What is the solution?

As a tongue-in-cheek solution, if you just wished to get rid of the “java.lang.OutOfMemoryError: GC overhead limit exceeded” message, adding the following to your startup scripts would achieve just that:

-XX:-UseGCOverheadLimit

I would strongly suggest NOT to use this option though – instead of fixing the problem you just postpone the inevitable: the application running out of memory and needing to be fixed. Specifying this option will just mask the original java.lang.OutOfMemoryError: GC overhead limit exceeded error with a more familiar message java.lang.OutOfMemoryError: Java heap space.

On a more serious note – sometimes the GC overhead limit error is triggered because the amount of heap you have allocated to your JVM is just not enough to accommodate the needs of your applications running on that JVM. In that case, you should just allocate more heap – see at the end of this chapter for how to achieve that.

In many cases however, providing more Java heap space will not solve the problem. For example, if your application contains a memory leak, adding more heap will just postpone the java.lang.OutOfMemoryError: Java heap space error. Additionally, increasing the amount of Java heap space also tends to increase the length of GC pauses affecting your application’s throughput or latency.

If you wish to solve the underlying problem with the Java heap space instead of masking the symptoms, you need to figure out which part of your code is responsible for allocating the most memory. In other words, you need to answer these questions:

  1. Which objects occupy large portions of heap
  2. where these objects are being allocated in source code

At this point, make sure to clear a couple of days in your calendar (or – see an automated way below the bullet list). Here is a rough process outline that will help you answer the above questions:

  • Get clearance for acquiring a heap dump from your JVM-to-troubleshoot. “Dumps” are basically snapshots of heap contents that you can analyze, and contain everything that the application kept in memory at the time of the dump. Including passwords, credit card numbers etc.
  • Instruct your JVM to dump the contents of its heap memory into a file. Be prepared to get a few dumps, as when taken at a wrong time, heap dumps contain a significant amount of noise and can be practically useless. On the other hand, every heap dump “freezes” the JVM entirely, so don’t take too many of them or your end users start swearing.
  • Find a machine that can load the dump. When your JVM-to-troubleshoot uses for example 8GB of heap, you need a machine with more than 8GB to be able to analyze heap contents. Fire up dump analysis software (we recommend Eclipse MAT, but there are also equally good alternatives available).
  • Detect the paths to GC roots of the biggest consumers of heap. We have covered this activity in a separate post here. Don’t worry, it will feel cumbersome at first, but you’ll get better after spending a few days digging.
  • Next, you need to figure out where in your source code the potentially hazardous large amount of objects is being allocated. If you have good knowledge of your application’s source code you’ll hopefully be able to do this in a couple searches. When you have less luck, you will need some energy drinks to assist.

Alternatively, we suggest Plumbr, the only Java monitoring solution with automatic root cause detection. Among other performance problems it catches all java.lang.OutOfMemoryErrors and automatically hands you the information about the most memory-hungry data structres. It takes care of gathering the necessary data behind the scenes – this includes the relevant data about heap usage (only the object layout graph, no actual data), and also some data that you can’t even find in a heap dump. It also does the necessary data processing for you – on the fly, as soon as the JVM encounters an java.lang.OutOfMemoryError. Here is an example java.lang.OutOfMemoryError incident alert from Plumbr:
Plumbr OutOfMemoryError incident alert
Without any additional tooling or analysis you can see:

  • Which objects are consuming the most memory (271 com.example.map.impl.PartitionContainer instances consume 173MB out of 248MB total heap)
  • Where these objects were allocated (most of them allocated in the MetricManagerImpl class, line 304)
  • What is currently referencing these objects (the full reference chain up to GC root)

Equipped with this information you can zoom in to the underlying root cause and make sure the data structures are trimmed down to the levels where they would fit nicely into your memory pools.

However, when your conclusion from memory analysis or from reading the Plumbr report are that memory use is legal and there is nothing to change in the source code, you need to allow your JVM more Java heap space to run properly. In this case, alter your JVM launch configuration and add (or increase the value if present) just one parameter in your startup scripts:

java -Xmx1024m com.yourcompany.YourClass

In the above example the Java process is given 1GB of heap. Modify the value as best fits to your JVM. However, if the result is that your JVM still dies with OutOfMemoryError, you might still not be able to avoid the manual or Plumbr-assisted analysis described above.