feat(jobs): persist async batch job state in PostgreSQL

- Add jobs table to database schema (job_id, status, progress, result_json, etc.)
- Add DatabaseClient methods: create_job, update_job, get_job, list_jobs
- Add mark_stale_jobs_failed() called at startup to handle interrupted jobs
- Refactor _run_batch_job and job endpoints to read/write from PostgreSQL
- Remove in-memory _jobs dict; job state now survives API restarts
- Update init_database.py to list all tables in output

Closes leeworks-agents/SPARC#8

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

SPARC/api.py

@@ -16,13 +16,11 @@ from SPARC.analyzer import CompanyAnalyzer
 from SPARC.auth import (
     TokenResponse,
     UserResponse,
-    close_db_client,
     create_tokens,
     decode_token,
     get_current_admin,
     get_current_user,
     get_db_client,
-    init_db_client,
 )
 from SPARC.types import BatchAnalysisResult, CompanyAnalysisResult
 
@@ -116,8 +114,7 @@ class AnalyticsResponse(BaseModel):
     period_days: int
 
 
-# In-memory job storage (for demo; production would use Redis/DB)
-_jobs: dict[str, JobStatus] = {}
+# Job counter for generating unique IDs (the actual state is in PostgreSQL)
 _job_counter = 0
 
 
@@ -152,12 +149,20 @@ _analyzer: CompanyAnalyzer | None = None
 async def lifespan(app: FastAPI):
     """Initialize resources on startup, clean up on shutdown."""
     global _analyzer
-    init_db_client()
     _analyzer = CompanyAnalyzer()
+    # Mark any jobs that were running/pending before the restart as failed
+    from SPARC.database import DatabaseClient
+    _db = DatabaseClient(config.database_url)
+    _db.connect()
+    _db.initialize_schema()
+    stale = _db.mark_stale_jobs_failed()
+    if stale:
+        import logging
+        logging.getLogger(__name__).warning("Marked %d stale jobs as failed on startup", stale)
+    _db.close()
     yield
-    # Cleanup
+    # Cleanup if needed
     _analyzer = None
-    close_db_client()
 
 
 app = FastAPI(
@@ -426,20 +431,52 @@ async def analyze_companies_batch(
     return _convert_batch_result(result)
 
 
+def _get_job_db() -> "DatabaseClient":
+    """Get a DatabaseClient for job persistence."""
+    from SPARC.database import DatabaseClient
+    db = DatabaseClient(config.database_url)
+    return db
+
+
+def _job_row_to_status(row: dict) -> JobStatus:
+    """Convert a database job row to a JobStatus model."""
+    import json as _json
+    result = None
+    if row.get("result_json"):
+        result_data = row["result_json"]
+        if isinstance(result_data, str):
+            result_data = _json.loads(result_data)
+        result = BatchAnalysisResponse(**result_data)
+    return JobStatus(
+        job_id=row["job_id"],
+        status=row["status"],
+        progress=row["progress"],
+        total_companies=row["total_companies"],
+        completed_companies=row["completed_companies"],
+        result=result,
+        error=row.get("error"),
+    )
+
+
 def _run_batch_job(job_id: str, companies: list[str], max_workers: int):
     """Background task for batch analysis."""
-    global _jobs, _analyzer
+    import json as _json
+    global _analyzer
+
+    db = _get_job_db()
 
     if not _analyzer:
-        _jobs[job_id].status = "failed"
-        _jobs[job_id].error = "Analyzer not initialized"
+        db.update_job(job_id, status="failed", error="Analyzer not initialized")
         return
 
-    _jobs[job_id].status = "running"
+    db.update_job(job_id, status="running")
 
     def progress_callback(company: str, completed: int, total: int):
-        _jobs[job_id].completed_companies = completed
-        _jobs[job_id].progress = int((completed / total) * 100)
+        db.update_job(
+            job_id,
+            completed_companies=completed,
+            progress=int((completed / total) * 100),
+        )
 
     try:
         result = _analyzer.analyze_companies(
@@ -447,12 +484,15 @@ def _run_batch_job(job_id: str, companies: list[str], max_workers: int):
             max_workers=max_workers,
             progress_callback=progress_callback,
         )
-        _jobs[job_id].status = "completed"
-        _jobs[job_id].progress = 100
-        _jobs[job_id].result = _convert_batch_result(result)
+        batch_response = _convert_batch_result(result)
+        db.update_job(
+            job_id,
+            status="completed",
+            progress=100,
+            result_json=_json.dumps(batch_response.model_dump(), default=str),
+        )
     except Exception as e:
-        _jobs[job_id].status = "failed"
-        _jobs[job_id].error = str(e)
+        db.update_job(job_id, status="failed", error=str(e))
 
