2.6. Object Storage

Previously, data was stored in either a dedicated service account (single project) or in the end-user's account (multi project). If data was stored in the service account, there could be issues with container or service-user deletion, and passwords and tokens were fragile. If data was stored in the end-user account, issues arose if the user picked the same name as the service, and the Object Storage service had to deal with users violating the integrity of tokens.

To solve these issues:

Previously, in globally distributed clusters, replicas were pushed out to each node in each region. With this update, the Object Storage service now only pushes replicas to just one remote node in each region (based on affinity); the remote node then spreads the replica out to primary nodes in the same region. By reducing the amount of data transfer between regions, this update stabilizes performance and lowers both the chance of replication delays as well as the cost of data transfers.