GitHub Repo
Dify Community Reports Missing LogStore Interface Methods; Human Input Flows May Not Recover
A November 16 Dify issue report says that, under certain LogStore configurations, Human Input form submissions and timeout handling may fail, while workflow event streams may return 404. The report includes simulated tests and a proposed fix, but its evidence is limited to the main branch; the impact on released versions remains unconfirmed.

On November 16, a Dify community member filed an issue reporting that self-hosted environments may leave workflows using Human Input v2 stuck in a waiting state when API_WORKFLOW_RUN_REPOSITORY points to a LogStore implementation. Form submissions return not_found, and the workflow does not continue after a timeout. The report identifies commit 5244c1b46 on the main branch, so it does not establish that all released versions are affected. Issue report
The reporter attributes the problem to missing implementations in the repository interface: the LogStore class explicitly inherits a protocol with 41 members but overrides only 17. The remaining methods execute protocol bodies containing only an ellipsis and return None. Form submission relies on get_workflow_run_by_id_for_update to find and lock the workflow run record; as a result, an existing workflow is treated as missing. The event stream may also return 404 during lookup. Technical analysis
This could affect the availability of AI workflows that use human review. According to Dify’s official documentation, a Human Input node pauses a workflow so a user can review model output, edit it, or choose a subsequent branch. If there is no response, the workflow continues or ends according to its timeout policy. If both submission and timeout paths fail, a completed human decision cannot advance the subsequent nodes, and the application may not finish as designed. Human Input documentation
The reporter says they compared six paths using the same SQLite data with the default SQL repository and the LogStore class: the former succeeded, while the latter failed. They also provided a minimal example that does not require connecting to Alibaba Cloud SLS. The proposed fix is to delegate workflow control, retention, and archival operations to the existing SQL repository, and remove type-checking exemptions so future interface additions can reveal missing implementations. These are the submitter’s tests and proposal; engineering teams should first check their own backend configuration, then monitor upstream validation, merge status, and release announcements before deciding whether to change their deployments. Reproduction steps and proposed fix