Design a Cloud Storage Service (Dropbox/Google Drive)

To design a robust cloud storage sync service, I would start by defining our consistency model as eventually consistent to prioritize availability during network partitions, which aligns with Microsoft's focus on reliability at scale. First, we implement a versioned object store where every file modification generates a unique immutable blob ID. Clients maintain a local manifest using vector clocks to track causality. When syncing, the client calculates a Merkle tree hash of its local directory structure and compares it against the server's root hash. If they differ, the client downloads only the specific leaf nodes representing changed blocks, achieving differential synchronization. For conflicts, if two users edit the same file simultaneously offline, their vector clocks will show incomparable timestamps. Instead of simple overwrites, we employ a mergeable CRDT approach for text files to automatically combine changes, while binary files trigger a 'conflict copy' mechanism, creating a separate file version for manual resolution. This ensures no data is silently lost. Finally, we introduce a background reconciliation service that periodically scans for unresolved conflicts and notifies users via the UI, balancing system automation with user control.