如何在 python 中使用 pyarrow 从 S3 读取分区的镶木地板文件

How to read partitioned parquet files from S3 using pyarrow in python

我正在寻找使用 python 从 s3 的多个分区目录读取数据的方法。

data_folder/serial_number=1/cur_date=20-12-2012/abcdsd0324324.snappy.parquet data_folder/serial_number=2/cur_date=27-12-2012/asdsdfsd0324324.snappy.parquet

pyarrow 的 ParquetDataset 模块具有从分区读取的能力。所以我尝试了以下代码:

>>> import pandas as pd
>>> import pyarrow.parquet as pq
>>> import s3fs
>>> a = "s3://my_bucker/path/to/data_folder/"
>>> dataset = pq.ParquetDataset(a)

它抛出以下错误:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/my_username/anaconda3/lib/python3.6/site-packages/pyarrow/parquet.py", line 502, in __init__
    self.metadata_path) = _make_manifest(path_or_paths, self.fs)
  File "/home/my_username/anaconda3/lib/python3.6/site-packages/pyarrow/parquet.py", line 601, in _make_manifest
    .format(path))
OSError: Passed non-file path: s3://my_bucker/path/to/data_folder/

根据 pyarrow 的文档,我尝试使用 s3fs 作为文件系统,即:

>>> dataset = pq.ParquetDataset(a,filesystem=s3fs)

抛出以下错误:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/my_username/anaconda3/lib/python3.6/site-packages/pyarrow/parquet.py", line 502, in __init__
    self.metadata_path) = _make_manifest(path_or_paths, self.fs)
  File "/home/my_username/anaconda3/lib/python3.6/site-packages/pyarrow/parquet.py", line 583, in _make_manifest
    if is_string(path_or_paths) and fs.isdir(path_or_paths):
AttributeError: module 's3fs' has no attribute 'isdir'

我仅限于使用 ECS 集群,因此 spark/pyspark 不是一个选项

有没有一种方法可以让我们在 python 中轻松地从 s3 中的此类分区目录轻松读取 parquet 文件?我觉得列出所有目录然后阅读并不是本 中建议的好做法。我需要将读取的数据转换为 pandas 数据帧以进行进一步处理,因此更喜欢与 fastparquet 或 pyarrow 相关的选项。我也对 python 中的其他选项持开放态度。

我设法让它与最新版本的 fastparquet 和 s3fs 一起工作。下面是相同的代码:

import s3fs
import fastparquet as fp
s3 = s3fs.S3FileSystem()
fs = s3fs.core.S3FileSystem()

#mybucket/data_folder/serial_number=1/cur_date=20-12-2012/abcdsd0324324.snappy.parquet 
s3_path = "mybucket/data_folder/*/*/*.parquet"
all_paths_from_s3 = fs.glob(path=s3_path)

myopen = s3.open
#use s3fs as the filesystem
fp_obj = fp.ParquetFile(all_paths_from_s3,open_with=myopen)
#convert to pandas dataframe
df = fp_obj.to_pandas()

感谢马丁通过我们的

为我指明了正确的方向

NB :这会比使用 pyarrow 慢,基于 benchmark . I will update my answer once s3fs support is implemented in pyarrow via ARROW-1213

我使用 pyarrow 对单个迭代进行了快速基准测试,并将文件列表作为 glob 发送到 fastparquet。 fastparquet 使用 s3fs 与 pyarrow + 我的 hackish 代码相比更快。但我认为 pyarrow +s3fs 实施后会更快。

代码和基准如下:

>>> def test_pq():
...     for current_file in list_parquet_files:
...         f = fs.open(current_file)
...         df = pq.read_table(f).to_pandas()
...         # following code is to extract the serial_number & cur_date values so that we can add them to the dataframe
...         #probably not the best way to split :)
...         elements_list=current_file.split('/')
...         for item in elements_list:
...             if item.find(date_partition) != -1:
...                 current_date = item.split('=')[1]
...             elif item.find(dma_partition) != -1:
...                 current_dma = item.split('=')[1]
...         df['serial_number'] = current_dma
...         df['cur_date'] = current_date
...         list_.append(df)
...     frame = pd.concat(list_)
...
>>> timeit.timeit('test_pq()',number =10,globals=globals())
12.078817503992468

>>> def test_fp():
...     fp_obj = fp.ParquetFile(all_paths_from_s3,open_with=myopen)
...     df = fp_obj.to_pandas()

>>> timeit.timeit('test_fp()',number =10,globals=globals())
2.961556333000317

2019 年更新

所有PR后,Arrow-2038 & Fast Parquet - PR#182等Issues已解决

使用 Pyarrow 读取 parquet 文件

# pip install pyarrow
# pip install s3fs

>>> import s3fs
>>> import pyarrow.parquet as pq
>>> fs = s3fs.S3FileSystem()

>>> bucket = 'your-bucket-name'
>>> path = 'directory_name' #if its a directory omit the traling /
>>> bucket_uri = f's3://{bucket}/{path}'
's3://your-bucket-name/directory_name'

>>> dataset = pq.ParquetDataset(bucket_uri, filesystem=fs)
>>> table = dataset.read()
>>> df = table.to_pandas() 

使用 Fast parquet 读取 parquet 文件

# pip install s3fs
# pip install fastparquet

>>> import s3fs
>>> import fastparquet as fp

>>> bucket = 'your-bucket-name'
>>> path = 'directory_name'
>>> root_dir_path = f'{bucket}/{path}'
# the first two wild card represents the 1st,2nd column partitions columns of your data & so forth
>>> s3_path = f"{root_dir_path}/*/*/*.parquet"
>>> all_paths_from_s3 = fs.glob(path=s3_path)

