Pickle is slow with large objects on HDFS

[belldandyxtq]

A related issue to #36, As peek is not in the HdfsFile, the same issue also happens when using HDFS. Unlike the #36, where the peek in POSIX was just hiden by the FileObject, peek needs to be added in case of HDFS.

$ python test.py
SLF4J: Class path contains multiple SLF4J bindings.
SLF4J: Found binding in [jar:file:/usr/share/java/slf4j-simple.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: Found binding in [jar:file:/usr/lib/zookeeper/lib/slf4j-log4j12-1.7.5.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: See http://www.slf4j.org/codes.html#multiple_bindings for an explanation.
SLF4J: Actual binding is of type [org.slf4j.impl.SimpleLoggerFactory]
19/08/16 14:53:56 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
chainerio:  186.07044076919556

$ cat test.py
import time
import pickle
import chainerio

chainerio.set_root("hdfs")
cache_path = 'a_large_file.pkl'

start = time.time()
with chainerio.open(cache_path, 'rb') as f:
    data = pickle.load(f)
print('chainerio: ', time.time() - start)

Wed Jul 24 19:43:47 2019    profile

         466895442 function calls (466887441 primitive calls) in 161.088 seconds

   Ordered by: internal time
   List reduced from 2941 to 10 due to restriction <10>

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1   74.242   74.242  161.008  161.008 {built-in method _pickle.load}
233293409   46.669    0.000   84.175    0.000 
xxx/versions/3.7.2/lib/python3.7/site-packages/chainerio/fileobject.py:50(read)
233293409   37.505    0.000   37.505    0.000 {method 'read' of '_io.BufferedReader' objects}
     1452    0.384    0.000    0.388    0.000 <frozen importlib._bootstrap>:157(_get_module_lock)