With 0.11.0, we have added a checksum mechanism for validating the
hoodie.proerties, which introduces a new table version,
Whenever a Hudi job is launched with this release on a table with older table version, an upgrade step is executed automatically to upgrade the table to table version
This automatic upgrade step happens just once per Hudi table as the hoodie.table.version will be updated in property file after upgrade is completed.
Similarly, a command line tool for Downgrading (command - downgrade) is added if in case some users want to downgrade Hudi
from table version
3 or move from Hudi 0.11.0 to pre 0.11.0. This needs to be executed from a 0.11.0 hudi-cli binary/script.
Bundle usage updates
- Spark bundle for 3.0.x is no longer officially supported. Users are encouraged to upgrade to Spark 3.2 or 3.1.
- Users are encouraged to use bundles with specific Spark version in the name (
hudi-sparkX.Y-bundle) and move away from the legacy bundles (
- Spark or Utilities bundle no longer requires additional
spark-avropackage at runtime; the option
--package org.apache.spark:spark-avro_2.1*:*can be dropped.
- For MOR tables,
hoodie.datasource.write.precombine.fieldis required for both write and read.
- Only set
hoodie.datasource.write.drop.partition.columns=truewhen work with BigQuery integration.
- For Spark readers that rely on extracting physical partition path,
hoodie.datasource.read.extract.partition.values.from.path=trueto stay compatible with existing behaviors.
- Default index type for Spark was changed from
SIMPLE(HUDI-3091). If you currently rely on the default
BLOOMindex type, please update your configuration accordingly.
In 0.11.0, we enable the metadata table with synchronous updates and metadata-table-based file listing
by default for Spark writers, to improve the performance of partition and file listing on large Hudi tables. On the
reader side, users need to set it to
true benefit from it. The metadata table and related file listing functionality
can still be turned off by setting
hoodie.metadata.enable=false. Due to this, users deploying Hudi with async table
services need to configure a locking service. If this feature is not relevant for you, you can set
hoodie.metadata.enable=false additionally and use Hudi as before.
We introduce a multi-modal index in metadata table to drastically improve the lookup performance in file index and query latency with data skipping. Two new indices are added to the metadata table
- bloom filter index containing the file-level bloom filter to facilitate key lookup and file pruning as a part of bloom index during upserts by the writers
- column stats index containing the statistics of all/interested columns to improve file pruning based on key and column value range in both the writer and the reader, in query planning in Spark for example.
They are disabled by default. You can enable them by setting
Refer to the metadata table guide for detailed instructions on upgrade and deployment.
Data Skipping with Metadata Table
With the added support for Column Statistics in metadata table, Data Skipping is now relying on the metadata table's
Column Stats Index (CSI) instead of its own bespoke index implementation (comparing to Spatial Curves added in 0.10.0),
allowing to leverage Data Skipping for all datasets regardless of whether they execute layout optimization procedures (
like clustering) or not. To benefit from Data Skipping, make sure to set
hoodie.enable.data.skipping=true on both
writer and reader, as well as enable metadata table and Column Stats Index in the metadata table.
Data Skipping supports standard functions (as well as some common expressions) allowing you to apply common standard
transformations onto the raw data in your columns within your query's filters. For example, if you have column "ts" that
stores timestamp as string, you can now query it using human-readable dates in your predicate like
date_format(ts, "MM/dd/yyyy" ) < "04/01/2022".
Note: Currently Data Skipping is only supported in COW tables and MOR tables in read-optimized mode. The work of full support for MOR tables is tracked in HUDI-3866
Refer to the performance guide for more info.
In 0.11.0, we added a new asynchronous service for indexing to our rich set of table services. It allows users to create
different kinds of indices (e.g., files, bloom filters, and column stats) in the metadata table without blocking
ingestion. The indexer adds a new action
indexing on the timeline. While the indexing process itself is asynchronous
and non-blocking to writers, a lock provider needs to be configured to safely co-ordinate the process with the inflight
See the indexing guide for more details.
Spark DataSource Improvements
Hudi's Spark low-level integration got considerable overhaul consolidating common flows to share the infrastructure and bring both compute and data throughput efficiencies when querying the data.
- MOR queries with no log files (except for incremental queries) tables are now leveraging Vectorized Parquet reader while reading the data, meaning that Parquet reader is now able to leverage modern processors vectorized instructions to further speed up decoding of the data. Enabled by default.
- When standard Record Payload implementation is used (e.g.,
OverwriteWithLatestAvroPayload), MOR table will only fetch strictly necessary columns (primary key, pre-combine key) on top of those referenced by the query, substantially reducing wasted data throughput as well as compute spent on decompressing and decoding the data. This is significantly beneficial to "wide" MOR tables with 1000s of columns, for example.
See the migration guide for the relevant configuration updates.
Schema-on-read for Spark
In 0.11.0, users can now easily change the current schema of a Hudi table to adapt to the evolving data schema over
time. Spark SQL DDL support (experimental) was added for Spark 3.1.x and Spark 3.2.1 via
ALTER TABLE syntax.