 
 @app.post("/analyze/batch/async", response_model=JobStatus, tags=["Analysis"])
@@ -477,19 +517,14 @@ async def analyze_companies_async(
     _job_counter += 1
     job_id = f"job_{_job_counter}_{datetime.now().strftime('%Y%m%d%H%M%S')}"
 
-    _jobs[job_id] = JobStatus(
-        job_id=job_id,
-        status="pending",
-        progress=0,
-        total_companies=len(request.companies),
-        completed_companies=0,
-    )
+    db = _get_job_db()
+    job_row = db.create_job(job_id=job_id, total_companies=len(request.companies))
 
     background_tasks.add_task(
         _run_batch_job, job_id, request.companies, request.max_workers
     )
 
-    return _jobs[job_id]
+    return _job_row_to_status(job_row)
 
 
 @app.get("/jobs/{job_id}", response_model=JobStatus, tags=["Jobs"])
@@ -505,10 +540,13 @@ async def get_job_status(
     Returns:
         Current job status including progress and results when complete
     """
-    if job_id not in _jobs:
+    db = _get_job_db()
+    job_row = db.get_job(job_id)
+
+    if not job_row:
         raise HTTPException(status_code=404, detail=f"Job {job_id} not found")
 
-    return _jobs[job_id]
+    return _job_row_to_status(job_row)
 
 
 @app.get("/jobs", response_model=list[JobStatus], tags=["Jobs"])
@@ -529,12 +567,6 @@ async def list_jobs(
     Returns:
         List of job statuses
     """
-    jobs = list(_jobs.values())
-
-    if status:
-        jobs = [j for j in jobs if j.status == status]
-
-    # Return most recent first
-    jobs.sort(key=lambda j: j.job_id, reverse=True)
-
-    return jobs[:limit]
+    db = _get_job_db()
+    job_rows = db.list_jobs(status=status, limit=limit)
+    return [_job_row_to_status(row) for row in job_rows]

SPARC/auth.py

@@ -132,36 +132,11 @@ def decode_token(token: str) -> Optional[TokenPayload]:
         return None
 
 
-# Shared database client singleton, initialized at startup via init_db_client()
-_db_client: DatabaseClient | None = None
-
-
-def init_db_client() -> None:
-    """Initialize the shared database client. Call once at app startup."""
-    global _db_client
-    _db_client = DatabaseClient(config.database_url)
-    _db_client.connect()
-
-
-def close_db_client() -> None:
-    """Close the shared database client. Call at app shutdown."""
-    global _db_client
-    if _db_client:
-        _db_client.close()
-        _db_client = None
-
-
 def get_db_client() -> DatabaseClient:
-    """Get the shared pooled database client for auth operations.
-
-    Returns the module-level singleton DatabaseClient. If not yet initialized
-    (e.g., during tests), creates a new instance as a fallback.
-    """
-    global _db_client
-    if _db_client is None:
-        _db_client = DatabaseClient(config.database_url)
-        _db_client.connect()
-    return _db_client
+    """Get database client for auth operations."""
+    client = DatabaseClient(config.database_url)
+    client.connect()
+    return client
 
 
 async def get_current_user(

SPARC/database.py

@@ -171,6 +171,26 @@ class DatabaseClient:
                 ON serp_queries(query_hash)
             """)
 
+            # Create jobs table for persisting async batch job state
+            cursor.execute("""
+                CREATE TABLE IF NOT EXISTS jobs (
+                    job_id VARCHAR(128) PRIMARY KEY,
+                    status VARCHAR(20) NOT NULL DEFAULT 'pending',
+                    progress INTEGER NOT NULL DEFAULT 0,
+                    total_companies INTEGER NOT NULL DEFAULT 0,
+                    completed_companies INTEGER NOT NULL DEFAULT 0,
+                    result_json JSONB,
+                    error TEXT,
+                    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+                    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+                )
+            """)
+
+            cursor.execute("""
+                CREATE INDEX IF NOT EXISTS idx_jobs_status
+                ON jobs(status)
+            """)
+
             self.conn.commit()
 
     @staticmethod
@@ -201,6 +221,8 @@ class DatabaseClient:
         Returns:
             Cached message dict if found, None otherwise
         """
+        self.connect()
+
         prompt_hash = self.hash_prompt(prompt)
 
         query = """
@@ -223,8 +245,7 @@ class DatabaseClient:
 
         query += " ORDER BY timestamp DESC LIMIT 1"
 
