[pve-devel] [PATCH container] fix #4765: lxc: report cpu usage correctly

[pve-devel] [PATCH container] fix #4765: lxc: report cpu usage correctly

[Maximiliano Sandoval]

When running `pct status VMID` the variable
$last_proc_vmid_stat->{$vmid} is not set and pct reports no cpu usage.

We address this by computing the used cpu time over the total uptime of
the container.

Signed-off-by: Maximiliano Sandoval <m.sandoval@proxmox.com>
---
 src/PVE/LXC.pm | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/src/PVE/LXC.pm b/src/PVE/LXC.pm
index a531ea5..17a7cd7 100644
--- a/src/PVE/LXC.pm
+++ b/src/PVE/LXC.pm
@@ -276,10 +276,13 @@ sub vmstatus {
 
 	    my $old = $last_proc_vmid_stat->{$vmid};
 	    if (!$old) {
+		my $uptime_ms =  $d->{uptime} * $cpucount * 1000.0;
+		$d->{cpu} = (($used_ms / $uptime_ms) * $cpucount) / $d->{cpus};
+
 		$last_proc_vmid_stat->{$vmid} = {
 		    time => $cdtime,
 		    used => $used_ms,
-		    cpu => 0,
+		    cpu => $d->{cpu},
 		};
 		next;
 	    }
-- 
2.39.2

Re: [pve-devel] [PATCH container] fix #4765: lxc: report cpu usage correctly

[Fiona Ebner]

Am 20.06.23 um 17:46 schrieb Maximiliano Sandoval:
> When running `pct status VMID` the variable

Issue is only present with --verbose.

> $last_proc_vmid_stat->{$vmid} is not set and pct reports no cpu usage.
> 
> We address this by computing the used cpu time over the total uptime of
> the container.
> 

qm would have the same issue, but there, we don't even output a line for
"cpu:" ;)

IMHO, the average of the CPU usage across the whole uptime is not what
most people would expect this value to be. Maybe we should just remove
the line from the output like in qm (technically a breaking change) or
maybe query the CPU stats twice with a small time window in between?

No matter what approach we go for, the behavior of qm and pct should be
made consistent here (if no good reason against it pops up).

Re: [pve-devel] [PATCH container] fix #4765: lxc: report cpu usage correctly

[Thomas Lamprecht]

Am 23/06/2023 um 13:54 schrieb Fiona Ebner:
> IMHO, the average of the CPU usage across the whole uptime is not what
> most people would expect this value to be.

Yeah, this is really not that useful in most cases.

> Maybe we should just remove the line from the output like in qm (technically
> a breaking change)

Luckily we have a more pragmatic approach to breaking changes that isn't
strictly bound by technicalities.

So yes, as it was hardcoded 0 and thus completely useless I'd also just drop
it. You can 100% redirect all fault to me if somebody complaints.