Breakfast and learn: Bigtable in action (Google Cloud Next '17)

Here is the link. 

19:38/ 55:07

Rows

• updates are atomic - but only at the row level
• store items related to a given entity in a single row
• where atomic updates aren't needed and entity is large - then split it up
• rows are sorted lexicographically by row-key
• store related entities in adjacent rows

Row-key

• determine a key strategy that facilitates common queries
• choose keys that help distribute read/writes and avoids hotspots
• avoid solely monotonically increasing keys (timestamp or sequence)
• a combined-key strategy is helpful