Add CI/CD workflow

database.py

@@ -111,6 +111,27 @@ BEGIN
                    WHERE table_name = 'pending_runs' AND column_name = 'quality_playlists') THEN
         ALTER TABLE pending_runs ADD COLUMN quality_playlists JSONB;
     END IF;
+    -- Checkpoint columns for resumable renders
+    IF NOT EXISTS (SELECT 1 FROM information_schema.columns
+                   WHERE table_name = 'pending_runs' AND column_name = 'checkpoint_frame') THEN
+        ALTER TABLE pending_runs ADD COLUMN checkpoint_frame INTEGER;
+    END IF;
+    IF NOT EXISTS (SELECT 1 FROM information_schema.columns
+                   WHERE table_name = 'pending_runs' AND column_name = 'checkpoint_t') THEN
+        ALTER TABLE pending_runs ADD COLUMN checkpoint_t FLOAT;
+    END IF;
+    IF NOT EXISTS (SELECT 1 FROM information_schema.columns
+                   WHERE table_name = 'pending_runs' AND column_name = 'checkpoint_scans') THEN
+        ALTER TABLE pending_runs ADD COLUMN checkpoint_scans JSONB;
+    END IF;
+    IF NOT EXISTS (SELECT 1 FROM information_schema.columns
+                   WHERE table_name = 'pending_runs' AND column_name = 'total_frames') THEN
+        ALTER TABLE pending_runs ADD COLUMN total_frames INTEGER;
+    END IF;
+    IF NOT EXISTS (SELECT 1 FROM information_schema.columns
+                   WHERE table_name = 'pending_runs' AND column_name = 'resumable') THEN
+        ALTER TABLE pending_runs ADD COLUMN resumable BOOLEAN DEFAULT TRUE;
+    END IF;
 END $$;
 
 CREATE INDEX IF NOT EXISTS idx_pending_runs_status ON pending_runs(status);
@@ -1530,7 +1551,9 @@ async def get_pending_run(run_id: str) -> Optional[dict]:
     async with pool.acquire() as conn:
         row = await conn.fetchrow(
             """
-            SELECT run_id, celery_task_id, status, recipe, inputs, dag_json, plan_cid, output_name, actor_id, error, ipfs_playlist_cid, quality_playlists, created_at, updated_at
+            SELECT run_id, celery_task_id, status, recipe, inputs, dag_json, plan_cid, output_name, actor_id, error,
+                   ipfs_playlist_cid, quality_playlists, checkpoint_frame, checkpoint_t, checkpoint_scans,
+                   total_frames, resumable, created_at, updated_at
             FROM pending_runs WHERE run_id = $1
             """,
             run_id
@@ -1544,6 +1567,10 @@ async def get_pending_run(run_id: str) -> Optional[dict]:
             quality_playlists = row.get("quality_playlists")
             if isinstance(quality_playlists, str):
                 quality_playlists = _json.loads(quality_playlists)
+            # Parse checkpoint_scans if it's a string
+            checkpoint_scans = row.get("checkpoint_scans")
+            if isinstance(checkpoint_scans, str):
+                checkpoint_scans = _json.loads(checkpoint_scans)
             return {
                 "run_id": row["run_id"],
                 "celery_task_id": row["celery_task_id"],
@@ -1557,6 +1584,11 @@ async def get_pending_run(run_id: str) -> Optional[dict]:
                 "error": row["error"],
                 "ipfs_playlist_cid": row["ipfs_playlist_cid"],
                 "quality_playlists": quality_playlists,
+                "checkpoint_frame": row.get("checkpoint_frame"),
+                "checkpoint_t": row.get("checkpoint_t"),
+                "checkpoint_scans": checkpoint_scans,
+                "total_frames": row.get("total_frames"),
+                "resumable": row.get("resumable", True),
                 "created_at": row["created_at"].isoformat() if row["created_at"] else None,
                 "updated_at": row["updated_at"].isoformat() if row["updated_at"] else None,
             }
@@ -1666,6 +1698,109 @@ async def update_pending_run_playlist(run_id: str, ipfs_playlist_cid: str, quali
         return "UPDATE 1" in result
 
