The chunk lifecycle

These pages describe a chunk from the moment it is placed to the moment the cluster takes the disk that stores the chunk out of use.

Allocation and space selection covers the choice made when a chunk is created: what counts as free space, which Chunkservers are skipped, and what a client is told when none is left. Chunk versions and states covers the number that decides which stored copy holds current data. The same page covers the state that each copy has relative to that number.

The remaining pages cover repair and upkeep. Replication priorities covers which chunk of a backlog is repaired first and how much traffic the repair is allowed. Rebalancing covers how chunks that already exist are moved so that a new disk or a new server gets its share of the load. Scrubbing and CRC verification covers the checksum stored with every block and the background pass that keeps reading data that nothing else reads. Disk error handling and disk removal covers how many errors a disk is allowed, what happens to the copies on a disk taken out of use, and how to remove a disk without any chunk losing redundancy.