Fix concurrency issues across storage layer

.github/workflows/ci.yml

@@ -63,7 +63,7 @@ jobs:
 
     - name: Test with pytest
       run: |
-        pytest tests/ -v --cov=sugar --cov-report=xml --cov-report=term-missing --tb=short --ignore=tests/plugin/
+        pytest tests/ -v --cov=sugar --cov-report=xml --cov-report=term-missing --tb=short --ignore=tests/plugin/ -m 'not benchmark'
 
     - name: Upload coverage to Codecov
       uses: codecov/codecov-action@v4

pyproject.toml

@@ -161,9 +161,10 @@ testpaths = ["tests"]
 python_files = ["test_*.py", "*_test.py"]
 addopts = "-v --cov=sugar --cov-branch --cov-report=term-missing --cov-report=xml"
 markers = [
-    "unit: Unit tests",
-    "integration: Integration tests",
-    "slow: Slow running tests"
+    "unit: Unit tests (no I/O, no database)",
+    "integration: Integration tests (real database, aiosqlite)",
+    "slow: Slow running tests (throughput, load)",
+    "benchmark: Throughput benchmarks (skip in CI, run with: pytest -m benchmark)"
 ]
 
 # MCP Registry identification

sugar/agent/subagent_manager.py

@@ -316,10 +316,17 @@ async def cancel_all(self) -> None:
 
         Note: This will attempt graceful shutdown but may not
         interrupt tasks that are already executing.
+
+        Takes a snapshot of the dict before iterating to avoid
+        RuntimeError if spawn()'s finally block mutates the dict
+        concurrently (e.g., during shutdown while tasks are in-flight).
         """
         logger.warning(f"Cancelling {len(self._active_subagents)} active sub-agents")
 
-        for task_id, subagent in self._active_subagents.items():
+        # Snapshot to avoid RuntimeError: dictionary changed size during iteration
+        active_snapshot = dict(self._active_subagents)
+
+        for task_id, subagent in active_snapshot.items():
             try:
                 await subagent.end_session()
                 logger.debug(f"Cancelled sub-agent task: {task_id}")

sugar/main.py

@@ -210,7 +210,10 @@ def signal_handler(signum, frame):
         shutdown_event.set()
         logger.info("🔔 Shutdown event triggered")
     else:
-        logger.warning("⚠️ Shutdown event not available")
+        # Fallback: if shutdown_event isn't ready yet (shouldn't happen
+        # now that we create it before registering handlers), exit cleanly.
+        logger.warning("⚠️ Shutdown event not available, forcing exit")
+        sys.exit(128 + signum)
 
 
 @click.group(invoke_without_command=True)
@@ -2035,6 +2038,11 @@ def run(ctx, dry_run, once, validate):
             asyncio.run(validate_config(sugar_loop))
             return
 
+        # Create shutdown event BEFORE registering signal handlers to avoid
+        # a window where a signal arrives but the event doesn't exist yet.
+        global shutdown_event
+        shutdown_event = asyncio.Event()
+
         # Set up signal handlers for graceful shutdown
         signal.signal(signal.SIGINT, signal_handler)
         signal.signal(signal.SIGTERM, signal_handler)
@@ -2117,9 +2125,10 @@ async def run_once(sugar_loop):
 
 
 async def run_continuous(sugar_loop):
-    """Run Sugar continuously"""
+    """Run Sugar continuously (shutdown_event created before signal handlers)"""
     global shutdown_event
-    shutdown_event = asyncio.Event()
+    if shutdown_event is None:
+        shutdown_event = asyncio.Event()
 
     # Create PID file for stop command
     import os

sugar/memory/store.py

@@ -7,6 +7,7 @@
 import json
 import logging
 import sqlite3
+import threading
 import uuid
 from datetime import datetime, timezone
 from pathlib import Path
@@ -59,6 +60,7 @@ def __init__(
         self.embedder = embedder or create_embedder()
         self._has_vec = self._check_sqlite_vec()
         self._conn: Optional[sqlite3.Connection] = None
+        self._lock = threading.Lock()
 
         self._init_db()
 
