FROSADO / onecloudriver / 31945758111

78%

Build Type

push

github

Committed by web-flow

Commit Message

fix(fs): release in-flight slot on failed uploads; keep dirty flag on empty flush (#87) (#88)

Uploads of newly created files silently never reached OneDrive. Root
cause found live: executeUpload only released its in-flight slot via
finishSession, which runs on success/abandon — a FAILED upload kept
its slot consumed forever. With the cap (5) reached by errored
sessions, retries could never re-launch (retries stuck at 1) and new
uploads stayed queued indefinitely: files remained local-* with their
content stranded in ContentCache.

- upload_manager.go: executeUpload releases its in-flight slot on
  every exit path (defer releaseInFlight); finishSession no longer
  decrements (the goroutine owns its slot, also for dedupe/cancel).
  QueueUpload now returns bool and only warns on a genuine read
  failure (file with content on disk), not on the empty first flush
  of a brand-new file (FUSE can process flush before the write).
- drive_item_node.go: Fsync clears hasChanges only when the upload
  was really enqueued or the file is empty, so a transient empty
  flush can never lose a file that is written afterwards.
- Tests: TestUploadManager_FailedUpload_ReleasesInFlightSlot (fails
  without the fix: slot leak stalls the pipeline), QueueUpload bool,
  and node-level create→empty-flush→write→flush enqueues the upload.
- docs/api/fs.md regenerated (QueueUpload signature).

Verified live: with the real (copied) cache state, a new file now
completes in ~3s (was: queued forever), stale sessions recover on
restart, and the false "ReadAll returned empty" warnings are gone.

Generated with Codebuff 🤖

Co-authored-by: Codebuff <noreply@codebuff.com>

Coverage Stats

20 of 22 new or added lines in 2 files covered. (90.91%)

3 existing lines in 2 files now uncovered.

3858 of 4984 relevant lines covered (77.41%)

489.31 hits per line