| commit | da47be597be6a59e9e86202d041a85c274308edf | |
| author | Lucian Wischik <ljw@fb.com> | |
| Tue, 31 May 2022 15:42:33 +0000 (31 08:42 -0700) | ||
| committer | Facebook GitHub Bot <facebook-github-bot@users.noreply.github.com> | |
| Tue, 31 May 2022 15:42:33 +0000 (31 08:42 -0700) | ||
| tree | b2bd43e7c3ddf72ff43b3c57ba18c32528579925 | treesnapshot (tar.gz zip) |
| parent | 9cb61feb5bf14bb275e28acfdda0575d4b30cd38 | commitdiff |
Gather a bit more information from sysinfo()
Summary:
What I've been seeing is that hh_server memory.current + memory.swap.current was only a modest 30gb (and its memory.swap.current was only 2gb), but the machine as a whole -- which was supposed to be running solely hack -- nevertheless had 16gb of swap used and it OOMed.
It's an ongoing quest to find out what's up. But as a starting point, I'd like hack telemetry to be able to show machinewide memory+swap. That'll at least let us point fingers and say "not us". It'll also give us an idea of how heavily occupied the machine is other than hh_server's own contribution.
This diff is just a small first step. It changes sysinfo.c to return most of the fields from sysinfo(), rather than just two. I believe that sysinfo() is the right way to gather information about machine ran and swap use.
I also moved nproc.c into sysinfo.c since it felt pretty similar.
Differential Revision: D36715952
fbshipit-source-id: 186d2b46855955c161c67f6bd804e94e305df306
Summary:
What I've been seeing is that hh_server memory.current + memory.swap.current was only a modest 30gb (and its memory.swap.current was only 2gb), but the machine as a whole -- which was supposed to be running solely hack -- nevertheless had 16gb of swap used and it OOMed.
It's an ongoing quest to find out what's up. But as a starting point, I'd like hack telemetry to be able to show machinewide memory+swap. That'll at least let us point fingers and say "not us". It'll also give us an idea of how heavily occupied the machine is other than hh_server's own contribution.
This diff is just a small first step. It changes sysinfo.c to return most of the fields from sysinfo(), rather than just two. I believe that sysinfo() is the right way to gather information about machine ran and swap use.
I also moved nproc.c into sysinfo.c since it felt pretty similar.
Differential Revision: D36715952
fbshipit-source-id: 186d2b46855955c161c67f6bd804e94e305df306