Question
Your media pipeline uploads large video files to object storage via multipart upload. Users report that some uploads 'finish' in the UI but the files never appear in the library, and finance flags that object-storage *storage* cost has crept up 30% this month with no growth in visible objects. Investigation shows thousands of multipart uploads that were initiated but never completed or aborted — their parts are sitting in the bucket, billable but invisible. A deploy three weeks ago changed the upload worker to crash-restart on any part-upload error instead of retrying. How do you triage, recover correctness, and stop the silent cost and data leak?
Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.