From 5ba99c1c15c738b12ec380a18b68f59f8967705c Mon Sep 17 00:00:00 2001
From: Marsell Kukuljevic <mk@spearhead.systems>
Date: Mon, 17 Jun 2024 17:27:56 +0200
Subject: [PATCH] Add AMD GPU plugin.

---
 check_mk-amd-gpu/amd-gpu-0.1.0.mkp            | Bin 0 -> 2600 bytes
 .../base/plugins/agent_based/amd_gpu.py       | 106 ++++++++++++++++++
 .../plugins/wato/check_parameters/amd_gpu.py  |  85 ++++++++++++++
 .../custom/amd_gpu/lib/plugins/amd_gpu.ps1    |  20 ++++
 4 files changed, 211 insertions(+)
 create mode 100755 check_mk-amd-gpu/amd-gpu-0.1.0.mkp
 create mode 100644 check_mk-amd-gpu/local/lib/check_mk/base/plugins/agent_based/amd_gpu.py
 create mode 100644 check_mk-amd-gpu/local/lib/python3/cmk/gui/plugins/wato/check_parameters/amd_gpu.py
 create mode 100644 check_mk-amd-gpu/local/share/check_mk/agents/custom/amd_gpu/lib/plugins/amd_gpu.ps1

diff --git a/check_mk-amd-gpu/amd-gpu-0.1.0.mkp b/check_mk-amd-gpu/amd-gpu-0.1.0.mkp
new file mode 100755
index 0000000000000000000000000000000000000000..28c40eddd576864bbfc69ec6aefacaead9b321c4
GIT binary patch
literal 2600
zcmZwH`9Bkk1Hf^i$1O*W(&NZc%sI-jX-Uj2SB@=XL<q~2+mcOEZZl_YbCiYL#g`Io
z7`cyJkH;}=?lbNC<MaAHukTNvf8hQ4NWJD{nYTlEu`)LwFaxI>g{V~49SDjRH-UTK
zi$I1LJSa4uQl0p5{<yCFrH=ZeSB)*bX=XOxKehj2FicDvJtzAK4YRY;QTMcj!aji{
zlc9S0h!zNUV>|HiZQ{={$T<pRGoZd<^ihb5e*|gBzxu$6=ocs$;l~y)YkEF@=EQq(
zm_PAvIRVhA=ZdnYy@mMG?~}@Dk1waz4;O^^{xSK)d5KffO!6pUOJ}QvmpxV3cs3s{
zKx<nXA0F?KOIpoJbz@gq+TWH5J3K&qE$WiL&5MnKEQrpWxZeJ0pbYM|(oSmbxf;rb
z`_@N@l}XIvVGuyUH2!byk8$<+fI_%*(HpH)wsmof5WA@d$rnNl)g;l7sr=;R%M{*k
z8sjB_mAudLxzIHnG9tlbIpLNV=_9MPRl{UkaXs2#(S9a8yw!#(_U1`n>AY1bDDb(Z
zujaI`)3f%WoFI(!L{Aov{!o7TD7=`n63vH(@5S{8#IS7W?@dS_$UA06N*!$sgpQn`
zCzxY_HuFr}X~(d?@+LEY`19*oN9tLiFUQTpS1n8b@k-1hFE}{TwyG4xxb3OKhB3&@
zlUiiyA+*My{;Nt8{W}s9ecI=<!+zX<jy-xKY%muKy5aV)dDyP6`<hY3RB-4Eg0+__
zxbw8EC0=M5zew^(qnjdaA+KelMDQohPPCQ3(j>V$O)l7VRv;3Lb;AXMLk@9c5N!!_
ze{4U4=>^jlJHULQ(-)3A1eNoAWw$$uL~kD2xsx6;`o}v)>Hj&=zMMa%FkX<N;vIhG
z+<JP#gl=!-{?rj_Go&+s%s8xOd{Q(e)C(-jVW1-%2~I8L%FTGkONi4D>dbilL|Me}