 
+async def update_pending_run_checkpoint(
+    run_id: str,
+    checkpoint_frame: int,
+    checkpoint_t: float,
+    checkpoint_scans: Optional[dict] = None,
+    total_frames: Optional[int] = None,
+) -> bool:
+    """Update checkpoint state for a streaming run.
+
+    Called at segment boundaries to enable resume after failures.
+
+    Args:
+        run_id: The run ID
+        checkpoint_frame: Last completed frame at segment boundary
+        checkpoint_t: Time value for checkpoint frame
+        checkpoint_scans: Accumulated scan state {scan_name: state_dict}
+        total_frames: Total expected frames (for progress %)
+    """
+    if pool is None:
+        raise RuntimeError("Database pool not initialized - call init_db() first")
+    async with pool.acquire() as conn:
+        result = await conn.execute(
+            """
+            UPDATE pending_runs SET
+                checkpoint_frame = $2,
+                checkpoint_t = $3,
+                checkpoint_scans = $4,
+                total_frames = COALESCE($5, total_frames),
+                updated_at = NOW()
+            WHERE run_id = $1
+            """,
+            run_id,
+            checkpoint_frame,
+            checkpoint_t,
+            _json.dumps(checkpoint_scans) if checkpoint_scans else None,
+            total_frames,
+        )
+        return "UPDATE 1" in result
+
+
+async def get_run_checkpoint(run_id: str) -> Optional[dict]:
+    """Get checkpoint data for resuming a run.
+
+    Returns:
+        Dict with checkpoint_frame, checkpoint_t, checkpoint_scans, quality_playlists, etc.
+        or None if no checkpoint exists
+    """
+    if pool is None:
+        raise RuntimeError("Database pool not initialized - call init_db() first")
+    async with pool.acquire() as conn:
+        row = await conn.fetchrow(
+            """
+            SELECT checkpoint_frame, checkpoint_t, checkpoint_scans, total_frames,
+                   quality_playlists, ipfs_playlist_cid, resumable
+            FROM pending_runs WHERE run_id = $1
+            """,
+            run_id
+        )
+        if row and row.get("checkpoint_frame") is not None:
+            # Parse JSONB fields
+            checkpoint_scans = row.get("checkpoint_scans")
+            if isinstance(checkpoint_scans, str):
+                checkpoint_scans = _json.loads(checkpoint_scans)
+            quality_playlists = row.get("quality_playlists")
+            if isinstance(quality_playlists, str):
+                quality_playlists = _json.loads(quality_playlists)
+            return {
+                "frame_num": row["checkpoint_frame"],
+                "t": row["checkpoint_t"],
+                "scans": checkpoint_scans or {},
+                "total_frames": row.get("total_frames"),
+                "quality_playlists": quality_playlists,
+                "ipfs_playlist_cid": row.get("ipfs_playlist_cid"),
+                "resumable": row.get("resumable", True),
+            }
+        return None
+
+
+async def clear_run_checkpoint(run_id: str) -> bool:
+    """Clear checkpoint data for a run (used on restart).
+
+    Args:
+        run_id: The run ID
+    """
+    if pool is None:
+        raise RuntimeError("Database pool not initialized - call init_db() first")
+    async with pool.acquire() as conn:
+        result = await conn.execute(
+            """
+            UPDATE pending_runs SET
+                checkpoint_frame = NULL,
+                checkpoint_t = NULL,
+                checkpoint_scans = NULL,
+                quality_playlists = NULL,
+                ipfs_playlist_cid = NULL,
+                updated_at = NOW()
+            WHERE run_id = $1
+            """,
+            run_id,
+        )
+        return "UPDATE 1" in result
+
+
 async def complete_pending_run(run_id: str) -> bool:
     """Remove a pending run after it completes (moves to run_cache)."""
     async with pool.acquire() as conn: