BigTable checkAndMutate atomicity

Question

What is the behaviour of the checkAndMutate function? Let's say I need to get row content before applying checkAndMutate. Is there a chance to retrieve stale data from BigTable? If there is no chance to retrieve stale data, is my thinking good:

1. Fetch row content by key,
2. Modify row content in application,
3. Apply checkAndMutate function on row.

Answer 1

CheckAndMutate is atomic. Here is the API definition. There is a chance that there was a change between step #1 and #3. Your checkAndMutate should ensure that the timestamp for the cells you care about are the same values as #3 to ensure that you're updating the same data that you read.