fix(files): review round 1, the server — the tray's window, the GC's locks, a file's origin, the upgrade path

- The queue keeps a done job by when it became done (ready_at), not by its last write: renaming,
  keeping or moving an old file, or the GC's expiry warning, no longer brings it back as just finished.
- GC expiry and eviction re-read each file under the asset→job locks the user routes take, and commit
  one file at a time: a delete, a re-download or a keep made while the GC runs is honoured, never a
  second release that frees a file another user still holds.
- A file records the origin it was placed under (download_jobs.source_key, migration 0105): an edit
  whose parent expired still goes back to its folder, not to Uploads.
- A folder a file is being put in is held (FOR SHARE) until it lands; a delete of it waits and takes
  that file home too, instead of failing on the foreign key.
- The hash route reads at most two files at once (429 past that) and holds no connection meanwhile.
- 0104's backfill dates old files by their request date only: no notice prod wrote names its job.
- The old DOWNLOAD_ROOT still finds the files; a files root that is not a mounted volume stops the
  worker and is logged by the API; the volume-copy recipe looks the project-prefixed names up.
- Stale comments about the /downloads mount and the Downloads page.
This commit is contained in:
2026-10-13 00:14:09 +02:00
parent 96509fe356
commit 815083bcc4
29 changed files with 593 additions and 92 deletions
@@ -8,9 +8,8 @@ Create Date: 2026-10-12
download that failed for days and then went through used to sit days down the list, under its
request date. Every path to done stamps it, held by `ck_download_jobs_ready_at`.
A done job from before has no record of that moment, except where its "ready to watch" notice is
still in the owner's inbox (read notices are pruned past a cap): that notice was written the moment
it finished. The rest take the request date, which is where they already sat.
A done job from before has no record of that moment (its "ready to watch" notice never named the
job, so it cannot be matched back), so it takes its request date, which is where it already sat.
"""
import sqlalchemy as sa
@@ -26,12 +25,7 @@ def upgrade() -> None:
op.add_column("download_jobs", sa.Column("ready_at", sa.DateTime(timezone=True)))
op.execute(
"""
UPDATE download_jobs j SET ready_at = COALESCE(
(SELECT max(n.created_at) FROM notifications n
WHERE n.type = 'download_done' AND n.user_id = j.user_id
AND n.data->>'job_id' = j.id::text),
j.created_at)
WHERE j.status = 'done'
UPDATE download_jobs SET ready_at = created_at WHERE status = 'done'
"""
)
op.create_check_constraint(
@@ -0,0 +1,37 @@
"""A file remembers where it came from
Revision ID: 0105_job_source_key
Revises: 0104_job_ready_at
Create Date: 2026-10-12
`download_jobs.source_key` is the origin a file came from — the source folder it may go back to —
recorded when it is placed. An edit used to work its origin out through its parent each time, and
once the parent was deleted or expired it fell back to "upload": moved into an own folder, it could
no longer go back to its folder, and deleting that folder sent it to Uploads to stay forever.
A file in a source folder takes that folder's key. One already in an own folder is left NULL and
keeps being worked out (`folders.source_key_for`), as before.
"""
import sqlalchemy as sa
from alembic import op
revision: str = "0105_job_source_key"
down_revision: str | None = "0104_job_ready_at"
branch_labels: str | None = None
depends_on: str | None = None
def upgrade() -> None:
op.add_column("download_jobs", sa.Column("source_key", sa.String(260)))
op.execute(
"""
UPDATE download_jobs j SET source_key = f.source_key
FROM file_folders f
WHERE f.id = j.folder_id AND f.kind = 'source'
"""
)
def downgrade() -> None:
op.drop_column("download_jobs", "source_key")