Summary
The user discusses the triggering of Flyte workflows based on external dependencies like dataset updates or new events. They note that this feature is partially supported in the Union variant of Flyte but not in the open-source version. Implementing it would require a webhook interface for monitoring files and initiating workflows, which involves complexities such as file watching, authentication, and authorization. The user suggests consulting Flyte documentation for deployment agent setup related to sensors and mentions that the current implementation is not truly reactive, complicating its justification as a Flyte feature. They also inquire about a scenario where a Flyte workflow scheduled to run every hour depends on an S3 file delayed by 2 hours, leading to the potential for two workflow instances to run if a custom sensor is created to monitor those events. They note that currently, yes, there would be two instances running, but there is an RFC about execution concurrency to address this issue.