zomR8+@rSes$-xwuHeVOfU0SYkm6GM=!Ov?HbOYqE7za`g(|u^O@qI~GH`Dg2seFn(
z-xt(md^0$RmGiXTfHr`S*QH^r=zc*@-=s>KN~moDj!HmYd0_zxbzScF^^J13Es>~F
ztV2Z)_6pWQM_nSz)?U`mxJ2Bl;N1=Uw}n2K!fXKo<hbv8v!<8g`f4uvcMZ2!TJ2VP
zC5(dzUD<VAe(&iE$y|FmPGlt7uB$=H#>6#sw<h>@i$}!2n<w_(wj~i&3n-!{*G))7
zQaR7iE0(hyzCnvI8_|hNVSn;eRVXsj;{4NBxH??i3xVjOw&sCs#&zuq!Kj_CrG&Ta
z2~`0*{b}0x^76SDnAbsRJ+xU3AlnS@?K3h^6i2+Lozs>^tLK@D@*yWwC!fFhSZPX7
zw4;bzhXE9!FEqN*W*E{$12Fldj@7}V6RDTCZf}cr%x*YClBN|0y>syF?~AB=fXcj*
zakexEm29y%1&UV;>3$!)!?{0kcmuz<JJkK8?Dx*2vauE8mDB!DB80E1u%M>L)hlEV
zG2To$Kwu5QS)yetlYNLhWB3%jj`x5Kz#BfT-xn*$E(z5uLyk#gqi`=Ns3aO4WS8zx
zaJ5yd-X*LKFghr8<D$s*^uuCMP#@-~p8Ib;7r=Dr=Q{~X#FX%;ygSZI4*&*j##Y#9
z#5Gx6T`}N@;5c9-37029>wZTF!G$>r-Hj;oaBHaMjhzUooL8n*I<ZB&L{)g9%?W#C
z*j{w^m<rO1v_B>K-S+bearAR5-~zn5jTTu*2R!>K{8YQnJQE*XGjjNm{Oj2WC7RNh
zzLl5;7?NJ2L~Gu=HfN;Rxk$c|5EZq&XV&xq3l!8LD&;kVaqGk@KE};JXJ?6R;F$OY
zS)B<V=eep{+-+$APIlIqf9GOa@F*$sq;h-pmv!6pCf)t_io(_hhGk^!J_662obmHo
zpPTZV{HfA5<TL$eA9IFXGIP1b9~=u=itLgF@;lvaw&p)Ly>?ZcM^C>r?O?xS$eMJj
zSGZRXz855Cs>Ow*${QM2tOE&y;@~7R`aNoH>Vm}+$sd07WG?wzZSp<dP;8Q%upH91
z>e+6?qM4P$AJ0u)1$R?$uIjQJQxV@`_&n=dC|#MHLAj=_cSp14y-_`{AE?OghYU~M
zY)rI~7JsTxy0e)|K8!tX_o$0jUwB(B(I<sDIS%c>95Cv4wR_HDojBb>1deRkXD_9n
z?bTxDhpy8wzHRUMPuq8k2VD9;hnFmTI6*CG(GGaFm@<A7@4vkUrH{;2Ta2Eh-cJym
zxI(R78g4Y`ZDOQfxnG8}WwDR~%6NBF&7_fvEfLfic46-5FJEO|U$OJ)g1TkgwRSR6
z9~w}6>O9XV&3RG7=sM!ScRqW&%se}9#x%i2n4bsjxXa_4&bRMxA7ojwX@LI1sH2N$
zil0Y^hmK=0!VZWwEi4~O;e4}K)w%$9$3=hEUm}DI#qU(Y;0n2;dfpx8>xm3@avAC1
zP%V<5+V{do{*zM6B}ZY=Yr7Z8l9l#KQ`FcA$5ZM?^~VD3TN@AWhq-PdOz7K?j|oej
zZ+#!9RpFC#N`?g*v%)XbWam0DAkIR-O6c}S-%i9U&rLZoCB~CPPp9)QjRZ}{EPYVD
zaLy}2aqtQ?@oejqPyUet@VXPVT3!#+K7tye9EO1)-Meer!}rPWQu|uJ51VG<90uDO
zm)57V60rYf_m4&7U*bK#))g=T`iji@haP9t#QApeI3o1Sag6zxu0n$}<@Bqx)s6=|
z(_o^7kAn6g4sr{LV$t_!(DydTz&rRa&b3Rf_U~B33nRDW&<6Qymw?bR-zu$^QvM7R
z7>DT%fQETBw!EKlSKy`5Et^a@k$hKVZ^)=+q{1T6u!hIpW<H1FU!A(Ev*+Y;J*XD?
z$L{uSYpTzy-jC{kzxY{vH%qnQr-Z%zK3GUCXsr)^gT%FaM_SQ-O^bR>8R?mGaalqA
z&m9e~7n+1-xH>dkv-ldZ3PtR0RYr=2!Lc&Jt+$%QX1eTLa)*oB&H5cQ7i7DiT;%ME
zBFv$*%Uu;lTu>B63izhOoYd+AQBRSE@gI(kxs*bGxTHf}Ho1i&0Z1M}Ee}1ii2zp?
