ProcessCPUUsage field values doesn't mean anything

[FlorianLeChat]

Recently, I switched from the old version of this module to the new one and I noticed that the ProcessCPUUsage field was giving completely ununderstandable values and didn't seem to make sense compared to the others.

For example, here is the result I can get (it happens to me all the time and is not exceptional):

> PrintTable(serverstat.All())...
LogicalCPUs = 8
PhysicalCPUs = 8
ProcessCPUUsage = 40300 -- ???
ProcessMemoryUsage = 572.24609375
SystemAvailableMemory = 19969.983398438
SystemCPUUsage = 41.920177459717
SystemMemoryUsage = 49319.202148438
SystemTotalMemory = 70362.319335938

The same thing happens with the asynchronous function (but sometimes it returns 0 without any reason).

> serverstat.async.All(function(data) PrintTable(data) end)...
LogicalCPUs = 8
PhysicalCPUs = 8
ProcessCPUUsage = 0 -- ???
ProcessMemoryUsage = 573.2109375
SystemAvailableMemory = 20000.5234375
SystemCPUUsage = 100 -- Why not
SystemMemoryUsage = 49288.662109375
SystemTotalMemory = 70362.319335938

For information, my server is on a dedicated machine hosted by a game provider (so I don't have control of the machine to check the stats). AFAIK, the server is running under Debian 10 64-bit.

] version
Protocol version 24
Exe version 2021.06.09 (garrysmod)
Exe build: 14:56:57 Jul 12 2021 (8283) (4000)
GMod version 2021.07.12, branch: unknown
Linux 32bit Dedicated Server

And here is the script I use to get the information from the module.

-- From : https://github.com/WilliamVenner/gmsv_serverstat/
if not file.Exists("bin/gmsv_serverstat_*.dll", "LUA") then return end

require("serverstat")

concommand.Add("server-stats", function(ply)

    local data = serverstat.All()
    local phrase = "CPU Usage : %s (%s/%s)\nMemory Usage : %s (%s)"

    phrase = string.format(phrase,

        math.Round(data.ProcessCPUUsage, 2) .. "%",

        data.LogicalCPUs, data.PhysicalCPUs,

        string.NiceSize(data.ProcessMemoryUsage * 10 ^ 6),

        string.NiceSize(data.SystemTotalMemory * 10 ^ 6)

    )

    Raven.Server.SendChatMessage(ply, color_red, phrase)

end)

[WilliamVenner]

Extract from the docs for the upstream function

Returns the total CPU usage (in %). Notice that it might be bigger than 100 if run on a multicore machine.

If you want a value between 0% and 100%, divide the returned value by the number of CPU processors.

Warning: If you want accurate CPU usage number, better leave a bit of time between two calls of this method (200 ms for example).

This is what I do:

gmsv_serverstat/src/sysinfo.rs

Line 105 in 4b42cde

*cpu_usage = (process.cpu_usage() as f64) / (*LOGICAL_CPUS as f64);

I believe this is sound. Can you check if you can reproduce this on Windows?

[FlorianLeChat]

On Windows (and on Windows Server with a dedicated server), it seems these values are perfectly normal.

] lua_run PrintTable(serverstat.All())
> PrintTable(serverstat.All())...
LogicalCPUs	=	8
PhysicalCPUs	=	4
ProcessCPUUsage	=	6.8244895935059
ProcessMemoryUsage	=	634.419921875
SystemAvailableMemory	=	7734.50390625
SystemCPUUsage	=	16.620286941528
SystemMemoryUsage	=	8981.4033203125
SystemTotalMemory	=	16715.907226563

] lua_run serverstat.async.All(function(data) PrintTable(data) end)
> serverstat.async.All(function(data) PrintTable(data) end)...
LogicalCPUs	=	8
PhysicalCPUs	=	4
ProcessCPUUsage	=	9.4401035308838
ProcessMemoryUsage	=	635.7470703125
SystemAvailableMemory	=	7742.8876953125
SystemCPUUsage	=	11.839296340942
SystemMemoryUsage	=	8973.01953125
SystemTotalMemory	=	16715.907226563

But sometimes the SystemCPUUsage field returns 0 even if it has been previously called. I also tested this on another dedicated server but this time under CentOS 8 and I experience exactly the same problem described above.

[WilliamVenner]

@FlorianLeChat Can you test with v0.2.4?

[FlorianLeChat]

lua_run serverstat.async.All(function(data) PrintTable(data) end)
> serverstat.async.All(function(data) PrintTable(data) end)...
LogicalCPUs     =       1
PhysicalCPUs    =       1
ProcessCPUUsage =       9100
ProcessMemoryUsage      =       548.36328125
SystemAvailableMemory   =       1802.69921875
SystemCPUUsage  =       5.363748550415
SystemMemoryUsage       =       1448.689453125
SystemTotalMemory       =       3727.6279296875

I still get the problem unfortunately. The server is a VPS with 1 vCore and running CentOS 8. On Windows (and Windows Server), everything works perfectly.

[WilliamVenner]

@FlorianLeChat Looks like sysinfo may have made some fixes since, please give v0.2.5 a try.

[FlorianLeChat]

The problem is still there, unfortunately.

lua_run serverstat.async.All(function(data) PrintTable(data) end)
> serverstat.async.All(function(data) PrintTable(data) end)...
LogicalCPUs     =       2
PhysicalCPUs    =       2
ProcessCPUUsage =       0
ProcessMemoryUsage      =       925.6171875
SystemAvailableMemory   =       1345.5234375
SystemCPUUsage  =       4.6325511932373
SystemMemoryUsage       =       2071.0537109375
SystemTotalMemory       =       3728.015625

lua_run serverstat.async.All(function(data) PrintTable(data) end)
> serverstat.async.All(function(data) PrintTable(data) end)...
LogicalCPUs     =       2
PhysicalCPUs    =       2
ProcessCPUUsage =       100
ProcessMemoryUsage      =       925.625
SystemAvailableMemory   =       1345.3671875
SystemCPUUsage  =       5.5462183952332
SystemMemoryUsage       =       2071.2099609375
SystemTotalMemory       =       3728.015625

Protocol version 24
Exe version 2021.06.09 (garrysmod)
Exe build: 14:56:57 Jul 12 2021 (8283) (4000)
GMod version 2021.07.12, branch: unknown
Linux 32bit Dedicated Server

(Always under CentOS 8)

[WilliamVenner]

Is the issue isolated to only CentOS? Do you know if it happens on other distros?

[FlorianLeChat]

This is what I also said to myself when I wrote my answer just before, I have a feeling that CentOS might be the problem. I'm very tempted to install Debian and/or Ubuntu Server on a virtual machine tomorrow to test that.

[WilliamVenner]

If we can only reproduce this on CentOS, I can let the maintainer of sysinfo know and he can implement a fix.

[FlorianLeChat]

I just did the test under Debian 11/Ubuntu 21.10 and I get the same problem (even when switching to the 64 bits version of the dedicated server).

[WilliamVenner]

This is just weird. My Ubuntu 16 is fine:

[FlorianLeChat]

This issue is really confusing 😕

For more information, the tests I did today were done under VMWare 16 with the latest OS images and I installed a fresh new Garry's Mod server using LinuxGSM.

[WilliamVenner]

Ok, mine's under WSL. I doubt LinuxGSM could be doing that.