有关将关系模式转换为Cassandra的建议

Question

I am hoping to get some suggestions on how to best approach converting a typical relational schema to Cassandra. The Relational Schema is:

CREATE TABLE IF NOT EXISTS sales (
   sale_id     bigint(20) UNSIGNED NOT NULL
                          AUTO_INCREMENT,
   create_time timestamp  NOT NULL
                          DEFAULT ‘0000-00-00 00:00:00’,
   account     bigint(20) UNSIGNED NOT NULL DEFAULT ‘0’,
   store       char(25)   NOT NULL DEFAULT ‘’,
   product     char(25)   NOT NULL DEFAULT ‘’,
   coupon      char(18)   NOT NULL DEFAULT ‘’,
   amount      decimal(8,2) NOT NULL,
   PRIMARY KEY (sale_id),
   KEY         create_time (create_time) )

The Cassandra schema I've come up with is:

CREATE TABLE sales (
            sale_id     uuid,
            create_time timestamp,
            account     text,
            store       int,
            coupon      text,
            product     text,
            amount      int,
            PRIMARY KEY ((create_time, store), coupon))

(with indexes created on non-key columns I need to query)

Typical query is to get all sales by product by product/coupon/account/store over some time period.

Does this make sense?

Any suggestions on how this may be improved for reasonable read/write performance?

Thanks in advance for any suggestions.

Answer 1

No, You want to model your Cassandra schema to answer each question to get the good performance. Let's say you want to find all (recent) sales by product with you want create your primary key as ( productID, created_time )

If your application normally wants to search for products that are sold recently, then you want to order the cluster factor( created_time in your example) as desc.

Likewise you might duplicating your sales data in multiple column families. Don't be scared to duplicate data while modeling in distributed environment. You want to de-normalize and look forward to get your results from partition itself.

Hope this helps.