memory.config_loader failed: config loader answered from the module's load-time snapshot: this module
re-reads no config file, so a settings change made after it loaded (Composio mode, sync cadence, a memory
source switched off) does not reach the engine in this process until the host reloads the module
Summary
Two host-visible defects in
tinymemory-moduleseen in OpenHuman's logs on 2026-09-03 (pinned artifact v1.13.7). Neither causes the memory-hang incident tracked in tinyhumansai/openhuman#6005, but both surface asreport_errorevents in module mode.Problem
Shutdownis a deliberate stub (crates/tinymemory-module/src/lib.rs, "Shutdown stays a stub: no bus interface serves it and no local answer"). The host'smemory.shutdown_hosttherefore fails in module mode:Consequence per the message itself: queue job locks are not released on a clean exit, and the next launch's ingestion waits out the lease.
config_loaderanswers from the load-time snapshot and reports it as an error whenever a setting changes after load:That is a known design limit, not a failure; as
report_errorit reaches Sentry on every settings change.Proposed fix
Shutdownon the bus (release queue job locks, stop the periodic schedulers) and have the host call it during shutdown.ReloadConfigmember the host can call after a settings change, or downgrade the snapshot answer to a warning and document the restart requirement.Both need a module release and a registry re-pin in the host before they take effect.
Related