RFR [XS] : 8204598 : add more thread-related system settings info to hs_error file on Linux

Thomas Stüfe thomas.stuefe at gmail.com
Fri Jun 8 09:42:44 UTC 2018


Hi Matthias,


Thanks for that change, this is useful.

--

I do not like lumping max_map_count together with the other variables
in ".._thread_info" since it has nothing to do with threads.

I would probably just rename the function to something different (e.g.
"print_procs_sys_info" as in "print information taken from kernel
variables in /proc/sys...") or spread them to other functions.

--

I dislike the "out->print("\n....\n") style. Could you please
reformulate like this:

out->cr();
out->print_cr("...");

to make newlines more explicit?

--

Can you please reformulate:

"kernel system-wide limit on the number of threads" -> "system-wide
limit on the number of kernel threads" or just "system-wide limit on
the number of threads"
"maximum number of unique process identifiers the system can support)"
-> "system-wide limit on number of process identifiers"

Thank you!

Thomas




On Fri, Jun 8, 2018 at 10:04 AM, Baesken, Matthias
<matthias.baesken at sap.com> wrote:
> Hi could you please review  this small Linux related change ?
>
> In linux os::print_os_info ,  I  print additional  info about a number of system parameters influencing thread creation on Linux.
>
> We noticed the influence of these parameters when looking into an application creating over 10.000 threads on Linux at the same time; there we got an OOM : unable to create new native thread
> which was caused by a failing pthread_create (error EAGAIN) .
>
> The machine had plenty of memory, so we looked into various kernel params and in the end noticed that /proc/sys/kernel/pid_max was too low.
> The other added parameters "threads-max" and "max_map_count" are also known to be related to problems when running with high thread numbers, so I add them too .
>
>
>
> Webrev :
>
> http://cr.openjdk.java.net/~mbaesken/webrevs/8204598/
>
> Bug :
>
> https://bugs.openjdk.java.net/browse/JDK-8204598
>
>
> Thanks, Matthias
>


More information about the hotspot-dev mailing list