
+Paimon so that today's data can be queried in a timely manner. You also have a
nightly batch processing job that overwrites
+the partitions in Paimon to ensure data accuracy. When you query this Paimon
table, you want to read first from the results
+of the batch processing job. However, if a specific partition (for example,
today's partition) is missing in its results,
+you want to read from the results of the streaming job. In this case, you can
create a branch for the streaming job and
+set 'scan.fallback-branch' to that streaming branch.
+
+```sql
+-- create a branch for streaming job (realtime)
+CALL sys.create_branch('default.T', 'rt');
+
+-- set primary key and bucket number for the branch
+ALTER TABLE `T$branch_rt` SET (
+ 'primary-key' = 'dt,name',
+ 'bucket' = '2',
+ 'changelog-producer' = 'lookup'
+);
+
+-- set fallback branch
+ALTER TABLE T SET (
+ 'scan.fallback-branch' = 'rt'
+);
+
+SELECT * FROM T;
+```
+
+## Universal Format
+
+Paimon's Universal Format allows you to use Iceberg clients or compute engines
to read data within Paimon. By using the
+'metadata.iceberg-compatible' option, Paimon automatically generates Iceberg
snapshots in the filesystem when creating
+snapshots, without requiring any additional dependencies or concerns about
governance-related issues.
+
+Notable points include:
+
+1. Iceberg metadata is stored in the file system (corresponding to Iceberg's
HadoopCatalog). For example, you can read
+ it using Spark DataFrame: spark.read.format("iceberg").load("path").
+2. Iceberg views are read-only, and writing through this method may corrupt
the table.
+3. For primary key tables, Iceberg views can only access the highest level
(LSM Level) files. You can configure
+ 'compaction.optimization-interval' to control the visibility of the data.
+
+## Caching Catalog
+
+Paimon's metadata is stored in the filesystem, which leads to frequent access
to the filesystem during the planning
+phase in compute engines, potentially impacting single-point performance,
especially in object storage where this cost
+is even higher.
+
+This version introduces the implementation of Caching Catalog by default,
which will be enabled automatically
+(it only caches manifest files smaller than 1MB by default). This can
accelerate the performance of OLAP queries.
+
+You can control the behavior of the cache using the following options:
+
+| option | default | description
|
+|-------------------------------------|---------|---------------------------------------------------------------------------------|
+| cache-enabled | true | Controls whether the catalog
will cache databases, tables and manifests. |
+| cache.expiration-interval | 1 min | Controls the duration for
which databases and tables in the catalog are cached. |
+| cache.manifest.max-memory | (none) | Controls the maximum memory
to cache manifest content. |
+| cache.manifest.small-file-memory | 128 mb | Controls the cache memory to
cache small manifest files. |
+| cache.manifest.small-file-threshold | 1 mb | Controls the threshold of
small manifest file. |
+
+## Deletion Vectors
+
+The Deletion Vectors mode is fully available in version 0.9.
+
+For primary key tables, the Deletion Vectors mode now supports asynchronous
compaction (defaulting to semi-synchronous),
+significantly enhancing its usability without heavily impacting checkpoints.
Since the DV mode requires local disk usage,
+we recommend using SSDs for local disks; performance can be quite poor with
lower-quality HDDs.
+
+For non-primary key tables (Append tables), version 0.9 supports DELETE,
UPDATE, and MERGE INTO operations via Spark SQL,
+making Paimon append tables resemble complete database tables, enabling
fine-grained modifications and deletions for users.
+
+Moreover, non-primary key tables also support the Deletion Vectors mode.
Before enabling it, deletions and modifications
+are done using copy-on-write; once the DV mode is enabled, deletions and
modifications switch to merge-on-write.
+Deletion files will be removed during compaction.
+
+## Core
+
+### New Aggregation Functions
+
+The new aggregation functions include hll_sketch, theta_sketch, rbm32, and
rbm64. You can use these sketch-related
+functions to estimate COUNT DISTINCT.
+
+Paimon does not support custom aggregation functions but encourages you to
propose enhancements to the built-in function library in the community.
+
+### Universal File Index
+
+The universal file index now supports the Bitmap type and also supports the
REWRITE CALL command, allowing you to regenerate the corresponding index.
+
+Bitmap indexes perform well under joint filter conditions across multiple
fields.
+
+### Historical Partition Compact
+
+Additionally, if your table is a partitioned table, although Paimon has
built-in automatic compaction, its historical
+partitions may not have undergone full compaction. In version 0.9, we
introduced partition_idle_time to automatically
+select partitions that haven't been updated for full compaction, reducing
small files and improving query performance.
