NvGridSvc — NVIDIA Grid Service (NGS)

WARNING

Deprecated for latest service bundles. In the newer C:\Asgard\services layout, there is no NvGridSvc service directory, while a full gs2 stack is present. Treat this page as legacy/compatibility documentation for older seats.

NvGridSvc is the main session management service on each GFN rig. It implements the state machine that drives game sessions from request to teardown.

Version

From the logs:

NvGridSvc version 1.4
Build: Fri, 01 Sep 2023 21:17:37

State Machine

Configured via GFN.xaml (XAML-based state machine file):

{
  "NvGridSvcStateMachine": {
    "stateMachineFile": "GFN.xaml"
  }
}

The state machine drives the full lifecycle:

IDLE
  │ ← PM allocates session
  ▼
SESSION_CREATED
  │ ← Setup app/game, configure display
  ▼
READY_TO_STREAM
  │ ← Client connects
  ▼
STREAMING
  │ ← Active game session
  ▼
PAUSED  ←→  STREAMING
  │ ← Session ends / error
  ▼
TEARDOWN
  │ ← Logs uploaded, state wiped
  ▼
IDLE

GSG (Game Seat Gateway) Connection

NGS communicates with the cloud orchestration layer via gRPC over the GSG:

{
  "GSGConnector": {
    "connectionTimeoutMs": 5000,
    "keepAliveIntervalMs": 3000,
    "heartbeatPeriodMs": 60000,
    "enableSeatHeartbeat": true,
    "enableSessionHeartbeat": true,
    "apiKeyRefreshPeriodSec": 300,
    "frontChannelResponseTimeoutMs": 6000,
    "backChannelAckResponseTimeoutMs": 500
  }
}

Retry logic:

• Initial backoff: 100ms → Max: 1000ms (multiplier 1.6)
• Max 5 retries over 10 seconds
• Retryable: UNKNOWN, UNAVAILABLE, DEADLINE_EXCEEDED

Timeouts

From the master config:

Timeout	Value	Description
clientConnectTimeoutSec	180s	Wait for client to connect
clientReconnectTimeoutSec	180s	Allow client to reconnect
launchAppTimeoutSec	180s	Wait for app to launch
readyToStreamTimeoutSec	30s	Wait for stream ready
setupAppTimeoutSec	30s	App setup timeout
resumeReadyTimeoutSec	30s	Session resume timeout
appShutdownTimeoutSec	360s	Wait for app shutdown
gamestreamShutdownTimeoutSec	60s	GameStream shutdown
sessionCreatedTimeoutSec	180s	Session creation timeout
prewarmAppTimeoutSec	15s	Prewarm timeout per app

App Manager Timeouts

Timeout	Value
contentControllerJoinTimeoutSec	15s
appPrewarmReadyTimeoutSec	360s
contentControllerPrewarmTimeoutSec	900s
contentControllerShutdownTimeoutSec	1800s
contentControllerLaunchTimeoutSec	900s

GCIS Plugin Loading

NGS initializes the GCIS plugin stack in this order:

"gcisPluginList": [
  "ElmPlugin",
  "GciPluginSysmon",
  "GcpPlugin",
  "LkmPlugin",
  "UadMLPlugin",
  "UadPlugin"
]

Initialization config:

{
  "GcisConnector": {
    "gcisJoinBusTimeoutMs": 5000,
    "gcisInitTimeoutMs": 5000,
    "gameStreamCreatingTimeoutMs": 5000,
    "lofnDataSendTimeoutMs": 3000,
    "enableGcis": true
  }
}

Volume Management

Disk space monitoring:

{
  "VolumeSetup": {
    "minDiskSpaceMB": 1024,
    "lowDiskSpaceMB": 32
  }
}

Crash Dump Management

NGS tracks crash dumps and reports them as telemetry events:

Component	File Pattern	Event	Error
NvStreamer	nvstreamer.exe.*.dmp	NGS_NvStreamerCrash	NGSE_GameStreamCrashDetected
NvStreamer (process)	nvstreamer.exe-*.dmp	NGS_NvStreamerProcessDump	NGSE_NvStreamerProcessDumpDetected
NvContainer	nvcontainer.exe.*.dmp	NGS_NvContainerCrash	NGSE_GameStreamCrashDetected
CTMT	crashdump-*.mdmp	NGS_CTMTCrash	NGSE_CTMTDumpdetected
System/BSOD	MEMORY.DMP	NGS_KernelCrash	NGSE_SystemCrashDetected
TAS	*.dmp	NGS_TasCrash	NGSE_TasDumpDetected

Log Upload

After each session, logs are uploaded to NVIDIA's backend. The uploader collects from dozens of paths:

%AG_LOGS%\*.txt
%AG_LOGS%\AutoOnboarder\**\*.*
%AG_LOGS%\GCIS\**\*.*
%AG_LOGS%\SessionMngr\*.log
C:\ProgramData\NVIDIA Corporation\NvStream\*.*
C:\ProgramData\NVIDIA Corporation\GCIS\*.*
C:\Windows\System32\winevt\Logs\Application.evtx
C:\Windows\System32\winevt\Logs\Security.evtx
C:\Windows\System32\winevt\Logs\Microsoft-Windows-Sysmon%4Operational.evtx

Log upload timeout: 600 seconds Compression: zlib_deflate (level -1 = default)

HTTP Client Settings

{
  "HttpClient": {
    "requestTimeoutMs": 60001,
    "connectTimeoutMs": 7000,
    "uploadTimeLimitSec": 500,
    "downloadTimeLimitSec": 120,
    "minBackoffMs": 150,
    "maxBackoffMs": 5000
  }
}

GameSeat2 Mode

NGS supports a new GameSeat2 architecture (Linux backend):

{
  "GameSeat2": {
    "isGameSeat2": false,
    "handlerDomainSocketPath": "/var/run/gs2/gs2-handler/handler.sock",
    "handlerConnectionTimeoutMs": 5000,
    "frontEndRebootTimeoutMs": 300000,
    "gsbeLogsDestination": "/var/run/gs2/vmlogs"
  }
}

Currently disabled (isGameSeat2: false) on the captured rig.

Telemetry

NGS sends telemetry to:

• OpenTelemetry: prod.otel.kaizen.nvidia.com:8282 (SSL)
• OpenTracing (LightStep): ls.dtrace.nvidia.com:443