I'd like to offline profile a YARN application. My application starts 10 containers. One of those starts consuming 100% CPU shortly before it crashes. I don't know which one is going to crash. And sadly, as this application starts multiple containers, they could potentially run on the same cluster hardware nodes. If I use a standard JProfiler config, I have to specify a filename for the savepoint. On my cluster, I can't use a relative path (relative to the working directory) as the folder is removed when the application crashes/finishes, so I need to specify an absolute path on the machine which then is the same for all containers overwriting their snapshots...
Is there a way to solve this issue? I'd like to offline profile all of my containers, let them write to a savepoint and pick the snapshot of the one which crashed.
Today, I found an old jprofiler-config.xml for JProfiler 7 on GitHub indicating that there are several more options available compared to the ones I see in my jprofiler-config.xml (JProfiler 11). Are you aware if there is any XSD describing the schema and what can be configured?
The "Save snapshot" action can add a number to the snapshot name to make it unique. It will check for existing files before writing the snapshot.
Since JProfiler 11, the config file only saves modified attributes and not all possible attributes. You can make changes in the GUI and observe the changes in the config file. There is no XSD for the config file format.