Classmethod Data Analytics Newsletter (AWS Data Analytics Edition) – August 2026 Issue
This page has been translated by machine translation. View original
This is Ishikawa from the Cloud Business Division Consulting Department. This month, Amazon OpenSearch Service introduced a log analysis-specific engine (Optimized), Amazon MSK Express brokers gained the ability to deliver Apache Kafka data directly to Amazon S3 and Apache Iceberg streaming tables, and AWS Glue Data Quality added support for anomaly detection and distribution statistics. My overall impression is that this was a month where both the direction of "delivering data from streaming to the lakehouse without writing glue code" and the automation of governance features such as data quality and lineage advanced simultaneously. There are other updates as well, so let me introduce them!
Amazon SageMaker Unified Studio
New Features & Updates
2026/07/02 - Amazon SageMaker Unified Studio now supports Terraform for provisioning
Amazon SageMaker Unified Studio now supports provisioning via Terraform. Using the open-source terraform-aws-sagemaker-unified-studio module, you can deploy domains from version-controlled templates.
This allows platform teams to incorporate SageMaker Unified Studio into existing IaC pipelines and maintain consistency across development, staging, and production accounts. The module handles domain infrastructure including IAM role provisioning, and submodules allow blueprint enablement, integration into project profiles, and project creation to be executed independently. Project creation using existing IAM roles is also supported. This is achieved through the Terraform AWS Cloud Control provider.
2026/07/07 - Amazon SageMaker Unified Studio now supports importing existing MWAA environments
You can now connect existing Amazon MWAA environments to SageMaker Unified Studio projects. Data engineers and platform teams already running MWAA can manage Airflow workflows from the same interface as analytics and machine learning, without rebuilding configurations or migrating DAGs.
To connect, open the Workflows tool in your project, select "Add connection" on the connection selection screen, and specify the Airflow configuration referencing the domain and project. Once connected, project members can sync, run, and monitor workflows directly from SageMaker Unified Studio. Environments running Apache Airflow 3 or later can also use drag-and-drop visual authoring.
2026/07/07 - Amazon SageMaker now supports data lineage in IAM-based domains
Amazon SageMaker Unified Studio now supports OpenLineage-compatible data lineage in IAM-based domains as well. It collects lineage events from Apache Spark jobs run on Amazon EMR, AWS Glue, SageMaker Visual ETL, and notebooks. Previously, this was only supported for IAM Identity Center-based domains.
The interactive lineage graph provides an overview of how data flows from source to consumption, with options to configure graph depth, an event timestamp mode for viewing column-level lineage in detail, and a view that shows only datasets. For both domain types, lineage can be published, queried, and managed programmatically from OpenLineage-compatible applications. The DeleteLineageEvent API also makes it possible to delete published events.
2026/07/08 - Amazon SageMaker Unified Studio Workflows now supports operators for Amazon Bedrock, S3 Tables, S3 Vectors, and Glue Catalog
19 new operators for Amazon Bedrock, Amazon S3 Tables, Amazon S3 Vectors, AWS Glue Data Catalog, and Amazon MWAA Serverless have been added to SageMaker Unified Studio Workflows. You can orchestrate these services simply by adding tasks from the visual workflow creator, without writing custom integration code.
Specifically, you can manage Bedrock guardrails, create and delete S3 Tables and S3 Vectors resources, manage Glue Data Catalog tables and databases, and trigger MWAA Serverless workflow executions. This reduces the need to switch between consoles or write custom DAGs. Available in all regions where SageMaker Unified Studio is available.
2026/07/09 - Amazon SageMaker Unified Studio adds custom asset types to the catalog in IAM-based domains
Custom asset types are now supported in IAM-based domains. Domain administrators can register assets of any format in the catalog, such as medical image files on Amazon S3, sales dashboards created in PowerBI, or PDF reports generated by third-party platforms.
Administrators create custom asset types by specifying a name, description, and optional metadata forms. From there, individual assets can be created and enriched with business context using glossary terms and README documents before being published. Once published, anyone in the domain can search by name, type, or term, and request subscriptions through the same governed workflow as other catalog assets. Available in all regions where SageMaker Unified Studio is available.
