推荐订阅源

1. Don't fetch any data that you don't need,or don't fetch any columns that you don't need. Because retrieving more data or more columns, which can increase network,I/O,memory and CPU overhead for the server. For example, if you need several columns you can use
   AT EPOCH LATEST
   SELECT fi.name, fi.InvestmentKey,id.VendorId,id.CUSIP,id.ISIN,id.DomicileCountryId,id.CurrencyId
   FROM dbo.FixedIncome fi
   INNER JOIN dbo.InvestmentIdDimension id ON id.InvestmentKey = fi.InvestmentKey
   WHERE id.InvestmentId = 'B000023K1X'
   But do not use:
   AT EPOCH LATEST
   SELECT fi.*, id.*
   FROM dbo.FixedIncome fi
   INNER JOIN dbo.InvestmentIdDimension id ON id.InvestmentKey = fi.InvestmentKey
   WHERE id.InvestmentId = 'B000023K1X'
2. To avoid blocking Vertica write process, we alway add the "AT EPOCH LATEST" for query,which is snapshot read. for example, You can use
   AT EPOCH LATEST SELECT ... FROM ...,
   But do not use:
   SELECT ... FROM ...
3. Chop up a complex query to many simpler queries.
4. Join decomposition, if posible, Sometimes, Using "In" clause or sub query clause instead of a complex "JOIN" clause. like this, we can use
   AT EPOCH LATEST
   SELECT s1.CompanyId, id.InvestmentId, s1.InvestmentKey,id.VendorId,id.CUSIP,id.ISIN,id.DomicileCountryId,id.CurrencyId
   FROM ( SELECT CompanyId,InvestmentKey FROM dbo.FixedIncome WHERE CompanyId = '0C00000BDL') s1
   INNER JOIN dbo.InvestmentIdDimension id ON id.InvestmentKey = s1.InvestmentKey
   WHERE id.VendorId = 101 OR id.VendorId = 102;
   But do not use:
   AT EPOCH LATEST
   SELECT s1.CompanyId, id.InvestmentId, s1.InvestmentKey,id.VendorId,id.CUSIP,id.ISIN,id.DomicileCountryId,id.CurrencyId
   FROM dbo.FixedIncome fi
   INNER JOIN dbo.InvestmentIdDimension id ON id.InvestmentKey = s1.InvestmentKey
   WHERE fi.CompanyId = '0C00000BDL' AND( id.VendorId = 101 OR id.VendorId = 102 );
5. Try to use the temporary table to cache data, which can avoid scan an physical table for times.
6. Try to push the outer predicate into the inner subquery clause, so that it is evaluated before the analytic computation
7. For Top-K query, if posible, we'd better omit the order by clause, Or we'd better adding a filter condition for it. 
8. For sort operation, We can create Pre-sorted projections, so the vertica can choose the faster Group By Pipeline over Group By Hash
9. Please refer to the "Optimizing Query Performance" chapter in reference manual of vertica, which doc's name is "Communiti Vertica Community Edition 6.0"
   [https://my.vertica.com/docs/CE/6.0.1/HTML/index.htm#12525.htm ]