java.lang.OutOfMemoryError from large Map marked with @Stringify

Question

How can I resolve a "java.lang.OutOfMemoryError: Java heap space" for the Map property timeseries that's apparently too big to retrieve from the Google App Engine datastore? The error occurs when I try to read the entity containing the large map from the datastore using an Objectify load() operation.

Here's the entity containing the large Map:

@com.googlecode.objectify.annotation.Entity
public class Insight {
    @com.googlecode.objectify.annotation.Id 
    public Long id;

    @Stringify(com.netbase.model.DateStringifier.class)
    public Map<Date, Double> timeseries;
}

Here is the Stringifier implementation:

public class DateStringifier implements Stringifier<Date> {

    @Override
    public String toString(Date obj) {
        return Long.toString(obj);
    }

    @Override
    public Date fromString(String str) {
        long timestamp_long = Long.parseLong(timestamp);
        Date date = new Date(timestamp_long);
        return date;
    }    
}

I've already tried running this in a Google Module with the largest memory size (Instance Class B8 as described at https://cloud.google.com/appengine/docs/java/modules/).

I've also tried using cursors, thinking garbage collection would happen after ever cursor refresh. But I'm actually get the exception on the very first cursor with a cursor size of 1, suggesting that a single Map itself is too big.

I'm not sure how I was able to persist such a large map in the first place but it happened weeks ago and the logs are lost now. It was an expensive operation to build the Map so I'd like to see if I can still retrieve it.

Since it was persisted with Objectify, I'm not sure how to read it with the underlying datastore. Any advice on how to resolve this would be most appreciated.

============================================================

Update: I managed to get my data out of Google App Engine by exporting it in pieces. I'm not sure how I managed to do it when the original problem seemed like it would have prevented me.

At any rate, I wanted to update you guys and let you know that the total amount of data was 23.5 MB of data. It's about 6K entities, each of which has the Map<Date, Double> timeseries. That means each entity had about 3.9 MB of data. According to @stickfigure in his solution below, that's about the limit of how big an Entity can be. So I will have to figure out some way to break the Entities up into smaller ones and process the data in separate sessions. I'm kind of disappointed because I thought Google App Engine would give me a lot more resources. I figure, if I can hold 23.5 MB in an Excel file and Excel on my local machine can process this much data without trouble, then Google App Engine should be able to handle many times more.

Good question @MoritzPetersen but the issue is I don't have a way to tell — Michael Osofsky

stickfigure stickfigure · Accepted Answer · 2015-01-30T20:09:46

There is something else going on here. GAE entities can't be larger than 1MB; even allowing for a multiple-X expansion factor (your POJO and the native low-level Entity exist in RAM simultaneously, String form plus Date form, etc) any single entity cannot represent more than a few megabytes.

It's likely that you're running at the edge of your heap and this couple-megabyte operation just happens to be the one that pushes you over the edge. The question is why, especially if you're already using a B8. I would run your app locally with a profiler.

One good candidate is bulk iteration through lots of entities in a single session. You can call ofy().clear() on iteration if you need it. You can read more in the session caching documentation.

java.lang.OutOfMemoryError from large Map marked with @Stringify

1 Answers