Partitioning on non-unique values - sql

Question

I have a table that lists events, operations in the events and the time of each operation. Event ID is not unique, as it is the same event, just happening on different times. Operations might differ for the same type of the event. The same event never runs twice in a row).
I want to populate three new columns as per given example. This will allow me to run analysis on the separate events as I'll be able to generate a unique "Event" ID.
Edit:
I already tried PARTITION function based on event and it haven't worked as SQL server assumes two events (A and B) and therefore gives the same start date to all "A" events, even if in reality I need to show them as separate events with different start dates.
Thank you!

Answer 1

This is just window functions:
select t.*,
min(operationtime) over (partition by event) as event_start_time,
max(operationtime) over (partition by event) as event_end_time,
concat(event, '-', min(operationtime) over (partition by event)) as event_id
from t;
Actually, for the event id, you probably want something like:
concat(event, '-', convert(varchar(255), min(operationtime) over (partition by event), 101)) as event_id
or whatever format for the date you really want. I recommend YYYY-MM-DD as a date format.

Answer 2

I understand this as a gaps-and-island problem, where you want to build groups of consecutive daily events.
One option uses the difference between row numbers to identify the groups:
select
t.*,
min(operation_time) over(partition by event, rn1 - rn2) event_start_time,
max(operation_time) over(partition by event, rn1 - rn2) event_end_time,
concat(event, '-', min(operation_time) over(partition by event, rn1 - rn2)) event_id
from (
select
t.*,
row_number() over(order by operation_time) rn1,
row_number() over(partition by event order by operation_time) rn2
from mytable t
) t
order by operation_time
If there is always one and only one event per day, as showned in your sample data, then one row_number() is sufficient, along with date arithmetics:
select
t.*,
min(operation_time) over(partition by event, grp) event_start_time,
max(operation_time) over(partition by event, grp) event_end_time,
concat(event, '-', min(operation_time) over(partition by event, grp)) event_id
from (
select
t.*,
dateadd(
day,
- row_number() over(partition by event order by operation_time),
operation_time
) grp
from mytable t
) t

Answer 3

This approach creates the event group explicitly, then it uses a windowing query very similar to the other answers. I created a simple sample table to show results.
Data
drop table if exists #tTEST;
go
select * INTO #tTEST from (values
('A', 'X', '2020-01-08'),
('A', 'Z', '2020-02-08'),
('B', 'X', '2020-03-08'),
('B', 'Z', '2020-04-08'),
('A', 'X', '2020-05-08'),
('A', 'Z', '2020-06-08')) V([Event], [Operation], operation_time);
Query
;with
grp_cte as (
select t.*, case when lag([Event], 1, 0) over (order by operation_time) != [Event] then 1 else 0 end grp_ind
from #tTEST t),
event_grp_cte as (
select gc.*, sum(grp_ind) over (order by operation_time) EventGroup
from grp_cte gc)
select
t.*,
min(operation_time) over(partition by EventGroup) event_start_time,
max(operation_time) over(partition by EventGroup) event_end_time,
concat(event, '-', min(operation_time) over(partition by EventGroup)) event_id
from event_grp_cte t
order by operation_time;
Results
Event Operation operation_time grp_ind EventGroup rn1 rn2 event_start_time event_end_time event_id
A X 2020-01-08 1 1 1 1 2020-01-08 2020-02-08 A-2020-01-08
A Z 2020-02-08 0 1 2 2 2020-01-08 2020-02-08 A-2020-01-08
B X 2020-03-08 1 2 3 1 2020-03-08 2020-04-08 B-2020-03-08
B Z 2020-04-08 0 2 4 2 2020-03-08 2020-04-08 B-2020-03-08
A X 2020-05-08 1 3 5 3 2020-05-08 2020-06-08 A-2020-05-08
A Z 2020-06-08 0 3 6 4 2020-05-08 2020-06-08 A-2020-05-08