Creating Date Cutoffs on continuous processes - postgresql

Question

I'm an intern databasing some CNC machine metrics and i'm a little stuck with a particular query, only started sql last week so please forgive me if this is a dumb question.
If a machine is running (state=on) past date,23:59, and I want to collect machine hours for that day, there is no logged off time, as the state=off column has not been recorded yet, thus I cannot collect that machine data. To work around this, I want to record a state off time of 23:59:59, and then create a new row with the same entity ID with a state_on time of day+1,00:00:01.
Here is what I have written so far, where am I going wrong? What combination of trigger, insert, procedure, case, etc should I use? Any suggestions are welcome, I've tried to look at some reference material, and want the first bit to look something like this.
CASE
WHEN min(stoff.last_changed) IS NULL
AND now() = '____-__-__ 23:59:59.062538+13'
THEN min(stoff.last_changed) IS now()
ELSE min(stoff.last_changed)
END
I know this is only the first component, but it fits into a larger select used within a view,let me know if I need to post anything else

Answer 1

This is a fairly complex query because there are a few possibilities you need to consider (given the physical setup of the CNC machine these may not both apply:
The machine might be running at midnight so you need to 'insert' a midnight start time (and treat midnight as the stop time for the previous day).
The machine might be running all day (so there are no records in your table for the day at all)
There are a number of ways you can implement this; I have chosen one that I think will be easiest for you to understand (and have avoided window functions). The response does use CTE's but I think these are easy enough to understand (it's really just a chain of queries; each one using the result of the previous one).
Lets setup some example data:
create table states (
entity_id int,
last_changed timestamp,
state bool
);
insert into states(entity_id, last_changed, state) values
(1, '2019-11-26 01:00', false),
(1, '2019-11-26 20:00', true),
(1, '2019-11-27 01:00', false),
(1, '2019-11-27 02:00', true),
(1, '2019-11-27 22:00', false);
Now the query (it's not as bad as it looks!):
-- Lets start with a range of dates that you want the report for (needed because its possible that there are
-- no entries at all in the states table for a particular date)
with date_range as (
select i::date from generate_series('2019-11-26', '2019-11-29', '1 day'::interval) i
),
-- Get all combinations of machine (entity_id) and date
allMachineDays as (
select distinct states.entity_id, date_range.i as started
from
states cross join date_range),
-- Work out what the state at the start of each day was (if no earlier state available then assume false)
stateAtStartOfDay as (
select
entity_id, started,
COALESCE(
(
select state
from states
where states.entity_id = allMachineDays.entity_id
and states.last_changed<=allMachineDays.started
order by states.last_changed desc limit 1
)
,false) as state
from allMachineDays
)
,
-- Now we can add in the state at the start of each day to the other state changes
statesIncludingStartOfDay as (
select * from stateAtStartOfDay
union
select * from states
),
-- Next we add the time that the state changed
statesWithEnd as (
select
entity_id,
state,
started,
(
select started from statesIncludingStartOfDay substate
where
substate.entity_id = states.entity_id and
substate.started > states.started
order by started asc
limit 1
) as ended
from
statesIncludingStartOfDay states
)
-- finally lets work out the duration
select
entity_id,
state,
started,
ended,
ended - started as duration
from
statesWithEnd states
where
ended is not null -- cut off the last midnight as its no longer needed
order by entity_id,started
Hopefully this makes sense and there are no errors in my logic! Note that I have made some assumptions about your data (i.e. last_changed is the time that the state began). If you just want a runtime for each day it's pretty easy to just add a group by to the last query and sum up duration.
It might help you to understand this if you run it one step at a time; gor example start with the following and then add in extra with clauses one at a time:
with date_range as (
select i::date from generate_series('2019-11-26', '2019-11-29', '1 day'::interval) i
)
select * from date_range