[krixkrix]

Problem: uploads hanging at 100 % when the finish request is slow

Summary

/upload/finish makes the server hash the whole stream, so it takes longer the bigger the file is. On slow hardware (e.g. a Raspberry Pi) a large file takes more than the default 10 s request timeout. axios aborts the request, the empty .catch swallows the error, and the upload widget stays at 100 % forever.

 performJobUpload / performImageUpload
   POST /upload/                 (init)
   POST /upload/<uuid>/...       (chunks, no timeout)
-  POST /upload/finish           (default 10 s timeout)
+  POST /upload/finish           (no timeout, like the chunks)
   .catch(err)
-    err;                        // swallowed, widget hangs
+    uploadState.failUpload(err) // calls _onError, the frontend shows the failure

Second commit: an upload widget reuses one AMCUploadState, and the user can start a new upload while one is still running. Now that failures are reported, a late failure of the replaced upload would stop the newer one. failUpload() therefore only acts while its upload is the active one, the same check the chunk handlers already do.

AMCUpload.failUpload(reason)
if this upload is still the active one
itemState.failUpload(reason)

Only Client/core changes, so both the Vue2 and the Svelte client get the fix.

Evidence

New Client/tests/uploadFinish.test.mjs (node:test, no new dependencies), covering build and image uploads:

• finish request is sent with timeout: 0
• a failed finish calls _onError and ends the upload
• a failed finish of a replaced upload leaves the newer upload running
• Before (test file run against develop): 0/6 pass (timeout of 10000ms exceeded, newer upload aborted)
  After: 6/6 pass; the full cd Client/tests && npm test passes 14/14

Merge Danger

• Client-only change with no API or data change. Revert the two commits to undo it.

• If the server never answers, a finish request now waits indefinitely instead of failing after 10 s. The chunk requests already behave that way. Upload init, chunk and finish failures that used to be silently ignored now reach the widget's _onError.