Databricks Runtime 6.3 (unsupported)

Article
01/15/2024

Databricks released this image in January 2020.

The following release notes provide information about Databricks Runtime 6.3, powered by Apache Spark.

New features

Delta Lake

This release introduces improved concurrency. The key features are:

Improved concurrency for all Delta Lake operations - You can now run more Delta Lake operations concurrently. Delta Lake's optimistic concurrency control has been improved by making conflict detection more fine-grained. This makes it easier to run complex workflows on Delta tables. For example:
- Running deletes (for example, for GDPR compliance) concurrently on older partitions while newer partitions are being appended.
- Running updates and merges concurrently on disjoint sets of partitions.
- Running file compactions concurrently with appends.
See Isolation levels and write conflicts on Azure Databricks.
Improved support for file compaction - You can now compact files by rewriting them with the DataFrameWriter option dataChange set to false. This option allows a compaction operation to run concurrently with other batch and streaming operations. See Compact files for details.
Improved performance for insert-only merge - Delta Lake now provides more optimized performance for merge operations that have only insert clauses and no update clauses. Furthermore, Delta Lake ensures that writes from such insert-only merges only append new data to the table. Hence, you can now use Structured Streaming and insert-only merges to do continuous deduplication of data (for example, logs). See Data deduplication when writing into Delta tables for details.

Add additional ports to allow list when process isolation is enabled

Note

This article contains references to the term whitelist, a term that Azure Databricks does not use. When the term is removed from the software, we'll remove it from this article.

When you enable security features like table ACLs or credential passthrough on the cluster, you can now specify additional outbound ports you want Python processes to be able to access by setting the Spark config spark.databricks.pyspark.iptable.outbound.whitelisted.ports to the ports to allow. The supported format of the configuration value is [port[:port][,port[:port]]...], for example: 21,22,9000:9999. The port must be within the valid range, i.e. 0-65535.

Disable job output for Scala jobs

Job output, such as log output emitted to stdout, is subject to a 20MB size limit. You can prevent stdout from being returned from the driver to Azure Databricks by setting the spark.databricks.driver.disableScalaOutput Spark configuration to true. The flag controls cell output for Scala JAR jobs and Scala notebooks. See Output size limits for Jar jobs.

Inline Matplotlib figures now supported in Azure Databricks notebooks (Public Preview)

You can now automatically display Matplotlib figures without needing to pass the figure to the display method. For details, see Matplotlib.

New parameter added to Azure Synapse Analytics

The parameter applicationName is set by default to Databricks-User-Query. The default value prevents the Azure DB Monitoring tool from raising spurious SQL injection alerts against queries. If not specified or the value is an empty string, the default value of the tag is added to the JDBC URL. See Azure Databricks Synapse connector options reference.

Improvements

shutil.copytree now work for /dbfs.
When running Structured Streaming production jobs on High Concurrency mode clusters, restarts of a job would occasionally fail, because the previously running job wasn't terminated properly. To ensure that the previous run stops, set the SQL configuration spark.sql.streaming.stopActiveRunOnRestart true on your cluster. In the case that the stream cannot be stopped properly, a timeout exception will be thrown in 15 seconds. In this case, we recommend restarting your cluster.
Upgraded several installed R libraries. See Installed R libraries.

Apache Spark

Databricks Runtime 6.3 includes Apache Spark 2.4.4. This release includes all Spark fixes and improvements included in Databricks Runtime 6.2 (unsupported), as well as the following additional bug fixes and improvements made to Spark:

[SPARK-30198][CORE] BytesToBytesMap does not grow internal long array as expected
[SPARK-29918][SQL] RecordBinaryComparator should check endianness when compared by long
[SPARK-30274][CORE] Avoid BytesToBytesMap lookup hang forever when holding keys reaching max capacity
[SPARK-25392][CORE][WEBUI] Prevent error page when accessing pools page from history server
[SPARK-30263][CORE] Don't log potentially sensitive value of non-Spark properties ignored in spark-submit
[SPARK-28637][SQL] Thriftserver support interval type
[SPARK-30238][SQL] hive partition pruning can only support string and integral types
[SPARK-30198][CORE] BytesToBytesMap does not grow internal long array as expected
[SPARK-24666][ML] Fix infinity vectors produced by Word2Vec when numIterations are large
[SPARK-30129][CORE] Set client's id in TransportClient after successful auth
[SPARK-30050][SQL] analyze table and rename table should not erase hive table bucketing info
[SPARK-28152][SQL] Mapped ShortType to SMALLINT and FloatType to REAL for MsSqlServerDialect
[SPARK-29971][CORE] Fix buffer leaks in TransportFrameDecoder/TransportCipher
[SPARK-29970][WEBUI] Preserver open/close state of Timelineview
[SPARK-29949][SQL] Fix formatting of timestamps by JSON/CSV datasources
[SPARK-27558][CORE] Gracefully cleanup task when it fails with OOM exception
[SPARK-29758][SQL] Fix truncation of requested string fields in json_tuple
[SPARK-29949][SQL] Fix formatting of timestamps by JSON/CSV datasources
[SPARK-29936][R] Fix SparkR lint errors and add lint-r GitHub Action
[SPARK-29904][SQL] Parse timestamps in microsecond precision by JSON/CSV datasources
[SPARK-29904][SQL] Parse timestamps in microsecond precision by JSON/CSV datasources
[SPARK-26499][SQL] JdbcUtils.makeGetter does not handle ByteType
[SPARK-29682][SQL] Resolve conflicting attributes in Expand correctly
[SPARK-29875][PYTHON][SQL] Avoid to use deprecated pyarrow.open_stream API in Spark 2.4.x
[SPARK-29850][SQL] sort-merge-join an empty table should not memory leak
[SPARK-28939][SQL] Propagate SQLConf for plans executed by toRdd

Maintenance updates

See Databricks Runtime 6.3 maintenance updates.

System environment

Operating System: Ubuntu 16.04.6 LTS
Java: 1.8.0_252
Scala: 2.11.12
Python: 3.7.3
R: R version 3.6.2 (2019-12-12)
Delta Lake: 0.5.0