2026/07/17 - Amazon SageMaker Unified Studio adds custom visual transforms
Custom visual transforms can now be created, shared, and reused in visual ETL flows. Data engineers can build business-specific transformation logic in a reusable format and make it available to their teams without requiring coding expertise.
When you publish your own transforms to the shared library, other team members can discover and use them directly from the visual ETL interface. Examples include transforms for normalizing customer phone numbers, masking personally identifiable information, or applying organization-standard data quality checks. The same logic can be reused across multiple ETL jobs, reducing duplicate work and improving consistency of results.
2026/07/21 - Amazon SageMaker Unified Studio now supports Amazon OpenSearch
Amazon OpenSearch is now supported as a data source. You can directly query and analyze search and log analytics data alongside other data assets. This allows you to combine OpenSearch operational search data with data from Amazon Redshift, Amazon S3, relational databases, and more, all within a single governed environment.
For example, you can join application logs and metrics stored in OpenSearch with transaction data to analyze system performance and user behavior. To use this feature, add an Amazon OpenSearch connection in the data section of your project. Once added, it appears in the data explorer, enabling direct querying in the query editor, exploration in notebooks, and integration into visual ETL jobs.
2026/07/30 - Amazon SageMaker Unified Studio brings richer Git version control to all project tools
Git version control is now available across all project tools: Query Editor, Visual ETL, Workflows, and Notebooks. Replacing the previous auto-sync approach, it provides flexible, file-level version control. Notebooks, which previously had no Git integration, are now also included.
By adding repositories from GitHub, GitLab, or Bitbucket, you can individually select which files to track. Source control is not enforced at the project level, so users decide what to version-control and when. Repositories are decoupled from project creation and can be added at any time afterward. You can connect to multiple repositories and branches simultaneously from a single project, and branch creation, pulling, and conflict resolution can all be done within the project. In JupyterLab and Code Editor, Git CLI is also available from the built-in terminal.
Amazon DataZone
New Features & Updates
API Changes
2026/07/23 - Amazon DataZone - 1 new 5 updated methods
APIs were added and updated to support notebook synchronization with ipynb files on S3.
Amazon Redshift
New Features & Updates
2026/07/08 - Amazon Redshift RG instances now available on the trailing track
As of July 7, 2026, rg.4xlarge and rg.xlarge are now available on the trailing track (P201) as well. The trailing track is designed for those who prioritize stability of production workloads, using versions already validated on the leading track.
With RG instances now available on both tracks, you can enjoy the Graviton benefits of up to 2.4x query performance compared to RA3 and 30% lower price per vCPU, regardless of your track selection. Available for provisioning new clusters or resizing existing ones via the Management Console, AWS CLI, or AWS SDK.
2026/07/16 - Amazon Redshift adds rg.large and rg.12xlarge instance sizes
Two new sizes, rg.large and rg.12xlarge, have been added to RG instances and are now generally available. They share the same Graviton-based performance characteristics as existing RG instances, expanding sizing options for provisioned clusters.
These two sizes are only available on the current track (P202). rg.xlarge and rg.4xlarge continue to be used on the trailing track (P201). Existing RA3 clusters can be migrated via Snapshot and Restore, Elastic Resize, or Classic Resize. Available in various regions including Asia Pacific (Tokyo).
2026/07/29 - Amazon Redshift Data API announces long polling, session management, and flexible batch execution
Three new features have been added to the Amazon Redshift Data API. Long polling allows you to specify WaitTimeSeconds in ExecuteStatement, BatchExecuteStatement, DescribeStatement, GetStatementResult, and GetStatementResultV2 to hold the response until SQL completion, retrieving metadata and results without repeated polling.
ListSessions enables applications that reuse sessions to enumerate active sessions and filter by status, compute target, and database, eliminating the need to manage session IDs externally. Flexible batch execution adds an ExecutionMode parameter to BatchExecuteStatement; in AUTO_COMMIT mode, each SQL statement executes independently, so a failure in one statement does not cause the entire batch to roll back.