@@ -73,9 +75,14 @@ def _check_sqlite_vec(self) -> bool:
             return False
 
     def _get_connection(self) -> sqlite3.Connection:
-        """Get or create database connection."""
+        """Get or create database connection.
+
+        Uses check_same_thread=False to allow safe cross-thread use when
+        called from asyncio's run_in_executor. Thread safety is ensured
+        by self._lock around all public methods that access the connection.
+        """
         if self._conn is None:
-            self._conn = sqlite3.connect(str(self.db_path))
+            self._conn = sqlite3.connect(str(self.db_path), check_same_thread=False)
             self._conn.row_factory = sqlite3.Row
 
             if self._has_vec:
@@ -209,96 +216,106 @@ def store(self, entry: MemoryEntry) -> str:
         if entry.created_at is None:
             entry.created_at = datetime.now(timezone.utc)
 
-        conn = self._get_connection()
-        cursor = conn.cursor()
+        with self._lock:
+            conn = self._get_connection()
+            cursor = conn.cursor()
 
-        # Store main entry
-        cursor.execute(
-            """
-            INSERT OR REPLACE INTO memory_entries
-            (id, memory_type, source_id, content, summary, metadata,
-             importance, created_at, last_accessed_at, access_count, expires_at)
-            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-        """,
-            (
-                entry.id,
+            # Store main entry
+            cursor.execute(
+                """
+                INSERT OR REPLACE INTO memory_entries
+                (id, memory_type, source_id, content, summary, metadata,
+                 importance, created_at, last_accessed_at, access_count, expires_at)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+            """,
                 (
-                    entry.memory_type.value
-                    if isinstance(entry.memory_type, MemoryType)
-                    else entry.memory_type
+                    entry.id,
+                    (
+                        entry.memory_type.value
+                        if isinstance(entry.memory_type, MemoryType)
+                        else entry.memory_type
+                    ),
+                    entry.source_id,
+                    entry.content,
+                    entry.summary,
+                    json.dumps(entry.metadata) if entry.metadata else None,
+                    entry.importance,
+                    entry.created_at.isoformat() if entry.created_at else None,
+                    (
+                        entry.last_accessed_at.isoformat()
+                        if entry.last_accessed_at
+                        else None
+                    ),
+                    entry.access_count,
+                    entry.expires_at.isoformat() if entry.expires_at else None,
                 ),
-                entry.source_id,
-                entry.content,
-                entry.summary,
-                json.dumps(entry.metadata) if entry.metadata else None,
-                entry.importance,
-                entry.created_at.isoformat() if entry.created_at else None,
-                entry.last_accessed_at.isoformat() if entry.last_accessed_at else None,
-                entry.access_count,
-                entry.expires_at.isoformat() if entry.expires_at else None,
-            ),
-        )
+            )
 
-        # Generate and store embedding if we have semantic search
-        if self._has_vec and not isinstance(self.embedder, FallbackEmbedder):
-            try:
-                embedding = self.embedder.embed(entry.content)
-                if embedding:
-                    cursor.execute(
-                        """
-                        INSERT OR REPLACE INTO memory_vectors (id, embedding)
-                        VALUES (?, ?)
-                    """,
-                        (entry.id, _serialize_embedding(embedding)),
-                    )
-            except Exception as e:
-                logger.warning(f"Failed to store embedding: {e}")
+            # Generate and store embedding if we have semantic search
+            if self._has_vec and not isinstance(self.embedder, FallbackEmbedder):
+                try:
+                    embedding = self.embedder.embed(entry.content)
+                    if embedding:
+                        cursor.execute(
+                            """
+                            INSERT OR REPLACE INTO memory_vectors (id, embedding)
+                            VALUES (?, ?)
+                        """,
+                            (entry.id, _serialize_embedding(embedding)),
+                        )
+                except Exception as e:
+                    logger.warning(f"Failed to store embedding: {e}")
 
-        conn.commit()
-        return entry.id
+            conn.commit()
+            return entry.id
 
     def get(self, entry_id: str) -> Optional[MemoryEntry]:
         """Get a memory entry by ID."""
-        conn = self._get_connection()
-        cursor = conn.cursor()
+        with self._lock:
+            conn = self._get_connection()
+            cursor = conn.cursor()
 
-        cursor.execute(
-            """
-            SELECT * FROM memory_entries WHERE id = ?
-        """,
-            (entry_id,),
-        )
+            cursor.execute(
+                """
+                SELECT * FROM memory_entries WHERE id = ?
+            """,
+                (entry_id,),
+            )
 
