Restructure query without sub-query but join?

Question

I am trying to rewrite the following query without using sub-query since I cannot afford hash join.

SELECT 
  COUNT(wins.bid) as wins, 
  to_char((TIMESTAMP 'epoch' + bidRequestTime * INTERVAL '1 Second'), 'DD') as date,
  COUNT(impression.bid) as impression, 
  lid, 
  COUNT(b.bid) as requests, 
  ROUND(SUM(spendCalc),2) as spend 
FROM (
  SELECT lid, bids.bid,bidRequestTime,price::float*COUNT(impression.bid)/1000::float/.80::float as spendCalc
  FROM bids
  LEFT OUTER JOIN wins on wins.bid = bids.bid
  LEFT OUTER JOIN impressions impression ON impression.bid = bids.bid
  WHERE bidRequestTime BETWEEN 1470614400 AND 1471219200 
  GROUP BY price, bids.bid, bidRequestTime, lid
) b
LEFT OUTER JOIN wins on wins.bid = b.bid
LEFT OUTER JOIN impressions impression ON impression.bid = b.bid
WHERE bidRequestTime BETWEEN 1470614400 AND 1471219200  
GROUP BY lid, date 

Any suggestions? Thanks in Advance.

Answer 1

Hard to tell exactly what your problem is... but here are some tricks removing the outer joins and using a CTE to reduce the search space.... may or may not work, would need to see more details about your model to be sure.

WITH mybid as
(
  SELECT bid, bidRequestTime
  FROM bids 
  WHERE bidRequestTime BETWEEN 1470614400 AND 1471219200  
)
SELECT 
  COUNT(wins_bid) as wins, 
  to_char((TIMESTAMP 'epoch' + bidRequestTime * INTERVAL '1 Second'), 'DD') as date,
  COUNT(imp_bid) as impression, 
  lid, 
  COUNT(b.bid) as requests, 
  ROUND(SUM(spendCalc),2) as spend 
FROM (
  SELECT wins.bid as wins_bid, lid, bids.bid, bidRequestTime, impression.bid as imp_bid,  price::float*COUNT(impression.bid)/1000::float/.80::float as spendCalc
  FROM mybid as bids
  LEFT OUTER JOIN wins on wins.bid = bids.bid
  LEFT OUTER JOIN impressions impression ON impression.bid = bids.bid
  GROUP BY price, bids.bid, bidRequestTime, lid
) b
GROUP BY lid, date