Airflow just made data pipelines resumable. version 3.3, out september 17, adds durable execution…
airflow just made data pipelines resumable. version 3.3, out september 17, adds durable execution with persistent state and crash recovery, and now covers ai agent runs too.
the scheduler stopped losing work when a node dies.
Context
The Apache Airflow blog post on 3.3.0, listed July 6, 2026, says 3.3.0 introduces a first-class state store for tasks and assets (AIP-103), a Language Task SDK for Java and Go (AIP-108), asset partitioning expansion and pluggable retry policies. The resumable tasks docs, listing 3.3.1 and marked added in 3.3.0, scope the feature to tasks that submit work to an external system and poll, such as Spark, BigQuery, Kubernetes and EMR. The common-ai provider docs say durable=True caches each LLM response and tool result so a retry replays completed steps, and that it only helps when the task has retries configured. The 3.3.1 release notes list ResumableJobMixin lines on Spark submit surviving worker failure, a durable toggle and crash-recovery metrics.
No source for a September 17 release was found, so the date is unverified, and it may reflect a later patch or provider release. The features are task-level retry and state replay for specific operators, and the agent coverage is cached replay on retry and not a scheduler-wide guarantee, so the scheduler stopped losing work when a node dies overstates it. This is not a guarantee that no work is lost when a node fails.
Watch next
- A dated provider release for durable agent execution.
Sources
- Apache Airflow: 3.3.0 releaseairflow.apache.org
- Airflow docs: resumable tasksairflow.apache.org
- Airflow common-ai provider: durable executionairflow.apache.org
- Airflow 3.3.1 release notesairflow.apache.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 22 September 2026 at 22:04 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →