Merge pull request #55 from graphistry/cuda-worker-args-hotfix

graphistry · Aug 20, 2023 · 0a86de8 · 0a86de8
2 parents 7d646bf + fe33391
commit 0a86de8
Show file tree

Hide file tree

Showing 8 changed files with 19 additions and 13 deletions.
diff --git a/.gitignore b/.gitignore
@@ -150,4 +150,4 @@ cluster-config/cluster-config.yaml
 charts/graphistry-helm/templates/postgres/pg-sts.yaml
 
 
-chart-bundle
+charts-aux-bundled
diff --git a/charts/graphistry-helm/templates/dask/dask-cluster.yml b/charts/graphistry-helm/templates/dask/dask-cluster.yml
@@ -29,17 +29,20 @@ spec:
         image: {{.Values.global.containerregistry.name}}/{{.Values.graphistry}}:{{.Values.daskcudaworker.repository}}-{{.Values.global.tag}}-{{.Values.cuda.version}}
         imagePullPolicy: {{.Values.global.imagePullPolicy }}
         args:
-            - dask-cuda-worker
+            - dask cuda worker
             - --interface
             - eth0
             - --dashboard-address localhost:8787
-            #- --resources GPU=10
-            #- dask-scheduler:8786
+            - --resources GPU=10,PROCESS=2
+            - --nthreads 2
+            - {{.Values.daskscheduler.location | quote }}
         command:
             - /entrypoints/rapids-entrypoint.sh
         env:
           - name: GRAPHISTRY_CPU_MODE
             value: {{.Values.graphistryCPUMode | quote }}
+          - name: SKIP_SERVER
+            value: "1"
    {{- range .Values.env }}
           - name: {{ .name  }}
             value: {{ .value | quote }}

diff --git a/charts/graphistry-helm/templates/dask/dask-cuda-worker-daemonset.yaml b/charts/graphistry-helm/templates/dask/dask-cuda-worker-daemonset.yaml
@@ -41,17 +41,20 @@ spec:
         - "nexus"
       containers:
         - args:
-            - dask-cuda-worker
+            - dask cuda worker
             - --interface
             - eth0
             - --dashboard-address localhost:8787
-            #- --resources GPU=10
-            - dask-scheduler:8786
+            - --resources GPU=10,PROCESS=2
+            - --nthreads 2
+            - {{.Values.daskscheduler.location | quote }}
           command:
             - /entrypoints/rapids-entrypoint.sh
           env:
           - name: GRAPHISTRY_CPU_MODE
             value: {{.Values.graphistryCPUMode | quote }}
+          - name: SKIP_SERVER
+            value: "1"
    {{- range .Values.env }}
           - name: {{ .name  }}
             value: {{ .value | quote }}

diff --git a/charts/graphistry-helm/templates/forge-etl/forge-etl-python-daemonset.yaml b/charts/graphistry-helm/templates/forge-etl/forge-etl-python-daemonset.yaml
@@ -97,7 +97,7 @@ spec:
               command:
                 - sh
                 - -c
-                - "curl -f http://localhost:8080/cudfhealth && curl -f http://localhost:8080/blazinghealth"
+                - "curl -f http://localhost:8080/cudfhealth && curl -f http://localhost:8080/dasksqlhealth"
             failureThreshold: 3
             initialDelaySeconds: 180
             periodSeconds: 120

diff --git a/charts/graphistry-helm/templates/nginx/nginx-deployment.yaml b/charts/graphistry-helm/templates/nginx/nginx-deployment.yaml
@@ -207,7 +207,7 @@ spec:
           - name: PROMETHEUS_PORT
             value: "4040"
           {{- if eq .Values.global.containerregistry.name "docker.io/graphistry" }}
-          image: quay.io/martinhelmich/prometheus-nginxlog-exporter:v1.9.2 #DockerHub
+          image: ghcr.io/martin-helmich/prometheus-nginxlog-exporter/exporter:v1.9.2 #DockerHub
           {{ else }}
           image: {{.Values.global.containerregistry.name}}/prometheus-nginxlog-exporter:v1.9.2
           {{- end }}

diff --git a/charts/graphistry-helm/values.yaml b/charts/graphistry-helm/values.yaml
@@ -406,7 +406,7 @@ dask:
 #dask-scheduler repository name 
 daskscheduler:
   repository: etl-server-python #dask-scheduler repository name 
-  location: "dask-scheduler:8786" #dask-scheduler location
+  location: dask-scheduler:8786 #dask-scheduler location
 
 vgpu: false #enables vgpu mode for bigger than memory workloads on VGPU
 

diff --git a/charts/postgres-cluster/templates/postgres-cluster.yaml b/charts/postgres-cluster/templates/postgres-cluster.yaml
@@ -48,11 +48,11 @@ spec:
       {{- if eq .Values.global.multiNode true  }}
         accessModes:
           - "ReadWriteMany"
-        storageClassName: postgres-longhorn-{{ .Release.Namespace }}
+        storageClassName: postgres-longhorn
       {{- else }}
         accessModes:
           - "ReadWriteOnce"
-        storageClassName: retain-sc-{{ .Release.Namespace }}
+        storageClassName: retain-sc
       {{- end }}
         resources:
           requests:

diff --git a/charts/values-overrides/internal/eks-dev-values.yaml b/charts/values-overrides/internal/eks-dev-values.yaml
@@ -30,7 +30,7 @@ k8sDashboard:
 
 global:
   provisioner: ebs.csi.aws.com
-  tag: v2.40.3
+  tag: v2.40.27
   nodeSelector: {"accelerator": "nvidia"}
   #change log levels
   #logs:
Original file line number	Diff line number	Diff line change
Expand Up		@@ -150,4 +150,4 @@ cluster-config/cluster-config.yaml
		charts/graphistry-helm/templates/postgres/pg-sts.yaml


		chart-bundle
		charts-aux-bundled