+
+## Flink
+
+### Cluster
+
+Clustering allows you to organize data in an Append table based on the values
of certain columns during the write process.
+This data organization method can significantly improve the efficiency of
downstream tasks when reading data, as it enables
+faster and more targeted data queries. This feature only supports Append
tables (with bucket = -1) and batch execution mode.
+
+```sql
+INSERT INTO my_table /*+ OPTIONS('sink.clustering.by-columns' = 'a,b') */
SELECT * FROM source;
+```
+
+### Partition Mark Done
+
+For partitioned tables, each partition may need to be scheduled to trigger
downstream batch computations. Therefore,
+it is essential to choose the right timing to indicate that a partition is
ready for scheduling, while minimizing
+data drift during the scheduling process. We refer to this process as "Marking
a Partition as Done."
+
+```sql
+CREATE TABLE my_partitioned_table (
+ f0 INT,
+ f1 INT,
+ f2 INT,
+ ...
+ dt STRING
+) PARTITIONED BY (dt) WITH (
+ 'partition.timestamp-formatter'='yyyyMMdd',
+ 'partition.timestamp-pattern'='$dt',
+ 'partition.time-interval'='1 d',
+ 'partition.idle-time-to-done'='15 m'
+);
+```
+
+1. First, it is necessary to define the time parsing for the partition and the
time interval between partitions to
+ determine when it is appropriate to mark a partition as complete.
+2. Second, idle time needs to be defined, which determines how long a
partition must wait without new data before it
+ can be marked as complete.
+3. Third, by default, marking a partition as complete will create a _SUCCESS
file. You can also configure
+ partition.mark-done-action to define specific actions.
+
+### Table Clone
+
+In Paimon 0.9, the clone table action is supported for data migration.
Currently, it only clones the table files used
+by the latest snapshot. If the table you are cloning hasn’t been modified
during this period, it is recommended to
+submit a Flink batch job for better performance. However, if you wish to clone
the table while writing, you should
+submit a Flink stream processing job for automatic fault recovery. This
command can assist you with convenient data
+backup and migration.
+
+### Procedures
+
+Paimon 0.9 introduces a large number of procedures. Additionally, Flink
procedures now support the latest version of
+named procedures, making your execution more convenient without the need to
specify all parameters forcefully.
+
+## Spark
+
+### Dynamic Options
+
+Historically, Spark SQL lacked the capability for dynamic parameters, while
Flink SQL offers dynamic options that are
+very convenient. In version 0.9, Spark introduces this capability through the
SET command. The SET command allows for
+specific Paimon configurations by requiring the prefix spark.paimon. to be
added.
+
+```sql
+-- set paimon conf
+SET spark.paimon.file.block-size=512M;
+
+-- reset conf
+RESET spark.paimon.file.block-size;
+```
+
+### Bucketed Join
+
+Hive has a feature called Bucketed Join. When two tables are bucketed by the
same column, a Bucketed Join can be
+performed without the need for shuffling the data; the join can be executed
directly on the buckets, making it very efficient.
+
+Buckets are one of the core concepts in Paimon. Previously, Paimon lacked
integration with compute engines, preventing
+the utilization of this feature for optimization.
+
+In version 0.9, Paimon achieves deep integration with Spark SQL, enabling this
optimization:
+
+- Optimized Execution: By leveraging the bucket structure during join
operations, Paimon can significantly reduce
+ the overhead of data shuffling, leading to better performance on join
queries.
+- Seamless Configuration: Users can easily define buckets when creating
tables, allowing for straightforward
+ implementation of bucketed joins in their Spark SQL queries.
+
+```sql
+-- Enable bucketing optimization
+SET spark.sql.sources.v2.bucketing.enabled=true;
+
+-- Bucketed Join
+SELECT * FROM t1 JOIN t2 on t1.id = t2.id
+```
+
+As long as both tables involved in the join are bucketed tables (whether they
are primary key tables or non-primary key
+tables) and the bucket-key field corresponds to the join field, an efficient
Bucketed Join will take place. This
+optimization minimizes data shuffling and enhances query performance, making
it a valuable feature for handling large
+datasets in Paimon.
+
+### Writing to Dynamic Bucket Tables
+
+- Optimized Writing: Spark SQL can now write to dynamic bucket tables with
optimizations that reduce data shuffling
+ during the first write operation.
+- Cross-Partition Updates: Spark SQL now supports writing to tables with
cross-partition updates, although the overall
+ efficiency for these operations is still not optimal.
+
+## Other Progress
+
+- Paimon Web UI: The Web UI is being released; feel free to try it out!
+- Paimon Python: The release process for version 0.1 is expected to start soon.
+- Paimon Rust: In development, with an expected release of a readable version
in 0.1.