zRv7DbYQ2#+uOqCfH@lOa+(zygIZ7I7nA5;rZ^k)YG}pkQzwMmO?XqS#xK&HY<x2fy
zTmvZ%X67_<R3J(IH*W#dFI2iYvN+E9jEq;R*=PaYvTl-##&zJUQMdZngB!Y=s@~3>
z1rYu4v~w7!8f>SwN7*feO|@~!4dKfXX?duW5q~C3-=DAe1`j4NISzpd8CASb*L*!5
zih!ho^j(~!Tgb_yme&<CbeS#`B)PTe@B_^5rZHyJL)u#g6};e^?$NYpHF7#*RM{9T
swX+($`Xh?fwfz?&rIdo`zFhGCM)i-F|DD)k!AmBqLSqz=rI3Z?zX?Yf0ssI2

literal 0
HcmV?d00001

diff --git a/check_mk-amd-gpu/local/lib/check_mk/base/plugins/agent_based/amd_gpu.py b/check_mk-amd-gpu/local/lib/check_mk/base/plugins/agent_based/amd_gpu.py
new file mode 100644
index 0000000..daf3add
--- /dev/null
+++ b/check_mk-amd-gpu/local/lib/check_mk/base/plugins/agent_based/amd_gpu.py
@@ -0,0 +1,106 @@
+#!/usr/bin/env python3
+#
+# Copyright 2024 Spearhead Systems SRL
+
+from cmk.base.plugins.agent_based.agent_based_api.v1 import (
+    register,
+    Service,
+    Result,
+    Metric,
+    State,
+)
+
+
+def discovery_amd_gpu(section):
+    name = section[0][0]
+    yield Service(item=name)
+
+
+def check_state(alert_percentages, measured_percent):
+    if alert_percentages:
+        if alert_percentages[1] <= measured_percent:
+            return State.CRIT
+        elif alert_percentages[0] <= measured_percent:
+            return State.WARN
+    return State.OK
+
+
+def get_levels(alert_levels, total=None):
+    if alert_levels == None:
+        return
+    
+    if total == None:
+        return alert_levels
+
+    return (alert_levels[0] / 100 * total, alert_levels[1] / 100 * total)
+
+
+def check_amd_gpu(item, params, section):
+    if item != section[0][0]:
+        return
+
+    gpu_percent      = int(float(section[1][0]))
+    vram_bytes_used  = int(section[2][0])
+    vram_bytes_total = int(section[3][0])
+    vram_bytes_free  = max(0, vram_bytes_total - vram_bytes_used)
+
+    vram_mb_used  = vram_bytes_used  // 1048576
+    vram_mb_total = vram_bytes_total // 1048576
+    vram_mb_free  = vram_bytes_free  // 1048576
+
+    alert_gpu_percent       = params.get("gpu_percent")
+    alert_vram_used_percent = params.get("vram_used_percent")
+    alert_vram_free_percent = params.get("vram_free_percent")
+
+    vram_used_percent = vram_bytes_used / vram_bytes_total * 100
+    vram_free_percent = 100 - vram_used_percent
+
+    yield Result(
+        state=check_state(alert_gpu_percent, gpu_percent),
+        summary=f"GPU: {gpu_percent}%"
+    )
+
+    yield Result(
+        state=check_state(alert_vram_free_percent, vram_free_percent),
+        summary=f"VRAM free: {vram_mb_free} MiB"
+    )
+
+    yield Result(
+        state=check_state(alert_vram_used_percent, vram_used_percent),
+        summary=f"VRAM used: {vram_mb_used} MiB"
+    )
+
+    yield Result(
+        state=State.OK,
+        summary=f"VRAM total: {vram_mb_total} MiB"
+    )
+
+    yield Metric(
+        name="gpu_percent",
+        value=gpu_percent,
+        levels=get_levels(alert_gpu_percent),
+        boundaries=(0, 100)
+    )
+
+    yield Metric(
+        name="vram_used",
+        value=vram_mb_used,
+        levels=get_levels(alert_vram_used_percent, vram_mb_total),
+        boundaries=(0, vram_mb_total)
+    )
+    
+    yield Metric(
+        name="vram_free",
+        value=vram_mb_free,
+        levels=get_levels(alert_vram_free_percent, vram_mb_total),
+        boundaries=(0, vram_mb_total)
+    )
+
+register.check_plugin(
+    name='amd_gpu',
+    service_name='AMD GPU - %s',
+    discovery_function=discovery_amd_gpu,
+    check_function=check_amd_gpu,
+    check_default_parameters={},
+    check_ruleset_name='amd_gpu',
+)
diff --git a/check_mk-amd-gpu/local/lib/python3/cmk/gui/plugins/wato/check_parameters/amd_gpu.py b/check_mk-amd-gpu/local/lib/python3/cmk/gui/plugins/wato/check_parameters/amd_gpu.py
new file mode 100644
index 0000000..b660603
--- /dev/null
+++ b/check_mk-amd-gpu/local/lib/python3/cmk/gui/plugins/wato/check_parameters/amd_gpu.py
@@ -0,0 +1,85 @@
+#!/usr/bin/env python3
+#
+# Copyright 2024 Spearhead Systems SRL
+
+from cmk.gui.i18n import _
+from cmk.gui.plugins.wato.utils import (
+    CheckParameterRulespecWithItem,
+    rulespec_registry,
+    RulespecGroupCheckParametersHardware,
+)
+from cmk.gui.valuespec import Dictionary, Percentage, Tuple
+
+
+def _parameter_valuespec_amd_gpu():
+    return Dictionary(
+        title=_("GPU utilization"),
+        help=_(
+            "These metrics are queried directly from the AMD GPU. "
+            "Upper and lower levels can be specified for individual metrics."
