How to incrementally write into a json file

Question

I am writing a program, which requires me to generate a very large json file. I know the traditional way is to dump a dictionary list using json.dump()

Answer 1

I know this is a year late, but the issue is still open and I'm surprised the json.iterencode() was not mentioned.

The potential problem with iterencode in this example, is that you would want to have an iterative handle on the large data set by using a generator, and json encode does not serialize generators.

The way around this is to the subclass list type and override the __iter__ magic method so that you could yield the output of your generator.

Here is an example of this list sub class.

class StreamArray(list):
    """
    Converts a generator into a list object that can be json serialisable
    while still retaining the iterative nature of a generator.

    IE. It converts it to a list without having to exhaust the generator
    and keep it's contents in memory.
    """
    def __init__(self, generator):
        self.generator = generator
        self._len = 1

    def __iter__(self):
        self._len = 0
        for item in self.generator:
            yield item
            self._len += 1

    def __len__(self):
        """
        Json parser looks for a this method to confirm whether or not it can
        be parsed
        """
        return self._len

The usage from here on is quite simple. Get the generator handle, pass it into the StreamArray class, pass the stream array object into iterencode() and iterate over the chunks. The chunks will be json formated output which can be directly written to file.

Example usage:

#Function that will iteratively generate a large set of data.
def large_list_generator_func():
    for i in xrange(5):
        chunk = {'hello_world': i}
        print 'Yielding chunk: ', chunk
        yield chunk

#Write the contents to file:
with open('/tmp/streamed_write.json', 'w') as outfile:
    large_generator_handle = large_list_generator_func()
    stream_array = StreamArray(large_generator_handle)
    for chunk in json.JSONEncoder().iterencode(stream_array):
        print 'Writing chunk: ', chunk
        outfile.write(chunk)

The output that shows yield and writes happen consecutively.

Yielding chunk:  {'hello_world': 0}
Writing chunk:  [
Writing chunk:  {
Writing chunk:  "hello_world"
Writing chunk:  : 
Writing chunk:  0
Writing chunk:  }
Yielding chunk:  {'hello_world': 1}
Writing chunk:  , 
Writing chunk:  {
Writing chunk:  "hello_world"
Writing chunk:  : 
Writing chunk:  1
Writing chunk:  }

Answer 2

You could also assume you have an iterable it that you want to write out to a file handle fh as a large JSON array of records, and do the following, which I think is the most straightforward approach:

def write_json_iter(it, fh):
    print("[", file=fh)
    for n, rec in enumerate(it):
        if n > 0:
            print(",", file=fh)
        json.dump(rec, fh)
    print("]", file=fh)

Answer 3

Sadly the json library does not have any incremental writing facilities, and therefore cannot do what you want.

That's clearly going to be a very large file - would some other representation be more appropriate?

Otherwise the best suggestion I can make is to dump each list entry to an in-memory structure and write them out with the necessary delimiters ([ at the beginning, ],[ between entries and ] at the end) to try and construct the JSON that you need.

If formatting is important you should know that the wrapper test your program writes will destroy correct indentation, but indentation is only for humans so it shouldn't make any difference to the semantics of the JSON structure.