These best practices reflect recommendations shared by a cross-functional team of seasoned Lookers. These insights come from years of experience working with Looker customers from implementation to long-term success. The practices are written to work for most users and situations, but as always use best judgment when implementing.
- Do: Define the
relationshipparameter for all joins.
This will ensure that metrics aggregate properly within Looker. By default, Looker will use a
many_to_onejoin relationship for any joins in which a relationship is not defined.
- Do: Define a primary key within each and every view, including derived tables.
All views, whether from the database directly or derived, should contain a primary key. This primary key should be a unique value, to enable Looker to uniquely identify any given record. This primary key can be a single column or a concatenation of columns ‐ it simply needs to be a unique identifier for the table or derived table.
- Do: Name dimensions, measures, and other LookML objects, using all lowercase letters and underscores for spaces.
labelparameter can be used for additional formatting of a name field, and can also be used to customize the appearance of view names, Explore names, and Model names.
- Do: Use datagroups to align generation of persistent derived tables (PDTs) and Explore caching with underlying ETL processes. Datagroups can also be used to trigger schedules to ensure that up-to-date data is sent to recipients.
- Don't: Use the
fromparameter for renaming views within an Explore.
view_labelparameter instead. For more on the difference between
view_label, check out our parameters documentation. The
fromparameter should primarily be used in the following situations:
- Polymorphic joins (joining the same table multiple times)
- Self-joins (joining a table to itself)
- Re-scoping an extended view back to its original view name
- Don't: Use the words "date" or "time" in a dimension group name.
Looker appends each timeframe to the end of the dimension group name. This means a dimension group named
created_dateresults in fields called
created_date_month, and so on. Simply use
createdas the dimension group name, because this results in fields named
created_month, and so forth.
- Don't: Use formatted timestamps within joins.
Instead, use the raw timeframe option for joining on any date or time fields. This will avoid the inclusion of casting and timezone conversion in join predicates.