Database Reference
In-Depth Information
When lightweight transactions aren't necessary
As we discussed earlier, concurrent insertions aren't a concern for any table that uses
UUIDs that are generated by the application or by Cassandra at row creation time. Simply
using UUIDs guarantees we'll never have a row key collision.
Another scenario in which we can skip conditional inserts is when we have a globally
unique natural key. For instance, if you're building an
RSS
(short for
Rich Site Summary
)
reader, you might identify individual posts by their canonical URL. There's no concern
about overwriting one post with another post's data because each row has a key that acts as
a global unique identifier. In that case, you may choose to periodically update a local copy
of the feed by downloading the latest posts from the RSS server.
It truly doesn't matter whether any given post already exists in your local table or not, since
you'll only be overwriting data with the same data. This is a situation in which Cassandra's
upsert behavior allows us to perform very efficient write-without-read updates.