Please refer to the schema evolution guide for more details.
Spark SQL Improvements
- Users can update or delete records in Hudi tables using non-primary-key fields.
- Time travel query is now supported via
timestamp as ofsyntax. (Spark 3.2+ only)
CALLcommand is added to support invoking more actions on Hudi tables.
Please refer to the Quick Start - Spark Guide for more details and examples.
Spark Versions and Bundles
- Spark 3.2 support is added; users who are on Spark 3.2 can use
hudi-spark3-bundle(legacy bundle name).
- Spark 3.1 will continue to be supported via
- Spark 2.4 will continue to be supported via
hudi-spark-bundle(legacy bundle name).
See the migration guide for usage updates.
Slim Utilities Bundle
In 0.11.0, a new
hudi-utilities-slim-bundle is added to exclude dependencies that could cause conflicts and
compatibility issues with other frameworks such as Spark.
hudi-utilities-slim-bundle is to work with a chosen Spark
hudi-utilities-slim-bundleworks with Spark 3.1 and 2.4.
hudi-utilities-bundlecontinues to work with Spark 3.1 as it does in Hudi 0.10.x.
Flink Integration Improvements
- In 0.11.0, both Flink 1.13.x and 1.14.x are supported.
- Complex data types such as
Arrayare supported. Complex data types can be nested in another component data type.
- A DFS-based Flink catalog is added with catalog identifier as
hudi. You can instantiate the catalog through API directly or use the
CREATE CATALOGsyntax to create it.
- Flink supports Bucket Index in normal
BULK_INSERToperations. Different from the default Flink state-based index, bucket index is in constant number of buckets. Specify SQL option
BUCKETto enable it.
Google BigQuery Integration
In 0.11.0, Hudi tables can be queried from BigQuery as external tables. Users can
org.apache.hudi.gcp.bigquery.BigQuerySyncTool as the sync tool implementation for
HoodieDeltaStreamer and make
the target Hudi table discoverable in BigQuery. Please refer to the BigQuery integration guide
page for more details.
Note: this is an experimental feature and only works with hive-style partitioned Copy-On-Write tables.
AWS Glue Meta Sync
In 0.11.0, Hudi tables can be sync'ed to AWS Glue Data Catalog via AWS SDK directly. Users can
org.apache.hudi.aws.sync.AwsGlueCatalogSyncTool as the sync tool implementation for
HoodieDeltaStreamer and make
the target Hudi table discoverable in Glue catalog. Please refer
to Sync to AWS Glue Data Catalog guide page for more details.
Note: this is an experimental feature.
DataHub Meta Sync
In 0.11.0, Hudi table's metadata (specifically, schema and last sync commit time) can be sync'ed
to DataHub. Users can set
org.apache.hudi.sync.datahub.DataHubSyncTool as the sync tool
HoodieDeltaStreamer and sync the target table as a Dataset in DataHub. Please refer
to Sync to DataHub guide page for more details.
Note: this is an experimental feature.
In 0.11.0, Spark 3.2 support has been added and accompanying that, Parquet 1.12 has been included, which brings encryption feature to Hudi (Copy-on-Write tables). Please refer to Encryption guide page for more details.
Bucket index, an efficient and light-weight index type, is added in 0.11.0. It distributes records to buckets using a
hash function based on the record keys, where each bucket corresponds to a single file group. To use this index, set the
index type to
BUCKET and set
For Flink, set
For more details, please refer to hoodie.bucket.index.* in the configurations page.
Savepoint & Restore
Disaster recovery is a mission critical feature in any production deployment. Especially when it comes to systems that store data. Hudi had savepoint and restore functionality right from the beginning for COW tables. In 0.11.0, we have added support for MOR tables.
More info about this feature can be found in Disaster Recovery.
Pulsar Write Commit Callback
Hudi users can use
org.apache.hudi.callback.HoodieWriteCommitCallback to invoke callback function upon successful
commits. In 0.11.0, we add
HoodieWriteCommitPulsarCallback in addition to the existing HTTP callback and Kafka
callback. Please refer to the configurations page for
org.apache.hudi.utilities.schema.HiveSchemaProvider is added for getting schema from user-defined hive
tables. This is useful when tailing Hive tables in
HoodieDeltaStreamer instead of having to provide avro schema files.
In 0.11.0 release, with the newly added support for Spark SQL features, the following performance regressions were inadvertently introduced:
- Partition pruning for some of the COW tables is not applied properly
- Spark SQL query caching (which caches parsed and resolved queries) was not working correctly resulting in additional
- overhead to re-analyze the query every time when it's executed (listing the table contents, etc.)
All of these issues have been addressed in 0.11.1 and are validated to be resolved by benchmarking the set of changes on TPC-DS against 0.10.1.
Raw Release Notes
The raw release notes are available here