Test reverting of CUDA installation on Windows staging

config/imagesets.yml

@@ -76,7 +76,6 @@ generic-worker-freebsd:
     genericWorker:
       config:
         ed25519SigningKeyLocation: /etc/generic-worker/ed25519_key
-        enableInteractive: true
         idleTimeoutSecs: 15
         shutdownMachineOnIdle: true
         shutdownMachineOnInternalError: true
@@ -96,7 +95,6 @@ generic-worker-ubuntu-24-04:
     genericWorker:
       config:
         ed25519SigningKeyLocation: /etc/generic-worker/ed25519_key
-        enableInteractive: true
         idleTimeoutSecs: 15
         shutdownMachineOnIdle: true
         shutdownMachineOnInternalError: true
@@ -118,7 +116,6 @@ generic-worker-ubuntu-24-04-arm64:
     genericWorker:
       config:
         ed25519SigningKeyLocation: /etc/generic-worker/ed25519_key
-        enableInteractive: true
         idleTimeoutSecs: 15
         shutdownMachineOnIdle: true
         shutdownMachineOnInternalError: true
@@ -134,10 +131,8 @@ generic-worker-ubuntu-24-04-staging:
     genericWorker:
       config:
         ed25519SigningKeyLocation: /etc/generic-worker/ed25519_key
-        enableInteractive: true
         idleTimeoutSecs: 15
         shutdownMachineOnIdle: true
-        shutdownMachineOnInternalError: false
         workerTypeMetadata:
           machine-setup:
             maintainer: [email protected]
@@ -198,7 +193,6 @@ generic-worker-win2022-staging:
         idleTimeoutSecs: 15
         livelogExecutable: C:\generic-worker\livelog.exe
         shutdownMachineOnIdle: true
-        shutdownMachineOnInternalError: true
         taskclusterProxyExecutable: C:\generic-worker\taskcluster-proxy.exe
         workerTypeMetadata:
           machine-setup:

config/projects/bugbug.yml

@@ -15,7 +15,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
     batch:
       owner: [email protected]
@@ -30,7 +29,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
     compute-smaller:
       owner: [email protected]
@@ -45,7 +43,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
     compute-small:
       owner: [email protected]
@@ -60,7 +57,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
     compute-large:
       owner: [email protected]
@@ -75,7 +71,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
     compute-super-large:
       owner: [email protected]
@@ -90,7 +85,6 @@ bugbug:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
             maxTaskRunTime: 87500
   secrets:
     bugbug/deploy: true

config/projects/mozci.yml

@@ -38,10 +38,6 @@ mozci:
       cloud: gcp
       minCapacity: 0
       maxCapacity: 5
-      workerConfig:
-        genericWorker:
-          config:
-            enableInteractive: true
   secrets:
     testing: true
     production: true

config/projects/relman.yml

@@ -62,10 +62,6 @@ relman:
       cloud: gcp
       minCapacity: 0
       maxCapacity: 50
-      workerConfig:
-        genericWorker:
-          config:
-            enableInteractive: true
 
   secrets:
     bugzilla-dashboard-backend/deploy-production: true

config/projects/taskcluster.yml

@@ -90,9 +90,11 @@ taskcluster:
       workerConfig:
         genericWorker:
           config:
-            enableInteractive: true
-            shutdownMachineOnInternalError: false
+            # this pool isn't in regular use, and when we use it, it is
+            # typically for testing bare metal stuff, so nice to have
+            # longer timeout
             idleTimeoutSecs: 3600
+            shutdownMachineOnInternalError: false
       # Use c5.metal to test kvm
       instanceTypes:
         m5d.metal: 1
@@ -104,11 +106,6 @@ taskcluster:
       cloud: gcp
       minCapacity: 0
       maxCapacity: 50
-      workerConfig:
-        genericWorker:
-          config:
-            enableInteractive: true
-            shutdownMachineOnInternalError: false
 
     gw-ubuntu-24-04-arm64:
       owner: [email protected]
@@ -117,10 +114,6 @@ taskcluster:
       cloud: gcp
       minCapacity: 0
       maxCapacity: 5
-      workerConfig:
-        genericWorker:
-          config:
-            enableInteractive: true
       machineType: "zones/{zone}/machineTypes/t2a-standard-4"
 
     gw-ubuntu-staging-aws:
@@ -136,7 +129,7 @@ taskcluster:
             # While iterating on the image building process for this worker
             # pool, useful for workers not to die immediately...
             idleTimeoutSecs: 3600
-            enableInteractive: true
+            shutdownMachineOnInternalError: false
 
     gw-ubuntu-staging-google:
       owner: [email protected]
@@ -159,7 +152,7 @@ taskcluster:
             # While iterating on the image building process for this worker
             # pool, useful for workers not to die immediately...
             idleTimeoutSecs: 3600
-            enableInteractive: true
+            shutdownMachineOnInternalError: false
 
     gw-windows-2022:
       owner: [email protected]
@@ -176,12 +169,16 @@ taskcluster:
       cloud: azure
       minCapacity: 0
       maxCapacity: 10
+      # Currently staging uses a GPU pool so we can also test GPU related changes
+      vmSizes:
+        Standard_NV12s_v3: 1
       workerConfig:
         genericWorker:
           config:
             # While iterating on the image building process for this worker
             # pool, useful for workers not to die immediately...
             idleTimeoutSecs: 3600
+            shutdownMachineOnInternalError: false
 
     gw-windows-2022-gpu:
       owner: [email protected]

generate/workers.py

@@ -588,6 +588,7 @@ def generic_worker(wp, **cfg):
                 "genericWorker": {
                     "config": {
                         "enableD2G": True,
+                        "enableInteractive": True,
                         "idleTimeoutSecs": 600,
                         "wstAudience": "communitytc",
                         "wstServerURL": "https://community-websocktunnel.services.mozilla.com",

imagesets/generic-worker-win2022-staging/azure_base_instance_type

@@ -1 +1 @@
-../generic-worker-win2022/azure_base_instance_type
+../generic-worker-win2022-gpu/azure_base_instance_type

imagesets/generic-worker-win2022-staging/bootstrap.ps1

@@ -225,8 +225,11 @@ if ($hasNvidiaGpu) {
   Start-Process "C:\nvidia_driver.exe" -ArgumentList "-s", "-noreboot" -Wait -NoNewWindow -RedirectStandardOutput "C:\nvidia-install-stdout.txt" -RedirectStandardError "C:\nvidia-install-stderr.txt"
   # install CUDA
   # https://github.com/taskcluster/community-tc-config/issues/713
-  $client.DownloadFile("https://developer.download.nvidia.com/compute/cuda/12.6.1/local_installers/cuda_12.6.1_560.94_windows.exe", "C:\cuda_installer.exe")
-  Start-Process "C:\cuda_installer.exe" -ArgumentList "-s", "-noreboot" -Wait -NoNewWindow -RedirectStandardOutput "C:\cuda-install-stdout.txt" -RedirectStandardError "C:\cuda-install-stderr.txt"
+
+  # Test removing this in staging to see if it fixes things...
+  # $client.DownloadFile("https://developer.download.nvidia.com/compute/cuda/12.6.1/local_installers/cuda_12.6.1_560.94_windows.exe", "C:\cuda_installer.exe")
+  # Start-Process "C:\cuda_installer.exe" -ArgumentList "-s", "-noreboot" -Wait -NoNewWindow -RedirectStandardOutput "C:\cuda-install-stdout.txt" -RedirectStandardError "C:\cuda-install-stderr.txt"
+
 }
 
 # now shutdown, in preparation for creating an image