fails in user acceptance - sql

Question

create table Minutes(Minute varchar2(5));
create table orders(OrderID varchar(54), Orderplaced TIMESTAMP ,
Ordercompleted TIMESTAMP);
insert into orders
VALUES
('#1',TO_TIMESTAMP('2018-01-15 00:12:20', 'YYYY-MM-DD HH24:MI:SS'),
TO_TIMESTAMP( '2018-01-15 00:12:42', 'YYYY-MM-DD HH24:MI:SS'));
insert into orders
VALUES
('#2',TO_TIMESTAMP('2018-01-15 01:15:20', 'YYYY-MM-DD HH24:MI:SS'),
TO_TIMESTAMP( '2018-01-15 02:56:20', 'YYYY-MM-DD HH24:MI:SS'));
insert into orders
VALUES
('#3',TO_TIMESTAMP('2018-01-15 01:20:20', 'YYYY-MM-DD HH24:MI:SS'),
TO_TIMESTAMP( '2018-01-15 03:00:20', 'YYYY-MM-DD HH24:MI:SS'));
insert into Minutes (Minute)
select to_char(trunc(sysdate) + interval '1' minute * (level - 1),
'HH24:MI') as minute
from dual
connect by level <= 1440;
select a.Minute, nvl(count(b.OrderID),0) as orders
from Minutes a
left join orders b
on a.Minute between to_char(cast( b.Orderplaced as date),'hh24:mi:ss') and
to_char(cast( b.Ordercompleted as date),'hh24:mi:ss')
where
a.Minute <= (select to_char(cast (sysdate as date),'hh24:mi:ss') from dual)
group by a.Minute
order by 1;
The processing time is too long and the result is undelivered as well.
It works fine with Integration testing. Please have a look once.

Answer 1

I ran your code, works OK for those test tables. However, I'd suggest a slight modification.
you don't have to CAST values from ORDERS table
Even worse is to CAST SYSDATE AS DATE - SYSDATE is a function that
returns DATE data type anyway
there's no need to select it from DUAL - you can use it "as is"
COUNT will return 0 even if there's nothing to return, so you can
omit NVL function
Here's the modified SELECT statement:
SELECT a.minute, COUNT (b.orderid) AS orders
FROM minutes a
LEFT JOIN orders b
ON a.minute BETWEEN TO_CHAR (b.orderplaced, 'hh24:mi:ss')
AND TO_CHAR (b.ordercompleted, 'hh24:mi:ss')
WHERE a.minute <= TO_CHAR (SYSDATE, 'hh24:mi:ss')
GROUP BY a.minute
ORDER BY 1;
What does it mean for you? I don't know. As I said, it works OK. Explain plan says that it performs full table scan of both MINUTES and ORDERS tables, so - if there's zillion rows in those tables, it might make a difference.
Consider creating indexes on columns you use; as you extract only time from the ORDERS table, those two would be function-based ones.
CREATE INDEX i1_min
ON minutes (minute);
CREATE INDEX i2_plac
ON orders (TO_CHAR (orderplaced, 'hh24:mi:ss'));
CREATE INDEX i3_compl
ON orders (TO_CHAR (ordercompleted, 'hh24:mi:ss'));
Then try again; hopefully, you'll see some improvement.

