Compartilhamento de tecnologia

Resumo do uso de dicas no Spark

2024-07-12

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

No spark sql, os usuários podem usar a dica Join para sugerir que tipo de Join Spark deve usar. Antes do Spark 3.0, apenas dicas de junção como BROADCAST eram suportadas. A partir do Spark 3.0, três dicas de junção, MERGE, SHUFFLE_HASH e SHUFFLE_REPLICATE_NL, foram adicionadas. A prioridade é BROADCAST > MERGE > SHUFFLE_HASH > SHUFFLE_REPLICATE_NL. Se BROADCAST ou SHUFFLE_HASH forem adicionados a ambos os lados do Join, o Spark escolherá qual lado construir com base no joinType e no tamanho de ambos os lados.

  1. -- Join Hints for broadcast join
  2. SELECT /*+ BROADCAST(t1) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  3. SELECT /*+ BROADCASTJOIN (t1) */ * FROM t1 left JOIN t2 ON t1.key = t2.key;
  4. SELECT /*+ MAPJOIN(t2) */ * FROM t1 right JOIN t2 ON t1.key = t2.key;
  5. -- Join Hints for shuffle sort merge join
  6. SELECT /*+ SHUFFLE_MERGE(t1) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  7. SELECT /*+ MERGEJOIN(t2) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  8. SELECT /*+ MERGE(t1) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  9. -- Join Hints for shuffle hash join
  10. SELECT /*+ SHUFFLE_HASH(t1) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  11. -- Join Hints for shuffle-and-replicate nested loop join
  12. SELECT /*+ SHUFFLE_REPLICATE_NL(t1) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;
  13. -- When different join strategy hints are specified on both sides of a join, Spark
  14. -- prioritizes the BROADCAST hint over the MERGE hint over the SHUFFLE_HASH hint
  15. -- over the SHUFFLE_REPLICATE_NL hint.
  16. -- Spark will issue Warning in the following example
  17. -- org.apache.spark.sql.catalyst.analysis.HintErrorLogger: Hint (strategy=merge)
  18. -- is overridden by another hint and will not take effect.
  19. SELECT /*+ BROADCAST(t1), MERGE(t1, t2) */ * FROM t1 INNER JOIN t2 ON t1.key = t2.key;

Relação de uso na dica spark: https://blog.51cto.com/u_15435003/5296344