2026/07/30 - Amazon Redshift RG large and 12xlarge instances now available on the trailing track
rg.large and rg.12xlarge are now available on the trailing track as well. This applies when running on patch P202 or later.
This means that users of the trailing maintenance track can now create rg.large / rg.12xlarge Redshift clusters in all regions where RG is generally available. Available for provisioning new clusters or resizing existing ones via the Management Console, AWS CLI, or AWS SDK.
API Changes
2026/07/21 - Amazon Redshift - 4 new methods
Four new APIs were added to manage IAM Identity Center applications for Query Editor V2: CreateQev2IdcApplication, DescribeQev2IdcApplications, ModifyQev2IdcApplication, and DeleteQev2IdcApplication.
2026/07/23 - Redshift Data API Service - 1 new 5 updated methods
The wait-time-seconds parameter for long polling was added to five APIs, along with the new ListSessions API and the execution-mode parameter for BatchExecuteStatement.
Amazon Redshift Serverless
New Features & Updates
2026/07/13 - Amazon Redshift Serverless now preserves zero-ETL and Amazon S3 event integrations during snapshot restores
Amazon Redshift Serverless now automatically preserves zero-ETL integrations and Amazon S3 event integrations when restoring from a snapshot or recovery point to the same serverless namespace.
Previously, associated integrations would enter a failed state after a restore and had to be manually recreated upon completion. This required time to reconfigure data pipelines and could result in gaps in data ingestion. With this improvement, integrations are maintained and resume operation as soon as the restore completes. Note that this only applies to restores to the same namespace; integrations are not preserved when restoring to a different namespace or when restoring provisioned clusters.
2026/07/24 - Amazon Redshift Serverless now offers All Upfront pricing for 3-year Serverless Reservations
An All Upfront payment option has been added for 3-year Serverless Reservations, enabling savings of up to 50% compared to on-demand pricing. Paying the full amount at the start of the term provides the maximum discount.
This adds to the existing options of 1-year No Upfront, 1-year All Upfront, and 3-year No Upfront, expanding cost optimization choices. Serverless Reservations commit to a number of RPUs, are managed at the AWS payer account level, and can be shared across multiple accounts. Usage exceeding the reserved RPUs is charged at standard on-demand rates. Available in all regions where Redshift Serverless is available, including commercial regions and AWS GovCloud (US).
2026/07/13 - Redshift Serverless - 2 updated methods
Support was added to preserve data sharing, zero-ETL, and S3 event integrations when restoring snapshots to a serverless namespace.
AWS Glue
New Features & Updates
2026/07/27 - AWS Glue Data Quality now supports distribution statistics for data profiling
A Distribution Analyzer that generates frequency distribution profiles of data has been added to AWS Glue Data Quality. From DQDL (Data Quality Definition Language), you can generate histograms for numeric columns and value distributions for categorical, date, and boolean columns. The number of bins is also customizable.
Distribution statistics allow you to quickly identify skew, outliers, and unexpected patterns without custom code. Since it is integrated into existing DQDL rulesets, you can add distribution profiling within the same evaluation run as your current data quality checks. Statistics are stored in Amazon S3 and can be queried from Amazon Athena and other tools, and can also be retrieved via API, making them easy to integrate into monitoring and visualization workflows including SageMaker Unified Studio. Available in all commercial regions and AWS GovCloud (US).
2026/07/27 - AWS Glue Data Quality now supports anomaly detection and writing results to the AWS Glue Data Catalog
AWS Glue Data Quality now supports anomaly detection for Data Catalog-based data quality evaluations and the ability to write evaluation results to AWS Glue Data Catalog tables. It works for both ETL jobs and catalog evaluations, providing a consistent data quality experience regardless of workflow type.
