How to handle a huge stream of JSON dictionaries?

Question

I have a file that contains a stream of JSON dictionaries like this:

{"menu": "a"}{"c": []}{"d": [3, 2]}{"e": "}"}

It also includes nested dictionaries and it looks like I cannot rely on a newline being a separator. I need a parser that could be used like this:

for d in getobjects(f):
  handle_dict(d)

The point is that it would be perfect if the iteration only happened at the root level. Is there a Python parser that would handle all JSON's quirks? I am interested in a solution that would work on files that wouldn't fit into RAM.

Answer 1

Here you go: a tested solution based on the answer from @Brien

This should be able to handle any arbitrary sized input file. It is a generator, so it yields up dictionary objects one at a time as it parses them out of the JSON input file.

If you run it as a stand-alone, it runs three test cases. (In the if __name__ == "__main__" block)

Of course, to make this read from standard input you would simply pass sys.stdin as the input file argument.

import json


_DECODER = json.JSONDecoder()

_DEFAULT_CHUNK_SIZE = 4096
_MB = (1024 * 1024)
_LARGEST_JSON_OBJECT_ACCEPTED = 16 * _MB  # default to 16 megabytes

def json_objects_from_file(input_file,
            chunk_size=_DEFAULT_CHUNK_SIZE,
            max_size=_LARGEST_JSON_OBJECT_ACCEPTED):
    """
    Read an input file, and yield up each JSON object parsed from the file.

    Allocates minimal memory so should be suitable for large input files.
    """
    buf = ''
    while True:
        temp = input_file.read(chunk_size)
        if not temp:
            break

        # Accumulate more input to the buffer.
        #
        # The decoder is confused by leading white space before an object.
        # So, strip any leading white space if any.
        buf = (buf + temp).lstrip()
        while True:
            try:
                # Try to decode a JSON object.
                x, i = _DECODER.raw_decode(buf)
                # If we got back a dict, we got a whole JSON object.  Yield it.
                if type(x) == dict:
                    # First, chop out the JSON from the buffer.
                    # Also strip any leading white space if any.
                    buf = buf[i:].lstrip()
                    yield x
            except ValueError:
                # Either the input is garbage or we got a partial JSON object.
                # If it's a partial, maybe appending more input will finish it,
                # so catch the error and keep handling input lines.

                # Note that if you feed in a huge file full of garbage, this will grow
                # very large.  Blow up before reading an excessive amount of data.

                if len(buf) >= max_size:
                    raise ValueError("either bad input or too-large JSON object.")
                break
    buf = buf.strip()
    if buf:
        if len(buf) > 70:
            buf = buf[:70] + '...'
        raise ValueError('leftover stuff from input: "{}"'.format(buf))

if __name__ == "__main__":
    from StringIO import StringIO

    jstring = '{"menu":\n"a"}{"c": []\n}\n{\n"d": [3,\n 2]}{\n"e":\n "}"}'
    f = StringIO(jstring)
    correct = [{u'menu': u'a'}, {u'c': []}, {u'd': [3, 2]}, {u'e': u'}'}]

    result = list(json_objects_from_file(f, chunk_size=3))
    assert result == correct

    f = StringIO(' ' * (17 * _MB))
    correct = []

    result = list(json_objects_from_file(f, chunk_size=_MB))
    assert result == correct

    f = StringIO('x' * (17 * _MB))
    correct = "ok"

    try:
        result = list(json_objects_from_file(f, chunk_size=_MB))
    except ValueError:
        result = correct
    assert result == correct

Answer 2

Here is a partial solution, but it keeps slowing down as input goes:

#!/usr/bin/env pypy

import json
import cStringIO
import sys

def main():
    BUFSIZE = 10240
    f = sys.stdin
    decoder = json.JSONDecoder()
    io = cStringIO.StringIO()

    do_continue = True
    while True:
        read = f.read(BUFSIZE)
        if len(read) < BUFSIZE:
            do_continue = False
        io.write(read)
        try:
            data, offset = decoder.raw_decode(io.getvalue())
            print(data)
            rest = io.getvalue()[offset:]
            if rest.startswith('\n'):
                rest = rest[1:]
            io = cStringIO.StringIO()
            io.write(rest)
        except ValueError, e:
            #print(e)
            #print(repr(io.getvalue()))
            continue
        if not do_continue:
            break

if __name__ == '__main__':
    main()

Answer 3

I think JSONDecoder.raw_decode may be what you're looking for. You may have to do some string formatting to get it in the perfect format depending on newlines and such, but with a bit of work, you'll probably be able to get something working. See this example.

import json
jstring = '{"menu": "a"}{"c": []}{"d": [3, 2]}{"e": "}"}'
substr = jstring
decoder = json.JSONDecoder()

while len(substr) > 0:
    data,index = decoder.raw_decode(substr)
    print data
    substr = substr[index:]

Gives output:

{u'menu': u'a'}
{u'c': []}
{u'd': [3, 2]}
{u'e': u'}'}