kaxil opened a new issue, #67236:
URL: https://github.com/apache/airflow/issues/67236

   ## Problem
   
   Users browsing the [Airflow Registry](https://airflow.apache.org/registry/) 
cannot tell which operators support resumption -- i.e., operators that can pick 
up from where they left off after a worker crash, restart, or retry, rather 
than re-running from scratch.
   
   This matters for users planning long-running or expensive workloads (data 
transfers, ML training, large queries) where re-running from scratch on retry 
is wasteful or unsafe.
   
   ## Proposal
   
   Add a visible indicator on operator pages (and a filter on listing pages) in 
the Airflow Registry that marks operators as resumable.
   
   Open questions to resolve in this issue:
   
   - **Definition.** What qualifies as "resumable"? Candidates: deferrable 
operators (trigger-based resumption), operators with explicit checkpointing, 
idempotent transfer operators that can resume from offset/cursor, or some 
combination. Pick one definition or model it as multiple flags.
   - **Source of truth.** How do we get this signal? Options: a class attribute 
on the operator (e.g. `resumable: ClassVar[bool] = True`), inference from 
`BaseOperator` subclass traits (e.g. presence of `defer()` / 
`execute_complete()`), or curated metadata in `provider.yaml`. A class 
attribute is the most discoverable and authoritative.
   - **Registry surfacing.** Badge on the operator detail page + a filter chip 
on the operator list page. Tooltip explaining what "resumable" means with a 
link to docs.
   
   Implementation likely spans two repos:
   - `apache/airflow` -- metadata on the operators
   - `apache/airflow-site` -- rendering in `landing-pages/` (registry)
   
   ## Out of scope
   
   - Implementing resumability in operators that don't have it. This issue is 
about *surfacing* the capability, not adding it.
   - Changing how deferrable operators work.
   
   ## Project tracking
   
   https://github.com/orgs/apache/projects/612


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to