Anomaly detection uses ML-based time-series forecasting to detect statistical changes such as a sudden drop in unique values or a sudden increase in row counts, without writing explicit threshold rules. This is useful for data engineers monitoring hundreds of tables in the catalog. For result storage, rule evaluation results, profiling metrics, and anomaly predictions with confidence intervals are written back to Data Catalog tables, creating a record that can be queried at any time with standard SQL. Available in all commercial regions and AWS GovCloud (US).
2026/07/29 - AWS Glue announces VPC support, filter pushdown, and partition support for the REST API connector
The AWS Glue REST API connector now supports VPC connectivity, filter pushdown, and partitioning. This connector can be used to ingest data from any source that exposes a REST API, including proprietary systems and emerging platforms that lack native connectors.
VPC support allows access to data sources in private subnets, over VPN, or via AWS PrivateLink, without exposing traffic to the internet. Filter pushdown translates query predicates into API-native parameters, transferring only matching records to improve data transfer costs and job performance. Partition support splits large datasets across multiple Spark workers using field-based or record-count-based strategies, reducing ingestion time for large volumes of data with pagination. Available in all commercial regions where AWS Glue is available.
API Changes
2026/07/27 - AWS Glue - 1 new 8 updated methods
The BatchGetDataQualityRulesetEvaluationRun API was added to retrieve multiple execution results in a single call. Other updates include ObservationScope / ObservationMode parameters for anomaly detection, writing evaluation results to Data Catalog tables, and support for custom log group paths for recommendation runs.
2026/07/29 - AWS Glue - 2 updated methods
Support for filtering, partitioning, and VPC was added to the AWS Glue REST API connector.
Amazon QuickSight
API Changes
2026/07/01 - Amazon QuickSight - 3 updated methods
FileSource is now supported in PhysicalTable, enabling handling of datasets with files as the source.
2026/07/10 - Amazon QuickSight - 2 new 8 updated methods
Two new APIs, CreateKnowledgeBase and UpdateKnowledgeBase, were added.
2026/07/20 - Amazon QuickSight - 7 updated methods
Custom permissions for triggers are now supported. Administrators can control user access to Schedule, Inbound Email, and Quick Event triggers.
2026/07/23 - Amazon QuickSight - 4 updated methods
The ability to control access to Amazon Quick via browser extensions and Microsoft Word / Outlook / Excel / PowerPoint add-ins was added to custom permission profiles.
2026/07/27 - Amazon QuickSight - 4 updated methods
Governance-related fields were added to the custom permissions API to implement Deny By Default behavior.
AWS Clean Rooms
New Features & Updates
2026/07/24 - AWS Clean Rooms supports larger worker types up to 32 vCPUs for SQL
Large worker types with 32 vCPUs and 244 GB of memory are now available for SQL analytics in AWS Clean Rooms. The previous limit was 16 vCPUs and 120 GB.
Larger workers improve performance for complex queries on large datasets and can also help optimize costs. For example, in collaborations between advertisers and publishers, this is expected to be effective in cases such as performing complex feature engineering on billions of records to train multi-touch attribution models.
API Changes
2026/07/27 - AWS Clean Rooms Service - 4 updated methods
Support for the CR.8X worker type (32 vCPUs) for SQL analytics was added.
2026/07/27 - AWS Clean Rooms ML - 4 updated methods
The CR.8X worker type (32 vCPUs) for SQL is now also supported in Clean Rooms ML.
Amazon EMR
New Features & Updates
2026/07/07 - Amazon EMR Serverless now supports larger worker sizes to run more compute and memory intensive workloads
Amazon EMR Serverless now supports large worker configurations with 32 vCPUs and up to 244 GB of memory. The previous limit was 16 vCPUs and up to 120 GB.
For shuffle-heavy workloads, inefficient data transfers between executors are reduced, and for jobs with data skew, the likelihood of OOM failures decreases. Jobs that cache data can also benefit from keeping more data in memory, leading to improved performance. Available in all regions where EMR Serverless is available.
2026/07/10 - Amazon EMR on EKS now supports Apache Spark troubleshooting agent
Amazon EMR on EKS now supports an Apache Spark troubleshooting agent. Without manually tracing distributed logs and Spark History Server data, you can receive natural language root cause analysis of job failures and improvement suggestions for PySpark code.
