๋ชฉ๋ก๋ฐ์ดํ„ฐ/Spark (22)

๐Ÿฅ

[Spark] Adaptive Query Execution(AQE)

Spark 2.x ์—์„œ๋Š” rule-based/cost-based์˜ ๋ฐฉ์‹์œผ๋กœ ์ฟผ๋ฆฌ๋ฅผ ์ตœ์ ํ™”ํ•œ๋‹ค. Spark 3.0๋ถ€ํ„ฐ๋Š” ๋Ÿฐํƒ€์ž„์— ์ตœ์ ํ™”ํ•  ์ˆ˜ ์žˆ๋Š” AQE๊ฐ€ ๋„์ž…๋˜์—ˆ๊ณ , Spark 3.2 ๋ถ€ํ„ฐ๋Š” AQE ํ™œ์„ฑํ™”๊ฐ€ ๋””ํดํŠธ ๋ฒ„์ „์ด๋‹ค. AQE์—์„œ๋Š” ์ฟผ๋ฆฌ๋ฅผ ์ตœ์ ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ํฌ๊ฒŒ ์•„๋ž˜์˜ ์„ธ ๊ฐ€์ง€์˜ ๊ธฐ๋Šฅ์„ ์ง€์›ํ•œ๋‹ค. Coalescing Post shuffle partitions Switching join strategies Optimizing Skew Join Coalescing Shuffle Partitions ์…”ํ”Œ ํŒŒํ‹ฐ์…˜์˜ ์ˆ˜๋ฅผ ์ตœ์ ํ™”ํ•ด์ฃผ๋Š” ๊ธฐ๋Šฅ์ด๋‹ค. ์…”ํ”Œ์€ ์ŠคํŒŒํฌ ์–ดํ”Œ๋ฆฌ์ผ€์ด์…˜์˜ ์„ฑ๋Šฅ์— ๋งค์šฐ ํฐ ์—ญํ• ์„ ์ฐจ์ง€ํ•˜๊ณ , ๊ธฐ์กด์—๋Š” ์‚ฌ์šฉ์ž๊ฐ€ ์…”ํ”Œ ํฌ๊ธฐ ๋ฐ ํŒŒํ‹ฐ์…˜์„ ์ง์ ‘ ํ™•์ธํ•˜๋ฉฐ ์ˆ˜๋™์œผ๋กœ ํŒŒํ‹ฐ์…˜ ์ˆ˜๋ฅผ ์กฐ์ •ํ•ด์ฃผ์–ด์•ผ ํ–ˆ๋‹ค. AQE์—..

๋ฐ์ดํ„ฐ/Spark 2024. 3. 23. 15:45