A time quantum is a feature for
STRINGSETQ type columns that allows you to associate a time (or multiple times) with each value in the column. Setting a time quantum creates views on the column that allow range queries down to the time granularity specified. You can think of a view as a rollup of your data based on the granularity of time you specify. If no time quantums are set, your data has one “standard” view by default.
You should use time quantums when you want to associate a time with each value in
STRINGSET type columns, in addition to querying by that time.
You should avoid time quantums if you don’t have a time you want to associate with a value, if you aren’t interested in deleting values over time to save space, if you are trying to count the number of distinct time quantums associated to a particular value, and if you are looking to pull out time values as opposed to filtering by them.
When creating a column, you specify the granularity of time you want views created for. FeatureBase supports hour (
H), day (
D), month (
M), or year (
Y) or any combination of the four (in descending order with no gaps in time. i.e.
YMD but not
YD). Setting these allows for lower latency queries depending on the period of time you are querying over, but at the cost of increased storage. For example, If you plan to have queries with a range of multiple months,
MD is the best option, but if you will be querying over only a couple of days,
D will be preferred. Note you can set just
D and still query over multiple months, but it will not be as fast as using
Once created, a timestamp must be passed with each record during ingest that will be associated with all time quantum columns. Note this means you can only pass one time for all the time quantums in a record. For more information on configuring ingest, see the appropriate section in “Data Ingestion” navigation.
Querying using time quantums is only supported in PQL Rows Queries. You can pass a timestamp in the
from arguments. In the example below, the
customer table will pull back the customer IDs and what stores they visited between
[customer]Extract(All(), Rows(stores_visited,from='2018-08-31', to='2022-02-18'))
You can associate multiple times with each value, so a value only has to exist in one view to be returned. This will not return the value twice and will only be counted once. You cannot return the underlying timestamps associated with each value.
Whenever a record with time quantums is ingested, a view is created for each level of granularity specified. This is essentially a copy of the column over a specific time range. If
YMDH is specified and the time
2018-08-31T22:30:00Z is ingested, a time view will exist for
2018-08-31T22. This means data which has times for every hour for two days (say May 2nd and 3rd) in a column with
YMDH time quantums configured will have 48+2+1+1+1 views (53) in total. 48 hours, 2 days, 1 month, 1 year, and the standard view.