VERA 4.5 — May 11, 2026

Compute Engine: VERA 4.5

Built-on: Apache Flink® 1.20

Overview

VERA 4.5 delivers major enhancements across SQL capabilities, AI integration, and deployment operations. Key additions include native BITMAP type support for exact real-time deduplication, seven new AI SQL functions for intelligent text processing, JDK 17 support, Expert Mode for fine-grained resource allocation, and significant connector updates including YAML CDC configuration support.

New Features

BITMAP Type and Functions

VERA 4.5 introduces the native BITMAP type and a library of accompanying functions for efficient real-time exact deduplication. BITMAP stores sets of 32-bit integers using the RoaringBitmap format and supports scalar operations (construction, cardinality, set algebra) and aggregate operations for use cases such as Unique Visitor (UV) counting.

Benefits:

• Enables exact UV counting without sampling or approximation.
• Provides set algebra across multiple groups or time windows.
• Reduces the need for custom UDFs for deduplication workloads.

Supported Deployments: BYOC, Managed Cloud, Self-Managed (On-Prem)

Learn More: BITMAP Type and Functions

AI SQL Functions

VERA 4.5 expands intelligent text processing and analysis capabilities in Flink SQL with seven new AI SQL functions. These functions use the LATERAL TABLE syntax and call LLMs directly from SQL, enabling classification, sentiment analysis, structured extraction, summarization, embedding, translation, and data masking — all within a streaming or batch pipeline.

Benefits:

• Brings LLM-powered text analytics directly into SQL pipelines without custom integration code.
• Supports a wide range of use cases: content tagging, public opinion monitoring, multilingual processing, PII masking, and semantic retrieval.
• Works with any model endpoint configured via CREATE TEMPORARY MODEL.

Supported Deployments: BYOC, Managed Cloud, Self-Managed (On-Prem)

Learn More: AI SQL Functions

JDK 17 Support

JDK 17 support is now available. Customers can select the new vera-4.5-flink-1.20-jdk17 engine version for deployments that require Java 17. The default vera-4.5-flink-1.20-jdk11 engine version remains available for backward compatibility.

Benefits:

• Enables deployments built with Java 17 to run without compatibility shims.
• Provides access to JDK 17 language features and performance improvements.

Supported Deployments: BYOC, Managed Cloud, Self-Managed (On-Prem)

Learn More: VERA Engine Versions and Java Compatibility

Configure Deployment Resources (Expert Mode)

Expert Mode is a fine-grained resource allocation method that lets you configure resources per Slot Sharing Group (SSG) within a SQL deployment. For complex deployments, the default resource mode may result in low resource utilization. Expert Mode lets you tune JobManager CPU and memory, set parallelism per operator group, and import or export execution plans as JSON.

Benefits:

• Improves resource utilization for complex deployments with varied workload profiles.
• Provides direct control over JobManager sizing and per-SSG parallelism.
• Enables plan portability through JSON import and export.

Supported Deployments: BYOC, Managed Cloud

Learn More: Configure Deployment Resources (Expert Mode)

Operator-Level State TTL

Users can now configure different state time-to-live (TTL) settings for individual operators within the same job. Previously, a single job-level TTL applied uniformly to all operators. This caused resource waste for operators with fast-changing data and broken correctness for operators that require long state retention — for example, logistics data associated with orders over weeks or months.

Benefits:

• Reduces state size and memory consumption for high-volume, short-lived operators.
• Preserves correctness for operators that need extended state retention.
• Eliminates the trade-off forced by a single job-level TTL.

Supported Deployments: BYOC, Managed Cloud

Learn More: Configure Deployment Resources (Expert Mode)

Optimized State Compatibility Check for SQL Deployments

When resuming a SQL deployment from existing state after making changes, VERA now automatically detects modifications to SQL statements, runtime parameters, and engine version, and provides a compatibility check to determine whether state data can be safely reused. This maximizes state reuse and reduces redeployment effort.

Benefits:

• Prevents deployment failures caused by incompatible state data.
• Supports lazy state migration via the Gemini state backend, reducing startup latency.
• Provides clear recommended actions for each compatibility result.

Supported Deployments: BYOC, Managed Cloud

Learn More: State Compatibility Check

SQL Console

The SQL Console provides a dedicated environment for executing ad hoc SQL commands directly within Ververica Cloud. Unlike SQL Drafts, which compile into persistent deployments, the SQL Console is designed for one-time operations: DDL, DML, DQL, CALL commands, and EXPLAIN statements in a single, unified interface. This resolves a key usability gap — previously there was no clear separation between long-running SQL jobs and one-time command execution.

Benefits:

• Simplifies catalog and table management with direct DDL execution.
• Supports CALL commands for Paimon table operations such as compaction.
• Enables inline use of EXPLAIN to inspect and optimize execution plans.

Supported Deployments: BYOC, Managed Cloud

Learn More: SQL Console

Connectors & Catalogs

YAML CDC Support

YAML configuration support for CDC connectors simplifies data pipeline development, compatible with community Flink CDC version 3.5. New additions include PostgreSQL CDC YAML jobs, MongoDB CDC YAML jobs, and native Paimon source integration. Additional enhancements support catalog-registered source and sink metadata references, regular expression-based routing for table merging and database sharding, dirty data handling for JSON format parsing, and empty schema fault tolerance for Kafka CDC YAML.

Benefits:

• Reduces boilerplate in CDC pipeline configuration.
• Improves job robustness with dirty data handling and empty schema fault tolerance.
• Enables complex routing scenarios through standard regular expression support.

Supported Deployments: BYOC, Managed Cloud

Learn More: Data Ingestion

Connector Enhancements

This release includes improvements across multiple built-in connectors:

• Kafka: Canal-JSON format extension supports parsing source database index events and extracting raw key/value metadata fields. Automatic removal of redundant Flink-created consumer groups for open-source Kafka.
• Paimon Sink: Supports writing to Paimon tables with cross-partition upsert and generates random commit user information automatically.
• MongoDB: Improved handling of ObjectId-type fields for non-_id primary keys. Sink tables now support Partial Update mode.
• Redis: Enhanced cluster mode support, optimized write performance, and batched write support.
• Elasticsearch: Supports explicit configuration of doc_as_upsert and a new connection timeout parameter.

Supported Deployments: BYOC, Managed Cloud

Learn More: Connectors

Azure CosmosDB Connector

The new Azure CosmosDB connector enables writing to CosmosDB from Flink SQL. It supports auto-creation of containers, partial updates, configurable throughput (RU/s), and built-in exactly-once semantics through a two-phase commit protocol.

Benefits:

• Enables direct Flink SQL integration with CosmosDB without custom connector code.
• Supports serverless and provisioned CosmosDB accounts.
• Provides exactly-once delivery guarantees for reliable data pipelines.

Supported Deployments: BYOC, Managed Cloud, Self-Managed (On-Prem)

Learn More: Azure CosmosDB Connector

MongoDB Catalog

The MongoDB Catalog enables direct integration between Flink SQL and MongoDB databases, allowing you to register MongoDB as a catalog and reference collections as tables in SQL deployments.

Supported Deployments: BYOC, Managed Cloud, Self-Managed (On-Prem)

Learn More: MongoDB Catalog