Custom Iterator for Processing Large Files

Question

I frequently work with large files and implemented a custom iterator to process these files efficiently while minimizing memory usage. The iterator reads each line from a file, parses it into a Message object, and skips invalid entries.

Are there any optimizations or best practices I can apply to this custom iterable implementation to further reduce memory usage or improve its performance? Below is the implementation for reference:

public class MessageIterable implements Iterable<Message> {

private final Logger logger = MessageFileProcessorSingleton.getInstance().getLog();

private final BufferedReader bufferedReader;
private final MessageParserService parserService;
private final PostExceptionTasksHandler postExceptionTasksHandler;

public MessageIterable(
        BufferedReader bufferedReader,
        PostExceptionTasksHandler postExceptionTasksHandler,
        MessageParserService service) {
    this.bufferedReader = bufferedReader;
    this.postExceptionTasksHandler = postExceptionTasksHandler;
    this.parserService = service;
}

@Override
public Iterator<Message> iterator() {

    return new Iterator<>() {
        Message next;

        @Override
        public boolean hasNext() {
            if (next != null) {
                return true;
            }

            try {
                String messageRowJson;
                while ((messageRowJson = bufferedReader.readLine()) != null) {
                    next = parserService.getMessageFromRowJson(messageRowJson);
                    if (next == null) {
                        logger.log(Level.SEVERE, "Unable to parse messaeg row");
System.exit(-1);
                    } else {
                        return true;
                    }
                }
            } catch (IOException e) {
                handleInvalidMessage();
            }
            return false;
        }

        @Override
        public Message next() {
            Message current = next;
            next = null;
            return current;
        }
    };
}

private void handleInvalidMessage() {
    logger.warn("Invalid message encountered. Executing post-exception tasks.");
    postExceptionTasksHandler.handle();
}

Answer 1

PostExceptionTasksHandler seems like an unnecessary type. It is basically a Runnable that is executed when an exception occurs. It would likely be more useful if you passed the exception that caused handler invocation to the handler itself. If you do that, the handler type could be replaced with Consumer<IOException> and that already contains enough information in it's name to suggest to the caller what it is used for.

If you have access to the MessageParserService.getMessageFromRowJson(String) method source code, I would change it to throw an exception on an invalid message instead of returning a null. That way you provide the caller with more information for deciding on how to handle the error. Of if you want to use it in a context where exceptions are not desired, maybe change the Message class so that it contains information about the validity so that you can return an error status from the conversion method.

MessageParserService seems like it's more of a mapper than an actual service. The method signature suggests that it converts a JSON string to a data object. These are of course just naming conventions and thus highly opinion based, but I prefer that classes that just convert data types to follow naming pattern like MessageMapper.from(String source). This has benefits because the name restricts the class' scope to creating Message objects and the method name clearly tells that it creates a Message from a String. Only having methods named "from" in your mapper helps you think twice before you start loading responsibilities to it.

Do not call System.exit() to handle an error unless you know exactly why you want to do that. Based on the fact that you describe the code as "skipping invalid entries" I would guess that you don't know. Because the System.exit doesn't skip invalid entries. Instead it does the equivalent of yanking the power cord from the computer when it encounters a spelling error... There are a lot of related questions on Stack Overflow about this. You can start from here: https://stackoverflow.com/questions/3715967/when-should-we-call-system-exit-in-java

For what it's worth, maybe you could use Files.lines(Path, Charset) method instead of writing a special iterator for this? Something like this:

Files.lines(path, charset)
    .map(parserService::getMessageFromRowJson)
    .filter(Objects::nonNull)
    .forEach([ message processing method ]);

Regarding the question about improving memory usage and performance. You didn't provide any information about how the memory usage is a problem and there is very little to be done with the code we see. If memory usage really is a problem, it is likely in the code that processes the Message objects after they have been read.

Regarding runtime performance you're better of with using Files.lines(...) as the generated stream is backed by a memory mapped file and it is very hard to improve the performance from that. You can even use a parallel stream with it to see if taking advantage of a multi-processor computer pays off in your use case.

But the improved runtime performance is paid off with increased memory consumption. There really is no end-all answer to this. Here is a good article about performance optimization: https://ericlippert.com/2012/12/17/performance-rant/

Answer 2

I haven't looked too deeply at this, but I notice that the Iterator only generates the next value in hasNext, and next just assumes this has already been done. This means that if there are ever two next() calls without a hasNext() in between, the second one will always return null as opposed to the value it ought return. Instead of assuming every call to next will be preceded by a hasNext call, make it so that next may itself advance the generation if needed (one way to do this could be having it call hasNext. Breaking the advancing logic into a helper method which both next and hasNext individually call is another).

Additionally, Iterator::next() should throw a NoSuchElementException when the underlying collection is exhausted. This implementation doesn't seem to do this.

Finally, it seems we expect message parsing to indicate failure in two different ways - either by throwing an IOException or by returning null. The former is handled well. The latter is handled by forcibly terminating the whole entire program with System::exit, leaving the caller no chance to recover or even clean up. Prefer handling this case in a manner more like the way you handle the other exception path.

Answer 3

Disclaimer: I'm not a Java programmer (last time was ~14 years ago)

Buffer lines
It's a good choice to use a Buffered Reader here as FileIO is generally a bottleneck.

Don't terminate on non-critical errors
I would never expect that a function that's called hasNext has the ability to nuke the running application (System.exit(-1)). That's an absolute no-go.
Definitely prefer exceptions over program termination.

Don't loop if you don't want a loop It looks like this is just copy/pasted because why is there a while loop here? Either you terminate the program or you return from the function.

while ((messageRowJson = bufferedReader.readLine()) != null) {
  next = parserService.getMessageFromRowJson(messageRowJson);
  if (next == null) {
    logger.log(Level.SEVERE, "Unable to parse messaeg row");
    System.exit(-1);
  } else {
    return true;
  }
}

If you decide to implement the buffer system, use a for loop instead (for reading lines) or a while loop (for reading byte sizes).

User expectation
Typically the hasNext implementation should be very light. In your case it contains FileIO. Initially I found this weird but decided not to raise a comment about it, since you can't know if there is a nextValue unless when you read the file. Except that you can. The only way that there is no further data is when the file reader has encountered EOF. Otherwise there is another line to be read. If the line read is invalid, you can still throw an exception. That is, move the file reading code into the next function and just check if another read will succeed. Potentially couple this with the buffering (check if there are still lines on the buffer first).

Typos
"Unable to parse messaeg row" contains a typo.

Answer 4

The field next of the Iterator anonymous class should be private...

private Message next;

The narative of next method could get more expressive refactored from...

@Override
public Message next() {
    Message current = next;
    next = null;
    return current;
}

...to...

@Override
public Message next() {
    return message(next, null);
}

private Message message(Message returnee, Message next) {
    this.next = next;
    return returnee;
}