Shards, RecordIDs, Main Storage, Delta Storage

From MemCP
Jump to navigation Jump to search
  • Datasets of a table are organized in shards of roughly 60,000 items
  • Inside a shard with n items, the datasets are numbered from 0..n-1
  • The first few datasets of each shard are organized in a column store
  • The rest of the datasets are organized in a row store
  • The delta store consists of the insert list as well as the deletion mask