v0.5

Released September 16, 2026 · Preview v0.5.0

v0.5 expands tapstate's single-node preview with official enterprise database capture for Oracle and SQL Server, real-time SQL stream joins, zero-data-loss pipeline pausing, schema drift reconciliation, and AI agent integration. Use it with non-production systems and sample data.

See the GitHub releases for the published CLI archives, individual SHA-256 files, checksums.txt, and platform-minimums.txt.

What changed

v0.5.0

Connectors and sources

• Capture enterprise data from Oracle and SQL Server: Stream live data from Oracle Database and Microsoft SQL Server into tapstate pipelines. The newly certified oracle and sqlserver source connectors support initial table snapshots and continuous CDC with exact numeric precision preservation.
• Scale MongoDB change streams and maintain exact precision: Configure multi-threaded change-stream decoding with decodeThreads and decodeQueueSize for busy replica sets, and preserve exact monetary values with native BSON Decimal128 support.

Pipelines and stream processing

• Enrich live streams with SQL joins (Preview): Combine streaming facts with dimension tables using standard ANSI SQL (INNER JOIN, LEFT JOIN) directly in pipelines with the join transform under engine: builtin, eliminating the need for custom enrichment scripts.
• Pause pipelines without reloading historical data: Safely pause running pipelines for configuration or schema updates using tapstate stop --keep-state. It preserves CDC offset checkpoints so the pipeline resumes from where it stopped instead of requiring a costly full re-read.
• Reconcile upstream schema changes safely: Inspect schema drift across pipeline steps with tapstate derived-schema <pipeline-id>. Review a three-way comparison and adopt updated field mappings with --accept before restarting data flow.

Developer tools and platform

• Scaffold resources with dedicated CLI commands: Bootstrap clean, validated configurations for source, pipeline, transform, view, and serve resources using tapstate add <kind>. The legacy tapstate new --kind syntax is deprecated.
• Persist workbench canvas layouts and avoid edit collisions: Visual layout positions are saved across sessions via pipeline layout endpoints, and pipeline updates enforce optimistic If-Match concurrency control to prevent accidental overwrites.
• Integrate AI developer agents: Connect AI coding assistants to your workspace through Model Context Protocol (MCP) using read-only MCP tools like source_list and pipeline_list to discover declared sources and pipeline topologies.

Release components

The release includes native CLI bundles for macOS and glibc-based Linux on arm64 and x64. The release metadata recommends macOS 15.0 and glibc 2.34 or newer; these are warnings from the installer, not compatibility gates. Verify a manually downloaded archive against its .sha256 file or checksums.txt before extracting it.

Server container images (ghcr.io/tapstate/tapstate) do not bundle database drivers by default to keep images lightweight and isolate vendor licenses. Official connector JARs (oracle, sqlserver, mysql, postgres, and mongodb) can be installed using tapstate register <id> or downloaded directly from the connector preview release assets. In air-gapped or pre-seeded environments, mount downloaded JARs into TAPSTATE_CONNECTORS_SEED_DIR for automatic startup registration. See Register and test connectors.

Known limitations and issues

This remains a preview rather than a production deployment target. Delivery for a change the connector reads is at least once, not exactly once. An insert-only target can receive duplicates when snapshot and CDC overlap, and a change written while a change stream positions itself around the initial snapshot can be missed.

• SQL joins are in Preview and require exact 1:1 dimension keys. Joining against non-unique dimension rows causes row displacement and emits diagnostic warnings. Complex SQL features including WHERE, GROUP BY, aggregation, subqueries, and outer-dimension fanout are not supported in this release.
• Connectors supported as write targets. In this preview, MongoDB is the supported pipeline synchronization target. Oracle and SQL Server are certified as data sources (snapshots and CDC) and are not supported as write targets.
• In-place upgrades do not resume a pipeline. After an in-place upgrade, a pipeline can report RUNNING without carrying data and cannot start again. Stop it and start it again to reload from the beginning. Track progress in issue #98.
• Purge and reload cannot remove rows deleted while the pipeline was down. A new full load writes current source rows but has no event for a historical deletion. If retaining delete propagation matters, stop without purging state (--keep-state) so the next run resumes from its recorded position. Track progress in issue #104.

Get started

• Install the CLI for offline authoring or to connect to an existing server.
• Run the Quickstart playground to evaluate local pipelines with sample databases.
• Register connectors to download and register official connector JARs for Oracle, SQL Server, MySQL, PostgreSQL, and MongoDB.
• Read the CLI reference for scaffolding commands and checkpoint operations.
• Read the transforms reference to configure join, nest, and js transforms.
• Read the Oracle connector guide and SQL Server connector guide for database prerequisites and CDC configuration.
• Explore the Connectors directory for configuration schemas and supported capture modes across all connectors.
• Read the REST API reference for pipeline layout and optimistic concurrency controls.
• Use the v0.5.0 release assets when you need this exact preview build.