Welcome!

Related Topics: Java IoT, Industrial IoT, Microservices Expo, IoT User Interface, Apache, @BigDataExpo

Java IoT: Article

Fix Memory Leaks in Java Production Applications

A memory diagnostics approach for production identifies and fixes the root cause of the problem

Adding more memory to your JVMs (Java Virtual Machines) might be a temporary solution to fixing memory leaks in Java applications, but it for sure won't fix the root cause of the issue. Instead of crashing once per day it may just crash every other day. "Preventive" restarts are also just another desperate measure to minimize downtime, but, let's be frank: this is not how production issues should be solved.

One of our customers - a large online retail store - ran into such an issue. They run one of their online gift card self-service interfaces on two JVMs. During peak holiday seasons when users are activating their gift cards or checking the balance, crashes due to OOM (Out Of Memory) were more frequent, which caused bad user experience. The first "measure" they took was to double the JVM Heap Size. This didn't solve the problem as JVMs were still crashing, so they followed the memory diagnostics approach for production as explained in Java Memory Leaks to identify and fix the root cause of the problem.

Before we walk through the individual steps, let's look at the memory graph that shows the problems they had in December during the peak of the holiday season. The problem persisted even after increasing the memory. They could fix the problem after identifying the real root cause and applying specific configuration changes to a third-party software component.

After identifying the actual root cause and applying necessary configuration changes did the memory leak issue go away? Increasing Memory was not even a temporary solution that worked.

Step 1: Identify a Java Memory Leak
The first step is to monitor the JVM/CLR Memory Metrics such as Heap Space. This will tell us whether there is a potential memory leak. In this case we see memory usage constantly growing, resulting in an eventual runtime crash when the memory limit is reached.

Java Heap Size of both JVMs showed significant growth starting Dec 2nd and Dec 4th resulting in a crash on Dec 6th for both JVMs when the 512MB Max Heap Size was exceeded.

Step 2: Identify problematic Java Objects
The out-of-memory exception automatically triggers a full memory dump that allows for an analysis of which objects consumed the heap and are most likely to be the root cause of the out-of-memory crash. Looking at the objects that consumed most of the heap below indicates that they are related to a third-party logging API used by the application.

Sorting by GC (Garbage Collection) Size and focusing on custom classes (instead of system classes) shows that 80% of the heap is consumed by classes of a third-party logging framework

A closer look at an instance of the VPReportEntry4 shows that it contains five strings - with one consuming 23KB (as compared to several bytes of other string objects).This also explains the high GC Size of the String class in the overall Heap Dump.

Individual very large String objects as part of the ReportEntry object

Following the referrer chain further up reveals the complete picture. The EventQueue keeps LogEvents in an Array, which keeps VPReportEntrys in an Array. All of these objects seem to be kept in memory as the objects are being added to these arrays but never removed and therefore not garbage collected:

Following the referrer tree reveals that global EventQueue objects hold on to the LogEvent and VPReportEntry objects in array lists which are never removed from these arrays

Step 3: Who allocates these objects?
Analyzing object allocation allows us to figure out which part of the code is creating these objects and adding them to the queue. Creating what is called a "Selective Memory Dump" when the application reached 75% Heap Utilization showed the customer that the ReportWriter.report method allocated these entries and that they have been "living" on the heap for quite a while.

It is the report method that allocates the VPReportEntry objects that stay on the heap for quite a while

Step 4: Why are these objects not removed from the Heap?
The premise of the third-party logging framework is that log entries will be created by the application and written in batches at certain times by sending these log entries to a remote logging service using JMS. The memory behavior indicates that even though these log entries might be sent to the service, these objects are not always removed from the EventQueue leading to the out-of-memory exception.

Further analysis revealed that the background batch writer thread calls a logBatch method, which loops through the event queue (calling EventQueue.next) to send current log events in the queue. The question is whether as many messages were taken out of the queue (using next) vs put into the queue (using add) and whether the batch job is really called frequently enough to keep up with the incoming event entries. The following chart shows the method executions of add, as well as the call to logBatch highlighting that logBatch is actually not called frequently enough and therefore not calling next to remove messages from the queue:

The highlighted area shows that messages are put into the queue but not taken out because the background batch job is not executed. Once this leads to an OOM and the system restarts it goes back to normal operation but older log messages will be lost.

