What changed in the plugins.

Retries sign in the way each tool expects.

• Airflow, Dagster and Prefect: the retry executors now sign in to each tool's own API the way that tool expects.
• The retry job itself is no longer reported as a run.

Release 0.1.7 on GitHub

Prefect and Airflow send more of what a run did.

• Prefect: a task's short result and the runs it launched now travel with the event.
• Prefect 2: results are read from the cache Prefect 2 actually keeps.
• Airflow: a task's group id is sent.

Release 0.1.6 on GitHub

Every tool tested back through its older versions, and values stay private by default.

• A PySpark driver-failure plugin (convalesce-emit-pyspark) was added to the source; it is not published to PyPI yet.
• Spark: every listener event is delivered.
• Great Expectations: column values are redacted by default and every datasource type is named correctly.
• Prefect: parameter values are withheld by default, secrets are redacted and assets are forwarded.
• Airflow and Dagster: asset checks and column lineage are forwarded, Airflow 3.2 events are fixed, and listener secrets are redacted.
• Dagster: asset groups and asset query metadata for SQL lineage.
• Backward-coverage tests for each tool run in CI, including Airflow 3.2 and Dagster 1.9 to 1.13.24.

Release 0.1.5 on GitHub

A bad value or a down network never breaks your pipeline.

• Batches that cannot be delivered wait on disk and go out after the next send that works; refused batches are split and kept aside.
• Spark events are flushed in the background.
• Failed tasks are forwarded.
• An unreadable value costs that one field, not the whole event, and the library never raises into the tool.

Release 0.1.4 on GitHub

Version string fix.

• The Java client reports the right version.

Release 0.1.3 on GitHub

Retries, and bigger payloads.

• Retry executors for Airflow, Dagster and Prefect: an approved fix can clear a task, re-execute a run from failure, or reschedule a flow run.
• A payload too large for one request is split, and the pieces share one id.
• Airflow: asset events are forwarded.
• Operation and assertion outcomes can be recorded explicitly.

Release 0.1.2 on GitHub

Each tool names its run, task and group.

• Airflow: events name the DAG and run they belong to.
• Dagster: the run, its events and what the run points at.
• Prefect: a task run says which flow and run it belongs to.
• Spark: the run is sent rather than every task, and every event names its application.
• Batches are capped by size and nesting is counted the way each tool nests.

Release 0.1.1 on GitHub

First release.

• Plugins for Airflow, Dagster, Prefect and Great Expectations, and a Spark listener.
• Each forwards the tool's own output unchanged; the client has no dependencies of its own.

Release 0.1.0 on GitHub