本节介绍的DataStream API,则使用了类似的结构。
为了方便,我们依然使用from_collection从内存中读取数据。
和使用Table API类似,我们给from_collection传递的第二参数是每行数据类型。本例中是String,即“A C B”的类型。
from pyflink.common import Types
from pyflink.datastream import StreamExecutionEnvironment, RuntimeExecutionMode
word_count_data = ["A C B",
"A E B",
"E C D"]
def word_count():
env = StreamExecutionEnvironment.get_execution_environment()
env.set_runtime_mode(RuntimeExecutionMode.BATCH)
# write all the data to one file
env.set_parallelism(1)
source_type_info = Types.STRING()
# define the source
source = env.from_collection(word_count_data, source_type_info)
可以使用下面指令输出source内容
source.print()
A C B
A E B
E C D
和上图一样,Map由Splitting和Mapping组成。它们分别将数据切割成做小运算单元,和生成map结构。
def split(line):
for s in line.split():
yield s
splitted = source.flat_map(split)
上述splitted的结构输出是
A
C
B
A
E
B
E
C
D
Mapping的操作就是将之前的数组结构转换成map结构
mapped=splitted.map(lambda i: (i, 1), Types.TUPLE([Types.STRING(), Types.INT()]))
mapped的输出值如下,可以看到它还是按我们输入数据的顺序排列的。
(A,1)
(C,1)
(B,1)
(A,1)
(E,1)
(B,1)
(E,1)
(C,1)
(D,1)
这一步对应于上图中的Shuffling&Sorting,它会将相同key的数据进行分区,以供后面reducing操作使用。
keyed=mapped.key_by(lambda i: i[0])
可以看到keyed数据已经经过排序和聚合了。
(A,1)
(A,1)
(B,1)
(B,1)
(C,1)
(C,1)
(D,1)
reduced=keyed.reduce(lambda i, j: (i[0], i[1] + j[1]))
reduce的方法有如下注释
Applies a reduce transformation on the grouped data stream grouped on by the given key position. The
ReduceFunction
will receive input values based on the key value. Only input values with the same key will go to the same reducer.
特别是最后一句非常有用“Only input values with the same key will go to the same reducer”(只有相同Key的输入数据才会进入相同的Reducer中)。这句话意味着上述Keyed的数据会被分组执行,于是就不会出现计算错乱。
(A,2)
(B,2)
(C,2)
(D,1)
(E,2)
from pyflink.common import Types
from pyflink.datastream import StreamExecutionEnvironment, RuntimeExecutionMode
word_count_data = ["A C B",
"A E B",
"E C D"]
def word_count():
env = StreamExecutionEnvironment.get_execution_environment()
env.set_runtime_mode(RuntimeExecutionMode.BATCH)
# write all the data to one file
env.set_parallelism(1)
source_type_info = Types.STRING()
# define the source
source = env.from_collection(word_count_data, source_type_info)
# source.print()
def split(line):
for s in line.split():
yield s
splitted = source.flat_map(split)
# splitted.print()
mapped=splitted.map(lambda i: (i, 1), Types.TUPLE([Types.STRING(), Types.INT()]))
# mapped.print()
keyed=mapped.key_by(lambda i: i[0])
# keyed.print()
reduced=keyed.reduce(lambda i, j: (i[0], i[1] + j[1]))
# define the sink
reduced.print()
# submit for execution
env.execute()
if __name__ == '__main__':
word_count()