Michael Simacek pointed out that koschei-backend01.phx2 has 20 GB of physical memory, while Ansible inventory vars specify only 4 GB.
Does anyone know where this difference comes from? Did anyone increase VM settings without committing them to Ansible git repo? If so, what was the reason?
I would expect it is trying to deal with a large number of OOM's on the service
On 16 November 2016 at 08:17, Mikolaj Izdebski mizdebsk@redhat.com wrote:
Michael Simacek pointed out that koschei-backend01.phx2 has 20 GB of physical memory, while Ansible inventory vars specify only 4 GB.
Does anyone know where this difference comes from? Did anyone increase VM settings without committing them to Ansible git repo? If so, what was the reason?
-- Mikolaj Izdebski Software Engineer, Red Hat IRC: mizdebsk _______________________________________________ infrastructure mailing list -- infrastructure@lists.fedoraproject.org To unsubscribe send an email to infrastructure-leave@lists.fedoraproject.org
On Wed, 16 Nov 2016 14:17:46 +0100 Mikolaj Izdebski mizdebsk@redhat.com wrote:
Michael Simacek pointed out that koschei-backend01.phx2 has 20 GB of physical memory, while Ansible inventory vars specify only 4 GB.
Does anyone know where this difference comes from? Did anyone increase VM settings without committing them to Ansible git repo? If so, what was the reason?
It might have been me... bumping the memory up to see if it helped it with memory usage. ;( But usually when I make those changes, I also make them in ansible, and I don't recall making this one, so perhaps it was someone else. ;(
Anyhow, the way we setup hosts in ansible they have the memory they are set for in vars, and also we set max_memory to 5* that value. ie, this host was set for 4GB memory, so max was 20GB and it was dynamically increased to that.
Shall I decrease it back to 4GB? (That may need a reboot).
kevin
On 11/16/2016 04:17 PM, Kevin Fenzi wrote:
On Wed, 16 Nov 2016 14:17:46 +0100 Mikolaj Izdebski mizdebsk@redhat.com wrote:
Michael Simacek pointed out that koschei-backend01.phx2 has 20 GB of physical memory, while Ansible inventory vars specify only 4 GB.
Does anyone know where this difference comes from? Did anyone increase VM settings without committing them to Ansible git repo? If so, what was the reason?
It might have been me... bumping the memory up to see if it helped it with memory usage. ;( But usually when I make those changes, I also make them in ansible, and I don't recall making this one, so perhaps it was someone else. ;(
That won't help much with swap usage by itself. We are limiting RSS for systemd services, any virtual memory beyond limit (currently 3 GB) is swapped to disk. So it ends up with swap usage above 80 %, but mostly free physical memory - below 25 %.
Anyhow, the way we setup hosts in ansible they have the memory they are set for in vars, and also we set max_memory to 5* that value. ie, this host was set for 4GB memory, so max was 20GB and it was dynamically increased to that.
Right, now I remember seeing this.
Shall I decrease it back to 4GB? (That may need a reboot).
Not needed right now. I'll be reinstalling it as Fedora 25 after GA.
When reinstalling I would like to add more swap (4 GB instead of 2 GB; that would require a custom kickstart, IIUC) and set max_mem_size to 4 GB. Any reason not to do this?
Thanks,
On Wed, 16 Nov 2016 16:47:21 +0100 Mikolaj Izdebski mizdebsk@redhat.com wrote:
On 11/16/2016 04:17 PM, Kevin Fenzi wrote:
...snip...
Shall I decrease it back to 4GB? (That may need a reboot).
Not needed right now. I'll be reinstalling it as Fedora 25 after GA.
ok.
When reinstalling I would like to add more swap (4 GB instead of 2 GB; that would require a custom kickstart, IIUC) and set max_mem_size to 4 GB. Any reason not to do this?
I guess not. Perhaps we could just change the default f25 kickstart to do that. I don't see a big problem using a few more GB of disk space on them all and then we don't have to keep yet another custom one.
kevin
infrastructure@lists.fedoraproject.org