hdf5 AttributeError: 'UnImplemented' object has no attribute 'read'

问题现象：
最近在用pandas分析数据时，用hdf5存储结果，当我监听不同文件时，多个进程同时写入hdf5（save到不同group）时，报hdf5 AttributeError: 'UnImplemented' object has no attribute 'read'，之后用ipython去查数据，报了以下错误
HDF5ExtError: HDF5 error back trace 
  File "H5Dio.c", line 173, in H5Dread
    can't read data
  File "H5Dio.c", line 554, in H5D__read
    can't read data
  File "H5Dchunk.c", line 1875, in H5D__chunk_read
    unable to read raw data chunk
  File "H5Dchunk.c", line 2905, in H5D__chunk_lock
    data pipeline read failed
  File "H5Z.c", line 1372, in H5Z_pipeline
    filter returned failure during read
End of HDF5 error back trace
Problems reading records.

解决方案：
利用文件锁（fcntl模块）来保证数据的同步。
后来买了本书《Python和HDF5大数据应用》，里面介绍可以用mpio驱动来处理多进程的问题