Hi team, I'm encountering an issue with Prefect version 2.10.18 running on a gRPC server. Initially, I was using a

DaskTaskRunner

that spun up a local Dask cluster, but we are transitioning to using a separate Dask cluster. The Setup: • Most of the task and flow results are protobuf messages. • I'm using caching to reuse outputs between runs. The Issue: When caching is enabled and I retrieve cached results, the outputs are unexpectedly being converted to DataFrames. This is causing issues because I expect them to remain in their original protobuf message format. What I’ve Tried: • Switching from a local

DaskTaskRunner

to a remote Dask cluster. • Ensuring that serialization is handled properly for protobuf messages. My Questions: 1. Is there a known issue or behavior in Prefect 2.10.18 where cached results might be implicitly converted to DataFrames? 2. Are there specific steps to ensure protobuf messages are serialized/deserialized correctly when using caching and a remote Dask cluster? 3. Should I be using a custom serializer or adjusting how results are stored to prevent this conversion? Any insights or guidance would be greatly appreciated. Let me know if I can provide additional context! Thanks in advance!

hi @Babaian Daniel - a couple points • result caching / serialization does not depend on choice of task runner (i.e. dask) • you can customize serialization/deserialization of results by writing your own serializer, though most often the default cloudpickle serializer is most compatible