Step 5: Fixing the Java Memory Leak problem
After providing this information to the third-party provider and discussing with them the number of log entries and their system environment the conclusion was that our customer used a special logging mode that was not supposed to be used in high-load production environments. It's like running with DEBUG log level in a high load or production environment. This overwhelmed the remote logging service and this is why the batch logging thread was stopped and log events remained in the EventQueue until the out of memory occurred.

After making the recommended changes the system could again run with the previous heap memory size without experiencing any out-of-memory exceptions.

The Memory Leak issue has been solved and the application now runs even with the initial 512MB Heap Space without any problem.

They still use the same dashboards they have built to troubleshoot this issue, and to monitor for any future excessive logging problems.

These dashboards allow them to verify that the logging framework can keep up with log messages after they applied the changes.

Conclusion
Adding additional memory to crashing JVMs is most often not a temporary fix. If you have a real Java memory leak it will just take longer until the Java runtime crashes. It will even incur more overhead due to garbage collection when using larger heaps. The real answer to this is to use the simple approach explained here. Look at the memory metrics to identify whether you have a leak or not. Then identify which objects are causing the issue and why they are not collected by the GC. Working with engineers or third-party providers (as in this case) will help you find a permanent solution that allows you to run the system without impacting end users and without additional resource requirements.

Next Steps
If you want to learn more about Java Memory Management or general Application Performance Best Practices check out our free online Java Enterprise Performance Book. Existing customers of our APM Solution may also want to check out additional best practices on our APM Community.

More Stories By Andreas Grabner

Andreas Grabner has been helping companies improve their application performance for 15+ years. He is a regular contributor within Web Performance and DevOps communities and a prolific speaker at user groups and conferences around the world. Reach him at @grabnerandi

Comments (0)

Share your thoughts on this story.

Add your comment
You must be signed in to add a comment. Sign-in | Register

In accordance with our Comment Policy, we encourage comments that are on topic, relevant and to-the-point. We will remove comments that include profanity, personal attacks, racial slurs, threats of violence, or other inappropriate material that violates our Terms and Conditions, and will block users who make repeated violations. We ask all readers to expect diversity of opinion and to treat one another with dignity and respect.


