github.com/hatchet-dev/hatchet

Hatchet v0.98.9 is a feature release. It adds idempotency keys for tasks and workflows, task batching, and an embedded engine the Go SDK can run in-process, alongside a new default for concurrency scheduling and a batch of dashboard and hardening fixes.

Highlights

• Tasks can declare a slot cost, so a task that needs more memory or CPU consumes more than one worker slot. See Task Slot Cost.
• Engine rows in the run trace view now carry a badge with their workflow, task, or event name, and the retry number on retried tasks, so repeated spans such as hatchet.engine.workflow_run are distinguishable at a glance.
• CLI commands that take --profile now use the configured default or only profile without prompting, and hatchet server start sets its new profile as the default when none is configured. In sessions without a terminal, such as CI, a selection that would still need a prompt fails with an error that explains how to proceed.
• Tasks and workflows can declare an idempotency key, so duplicate triggers no longer produce duplicate runs. Keys are held for a TTL or until the claiming run reaches a terminal status, in all four SDKs. Note, that idempotency is currently in beta and may be subject to change. See Idempotency Key Expression for more information.
• The Go SDK can run a full engine in-process, pointed at your Postgres database, via hatchet.WithEmbedded(hatchet.WithEmbeddedDatabaseURL(databaseURL)).
• Concurrency strategies are now evaluated against the in-memory index by default rather than querying Postgres on every scheduling pass, where SERVER_CONCURRENCY_IN_MEMORY_INDEX_ENABLED now defaults to true.
• Best-effort pub/sub is now configured independently of the durable queue and can run on Postgres while durable messages stay on RabbitMQ. Existing deployments need no new configuration, see Task Queue Configuration for full list of options.
• Rate limits can be managed from the CLI with hatchet rate-limits, and interactively from the TUI.
• The CLI now reports anonymized usage data on invocation. See Anonymous Telemetry for more information.
• Added worker label filters, and reverted Monaco editor bundle to once again be fetched from a CDN.
• Sensitive request data and RabbitMQ credentials are now kept out of logs, and rate limiting is re-enabled on incoming webhook requests.

Hatchet v0.94.10 headlines two DevEx improvements: development images that run without authentication, and use-case templates for the hatchet quickstart command. The release also adds independent OLAP and core data retention settings to the engine, tenant tagging and consolidated settings pages in Hatchet Cloud, automatic stream listener reconnection in the Go SDK, and RSS feeds for newly published changelog and cookbooks entries.

Highlights

• Introduces a set of development images that allow Hatchet to be run without authentication, removing the need to manage API tokens in development and testing environments. See Running without authentication.
• Hatchet Cloud now supports tagging tenants and organization user groups, automatically adding group members to tenants with matching tags. Additionally, the tenant and organization-scoped settings pages have been consolidated into a single Settings section.
• OLAP and core data retention can be configured independently on the Hatchet engine via SERVER_LIMITS_OLAP_PARTITION_RETENTION and SERVER_LIMITS_CORE_PARTITION_RETENTION, falling back to SERVER_LIMITS_DEFAULT_TENANT_RETENTION_PERIOD when unset.
• Exposes additional Prometheus Tenant Metrics for tracking slot utilization per worker pool, grouped by labels: hatchet_tenant_worker_label_slots (total), hatchet_tenant_used_worker_label_slots (in-use), and hatchet_tenant_available_worker_label_slots (free).
• The Go SDK's streaming listeners now reconnect automatically with jittered backoff: Workflow.Result() no longer hangs if its listener dies, and transient network failures no longer kill workers.
• The Hatchet documentation website now supports RSS feeds for new cookbooks and releases. These can be accessed at https://docs.hatchet.run/cookbooks/feed.xml and /reference/changelog/<component>/feed.xml (e.g. platform), or via autodiscovery by adding a page URL directly to your RSS reader.
• Fixed timeouts in the UserSession cleanup job (introduced in v0.90.13) by indexing the relevant columns; the migration is applied automatically on upgrade.
• hatchet quickstart supports use-case templates via a new --use-case flag, starting with scheduled: a Go template whose workflow runs on a cron schedule and can also be run on demand. See the quickstart CLI docs.

Upgrade Notes

Running without authentication

The Hatchet development images can be pulled directly as hatchet-api-dev, hatchet-engine-dev, hatchet-admin-dev, and hatchet-dashboard-dev, or spun up via the Hatchet CLI with hatchet server start --disable-auth, which runs hatchet-lite-dev in place of hatchet-lite.

This approach embeds a single global worker API key in the Hatchet binaries themselves, so it should only be used in development and testing environments.

See the docs for running without authentication via the CLI and via Docker Compose.