On 04/06/2024 21:52, Iñigo Mediavilla wrote:
Hello,
While there's ongoing work on:
https://github.com/openjdk/jdk/pull/19482
to add the stack trace of mounted virtual threads to the `jcmd <pid>
Thread.print` command, I'm starting to think about how I could do to
print the stack trace for virtual threads from `jstack` but I'm not
sure about what would be the best approach, so I'd be happy to get
people's thoughts on the topic.
JDK-8330846 is about showing the stack traces of mounted virtual
threads, I don't think pull/19482 should be expanded to do anything
else. For the PR then I think you are just waiting for Thomas Stuefe's
change with support for automatic indentation and of course converging
on some details of the output.
If you are asking if it should include the stack traces of unmounted
virtual threads then we decided some time ago to not do that. The
Thread.print commands runs as a VM operation so the at a safepoint.
Adding more work to that, to walk the heap to find virtual thread
objects, would just increase the STW time. The other issue is that the
output (text format) doesn't scale to hundreds of thousands of threads.
You might ask about de-duplication but that is just adding more overhead
to the thread dump and may be better left to tools that process the
thread dump. These are issues/reasons for Thread.dump_to_file. It uses
the tree of thread groupings to find all threads.
-Alan
