从 Azure Databricks 中的 Azure Datalake Gen2 读取 .nc 文件

Question

尝试读取 Azure Databricks 中的 .nc (netCDF4) 文件。

从未使用过 .nc 文件

1. 所有必需的 .nc 文件都在 Azure Datalake Gen2 中
2. 在“/mnt/eco_dailyRain”处将上述文件挂载到 Databricks 中

3. 可以使用dbutils.fs.ls("/mnt/eco_dailyRain")列出mount的内容 输出：

   Out[76]: [FileInfo(path='dbfs:/mnt/eco_dailyRain/2000.daily_rain.nc', name='2000.daily_rain.nc', size=429390127),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc', name='2001.daily_rain.nc', size=428217143),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2002.daily_rain.nc', name='2002.daily_rain.nc', size=428218181),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2003.daily_rain.nc', name='2003.daily_rain.nc', size=428217139),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2004.daily_rain.nc', name='2004.daily_rain.nc', size=429390143),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2005.daily_rain.nc', name='2005.daily_rain.nc', size=428217137),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2006.daily_rain.nc', name='2006.daily_rain.nc', size=428217127),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2007.daily_rain.nc', name='2007.daily_rain.nc', size=428217143),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2008.daily_rain.nc', name='2008.daily_rain.nc', size=429390137),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2009.daily_rain.nc', name='2009.daily_rain.nc', size=428217127),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2010.daily_rain.nc', name='2010.daily_rain.nc', size=428217134),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2011.daily_rain.nc', name='2011.daily_rain.nc', size=428218181),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2012.daily_rain.nc', name='2012.daily_rain.nc', size=429390127),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2013.daily_rain.nc', name='2013.daily_rain.nc', size=428217143),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2014.daily_rain.nc', name='2014.daily_rain.nc', size=428218104),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2015.daily_rain.nc', name='2015.daily_rain.nc', size=428217134),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2016.daily_rain.nc', name='2016.daily_rain.nc', size=429390127),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2017.daily_rain.nc', name='2017.daily_rain.nc', size=428217223),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2018.daily_rain.nc', name='2018.daily_rain.nc', size=418143765),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/2019.daily_rain.nc', name='2019.daily_rain.nc', size=370034113),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/Consignments.parquet', name='Consignments.parquet', size=237709917),
    FileInfo(path='dbfs:/mnt/eco_dailyRain/test.nc', name='test.nc', size=428217137)]
   

只是为了测试是否可以从挂载中读取。

spark.read.parquet('dbfs:/mnt/eco_dailyRain/Consignments.parquet')

确认可以读取 parquet 文件。

输出

Out[83]: DataFrame[CONSIGNMENT_PK: int, CERTIFICATE_NO: string, ACTOR_NAME: string, GENERATOR_FK: int, TRANSPORTER_FK: int, RECEIVER_FK: int, REC_POST_CODE: string, WASTEDESC: string, WASTE_FK: int, GEN_LICNUM: string, VOLUME: int, MEASURE: string, WASTE_TYPE: string, WASTE_ADD: string, CONTAMINENT1_FK: int, CONTAMINENT2_FK: int, CONTAMINENT3_FK: int, CONTAMINENT4_FK: int, TREATMENT_FK: int, ANZSICODE_FK: int, VEH1_REGNO: string, VEH1_LICNO: string, VEH2_REGNO: string, VEH2_LICNO: string, GEN_SIGNEE: string, GEN_DATE: timestamp, TRANS_SIGNEE: string, TRANS_DATE: timestamp, REC_SIGNEE: string, REC_DATE: timestamp, DATECREATED: timestamp, DISCREPANCY: string, APPROVAL_NUMBER: string, TR_TYPE: string, REC_WASTE_FK: int, REC_WASTE_TYPE: string, REC_VOLUME: int, REC_MEASURE: string, DATE_RECEIVED: timestamp, DATE_SCANNED: timestamp, HAS_IMAGE: string, LASTMODIFIED: timestamp]

但尝试读取 netCDF4 文件时会显示 No such file or directory

代码：

import datetime as dt  # Python standard library datetime  module
import numpy as np
from netCDF4 import Dataset  # http://code.google.com/p/netcdf4-python/
import matplotlib.pyplot as plt

rootgrp = Dataset("dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc","r", format="NETCDF4")

错误

FileNotFoundError: [Errno 2] No such file or directory: b'dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc'

任何线索。

Answer 1

根据netCDF4 module类Dataset的API参考，如下图。

Dataset的path参数的值应该是unix目录格式的路径，但路径dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc是我知道的PySpark的格式，所以你得到了错误FileNotFoundError: [Errno 2] No such file or directory: b'dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc'。

解决方法是将路径值dbfs:/mnt/eco_dailyRain/2001.daily_rain.nc改成等价的unix路径/dbfs/mnt/eco_dailyRain/2001.daily_rain.nc，代码如下。

rootgrp = Dataset("/dbfs/mnt/eco_dailyRain/2001.daily_rain.nc","r", format="NETCDF4")

您可以通过下面的代码查看它。

%sh
ls /dbfs/mnt/eco_dailyRain

当然，你也可以通过dbutils.fs.ls('/mnt/eco_dailyRain')挂载你的netCDF4格式的数据文件。