The agent analyzes Spark History Server data, executor logs, and cluster configurations to identify issues such as memory errors, data skew, resource contention, and connectivity failures. This update brings coverage to all deployment types: EMR on EC2, EMR Serverless, and EMR on EKS. Available via "Troubleshoot with AI" in the console, as well as from Kiro, Claude Code, and Cursor via MCP. All operations are read-only and recorded in CloudTrail.
API Changes
2026/07/14 - Amazon EMR Containers - 10 updated methods
In conjunction with Spark Connect server support in EMR on EKS, five fields were added to three APIs: sessionIdleTimeoutInMinutes, sessionEnabled, endpointToken, authProxyUrl, and encryptionKeyArn.
2026/07/21 - Amazon EMR Containers - 1 new 3 updated methods
The DeleteSecurityConfiguration API was added to delete security configurations in EMR on EKS, and authenticationConfiguration was added to securityConfigurationdata.
2026/07/27 - Amazon EMR Containers - 1 new 3 updated methods
It is now possible to set a limit on the number of concurrent jobs for virtual clusters, enabling fine-grained control over the number of simultaneously running jobs and queue wait counts.
Amazon Kinesis
New Features & Updates
2026/07/24 - Amazon Kinesis Data Streams now supports scaling down ingest capacity with warm throughput
Amazon Kinesis Data Streams warm throughput now supports scaling down ingest capacity. Previously, warm throughput was primarily used to pre-increase capacity in anticipation of sudden traffic spikes, but write throughput can now be controlled in both directions.
Setting a lower warm throughput value for an on-demand stream adjusts it to the larger of the specified value and the peak ingest volume from the past hour. This ensures that capacity needed for current traffic is maintained while releasing unnecessary surplus. Available at no additional cost for on-demand streams with On-demand Advantage mode enabled.
API Changes
2026/07/17 - Amazon Kinesis Analytics - 8 updated methods
Updates accompanying Flink 2.3 support for Amazon Managed Service for Apache Flink (the API namespace remains kinesisanalytics).
Amazon MSK
New Features & Updates
2026/07/09 - Amazon MSK Replicator now supports replication from external Apache Kafka clusters to MSK Standard brokers
Amazon MSK Replicator now supports replication from external Apache Kafka clusters—including on-premises, self-managed on AWS, and other clouds—to MSK Standard brokers. Previously, MSK Express brokers were the only replication destination available for external clusters.
This enables workload migration to MSK Standard brokers, disaster recovery with MSK clusters as failover/backup targets, and data distribution in hybrid and multi-cloud environments. SASL/SCRAM or mTLS authentication is used to connect to external clusters. Replication from MSK Standard to external Kafka clusters is also supported, enabling failback as well.
2026/07/15 - Amazon MSK Express Brokers adds support for Apache Kafka version 4.2
Amazon MSK Express Brokers now supports Apache Kafka 4.2. This includes enhancements to Eligible Leader Replicas (ELR) for improved leader election accuracy, a new consumer rebalance protocol for faster and smoother group rebalances, and the Streams Rebalance Protocol for optimizing Kafka Streams task assignment.
You can select 4.2.x when creating a new cluster, or upgrade existing Express Brokers in-place via a rolling update. Available in all regions where MSK Express Brokers is available.
2026/07/30 - Amazon MSK Express brokers now delivers Apache Kafka data to Amazon S3
Amazon MSK Express brokers now supports a fully managed capability to deliver Apache Kafka data to general-purpose Amazon S3 buckets. This eliminates the need to operate your own connectors for use cases such as log archiving, compliance-driven retention, Kafka replay, and AI/ML model training.
Ingestion and delivery costs can be reduced by up to 60% compared to self-managed alternatives, and MSK Express automatically handles scaling, retries, and backpressure. S3 delivery supports throughput of up to 10 GB/s, and capacity scaling and version upgrades are performed without interrupting delivery.
