fix(reader): the Takeout import keeps a later upload, an un-mark and its memory bound

- An ad-hoc request made while its target's run is under way is the run after it: the scheduler
  dropped a second fire of a running id, so a second Takeout file was never imported and its key
  stayed queued until a restart.
- Un-marking or resetting a video is remembered (video_unmarks): an import of an older YouTube view
  no longer marks it watched again; un-marks take the user's Takeout lock.
- One user's failure in the Shorts job's Takeout pass is theirs: the job keeps its own result, the
  others are settled, and a failed lookup spends none of the daily share.
- The upload reads the user's id before giving the read connection back.
- A watched row waits on a discovery video (search or explore) until the catalog keeps it, so a
  mark never pins it against the cleanup jobs.
- The history is decoded a window at a time, a Watch later file over 2 MiB is refused unread, and
  two uploads at once across all users (takeout_server_busy).
- A Watch later list goes in as one presence read and one insert.
This commit is contained in:
2026-10-23 22:07:11 +02:00
parent 9f580fd8c8
commit 9c5f81fd94
17 changed files with 667 additions and 70 deletions
+2 -1
View File
@@ -75,7 +75,8 @@ DEFAULT_MAX_BODY_BYTES = 1024 * 1024
# A Google Takeout file (`reader.takeout`). A real watch history measured 3.7 MB for 9500 views, ~390
# bytes a view, so 32 MiB holds ~86k views. YouTube keeps the history 36 months by default on an
# account made since mid-2020 and for good on an older one, so a heavy viewer's can pass 16 MiB. The
# file is read an entry at a time (`takeout._json_list`), so it costs its own size, not a tree of it.
# file's text is decoded a window at a time and read an entry at a time (`takeout._json_list`), so
# it costs its bytes and a window, not a tree of it; the route takes two at once across every user.
TAKEOUT_MAX_BYTES = 32 * 1024 * 1024
ROUTE_MAX_BODY_BYTES: dict[tuple[str, str], int] = {