Kamu

Added
- Complete support for `arm64` architecture (including M-series Apple Silicon)
- `kamu-cli` now depends on multi-platform Datafusion, Spark, Flink, and Jupyter images allowing you to run data processing at native CPU speeds
Changed
- Spark engine is upgraded to latest version of Spark 3.5
- Spark engine is using [ANSI mode](https://spark.apache.org/docs/latest/sql-ref-ansi-compliance.html) by default which produces helpful errors instead of silently returning `null` in many built-in functions
Fixed
- Modeling mistake in Smart Transfer Protocol that creates unexpected push/pull transfer path for metadata blocks

Added
- Native support for `arm64` architecture (including M-series Apple Silicon) in `kamu-cli` and `kamu-engine-datafusion`
- Note: Flink and Spark engine images still don't provide `arm64` architecture and continue to require QEMU
Changed
- Flow system scheduling rules improved to respect system-wide throttling setting and take last successful run into account when rescheduling a flow or after a restart

Changed
- Using unique container and network names to prevent collisions when running concurrent `kamu` processes
- Improved error handling for all subprocesses

Fixed
- Pausing dataset flow configuration via the dedicated API should have impact on the currently running flow

Added
- New type `DatasetRefPattern` which allows CLI command to accept global pattern
- New GraphQL APIs for quick pausing/resuming of dataset flow configs preserving the scheduling rules
- New GraphQL APIs for server-side filtering of flow listings (by type, by status, and by initiator)
Changed
- `kamu deleted` and `kamu verify` now accepts global pattern expression
- Error handling for pipe subprocesses with file output
- Pagination implementation made more efficient for flows and tasks event stores

Added
- Datafusion-based interactive SQL shell
Changed
- Datafusion is now the default engine for `kamu sql` command