SQL - Using sum but optionally using default value for row - mysql

Question

Given tables
asset
col - id
date_sequence
col - date
daily_history
col - date
col - num_error_seconds
col - asset_id
historical_event
col - start_date
col - end_date
col - asset_id
I'm trying to count up all the daily num_error_seconds for all assets in a given time range in order to display "Percentage NOT in error" by day. The catch is if there is a historical_event involving an asset that has an end_date beyond the sql query range, then daily_history should be ignored and a default value of 86400 seconds (one day of error_seconds) should be used for that asset
The query I have that does not use the historical_event is:
select ds.date,
IF(count(dh.time) = 0,
100,
100 - (100*sum(dh.num_error_seconds) / (86400 * count(*)))
) percent
from date_sequence ds
join asset a
left join daily_history dh on dh.date = ds.date and dh.asset_id=a.asset_id
where ds.date >= in_start_time and ds.date <= in_end_time
group by ds.thedate;
To build on this is beyond my SQL knowledge. Because of the aggregate function, I cannot simply inject 86400 seconds for each asset that is associated with an event that has an end_date beyond the in_end_time.
Sample Data
Asset
1
2
Date Sequence
2013-09-01
2013-09-02
2013-09-03
2013-09-04
Daily History
2013-09-01, 1400, 1
2013-09-02, 1501, 1
2013-09-03, 1420, 1
2013-09-04, 0, 1
2013-09-01, 10000, 2
2013-09-02, 20000, 2
2013-09-03, 30000, 2
2013-09-04, 40000, 2
Historical Event
start_date, end_date, asset_id
2013-09-03 12:01:03, 2014-01-01 00:00:00, 1
What I would expect to see with this sample data is a % of time these assets are in error
2013-09-01 => 100 - (100*(1400 + 10000))/(86400*2)
2013-09-02 => 100 - (100*(1501 + 20000))/(86400*2)
2013-09-03 => 100 - (100*(1420 + 30000))/(86400*2)
2013-09-04 => 100 - (100*(0 + 40000))/(86400*2)
Except: There was a historical event which should take precendence. It happened on 9/3 and is open-ended (has an end date in the future, so the calculations would change to:
2013-09-01 => 100 - (100*(1400 + 10000))/(86400*2)
2013-09-02 => 100 - (100*(1501 + 20000))/(86400*2)
2013-09-03 => 100 - (100*(86400 + 30000))/(86400*2)
2013-09-04 => 100 - (100*(86400 + 40000))/(86400*2)
Asset 1's num_error_seconds gets overwritten with a full day of error seconds if there is a historical event that has a start_date before 'in_end_time' and an end_time after the in_end_time
Can this be accomplished in one query? Or do I need to stage data with an initial query?

Answer 1

I think you're after something like this:
Select
ds.date,
100 - 100 * Sum(
case
when he.asset_id is not null then 86400 -- have a historical_event
when dh.num_error_seconds is null then 0 -- no daily_history record
else dh.num_error_seconds
end
) / 86400 / count(a.id) as percent -- need to divide by number of assets
From
date_sequence ds
cross join
asset a
left outer join
daily_history dh
on a.id = dh.asset_id and
ds.date = dh.date
left outer join (
select distinct -- avoid counting multiple he records
asset_id
from
historical_event he
Where
he.end_date > in_end_time
) he
on a.id = he.asset_id
Where
ds.date >= in_start_time and
ds.date <= in_end_time -- I'd prefer < here
Group By
ds.date
Example Fiddle