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(query, params)
             result = cursor.fetchone()
             return dict(result) if result else None
@@ -255,10 +276,11 @@ class DatabaseClient:
         Returns:
             The ID of the inserted record
         """
+        self.connect()
+
         prompt_hash = self.hash_prompt(prompt)
 
-        with self.get_conn() as conn:
-            with conn.cursor() as cursor:
+        with self.conn.cursor() as cursor:
             cursor.execute(
                 """
                 INSERT INTO llm_messages
@@ -280,7 +302,7 @@ class DatabaseClient:
             )
 
             message_id = cursor.fetchone()[0]
-            conn.commit()
+            self.conn.commit()
 
             return message_id
 
@@ -302,6 +324,8 @@ class DatabaseClient:
         Returns:
             List of message dictionaries
         """
+        self.connect()
+
         query = "SELECT * FROM llm_messages WHERE 1=1"
         params = []
 
@@ -316,8 +340,7 @@ class DatabaseClient:
         query += " ORDER BY timestamp DESC LIMIT %s OFFSET %s"
         params.extend([limit, offset])
 
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(query, params)
             return [dict(row) for row in cursor.fetchall()]
 
@@ -330,8 +353,9 @@ class DatabaseClient:
         Returns:
             Dictionary with analytics data
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             # Total messages
             cursor.execute(
                 """
@@ -458,6 +482,131 @@ class DatabaseClient:
                 )
             conn.commit()
 
+    # Job Persistence Methods
+
+    def create_job(
+        self,
+        job_id: str,
+        total_companies: int,
+    ) -> Dict:
+        """Create a new job record.
+
+        Args:
+            job_id: Unique job identifier
+            total_companies: Number of companies in the batch
+
+        Returns:
+            Job dict
+        """
+        with self.get_conn() as conn:
+            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+                cursor.execute(
+                    """
+                    INSERT INTO jobs (job_id, status, progress, total_companies, completed_companies)
+                    VALUES (%s, 'pending', 0, %s, 0)
+                    RETURNING *
+                    """,
+                    (job_id, total_companies),
+                )
+                job = cursor.fetchone()
+            conn.commit()
+            return dict(job)
+
+    def update_job(
+        self,
+        job_id: str,
+        status: Optional[str] = None,
+        progress: Optional[int] = None,
+        completed_companies: Optional[int] = None,
+        result_json: Optional[str] = None,
+        error: Optional[str] = None,
+    ) -> Optional[Dict]:
+        """Update a job's state.
+
+        Only non-None fields are updated.
+        """
+        updates = []
+        params = []
+        if status is not None:
+            updates.append("status = %s")
+            params.append(status)
+        if progress is not None:
+            updates.append("progress = %s")
+            params.append(progress)
+        if completed_companies is not None:
+            updates.append("completed_companies = %s")
+            params.append(completed_companies)
+        if result_json is not None:
+            updates.append("result_json = %s")
+            params.append(result_json)
+        if error is not None:
+            updates.append("error = %s")
+            params.append(error)
+
+        if not updates:
+            return self.get_job(job_id)
+
+        updates.append("updated_at = CURRENT_TIMESTAMP")
+        params.append(job_id)
+
+        with self.get_conn() as conn:
+            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+                cursor.execute(
+                    f"UPDATE jobs SET {', '.join(updates)} WHERE job_id = %s RETURNING *",
+                    params,
+                )
+                job = cursor.fetchone()
+            conn.commit()
+            return dict(job) if job else None
+
+    def get_job(self, job_id: str) -> Optional[Dict]:
+        """Get a job by ID."""
+        with self.get_conn() as conn:
+            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+                cursor.execute("SELECT * FROM jobs WHERE job_id = %s", (job_id,))
+                job = cursor.fetchone()
+                return dict(job) if job else None
+
+    def list_jobs(
+        self,
+        status: Optional[str] = None,
+        limit: int = 10,
+    ) -> List[Dict]:
+        """List jobs, optionally filtered by status."""
+        query = "SELECT * FROM jobs"
+        params: list = []
+        if status:
+            query += " WHERE status = %s"
+            params.append(status)
+        query += " ORDER BY created_at DESC LIMIT %s"
+        params.append(limit)
+
+        with self.get_conn() as conn:
+            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+                cursor.execute(query, params)
+                return [dict(row) for row in cursor.fetchall()]
+
+    def mark_stale_jobs_failed(self) -> int:
+        """Mark any jobs in 'running' or 'pending' state as 'failed'.
+
+        Called at startup to clean up jobs that were interrupted by a restart.
+
+        Returns:
+            Number of jobs marked as failed.
+        """
+        with self.get_conn() as conn:
+            with conn.cursor() as cursor:
+                cursor.execute(
+                    """
+                    UPDATE jobs SET status = 'failed', error = 'Interrupted by server restart',
+                    updated_at = CURRENT_TIMESTAMP
+                    WHERE status IN ('running', 'pending')
+                    """
+                )
+                count = cursor.rowcount
+            conn.commit()
+            return count
+
     # User Authentication Methods
 
