-
Notifications
You must be signed in to change notification settings - Fork 234
Time Bucketizer
A time bucketizer in AQL specifies how to bucketize a time dimension according to the query timezone before grouping by. A time column must be bucketized, raw timestamps should not be used directly as group by conditions.
There are two types of bucketizers:
- Time series bucketizers that partition the timeline into non-overlapping and non-recurring buckets. A bucket corresponds to a single range of time in the timeline aligned to calendar unit boundaries using the specified timezone. The name of the bucket is formatted using the start of the bucket. This was also known as precision truncation based: precision of the timestamp is truncated to the desired level. Examples:
| Bucketizer | Example Bucket Name | Comment |
|---|---|---|
minute / m
|
2016-01-06 14:37 | beginning of the minute |
{2,3,4,5,6,10,12,15,20,30} minutes/m) |
2016-01-06 14:30 | beginning of the hour-aligned N minute |
quarter-hour |
2016-01-06 14:00 | beginning of a quarter hour, same as 15m
|
hour |
2016-01-06 14:00 | beginning of the hour |
{2,3,4,6,8,12} hours
|
2016-01-06 12:00 | beginning of the day-aligned N hour |
day |
2016-01-06 | beginning of the day (midnight) |
week |
2016-01-04 | beginning of the ISO week (Monday) |
month |
2016-01 | beginning of the month |
quarter |
2016-Q1 | beginning of the quarter |
year |
2016 | beginning of the year |
- Recurring unit bucketizers. The timeline is partitioned into non-overlapping ranges; the ranges are then mapped into recurring buckets. This was also known as component based: the buckets represent recurring time concepts, and are no longer single points/periods in the timeline. Here are some examples:
| Bucketizer | Example Bucket Name |
|---|---|
time of day |
14:37 |
{2,3,4,5,6,10,15,20,30} minutes of day
|
14:30 |
hour of day |
14:00 |
hour of week |
Wednesday 14:00 |
day of week |
Wednesday |
day of month |
6 |
day of year |
January 6 |
month of year |
January |
quarter of year |
Q1 |
Note this is different from the time series model in the Artemis query language, where start and end are first (optionally) snapped to calendar, and then each bucket/step/interval generated is aligned to either start or end using a specified size. In Apollo Query Language, time buckets are independent from the time range (time filter), and the buckets are always aligned to calendar unit boundaries. This significantly reduces the query space and made caching more effective.
By default the bucket names are reported as strings as shown in the above examples. You can specify a timeUnit to report the bucket name as a number in the amount of specified unit instead. For time series bucketizers, the number is the amount of specified unit since Unix Epoch time until the start of the bucket; for recurring unit bucketizers, the number is the amount of the unit since the beginning of day, week, month, year, until the start of the bucket. Time unit can be any of millisecond, second, minute, hour, day. Examples:
Example timeBucketizer
|
Default Bucket Name | Example timeUnit To Use With The Bucketizer |
Resulting Bucket Name |
|---|---|---|---|
day |
2016-01-06 | millisecond |
1452067200000 |
day |
2016-01-06 | second |
1452067200 |
time of day |
14:37 | minute |
877 |
day of week |
Wednesday | day |
2 |
day of year |
January 6 | day |
5 |