Spark TempView和GlobalTempView的區別

TempView和GlobalTempView在spark的Dataframe中經常使用，兩者的區別和應用場景有什麼不同。

我們以下麵的例子比較下兩者的不同。

from pyspark.sql import SparkSession
import numpy as np
import pandas as pd

spark = SparkSession.builder.getOrCreate()

d = np.random.randint(1,100, 5*5).reshape(5,-1)
data = pd.DataFrame(d, columns=list('abcde'))
df = spark.createDataFrame(data)
df.show()

+---+---+---+---+---+
|  a|  b|  c|  d|  e|
+---+---+---+---+---+
| 17| 30| 61| 61| 33|
| 32| 23| 24|  7|  7|
| 47|  6|  4| 95| 34|
| 50| 69| 83| 21| 46|
| 52| 12| 83| 49| 85|
+---+---+---+---+---+

從tempview中取數據

temp = df.createTempView('temp')
temp_sql = "select * from temp where a=50"
res = spark.sql(temp_sql)
res.show()

+---+---+---+---+---+
|  a|  b|  c|  d|  e|
+---+---+---+---+---+
| 50| 69| 83| 21| 46|
+---+---+---+---+---+

從globaltempview中取數據

glob = df.createGlobalTempView('glob')
glob_sql = "select * from global_temp.glob where a = 17"
res2 = spark.sql(glob_sql)
res2.show()

+---+---+---+---+---+
|  a|  b|  c|  d|  e|
+---+---+---+---+---+
| 17| 30| 61| 61| 33|
+---+---+---+---+---+

Globaltempview 數據可以在多個sparkSession中共用

# 創建新的sparkSession
spark2 = spark.newSession()
spark2 == spark

False

# 新的sparkSession可以獲取globaltempview中的數據
new_sql = "select * from global_temp.glob where a = 47"
temp = spark2.sql(new_sql)
temp.show()

+---+---+---+---+---+
|  a|  b|  c|  d|  e|
+---+---+---+---+---+
| 47|  6|  4| 95| 34|
+---+---+---+---+---+

# 新的sparkSession無法獲取tempview中的數據
# 會提示找不到temp表

new_sql2 = "select * from temp where a = 47"
temp = spark2.sql(new_sql2)
temp.show()

# 使用global_temp首碼也不行
new_sql2 = "select * from global_temp.temp where a = 47"
temp = spark2.sql(new_sql2)
temp.show()

---------------------------------------------------------------------------
Py4JJavaError                             Traceback (most recent call last)
# 此處多行刪除異常信息
AnalysisException: "Table or view not found: `global_temp`.`temp`; line 1 pos 14;\n'Project [*]\n+- 'Filter ('a = 47)\n   +- 'UnresolvedRelation `global_temp`.`temp`\n"

tempview刪除後無法使用

spark.catalog.dropTempView('temp')
spark.catalog.dropGlobalTempView('glob')

# 報錯，找不到table temp
temp_sql2 = "select * from temp where a = 47"
temp = spark.sql(temp_sql2)

# 報錯，找不到global_temp.glob，spark和spark2中均報錯
glob_sql2 = "select * from global_temp.glob where a = 47"
temp = spark.sql(glob_sql2)
temp = spark2.sql(glob_sql2)

總結

spark中有四個tempview方法

• df.createGlobalTempView
• df.createOrReplaceGlobalTempView
• df.createOrReplaceTempView
• df.createTempView

replace方法：不存在則直接創建，存在則替換

tempview刪除後無法使用

兩個刪除方法
spark.catalog.dropTempView('temp')
spark.catalog.dropGlobalTempView('glob')

TempView和GlobalTempView的異同

1. tempview只能在一個sparkSession中使用
2. GlobaltempView可以在多個sparkSession中共用使用
3. 但是他們都不能跨Application使用