Summarize streams with real progress: adapters reassemble SSE/NDJSON deltas and report 0-99% as tokens arrive

backend/shonar/services/processing.py

@@ -90,6 +90,49 @@ def _thread_progress_reporter(job_id):
 
     return report
 
+
+def _async_progress_writer(job_id):
+    """Progress callback for providers that stream on our own event loop
+    (the LLM adapters). Unlike _thread_progress_reporter there is no
+    worker thread: schedule the tiny write as a task on the running loop.
+
+    Same contract as the thread version — best-effort display state — and
+    identical values are coalesced so a chatty stream does not hammer
+    SQLite."""
+    loop = asyncio.get_running_loop()
+    last = -1
+
+    def report(pct: int) -> None:
+        nonlocal last
+        if int(pct) == last:
+            return
+        last = int(pct)
+
+        async def _write() -> None:
+            from sqlalchemy import update
+
+            from shonar.db.session import session_factory
+
+            try:
+                async with session_factory()() as s:
+                    await s.execute(
+                        update(ProcessingJob)
+                        .where(ProcessingJob.id == job_id)
+                        .values(progress=last)
+                    )
+                    await s.commit()
+            except Exception:  # pragma: no cover - display state only
+                logger.debug("summarize progress write failed for job %s",
+                             job_id, exc_info=True)
+
+        try:
+            loop.create_task(_write())
+        except RuntimeError:  # loop already gone (shutdown race)
+            logger.debug("summarize progress dropped for job %s", job_id)
+
+    return report
+
+
 MAX_TRIES = 3
 
 # A `running` job younger than this is treated as live work, not a crash
@@ -437,8 +480,9 @@ async def run_summarize(ctx: dict, recording_id: str) -> None:
         rec.processing_status = ProcessingStatus.processing
         await session.commit()  # visible before the long LLM call
         try:
-            result = await provider.summarize(text, title=rec.title,
-                                              tone=job.tone)
+            result = await provider.summarize(
+                text, title=rec.title, tone=job.tone,
+                on_progress=_async_progress_writer(job.id))
         except AIError as e:
             await _fail(session, rec, job, str(e), ctx, e)
             await session.commit()