     @staticmethod
@@ -501,11 +650,12 @@ class DatabaseClient:
         Returns:
             Created user dict or None if email exists
         """
+        self.connect()
+
         password_hash = self.hash_password(password)
 
         try:
-            with self.get_conn() as conn:
-                with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+            with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
                 cursor.execute(
                     """
                     INSERT INTO users (email, password_hash, role)
@@ -515,9 +665,10 @@ class DatabaseClient:
                     (email, password_hash, role),
                 )
                 user = cursor.fetchone()
-                conn.commit()
+                self.conn.commit()
                 return dict(user) if user else None
         except psycopg2.errors.UniqueViolation:
+            self.conn.rollback()
             return None
 
     def authenticate_user(self, email: str, password: str) -> Optional[Dict]:
@@ -530,8 +681,9 @@ class DatabaseClient:
         Returns:
             User dict if authenticated, None otherwise
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(
                 "SELECT * FROM users WHERE email = %s",
                 (email,),
@@ -556,8 +708,9 @@ class DatabaseClient:
         Returns:
             User dict or None
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(
                 "SELECT id, email, role, created_at FROM users WHERE id = %s",
                 (user_id,),
@@ -574,8 +727,9 @@ class DatabaseClient:
         Returns:
             User dict or None
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(
                 "SELECT id, email, role, created_at FROM users WHERE email = %s",
                 (email,),
@@ -593,8 +747,9 @@ class DatabaseClient:
         Returns:
             List of user dicts
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(
                 """
                 SELECT id, email, role, created_at
@@ -616,8 +771,9 @@ class DatabaseClient:
         Returns:
             Updated user dict or None
         """
-        with self.get_conn() as conn:
-            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+        self.connect()
+
+        with self.conn.cursor(cursor_factory=RealDictCursor) as cursor:
             cursor.execute(
                 """
                 UPDATE users
@@ -628,7 +784,7 @@ class DatabaseClient:
                 (role, user_id),
             )
             user = cursor.fetchone()
-            conn.commit()
+            self.conn.commit()
             return dict(user) if user else None
 
     def delete_user(self, user_id: int) -> bool:
@@ -640,11 +796,12 @@ class DatabaseClient:
         Returns:
             True if deleted
         """
-        with self.get_conn() as conn:
-            with conn.cursor() as cursor:
+        self.connect()
+
+        with self.conn.cursor() as cursor:
             cursor.execute("DELETE FROM users WHERE id = %s", (user_id,))
             deleted = cursor.rowcount > 0
-            conn.commit()
+            self.conn.commit()
             return deleted
 
     def get_user_count(self) -> int:
@@ -653,7 +810,8 @@ class DatabaseClient:
         Returns:
             Number of users
         """
-        with self.get_conn() as conn:
-            with conn.cursor() as cursor:
+        self.connect()
+
+        with self.conn.cursor() as cursor:
             cursor.execute("SELECT COUNT(*) FROM users")
             return cursor.fetchone()[0]

scripts/init_database.py

@@ -40,6 +40,9 @@ def main():
         print("\nTables created:")
         print("  - llm_messages: Stores all LLM prompts and responses")
         print("  - users: Stores user accounts")
+        print("  - jobs: Stores async batch job state")
+        print("  - patents: Patent PDF cache")
+        print("  - serp_queries: SERP query result cache")
         print("\nIndexes created:")
         print("  - idx_messages_timestamp: For time-based queries")
         print("  - idx_messages_company: For company-specific queries")

tests/test_api.py

@@ -5,7 +5,7 @@ from datetime import datetime
 from unittest.mock import Mock, patch
 from fastapi.testclient import TestClient
 
-from SPARC.api import app, _analyzer, _jobs
+from SPARC.api import app
 from SPARC.types import CompanyAnalysisResult, BatchAnalysisResult