Latest Stories
Between the mockups and specs produced by analysts, and resulting applications built by developers, there exists a gulf where projects fail, costs spiral, and applications disappoint. Methodologies like Agile attempt to address this with intensified communication, with partial success but many limitations. In his session at 18th Cloud Expo, Charles Kendrick, CTO & Chief Architect at Isomorphic Software, will present a revolutionary model enabled by new technologies. Learn how business and devel...
SYS-CON Events announced today that Peak 10, Inc., a national IT infrastructure and cloud services provider, will exhibit at SYS-CON's 18th International Cloud Expo®, which will take place on June 7-9, 2016, at the Javits Center in New York City, NY. Peak 10 provides reliable, tailored data center and network services, cloud and managed services. Its solutions are designed to scale and adapt to customers’ changing business needs, enabling them to lower costs, improve performance and focus inter...
You think you know what’s in your data. But do you? Most organizations are now aware of the business intelligence represented by their data. Data science stands to take this to a level you never thought of – literally. The techniques of data science, when used with the capabilities of Big Data technologies, can make connections you had not yet imagined, helping you discover new insights and ask new questions of your data. In his session at @ThingsExpo, Sarbjit Sarkaria, data science team lead ...
So, you bought into the current machine learning craze and went on to collect millions/billions of records from this promising new data source. Now, what do you do with them? Too often, the abundance of data quickly turns into an abundance of problems. How do you extract that "magic essence" from your data without falling into the common pitfalls? In her session at @ThingsExpo, Natalia Ponomareva, Software Engineer at Google, will provide tips on how to be successful in large scale machine lear...
If there is anything we have learned by now, is that every business paves their own unique path for releasing software- every pipeline, implementation and practices are a bit different, and DevOps comes in all shapes and sizes. Software delivery practices are often comprised of set of several complementing (or even competing) methodologies – such as leveraging Agile, DevOps and even a mix of ITIL, to create the combination that’s most suitable for your organization and that maximize your busines...
Struggling to keep up with increasing application demand? Learn how Platform as a Service (PaaS) can streamline application development processes and make resource management easy.
In his session at @ThingsExpo, Chris Klein, CEO and Co-founder of Rachio, will discuss next generation communities that are using IoT to create more sustainable, intelligent communities. One example is Sterling Ranch, a 10,000 home development that – with the help of Siemens – will integrate IoT technology into the community to provide residents with energy and water savings as well as intelligent security. Everything from stop lights to sprinkler systems to building infrastructures will run ef...
Whether your IoT service is connecting cars, homes, appliances, wearable, cameras or other devices, one question hangs in the balance – how do you actually make money from this service? The ability to turn your IoT service into profit requires the ability to create a monetization strategy that is flexible, scalable and working for you in real-time. It must be a transparent, smoothly implemented strategy that all stakeholders – from customers to the board – will be able to understand and comprehe...
Machine Learning helps make complex systems more efficient. By applying advanced Machine Learning techniques such as Cognitive Fingerprinting, wind project operators can utilize these tools to learn from collected data, detect regular patterns, and optimize their own operations. In his session at 18th Cloud Expo, Stuart Gillen, Director of Business Development at SparkCognition, will discuss how research has demonstrated the value of Machine Learning in delivering next generation analytics to im...
In the world of DevOps there are ‘known good practices’ – aka ‘patterns’ – and ‘known bad practices’ – aka ‘anti-patterns.' Many of these patterns and anti-patterns have been developed from real world experience, especially by the early adopters of DevOps theory; but many are more feasible in theory than in practice, especially for more recent entrants to the DevOps scene. In this power panel at @DevOpsSummit at 18th Cloud Expo, moderated by DevOps Conference Chair Andi Mann, panelists will dis...
See storage differently! Storage performance problems have only gotten worse and harder to solve as applications have become largely virtualized and moved to a cloud-based infrastructure. Storage performance in a virtualized environment is not just about IOPS, it is about how well that potential performance is guaranteed to individual VMs for these apps as the number of VMs keep going up real time. In his session at 18th Cloud Expo, Dhiraj Sehgal, in product and marketing at Tintri, will discu...
Unless you don’t use the internet, don’t live in California, or haven’t been paying attention to the recent news… you should be aware that self-driving cars are on their way to becoming a reality. I have seen them – they are real. If you believe in the future reality of self-driving cars, then continue reading on. If you don’t believe in the future possibilities, then I am not sure what to do to convince you other than discuss the very real changes that will roll out with the consumer producti...
Up until last year, enterprises that were looking into cloud services usually undertook a long-term pilot with one of the large cloud providers, running test and dev workloads in the cloud. With cloud’s transition to mainstream adoption in 2015, and with enterprises migrating more and more workloads into the cloud and in between public and private environments, the single-provider approach must be revisited. In his session at 18th Cloud Expo, Yoav Mor, multi-cloud solution evangelist at Cloudy...
There is an ever-growing explosion of new devices that are connected to the Internet using “cloud” solutions. This rapid growth is creating a massive new demand for efficient access to data. And it’s not just about connecting to that data anymore. This new demand is bringing new issues and challenges and it is important for companies to scale for the coming growth. And with that scaling comes the need for greater security, gathering and data analysis, storage, connectivity and, of course, the...
This is not a small hotel event. It is also not a big vendor party where politicians and entertainers are more important than real content. This is Cloud Expo, the world's longest-running conference and exhibition focused on Cloud Computing and all that it entails. If you want serious presentations and valuable insight about Cloud Computing for three straight days, then register now for Cloud Expo.