personal memory agent
fix: persist stream name in sync history to prevent verification mismatches
The upload path saved files under a stream-derived directory but did not
record the stream name in the sync history. The query path re-derived
the stream independently, which could produce a different result when
_strip_hostname() behaved differently or the client sent a different
stream query param. This caused the April 9 circuit breaker outage.
Now the stream is persisted in the sync history record at upload time.
The query handler reads it from the record, falling back to derivation
for legacy records that lack the field.
Files changed:
- apps/observer/routes.py
- apps/observer/tests/test_routes.py
+5
-2
+4
-2
apps/observer/routes.py
+4
-2
apps/observer/routes.py
···
535
535
sync_record = {
536
536
"ts": now_ms(),
537
537
"segment": segment,
538
+
"stream": stream,
538
539
"files": file_records,
539
540
}
540
541
if segment != original_segment:
···
702
703
client_stream = request.args.get("stream", "").strip()
703
704
observer_name = observer.get("name", "unknown")
704
705
if client_stream and re.match(r"^[a-z0-9][a-z0-9._-]*$", client_stream):
705
-
stream = client_stream
706
+
fallback_stream = client_stream
706
707
else:
707
-
stream = stream_name(observer=observer_name)
708
+
fallback_stream = stream_name(observer=observer_name)
708
709
709
710
# Build response grouped by segment, deduplicating by sha256
710
711
# Later records overwrite earlier ones (most recent upload wins)
···
719
720
continue
720
721
721
722
segment = record.get("segment", "")
723
+
stream = record.get("stream", fallback_stream)
722
724
segment_original = record.get("segment_original")
723
725
724
726
if segment not in segments:
+1
apps/observer/tests/test_routes.py
+1
apps/observer/tests/test_routes.py