If spark isn't a storage system, how do tables work?

I am reading "Learning spark, 2nd edition" and I find two confusing parts:
"Unlike Hadoop, which included both storage and compute, Spark decouples the two. That means you can (...) read data from sources, (...) process it in memory"
Also on reddit and in courses I found info that spark doesn't store data, it processes it only.

Now I am reading the spark sql part and it says that I can create tables that will be managed by spark and will hold data (in contrast to views, which do not hold data). I have trouble understanding it.

"To enable you to query structured data (...) Spark manages (...) creating and managing views and tables, both in memory and on disk"

I tried googling and reading, but it only mentions that the data can be held in hdfs etc, but the examples in the book don't mention that. So I would like to clarify - let's say I install spark only, without separate hdfs, hive installation - can I create the tables and hold data in them? What will be the difference between those tables and dataframes? How are the tables stored? Metadata is stored in the catalog, but what about the actual data? If I take a csv, create a table with data from the csv, then delete the csv, can I still read the data? If so, then where is the data? And if not, then maybe tables do not hold data and are therefore no different than views?

Spark is a tool specifically designed for processing data, not storing it. In the context of Hadoop, it's more akin to MapReduce, the processing engine, rather than HDFS, the storage layer. In fact, Spark can operate on data stored in HDFS, effectively serving as a replacement for MapReduce in the Hadoop ecosystem.

Regarding tables, Spark creates a (temporary) view on top of the storage system, facilitating the processing of that data. When you utilize dataframes or SQL in your code, Spark translates this into its own internal format for processing. This is somewhat comparable to how Pandas reads data from a file system and lets you query the dataframe using SQL. Similarly, Spark reads data from a storage layer (like HDFS or S3), then generates an internal table/view based on your instructions.

There are two kinds of tables in Spark:

• Global Temporary View: This type of table is accessible in all
  sessions, but it disappears when the application is closed.

• Local Temporary View: This table is only available in the current
  session and disappears when that session ends.

You can also construct tables using Spark SQL, which are akin to tables in a standard database. This offers a familiar structure for querying and manipulating the data.

In simple words: Spark is just a data processing engine. It is used to read files from the local file system, HDFS, S3 or a database table, and write out an output to another or to the same destination.

The sentence you have found "creating and managing views and tables, both in memory and on disk" most likely refers to the fact that you can cache data during the process in memory and/or on disk. This is usually done to reuse data that you may need across multiple steps in your job, without creating any temporary output that you need to clean up manually later. Besides, using cached data it's much faster than reading a file or a table twice.