On Mon, 23 Oct 2023 05:25:50 GMT, Thomas Stuefe <stu...@openjdk.org> wrote:
> > The hs_err file includes a dump of the maps file. It seems these > > annotations would be useful there also. > > I thought about this too. The problem is the printing may take long since it > is O^2 (iterate through VMAs, for each one iterate NMT regions). We had cases > where the process ran with >20 mio VMAs (ZGC with a 5TB heap). > > But thinking further, this is a problem anyway, since the hs-err file was > >2GB. And we do have per-step timeouts in the hs-err writer. So, stopping > after a set number of VMAs (lets say 100k, that covers 99% of all processes) > and otherwise relying on the hs-err timeout code may be ok. > > I'll give it a shot. IIRC, both lists are sorted, should be a way to reduce time complexity. ------------- PR Comment: https://git.openjdk.org/jdk/pull/16301#issuecomment-1775748234