fix: address critical architectural issues identified by reviewers

- Fix LiteLLM provider TypeError by stripping 'provider' key from config
- Fix DB adapter write-side gap allowing state store writes without conversation store
- Fix tool stacks cleanup on LangfuseSink shutdown
- Fix limit=0 logic error in get_runs/get_traces
- Add exception chaining for ImportError
- Add error handling for state store reads

Co-authored-by: Mervin Praison <MervinPraison@users.noreply.github.com>

Author: praisonai-triage-agent[bot]

src/praisonai/praisonai/db/adapter.py

@@ -290,10 +290,7 @@ def on_run_start(
         metadata: Optional[Dict[str, Any]] = None,
     ) -> None:
         """Called when a new run (turn) starts."""
-        if not self._conversation_store:
-            return
-        
-        # Store run start in state if available
+        # Store run start in state if available (even without conversation store)
         if self._state_store:
             run_key = f"run:{session_id}:{run_id}"
             self._state_store.set(run_key, {
@@ -304,6 +301,10 @@ def on_run_start(
                 "status": "running",
                 "metadata": metadata or {}
             })
+        
+        # Early return only applies to conversation store operations
+        if not self._conversation_store:
+            return
 
     def on_run_end(
         self,
@@ -347,12 +348,16 @@ def get_runs(
 
         runs = []
         for k in keys:
-            run_data = self._state_store.get(k)
-            if run_data:
+            try:
+                run_data = self._state_store.get(k)
+            except Exception as exc:
+                logger.warning("Failed to load run key '%s': %s", k, exc)
+                continue
+            if isinstance(run_data, dict):
                 runs.append(run_data)
 
         runs.sort(key=lambda r: r.get("started_at", 0), reverse=True)
-        return runs[:limit] if limit else runs
+        return runs[:limit] if limit is not None else runs
 
     def export_session(
         self,
@@ -530,16 +535,20 @@ def get_traces(
 
         traces = []
         for k in keys:
-            trace_data = self._state_store.get(k)
-            if trace_data:
+            try:
+                trace_data = self._state_store.get(k)
+            except Exception as exc:
+                logger.warning("Failed to load trace key '%s': %s", k, exc)
+                continue
+            if isinstance(trace_data, dict):
                 traces.append(trace_data)
 
         if session_id is not None:
             traces = [t for t in traces if t.get("session_id") == session_id]
         if user_id is not None:
             traces = [t for t in traces if t.get("user_id") == user_id]
         traces.sort(key=lambda t: t.get("started_at", 0), reverse=True)
-        return traces[:limit] if limit else traces
+        return traces[:limit] if limit is not None else traces
 
     def close(self) -> None:
         """Close all database connections."""

src/praisonai/praisonai/llm/registry.py

@@ -279,16 +279,19 @@ def __init__(self, model_id: str, config: Optional[Dict[str, Any]] = None):
         def generate(self, prompt: str, **kwargs):
             try:
                 import litellm  # lazy
-            except ImportError:
+            except ImportError as err:
                 raise ImportError(
                     "LiteLLM is required for built-in providers. "
                     "Install with: pip install litellm"
-                )
-            full_model = f"{self.config.get('provider', '')}/{self.model_id}".strip("/")
+                ) from err
+            provider_prefix = self.config.get("provider", "")
+            full_model = f"{provider_prefix}/{self.model_id}".strip("/") if provider_prefix else self.model_id
+            # Exclude internal 'provider' key from litellm.completion kwargs
+            passthrough_config = {k: v for k, v in self.config.items() if k != "provider"}
             return litellm.completion(
-                model=full_model or self.model_id,
+                model=full_model,
                 messages=[{"role": "user", "content": prompt}],
-                **{**self.config, **kwargs},
+                **{**passthrough_config, **kwargs},
             )
 
     def _make_litellm_factory(provider_prefix: str):

src/praisonai/praisonai/observability/langfuse.py

@@ -297,6 +297,16 @@ def close(self) -> None:
                 try:
                     # Close any remaining spans
                     with self._lock:
+                        # End any in-flight tool spans
+                        for stack in self._tool_stacks.values():
+                            while stack:
+                                try:
+                                    stack.pop().end()
+                                except Exception:
+                                    pass
+                        self._tool_stacks.clear()
+                        
+                        # End root spans
                         for span in self._spans.values():
                             try:
                                 span.end()