After BigTable introduction by Google in 2004 a large number of systems has been designed following its ideas named as NoSQL. Recently Google has introduced Spanner. Do you see this as a move back?!
I don't think that this is a move back as Spanner still has a number of NoSQL concepts implemented. Remember that NoSQL is "not only SQL" - hence a solution closer to DBMS's following ACID principles is also a valid. Surely the "hype" around NoSQL will decline, but the open-source community is still strong, as various projects prove. Hence, I think that choosing a database will be a question of "what suits best for a certain purpose" - therefore I appreciate to have more types of database concepts to choose from.
The NoSQL was always here even before SQL, did not have that nice name maybe. I see it as a tool that permits you to master your database. Kind of nodejs in web which permits you to write webserver from scratch, NoSQL give you a tool to create database from scratch. As Johannes said the "hype" will vanish in some time since NoSQL is not going to do a miracle but rather a good job in good hands. Google introduces Spanner since in some scenarios the Relational DBMS is an obligation, where the strong consistence is precious.
I am seeing a dramatic increase in NoSQL usage. This is driven by Web apps and when the web presentation tier scales out (to multiple servers in the farm), the fan in effect into RDMBS (SQL databases) puts greater load on the backend systems.
You can scale the DB tier with read-replicals and multi-master setups, you will hit a point where you need to scale out the DB tier to accomodate scale out in Web Presentation, and Application Tiers. So the only way to do that is to start sharding your data - and this is the world of NoSQL.
I see a lot of folks with hybrid solutions, they still have RDBMS engines for highly normalised data, and then they have a couple of tables sharded in NoSQL configurations for higher performance access read/write scenarios.
I see NoSQL thrive as response to a need that SQL couldn't easily provide - managing very large data sets of unstructured information. Big Data initiatives will seal NoSQL success, but as the hype dies down I believe NoSQL will remain a viable alternative to traditional relational databases.