>>> fp_obj = fp.ParquetFile(all_paths_from_s3,open_with=myopen, root=root_dir_path)
>>> df = fp_obj.to_pandas()

快速基准测试

这可能不是对其进行基准测试的最佳方式。请阅读 blog post 以获得完整的基准测试

#pyarrow
>>> import timeit
>>> def test_pq():
...     dataset = pq.ParquetDataset(bucket_uri, filesystem=fs)
...     table = dataset.read()
...     df = table.to_pandas()
...
>>> timeit.timeit('test_pq()',number =10,globals=globals())
1.2677053569998407

#fastparquet
>>> def test_fp():
...     fp_obj = fp.ParquetFile(all_paths_from_s3,open_with=myopen, root=root_dir_path)
...     df = fp_obj.to_pandas()

>>> timeit.timeit('test_fp()',number =10,globals=globals())
2.931876824000028

关于 Pyarrow 的进一步阅读 speed

参考:

此问题已于 2017 年 this pull request 解决。

对于那些只想使用 pyarrow 从 S3 读取镶木地板的人,这里有一个例子:

import s3fs
import pyarrow.parquet as pq

fs = s3fs.S3FileSystem()
bucket = "your-bucket"
path = "your-path"

# Python 3.6 or later
p_dataset = pq.ParquetDataset(
    f"s3://{bucket}/{path}",
    filesystem=fs
)
df = p_dataset.read().to_pandas()

# Pre-python 3.6
p_dataset = pq.ParquetDataset(
    "s3://{0}/{1}".format(bucket, path),
    filesystem=fs
)
df = p_dataset.read().to_pandas()

对于那些只想读入分区 parquet 文件的 部分 的人,pyarrow 接受键列表以及部分目录路径以读入所有文件部分的分区。这种方法对于以有意义的方式(例如按年份或国家/地区)对 parquet 数据集进行分区的组织特别有用,允许用户指定他们需要文件的哪些部分。这将降低长期 运行 的成本,因为 AWS 在读取数据集时按字节收费。

# Read in user specified partitions of a partitioned parquet file 

import s3fs
import pyarrow.parquet as pq
s3 = s3fs.S3FileSystem()

keys = ['keyname/blah_blah/part-00000-cc2c2113-3985-46ac-9b50-987e9463390e-c000.snappy.parquet'\
         ,'keyname/blah_blah/part-00001-cc2c2113-3985-46ac-9b50-987e9463390e-c000.snappy.parquet'\
         ,'keyname/blah_blah/part-00002-cc2c2113-3985-46ac-9b50-987e9463390e-c000.snappy.parquet'\
         ,'keyname/blah_blah/part-00003-cc2c2113-3985-46ac-9b50-987e9463390e-c000.snappy.parquet']

bucket = 'bucket_yada_yada_yada'

# Add s3 prefix and bucket name to all keys in list
parq_list=[]
for key in keys:
    parq_list.append('s3://'+bucket+'/'+key)

# Create your dataframe
df = pq.ParquetDataset(parq_list, filesystem=s3).read_pandas(columns=['Var1','Var2','Var3']).to_pandas()

对于 python 3.6+ AWS 有一个名为 aws-data-wrangler 的库,它有助于 Pandas/S3/Parquet

之间的集成

安装做;

pip install awswrangler

要使用 awswrangler 1.x.x 及更高版本从 s3 读取分区镶木地板,执行;

import awswrangler as wr
df = wr.s3.read_parquet(path="s3://my_bucket/path/to/data_folder/", dataset=True)

通过设置 dataset=True awswrangler 需要分区的 parquet 文件。它将从您在 path.

中指定的 s3 密钥下方的分区中读取所有单独的镶木地板文件

PyArrow 7.0.0 对新模块 pyarrow.dataset 进行了一些改进,旨在从之前的 Parquet-specific pyarrow.parquet.ParquetDataset.[=16] 中抽象出数据集概念=]

假设您对从第一个文件推断出的数据集模式没有问题,example from the documentation for reading a partitioned dataset 应该就可以了。

这是一个 more-complete 示例,假设您要使用来自 S3 的数据:

import pyarrow.dataset as ds
from pyarrow import fs

s3 = fs.S3FileSystem()

dataset = ds.dataset(
    "my-bucket-name/my-path-to-dataset-partitions",
    format="parquet",
    filesystem=s3,
    partitioning="hive"
)

# Assuming your data is partitioned like year=2022/month=4/day=29
# this will only have to read the files for that day

expression = ((ds.field("year") == 2022) & (ds.field("month") == 4) & (ds.field("day") == 29))

pyarrow_table_2022_04_29 = dataset.to_table(filter=expression)

如果您自己定义数据集模式,请注意。上面使用分区参数 的推断会自动将分区添加到您的数据集模式

如果您希望分区与 manually-defined 数据集架构一起正常工作,您必须确保将分区添加到架构中:

import pyarrow as pa

my_manual_schema = pa.schema([])  # Some pyarrow.Schema instance for your dataset

# Be sure to add the partitions even though they're not in the dataset files
my_manual_schema.append(pa.field("year", pa.int16()))
my_manual_schema.append(pa.field("month", pa.int8()))
my_manual_schema.append(pa.field("day", pa.int8()))

dataset = ds.dataset(
    "my-bucket-name/my-path-to-dataset-partitions",
    format="parquet",
    filesystem=s3,
    schema=my_manual_schema,
    partitioning="hive"
)