Answer 2

You said you're trying to get "the number of orders count per minute on a particular day", and later clarified that should be the current day. Your query is only looking at times - converted to strings - so it's looking at the same time slot across all records in your orders table. Really you want to restrict the found orders to the day you're interested in. Presumably your UAT environment just has much more data, across more days, than you created in IT.
You could just add a filter to restrict it to orders placed today:
select a.Minute, nvl(count(b.OrderID),0) as orders
from Minutes a
left join orders b
on a.Minute between to_char(cast( b.Orderplaced as date),'hh24:mi:ss') and
to_char(cast( b.Ordercompleted as date),'hh24:mi:ss')
and b.Orderplaced > trunc(sysdate) -- added this filter
where
a.Minute <= (select to_char(cast (sysdate as date),'hh24:mi:ss') from dual)
group by a.Minute
order by 1;
though you don't need any of the casting or subquery or nvl(), as #Littlefoot mentioned, so can simplify that a bit to:
select a.Minute, count(b.OrderID) as orders
from Minutes a
left join orders b
on a.Minute between to_char(b.Orderplaced,'hh24:mi:ss') and
to_char(b.Ordercompleted,'hh24:mi:ss')
and b.Orderplaced > trunc(sysdate)
where a.Minute <= to_char(sysdate,'hh24:mi:ss')
group by a.Minute
order by 1;
You're still doing a lot of conversions and comparing strings rather than dates/timestamps. It might be simpler to generate the minutes for that specific day in a CTE instead of a permanent table, and join using those values as well, without doing any further data conversions
with minutes (minute) as (
select cast(trunc(sysdate) as timestamp) + interval '1' minute * (level - 1)
from dual
connect by level <= (sysdate - trunc(sysdate)) * 1440
)
select to_char(m.minute, 'HH24:MI') as minute, count(o.orderid) as orders
from minutes m
left join orders o
on o.orderplaced >= cast(trunc(sysdate) as timestamp)
and o.orderplaced <= m.minute
and (o.ordercompleted is null or o.ordercompleted >= m.minute)
group by m.minute
order by m.minute;
I've included rows with no ordercompleted date, though it isn't clear if you want to count those.
You could also join on just the orderplaced date being today, which looks a bit odd, and do a conditional count:
with minutes (minute) as (
select cast(trunc(sysdate) as timestamp) + interval '1' minute * (level - 1)
from dual
connect by level <= (sysdate - trunc(sysdate)) * 1440
)
select to_char(m.minute, 'HH24:MI') as minute,
count(case when o.orderplaced <= m.minute
and (o.ordercompleted is null or o.ordercompleted >= m.minute)
then o.orderid end) as orders
from minutes m
left join orders o
on o.orderplaced >= cast(trunc(sysdate) as timestamp)
group by m.minute
order by m.minute;
Either way this assumes you have an index on orderplaced.
Look at the execution plans for your original query and these options and any others suggested, and test with realistic data, to see which is the best approach for your actual data and requirements.
To look for records for a different, full, day, change the sysdate references to a date/timestamp literal like timestamp '2018-01-15 00:00:00' or something relative like trunc(sysdate-1), and include an end-date on the orderplaced; and remove the end-time filter in the CTE; e.g.:
with minutes (minute) as (
select cast(trunc(sysdate - 1) as timestamp) + interval '1' minute * (level - 1)
from dual
connect by level <= 1440
)
select to_char(m.minute, 'HH24:MI') as minute, count(o.orderid) as orders
from minutes m
left join orders o
on o.orderplaced >= cast(trunc(sysdate - 1) as timestamp)
and o.orderplaced < cast(trunc(sysdate - 1) as timestamp) + interval '1' day
and o.orderplaced <= m.minute
and (o.ordercompleted is null or o.ordercompleted >= m.minute)
group by m.minute
order by m.minute;
or
with minutes (minute) as (
select timestamp '2018-01-15 00:00:00' + interval '1' minute * (level - 1)
from dual
connect by level <= 1440
)
select to_char(m.minute, 'HH24:MI') as minute, count(o.orderid) as orders
from minutes m
left join orders o
on o.orderplaced >= timestamp '2018-01-15 00:00:00'
and o.orderplaced < timestamp '2018-01-16 00:00:00'
and o.orderplaced <= m.minute
and (o.ordercompleted is null or o.ordercompleted >= m.minute)
group by m.minute
order by m.minute;
If you want to include rows where the placed and completed times are in the same minute, but still otherwise want to exclude rows from the minute they were placed, you'll need a bit more logic; maybe something like:
with minutes (minute) as (
select timestamp '2018-01-15 00:00:00' + interval '1' minute * (level - 1)
from dual
connect by level <= 1440
)
select to_char(m.minute, 'HH24:MI') as minute, count(o.orderid) as orders
from minutes m
left join orders o
on o.orderplaced >= timestamp '2018-01-15 00:00:00'
and o.orderplaced < timestamp '2018-01-16 00:00:00'
and ((trunc(o.ordercompleted, 'MI') > trunc(o.orderplaced, 'MI')
and o.orderplaced <= m.minute)
or (trunc(o.ordercompleted, 'MI') = trunc(o.orderplaced, 'MI')
and o.orderplaced < m.minute + interval '1' minute))
and (o.ordercompleted is null or o.ordercompleted >= m.minute)
group by m.minute
order by m.minute;
If you need further refinements you'll need to modify the clauses to suit, which might need a bit of experimentation.