v0.6

Released September 28, 2026 · Preview v0.6.0

v0.6 expands tapstate with a bundled web application, opt-in cluster operation with automated capture failover, 15-day metrics history with root-cause explanations, flexible Nest state databases, and flat child embeds. 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.

Breaking changes and upgrade guidance

• Control-store upgrades are one-way: The first start of v0.6 applies four control-store migrations. A v0.5 server refuses to start against an upgraded store afterward. Back up your MongoDB control database before upgrading.
• Views no longer accept schema:: Remove schema: from local .tap.yml view definitions to avoid dsl.unknown-field. Stored resources are cleaned automatically during migration. Connector config.schema settings are unaffected.

Follow the Upgrade to v0.6 guide for step-by-step backup, verification, and rollback procedures.

What changed

v0.6.0

Web application and observability

• Manage pipelines visually in your browser: The server image now bundles a web application. Inspect connection discoveries, visualize pipeline execution graphs, monitor streaming throughput, and tail live logs without CLI setup. See Quickstart.
• Query 15-day metrics history and explanations: Inspect output rates and table lag over a 15-day retention window through CLI, REST, or MCP. Run explain <pipeline-id> to automatically distinguish between transient pauses, stale observations, and genuine stalls. See Observe a pipeline.
• Export metrics to Prometheus and OpenTelemetry: Push operational metrics to an OpenTelemetry collector via HTTP/protobuf or gRPC, or enable a Prometheus scrape listener.

Clustering and high availability (Preview)

• Scale capture with automated failover: Deploy tapstate across multiple nodes using TCP-IP or Kubernetes discovery. A single active member captures each source stream, and compatible members take over capture seamlessly if a node fails. See Configure a cluster.
• Enforce split-brain protection: The production-ha profile requires a strict majority quorum, preventing conflicting capture or state writes across network partitions.

Document assembly and transforms

• Isolate Nest state databases with state.database: Store any Nest step's working state, shape definitions, and dead letters in a dedicated MongoDB database. Relocate active state safely using Move Nest state.
• Flatten related records with as: flat: Merge one-to-one or many-to-one child fields directly into parent documents without nested wrappers, protected by cardinality and conflict guards. See Flat embeds.
• Strengthen stream joins and decimals: Joins validate input shapes early and recover matched state reliably across restarts. JavaScript and mapping transforms preserve exact decimal precision on target writes. See Join transforms.

Performance and connector improvements

• Share CDC readers across pipelines: Multiple pipelines reading the same source share a single CDC capture reader, decoding database logs (such as PostgreSQL WAL) once per cluster.
• Stream large snapshots with bounded memory: Initial table snapshot loads stream with bounded memory hand-off, avoiding oversized heap requirements, while PostgreSQL sources reuse pooled snapshot connections across tables.
• Pass node-level connector configuration: Node-specific connector settings reach runtimes reliably, including Kafka Enhanced schemaRegistryType and PostgreSQL WAL archive settings.

Release components

The release includes native CLI bundles for macOS and glibc-based Linux on arm64 and x64. Verify a download against its .sha256 or checksums.txt.

Server container images (ghcr.io/tapstate/tapstate:0.6.0) bundle the web application for Linux amd64 and arm64. Connector JARs can be installed with tapstate register <id> or downloaded from the connector preview release assets. See Register and test connectors.

Known limits and issues

This release remains an evaluation preview rather than a production deployment target.

• At-least-once delivery semantics: Delivery for read changes is at-least-once. For append-only targets, overlapping snapshot and CDC reads can produce duplicate rows; configure upsert or unique target keys where deduplication is required.
• Snapshot-to-stream positioning window: Because source stream positioning occurs around initial snapshot execution, changes written during that brief window can be missed on certain connectors.
• Cluster network security: Cluster member coordination communicates over a dedicated port (5701) without TLS or authentication. Restrict cluster ports to a private VPC or protected container network.
• Stream join relational constraints: The join transform operates on a streaming SQL subset without fan-out (1:N), subqueries, or aggregations. Dimension keys must be unique.
• PostgreSQL replication slot WAL retention: PostgreSQL CDC replication slots may delay advancing confirmed flush positions under specific workloads, causing temporary WAL buildup (tracked in issue #468 and #526).

Get started

• Install the CLI for offline workspace authoring and scripts.
• Run the Quickstart playground to explore visual management and assembled state in Docker.
• Read Upgrade to v0.6 for migration procedures and verification checks.
• Read Configure a cluster to set up multi-node deployments.
• Read Move Nest state to isolate or relocate operator-state databases.
• Read Observe a pipeline to query metrics history and configure export.
• Use the v0.6.0 release assets for native binaries and checksums.
• Explore the complete product changelog for the commit and PR history.