2026/07/30 - Amazon MSK Express brokers now deliver data to streaming tables for Apache Iceberg
Amazon MSK Express brokers now supports "streaming tables for Apache Iceberg," which continuously materializes Apache Kafka topics as Apache Iceberg tables on Amazon S3 Tables. Ingestion and delivery costs to S3 Tables are said to be reduced by up to 60%, and downstream query costs by up to 30%, compared to self-managed configurations.
For the problem of large numbers of small Parquet files generated by high-frequency ingestion, inline automatic compaction maintains query performance. Write conflicts from multiple high-throughput writers are also resolved by a built-in arbitration mechanism. Delivery to S3 Tables supports throughput of up to 10 GB/s.
API Changes
2026/07/30 - Managed Streaming for Kafka - 5 new methods
5 new methods have been added in conjunction with support for Apache Iceberg streaming tables on MSK Express brokers and support for data delivery to S3 general-purpose buckets.
Amazon MWAA
New Features & Updates
2026/07/24 - Amazon MWAA now supports Apache Airflow version 2.11.2
Amazon MWAA now supports Apache Airflow 2.11.2. This is a maintenance release that includes security improvements, bug fixes, and dependency package updates.
In addition to security patches and stability improvements for the Airflow webserver and task execution layer, it includes fixes for lifecycle management of queued tasks, enhanced masking of secrets in logs, UI fixes for the Task Instances list, and provider package updates for log delivery in S3 and CloudWatch. It is available in all regions where MWAA is offered by creating a new environment or upgrading an existing one.
API Changes
Amazon OpenSearch Service
New Features & Updates
2026/07/01 - Amazon OpenSearch Service optimized for log analytics
Amazon OpenSearch Service has added a new engine specialized for log analytics workloads. It achieves up to 4x price-performance improvement in internal benchmarks while still supporting ad-hoc full-text searches needed for incident investigations, just as before.
A new columnar storage for aggregation workloads can reduce storage by up to 70%, allowing you to retain up to 3x more data at the same cost. In addition, you get up to 2x ingestion throughput and up to 2x faster analytics queries on the same hardware. To use it, create a new domain with OpenSearch 3.5 or later, select observability for the use case and optimized for the engine mode.
2026/07/15 - Amazon OpenSearch Service now supports the Agent Toolkit for AWS with a curated skill
Amazon OpenSearch Service now supports the Agent Toolkit for AWS, enabling you to build, manage, and query OpenSearch Service domains and OpenSearch Serverless collections from AI coding agents such as Claude Code, Kiro, and Cursor.
The AWS MCP server executes API calls, and the amazon-opensearch-service skill routes natural language requests to the appropriate functionality. The five supported areas are: migration from self-managed, domain and collection operations, vector/semantic/hybrid/RAG search, log analytics with PPL and OpenSearch Ingestion, and trace analytics with OpenTelemetry. No changes to existing configurations are required, and there are no additional charges.
2026/07/17 - Amazon OpenSearch UI now supports one-click dashboard migration
You can now migrate from the traditional OpenSearch Dashboards to OpenSearch UI with a single click. This applies to both OpenSearch Service domains and Serverless collections.
You can migrate multiple tenants and large numbers of saved objects created in OpenSearch Dashboards to OpenSearch UI workspaces without manually recreating them. The migration destination can be either a new workspace or an existing one. If there are multiple tenants, you can choose to consolidate them into a single workspace or keep them separated by team. Available in all regions where OpenSearch UI is available.
2026/07/30 - Amazon OpenSearch Service now supports OpenSearch version 3.7
OpenSearch 3.7 is now available on Amazon OpenSearch Service. Vector search performance, search relevance, and Query Insights have been improved.
1-bit scalar quantization with the Faiss and Lucene engines compresses vectors, reducing storage and memory usage while maintaining search accuracy. Using doc values instead of document sources allows fast retrieval of vectors without reindexing. The Search Relevance Workbench has added evaluation metrics, CSV judgment file uploads, and hybrid search optimization. Query Insights now supports automatic generation of query recommendations, caching of completed queries, and exporting top queries to Amazon S3.