+        ),
+        elements=[
+            (
+                "gpu_percent",
+                Tuple(
+                    title=_("GPU Used"),
+                    help=_("If usage of total GPU compute goes above these percentages, issue alerts."),
+                    elements=[
+                        Percentage(
+                            title=_("Warn if above"),
+                            default_value=90
+                        ),
+                        Percentage(
+                            title=_("Crit if above"),
+                            default_value=100
+                        )
+                    ]
+                )
+            ),
+            (
+                "vram_free_percent",
+                Tuple(
+                    title=_("VRAM Free"),
+                    help=_("If free VRAM goes above these percentages, issue alerts."),
+                    elements=[
+                        Percentage(
+                            title="Warn if above",
+                            default_value=70
+                        ),
+                        Percentage(
+                            title="Crit if above",
+                            default_value=90
+                        )
+                    ]
+                )
+            ),
+            (
+                "vram_used_percent",
+                Tuple(
+                    title=_("VRAM Used"),
+                    help=_("If used VRAM goes above these percentages, issue alerts."),
+                    elements=[
+                        Percentage(
+                            title="Warn if above",
+                            default_value=70
+                        ),
+                        Percentage(
+                            title="Crit if above",
+                            default_value=90
+                        )
+                    ]
+                )
+            )
+        ]
+    )
+
+
+rulespec_registry.register(
+    CheckParameterRulespecWithItem(
+        check_group_name="amd_gpu",
+        group=RulespecGroupCheckParametersHardware,
+        match_type="dict",
+        parameter_valuespec=_parameter_valuespec_amd_gpu,
+        title=lambda: _("AMD GPU Metrics"),
+    )
+)
diff --git a/check_mk-amd-gpu/local/share/check_mk/agents/custom/amd_gpu/lib/plugins/amd_gpu.ps1 b/check_mk-amd-gpu/local/share/check_mk/agents/custom/amd_gpu/lib/plugins/amd_gpu.ps1
new file mode 100644
index 0000000..4cb5b74
--- /dev/null
+++ b/check_mk-amd-gpu/local/share/check_mk/agents/custom/amd_gpu/lib/plugins/amd_gpu.ps1
@@ -0,0 +1,20 @@
+# Copyright 2024 Spearhead Systems SRL
+#
+# This goes in C:\ProgramData\checkmk\agent\plugins. It should be added automatically by
+# baking a new MSI after setting "Agent Rules" > "Deploy Custom Files With Agent" with
+# "Deploy Custom Files With Agent" including "amd_gpu".
+
+foreach ($Item in Get-ChildItem "HKLM:\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}" -Name -Include 000*) {
+  $Name = Get-ItemPropertyValue "HKLM:\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\$Item" "DriverDesc"
+  if ($Name -match 'Radeon') {
+    $GpuBytesTotal = Get-ItemPropertyValue "HKLM:\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\$Item" "HardwareInformation.qwMemorySize"
+    $GpuRawName    = Get-ItemPropertyValue "HKLM:\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\$Item" "HardwareInformation.AdapterString"
+    break
+  }
+}
+
+$GpuName = [System.Text.Encoding]::Unicode.GetString($GpuRawName)
+$GpuPercent   = (((Get-Counter "\GPU Engine(*)\Utilization Percentage" ).CounterSamples).CookedValue | measure -sum).sum
+$GpuBytesUsed = (((Get-Counter "\GPU Process Memory(*)\Dedicated Usage").CounterSamples).CookedValue | measure -sum).sum
+
+Write-Output "<<<amd_gpu:sep(0)>>>", $GpuName, $GpuPercent, $GpuBytesUsed, $GpuBytesTotal