Based on Snowflake recommendations, how should the clustering key columns be ordered while defining the multi-column clustering key?

A table contains five columns and it has millions of records.

The cardinality distribution of the columns is shown below:

Column C4 and C5 are mostly used by SELECT queries in the GROUP BY and ORDER BY clauses.

Whereas columns C1, C2 and C3 are heavily used in filter and join conditions of SELECT queries.

The Architect must design a clustering key for this table to improve the query performance.

Based on Snowflake recommendations, how should the clustering key columns be ordered while defining the multi-column clustering key?
A . C5, C4, C2
B . C3, C4, C5
C . C1, C3, C2
D . C2, C1, C3

View Answer

Answer: D

Explanation:

According to the Snowflake documentation, the following are some considerations for choosing clustering for a table1:

Clustering is optimal when either:

You require the fastest possible response times, regardless of cost.

Your improved query performance offsets the credits required to cluster and maintain the table.

Clustering is most effective when the clustering key is used in the following types of query predicates:

Filter predicates (e.g. WHERE clauses)

Join predicates (e.g. ON clauses)

Grouping predicates (e.g. GROUP BY clauses)

Sorting predicates (e.g. ORDER BY clauses)

Clustering is less effective when the clustering key is not used in any of the above query predicates, or when the clustering key is used in a predicate that requires a function or expression to be applied to the key (e.g. DATE_TRUNC, TO_CHAR, etc.).

For most tables, Snowflake recommends a maximum of 3 or 4 columns (or expressions) per key.

Adding more than 3-4 columns tends to increase costs more than benefits.

Based on these considerations, the best option for the clustering key columns is C. C1, C3, C2, because:

These columns are heavily used in filter and join conditions of SELECT queries, which are the most effective types of predicates for clustering.

These columns have high cardinality, which means they have many distinct values and can help reduce the clustering skew and improve the compression ratio.

These columns are likely to be correlated with each other, which means they can help co-locate similar rows in the same micro-partitions and improve the scan efficiency.

These columns do not require any functions or expressions to be applied to them, which means they can be directly used in the predicates without affecting the clustering.

Reference: 1: Considerations for Choosing Clustering for a Table | Snowflake Documentation