TechPrufer
Get started
Atlassian
Atlassian logoAtlassian · Feature trail

JVM Continuous Profiling

1 post · Jun 2021

What this trail covers

Atlassian has developed and adopted a continuous profiling system for JVM applications to improve incident diagnosis and performance engineering. Initially, an in-house solution using AWS Systems Manager, Linux perf, and async-profiler was built to provide CLI-based profiling and flame graph visualization. Limitations in capturing transient issues and providing baselines led to the development of a continuous profiling system that scheduled regular profiling runs, stored data in Apache ORC format on S3, and enabled querying via AWS Athena. The team then evaluated and adopted Amazon CodeGuru Profiler as a platform-level solution, integrating it to provide continuous profiling capabilities to all Atlassian service teams. This has been used to identify and resolve issues such as excessive thread blocking due to synchronized EnumMaps by migrating to ConcurrentHashMap.

All Atlassian trails
  1. Continuous profiling of JVMThis post details the evolution of Atlassian's JVM profiling tooling, starting from an initial in-house solution for ad-hoc profiling using AWS Systems Manager, Linux perf, and async-profiler. It highlights the limitations of this approach, such as missing transient issues and lack of baselines, which drove the development of a continuous profiling system. This system involved scheduled profiling runs, data transformation to Apache ORC, storage in S3, and querying via AWS Athena. The post then discusses the adoption of Amazon CodeGuru Profiler as a platform-level solution, including a case study on using its HotSpot view to identify and resolve thread blocking issues by switching from synchronized EnumMap to ConcurrentHashMap.