API Changes
2026/07/01 - Amazon OpenSearch Service - 1 new 7 updated methods
When creating a Mustang domain with the AWS CLI, you need to specify OPTIMIZED for EngineMode and OBSERVABILITY or MIXED for UseCase. If not specified, it will be a regular GENERAL domain. An Insights Feedback API for sending feedback to the Insight API has also been added.
2026/07/06 - Amazon OpenSearch Service - 3 new methods
A Saved Object Migration API has been added. You can migrate saved objects such as dashboards, visualizations, and index patterns to OpenSearch Service application workspaces by specifying export filters and conflict resolution strategies.
2026/07/13 - Amazon Elasticsearch Service - 6 updated methods
You can now specify EngineMode (GENERAL / OPTIMIZED) and UseCase (SEARCH / VECTOR / OBSERVABILITY / MIXED) when creating or updating a domain.
Amazon Managed Service for Apache Flink
New Features & Updates
2026/07/14 - Amazon Managed Service for Apache Flink now offers AI Agent Skills to simplify building and operating Flink applications
Amazon Managed Service for Apache Flink has launched AI Agent Skills. These provide AI coding assistants with up-to-date guidance on tasks such as application creation, troubleshooting, scaling, monitoring, network configuration, and cost optimization.
Available from existing AI coding agents such as Kiro, Claude Code, and Cursor. You configure the Agent Toolkit for AWS with the AWS CLI and then ask questions like "How do I create a new Flink application on MSF?" or "My Flink application is not working well — what could be the cause?"
2026/07/20 - Amazon Managed Service for Apache Flink now supports Apache Flink 2.3
Amazon Managed Service for Apache Flink now supports Apache Flink 2.3. Adaptive partition selection, which improves backpressure handling, allows applications to run stably even under skewed load conditions.
Out-of-order update handling in CDC pipelines has also been improved for better data accuracy, and SQL functions have been added to convert between changelogs and regular streams. In addition to creating new applications, compatible existing applications can be migrated via in-place version upgrades. Available in all regions where Managed Service for Apache Flink is offered.
Amazon S3 Tables
New Features & Updates
2026/07/28 - Amazon S3 Tables now support the Variant data type for Apache Iceberg V3
Amazon S3 Tables now supports the Variant data type from the Apache Iceberg V3 specification. Semi-structured data like JSON can be written as-is without fixing a schema in advance, allowing data to land quickly while still ensuring analytics query performance.
Iceberg V3-compatible engines shred semi-structured data into hidden columns at write time and generate Parquet column statistics. Query engines use these statistics for optimizations such as file pruning, reducing the amount of data scanned. S3 Tables continuously performs table maintenance including compaction for Variant columns as well. Available in 15 regions including Asia Pacific (Tokyo).
Closing
In July 2026, two notable trends stood out: "streaming data flowing directly into the lakehouse" and "automation of governance features such as data quality and lineage." Amazon MSK Express brokers can now deliver directly to Apache Iceberg streaming tables on S3 general-purpose buckets and S3 Tables, making it practical to move away from managing Kafka Connect connectors. AWS Glue Data Quality's anomaly detection and write-back of results to the Data Catalog, along with OpenLineage-compatible lineage in SageMaker Unified Studio's IAM-based domains, also address the challenge of ensuring pipeline reliability through operations.
On the cost front, more options have emerged: up to 70% storage reduction with Amazon OpenSearch Service's log analytics-optimized engine, the removal of the 30-day minimum retention period for Amazon S3 Standard-IA / One Zone-IA, and 3-year All Upfront reservations for Amazon Redshift Serverless. In addition, the EMR on EKS Spark troubleshooting agent, OpenSearch Service's Agent Toolkit support, and Managed Service for Apache Flink's AI Agent Skills are all signs that the path to operating and diagnosing services from AI coding agents is becoming well-established across each service. If any of these updates catch your eye, be sure to give them a try.
