Hive - Is there a way to further optimize a HiveQL query? - sql

Question

I have written a query to find 10 most busy airports in the USA from March to April. It produces the desired output however I want to try to further optimize it.
Are there any HiveQL specific optimizations that can be applied to the query? Is GROUPING SETS applicable here? I'm new to Hive and for now this is the shortest query that I've come up with.
SELECT airports.airport, COUNT(Flights.FlightsNum) AS Total_Flights
FROM (
SELECT Origin AS Airport, FlightsNum
FROM flights_stats
WHERE (Cancelled = 0 AND Month IN (3,4))
UNION ALL
SELECT Dest AS Airport, FlightsNum
FROM flights_stats
WHERE (Cancelled = 0 AND Month IN (3,4))
) Flights
INNER JOIN airports ON (Flights.Airport = airports.iata AND airports.country = 'USA')
GROUP BY airports.airport
ORDER BY Total_Flights DESC
LIMIT 10;
The table columns are as following:
Airports
|iata|airport|city|state|country|
Flights_stats
|originAirport|destAirport|FlightsNum|Cancelled|Month|

Answer 1

Filter by airport(inner join) and do aggregation before UNION ALL to reduce dataset passed to the final aggregation reducer. UNION ALL subqueries with joins should run in parallel and faster than join with bigger dataset after UNION ALL.
SELECT f.airport, SUM(cnt) AS Total_Flights
FROM (
SELECT a.airport, COUNT(*) as cnt
FROM flights_stats f
INNER JOIN airports a ON f.Origin=a.iata AND a.country='USA'
WHERE Cancelled = 0 AND Month IN (3,4)
GROUP BY a.airport
UNION ALL
SELECT a.airport, COUNT(*) as cnt
FROM flights_stats f
INNER JOIN airports a ON f.Dest=a.iata AND a.country='USA'
WHERE Cancelled = 0 AND Month IN (3,4)
GROUP BY a.airport
) f
GROUP BY f.airport
ORDER BY Total_Flights DESC
LIMIT 10
;
Tune mapjoins and enable parallel execution:
set hive.exec.parallel=true;
set hive.auto.convert.join=true; --this enables map-join
set hive.mapjoin.smalltable.filesize=25000000; --size of table to fit in memory
Use Tez and vectorizing, tune mappers and reducers parallelism: https://stackoverflow.com/a/48487306/2700344

Answer 2

It might help if you do the aggregation before the union all:
SELECT a.airport, SUM(cnt) AS Total_Flights
FROM ((SELECT Origin AS Airport, COUNT(*) as cnt
FROM flights_stats
WHERE (Cancelled = 0 AND Month IN (3,4))
GROUP BY Origin
) UNION ALL
(SELECT Dest AS Airport, COUNT(*) as cnt
FROM flights_stats
WHERE Cancelled = 0 AND Month IN (3,4)
GROUP BY Dest
)
) f INNER JOIN
airports a
ON f.Airport = a.iata AND a.country = 'USA'
GROUP BY a.airport
ORDER BY Total_Flights DESC
LIMIT 10;

Answer 3

I don't think GROUPING SETS are applicable here because you are only grouping by one field.
From Apache Wiki:
"The GROUPING SETS clause in GROUP BY allows us to specify more than one GROUP BY option in the same record set."

Answer 4

You can test this but you are in the case where an Union maybe better, so You really need to test it and come back :
SELECT airports.airport,
SUM(
CASE
WHEN T1.FlightsNum IS NOT NULL THEN 1
WHEN T2.FlightsNum IS NOT NULL THEN 1
ELSE 0
END
) AS Total_Flights
FROM airports
LEFT JOIN (SELECT Origin AS Airport, FlightsNum
FROM flights_stats
WHERE (Cancelled = 0 AND Month IN (3,4))) t1
on t1.Airport = airports.iata
LEFT JOIN (SELECT Dest AS Airport, FlightsNum
FROM flights_stats
WHERE (Cancelled = 0 AND Month IN (3,4))) t2
on t1.Airport = airports.iata
GROUP BY airports.airport
ORDER BY Total_Flights DESC