-        row = cursor.fetchone()
-        if row:
-            return self._row_to_entry(row)
-        return None
+            row = cursor.fetchone()
+            if row:
+                return self._row_to_entry(row)
+            return None
 
     def delete(self, entry_id: str) -> bool:
         """Delete a memory entry."""
-        conn = self._get_connection()
-        cursor = conn.cursor()
+        with self._lock:
+            conn = self._get_connection()
+            cursor = conn.cursor()
 
-        cursor.execute("DELETE FROM memory_entries WHERE id = ?", (entry_id,))
+            cursor.execute("DELETE FROM memory_entries WHERE id = ?", (entry_id,))
 
-        if self._has_vec:
-            try:
-                cursor.execute("DELETE FROM memory_vectors WHERE id = ?", (entry_id,))
-            except Exception:
-                pass
+            if self._has_vec:
+                try:
+                    cursor.execute(
+                        "DELETE FROM memory_vectors WHERE id = ?", (entry_id,)
+                    )
+                except Exception:
+                    pass
 
-        conn.commit()
-        return cursor.rowcount > 0
+            conn.commit()
+            return cursor.rowcount > 0
 
     def search(self, query: MemoryQuery) -> List[MemorySearchResult]:
         """
         Search memories.
 
         Uses vector similarity if available, falls back to FTS5.
         """
-        if self._has_vec and not isinstance(self.embedder, FallbackEmbedder):
-            return self._search_semantic(query)
-        return self._search_keyword(query)
+        with self._lock:
+            if self._has_vec and not isinstance(self.embedder, FallbackEmbedder):
+                return self._search_semantic(query)
+            return self._search_keyword(query)
 
     def _search_semantic(self, query: MemoryQuery) -> List[MemorySearchResult]:
         """Search using vector similarity."""
@@ -638,6 +655,7 @@ def prune_expired(self) -> int:
 
     def close(self):
         """Close database connection."""
-        if self._conn:
-            self._conn.close()
-            self._conn = None
+        with self._lock:
+            if self._conn:
+                self._conn.close()
+                self._conn = None

sugar/storage/issue_response_manager.py

@@ -2,6 +2,7 @@
 Issue Response Manager - Track GitHub issue responses
 """
 
+import asyncio
 import json
 import logging
 import uuid
@@ -19,41 +20,46 @@ class IssueResponseManager:
     def __init__(self, db_path: str = ".sugar/sugar.db"):
         self.db_path = db_path
         self._initialized = False
+        self._init_lock = asyncio.Lock()
 
     async def initialize(self) -> None:
         """Create table if not exists"""
         if self._initialized:
             return
 
-        async with aiosqlite.connect(self.db_path) as db:
-            await db.execute(
+        async with self._init_lock:
+            if self._initialized:
+                return
+
+            async with aiosqlite.connect(self.db_path) as db:
+                await db.execute(
+                    """
+                    CREATE TABLE IF NOT EXISTS issue_responses (
+                        id TEXT PRIMARY KEY,
+                        repo TEXT NOT NULL,
+                        issue_number INTEGER NOT NULL,
+                        response_type TEXT NOT NULL,
+                        work_item_id TEXT,
+                        confidence REAL,
+                        posted_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+                        response_content TEXT,
+                        labels_applied TEXT,
+                        was_auto_posted BOOLEAN DEFAULT 0,
+                        UNIQUE(repo, issue_number, response_type)
+                    )
                 """
-                CREATE TABLE IF NOT EXISTS issue_responses (
-                    id TEXT PRIMARY KEY,
-                    repo TEXT NOT NULL,
-                    issue_number INTEGER NOT NULL,
-                    response_type TEXT NOT NULL,
-                    work_item_id TEXT,
-                    confidence REAL,
-                    posted_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
-                    response_content TEXT,
-                    labels_applied TEXT,
-                    was_auto_posted BOOLEAN DEFAULT 0,
-                    UNIQUE(repo, issue_number, response_type)
                 )
-            """
-            )
 
-            await db.execute(
+                await db.execute(
+                    """
+                    CREATE INDEX IF NOT EXISTS idx_issue_responses_repo_number
+                    ON issue_responses (repo, issue_number)
                 """
-                CREATE INDEX IF NOT EXISTS idx_issue_responses_repo_number
-                ON issue_responses (repo, issue_number)
-            """
-            )
+                )
 
-            await db.commit()
+                await db.commit()
 
-        self._initialized = True
+            self._initialized = True
         logger.debug(f"Issue response manager initialized: {self.db_path}")
 
     async def has_responded(