Skip to content

💡 Proposal: get_state() and set_state() on the Cache interface #680

@aaronsteers

Description

@aaronsteers

This proposal would give a public API for reading and writing stream state for a Cache interface.

Signature might be something like:

class CacheBase:
    ...

    def get_state(
        stream_name: str,
    ) -> StreamState | None:
        """Return a stream state object for the provided stream name.

        This is a thin wrapper around the internal `StateProvider` interface.
        """
        ...

    def set_state(
        stream_name: str,
        stream_state: StreamState | dict,
    ) -> None:
        """Set a stream state object for the provided stream name.

        This is a thin wrapper around the internal `StateWriter` interface.
        """
        ...


    def migrate_state(
        streams: list[str] | Literal["*"],
        to_cache: CacheBase,
    ) -> None:
        """Copies all matching stream states to the specified `Cache` object.

        This is a thin wrapper around the respective `Cache` objects'
        `get_state` and `set_state` methods.
        """
        ...

Other considerations:

  1. Since there are as of now many different strongly typed State classes (StateMessage, StreamState, etc.), we'd want to think carefully on which (if any) we feel comfortable to make a part of the public interface.
  2. It might actually be cleaner to get/set values as dict objects, since that would fully avoid needing to expose a public API for the state object itself. (Tradeoff: dict objects may be difficult to parse.)
  3. It should be noted that manually modifying a state artifact for a stream basically always "voids the warranty", and sources and not guaranteed to have stable state artifact interfaces over time.
  4. This feature is actually more appropriate for migration of state - such as during renames or moving from one state backend to another, or from one table name/alias to a new one. For this reason, I've included a possible migrate_state() method in the above which could streamline a "copy-all"-type operation.
  5. Internally, states are often cached in memory by the StateProvider class. We'd need to take care to invalidate or refresh the caches after a "set" action.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions