Stop Thinking, Just Do!

Sungsoo Kim's Blog

Column-Family Stores

tagsTags

19 May 2014


Article Source

  • Title: NoSQL Distilled; A Brief Guide to the Emerging World of Polyglot Persistence, 2013.
  • Authors: Pramod J. Sadalage, Martin Fowler

Column-Family Stores

One of the early and influential NoSQL databases was Google’s BigTable [Chang etc.]. Its name conjured up a tabular structure which it realized with sparse columns and no schema. As you’ll soon see, it doesn’t help to think of this structure as a table; rather, it is a two-level map. But, however you think about the structure, it has been a model that influenced later databases such as HBase and Cassandra.

These databases with a bigtable-style data model are often referred to as column stores, but that name has been around for a while to describe a different animal. Pre-NoSQL column stores, such as C-Store [C-Store], were happy with SQL and the relational model. The thing that made them different was the way in which they physically stored data. Most databases have a row as a unit of storage which, in particular, helps write performance. However, there are many scenarios where writes are rare, but you often need to read a few columns of many rows at once. In this situation, it’s better to store groups of columns for all rows as the basic storage unit—which is why these databases are called column stores.

Bigtable and its offspring follow this notion of storing groups of columns (column families) together, but part company with C-Store and friends by abandoning the relational model and SQL. In this book, we refer to this class of databases as column-family databases.

Perhaps the best way to think of the column-family model is as a two-level aggregate structure. As with key-value stores, the first key is often described as a row identifier, picking up the aggregate of interest. The difference with column-family structures is that this row aggregate is itself formed of a map of more detailed values. These second-level values are referred to as columns. As well as accessing the row as a whole, operations also allow picking out a particular column, so to get a particular customer’s name from Figure 2.5 you could do something like get(‘1234’, ‘name’).

Figure 2.5. Representing customer information in a column-family structure

Column-family databases organize their columns into column families. Each column has to be part of a single column family, and the column acts as unit for access, with the assumption that data for a particular column family will be usually accessed together.

This also gives you a couple of ways to think about how the data is structured.

  • Row-oriented: Each row is an aggregate (for example, customer with the ID of 1234) with column families representing useful chunks of data (profile, order history) within that aggregate.
  • Column-oriented: Each column family defines a record type (e.g., customer profiles) with rows for each of the records. You then think of a row as the join of records in all column families.

This latter aspect reflects the columnar nature of column-family databases. Since the database knows about these common groupings of data, it can use this information for its storage and access behavior. Even though a document database declares some structure to the database, each document is still seen as a single unit. Column families give a two-dimensional quality to column-family databases.

This terminology is as established by Google Bigtable and HBase, but Cassandra looks at things slightly differently. A row in Cassandra only occurs in one column family, but that column family may contain supercolumns—columns that contain nested columns. The supercolumns in Cassandra are the best equivalent to the classic Bigtable column families.

It can still be confusing to think of column-families as tables. You can add any column to any row, and rows can have very different column keys. While new columns are added to rows during regular database access, defining new column families is much rarer and may involve stopping the database for it to happen.

The example of Figure 2.5 illustrates another aspect of column-family databases that may be unfamiliar for people used to relational tables: the orders column family. Since columns can be added freely, you can model a list of items by making each item a separate column. This is very odd if you think of a column family as a table, but quite natural if you think of a column-family row as an aggregate. Cassandra uses the terms “wide” and “skinny.” Skinny rows have few columns with the same columns used across the many different rows. In this case, the column family defines a record type, each row is a record, and each column is a field. A wide row has many columns (perhaps thousands), with rows having very different columns. A wide column family models a list, with each column being one element in that list.

A consequence of wide column families is that a column family may define a sort order for its columns. This way we can access orders by their order key and access ranges of orders by their keys. While this might not be useful if we keyed orders by their IDs, it would be if we made the key out of a concatenation of date and ID (e.g., 20111027-1001).

Although it’s useful to distinguish column families by their wide or skinny nature, there’s no technical reason why a column family cannot contain both field-like columns and list-like columns—although doing this would confuse the sort ordering.


comments powered by Disqus