My professional journey in computers has involved C++, then Java, and now Python. SQL remains, at best, a foreign language. For my own sanity, therefore, I’ve brought some of my programming best practices to SQL. In particular, the WITH statement has been my friend.
I’ll use a public dataset of London bikeshares in Google BigQuery to demonstrate. Let’s say we want to find whether bikes get rented for longer durations on weekends.
1. Constants, not hardcoded numbers
A good first step is to define constants we will use throughout my query (See full query):
#standardsql WITH constants AS ( SELECT 600 AS SHORT_DUR, 1800 AS LONG_DUR, ['Sun', 'Mon', 'Tues', 'Wed', 'Thurs', 'Fri', 'Sat'] AS daysofweek ),
Here, I’m defining rides of less than 10 minutes as “short” and rides of over 30 minutes as “long.” Notice how, by defining these constants up-front, I can make it quite easy to try out different numbers. The use of named constants will also make the query a lot more readable.
2. Named result sets
Another thing you want to do to increase readability is to decompose the query into named result sets. Instead of writing queries and subqueries and counting parantheses, I tend to use WITH statements a lot. Like functions in languages like C++ or Python, named result sets allow for both reuse and logical separation.
I first define a query to pull out the fields I want, and name this result set as bikeshare (full query):
bikeshare AS ( SELECT IF(duration < SHORT_DUR, 1, 0) AS short_ride, IF(duration > LONG_DUR, 1, 0) AS long_ride, daysofweek[ORDINAL(EXTRACT(DAYOFWEEK FROM start_date))] AS dayofweek FROM `bigquery-public-data.london_bicycles.cycle_hire`, constants )
Notice that the FROM clause has to include the “constants” in order to use the defined constants.
3. SQL functions
You can decompose complex queries using the WITH keyword and create named result sets. But what about complex parsing? In the snippet above, the line pulling the day of the week and indexing into the daysofweek array is not readable, is it? And it is quite likely that this is something that you’d want in another place.
Use a SQL function so that you can reuse this expression:
CREATE TEMPORARY FUNCTION dayOfWeek(ts TIMESTAMP, days ARRAY<STRING>) AS ( days[ORDINAL(EXTRACT(DAYOFWEEK FROM ts))] );
I’m defining a function dayOfWeek that, given a timestamp and array of day names, will return the day of the week that the time in the timestamp corresponds to. Once we have this function defined, the named result set in the previous section becomes cleaner (full query):
bikeshare AS ( SELECT IF(duration < SHORT_DUR, 1, 0) AS short_ride, IF(duration > LONG_DUR, 1, 0) AS long_ride, dayOfWeek(start_date, daysofweek) AS dayofweek FROM `bigquery-public-data.london_bicycles.cycle_hire`, constants )
Once we have named constants and named result sets, the final query is simplicity itself:
SELECT dayofweek, SUM(short_ride)/COUNT(short_ride) AS frac_short_rides, SUM(long_ride)/COUNT(long_ride) AS frac_long_rides, COUNT(short_ride) AS num_all_rides FROM bikeshare GROUP BY dayofweek ORDER BY frac_long_rides DESC
Here’s the full query, and the ensuing result:
Weekdays are for quick, short commutes and weekends are for long, slow rides. Makes perfect sense!