--wip-- [skipci]

RareCompute · Nov 19, 2024 · 92a563d · 92a563d
1 parent f659fad
commit 92a563d
Show file tree

Hide file tree

Showing 37 changed files with 790 additions and 22 deletions.
diff --git a/.github/gitleaks.yaml → .github/workflows/gitleaks.yaml b/.github/gitleaks.yaml → .github/workflows/gitleaks.yaml
diff --git a/.github/kubeconform.yaml → .github/workflows/kubeconform.yaml b/.github/kubeconform.yaml → .github/workflows/kubeconform.yaml
diff --git a/kubernetes/arc1/apps/kube-system/nvidia-device-plugin/app/helmrelease.yaml b/kubernetes/arc1/apps/kube-system/nvidia-device-plugin/app/helmrelease.yaml
@@ -63,4 +63,4 @@ spec:
               resources:
                 - name: nvidia.com/gpu
                   replicas: 7
-      default: "single"
+      default: "default"
diff --git a/kubernetes/arc1/apps/machine-learning/colabfold/app/helmrelease.yaml b/kubernetes/arc1/apps/machine-learning/colabfold/app/helmrelease.yaml
@@ -26,7 +26,7 @@ spec:
       retries: 3
   values:
     controllers:
-      ollama:
+      colabfold:
         type: deployment
         annotations:
           reloader.stakater.com/auto: "true"
@@ -56,11 +56,11 @@ spec:
               requests:
                 cpu: 200m
                 memory: 4Gi
-                gpu.intel.com/i915: "4"
+                nvidia.com/gpu: 2
               limits:
                 cpu: 32000m
                 memory: 64Gi
-                gpu.intel.com/i915: "4"
+                nvidia.com/gpu: 4
     service:
       app:
         controller: *app
@@ -88,6 +88,7 @@ spec:
           - secretName: colabfold-tls
             hosts: [*host]
     persistence:
+      # TODO: Replace with existing PVC
       data:
         storageClass: local-nvme
         accessMode: ReadWriteMany

diff --git a/kubernetes/arc1/apps/machine-learning/kustomization.yaml b/kubernetes/arc1/apps/machine-learning/kustomization.yaml
@@ -4,5 +4,6 @@ apiVersion: kustomize.config.k8s.io/v1beta1
 kind: Kustomization
 resources:
   - ./namespace.yaml
+  - ./qlora/ks.yaml
   #- ./ollama/ks.yaml
   #- ./jupyterhub/ks.yaml
diff --git a/kubernetes/arc1/apps/machine-learning/mmseqs2/app/helmrelease.yaml b/kubernetes/arc1/apps/machine-learning/mmseqs2/app/helmrelease.yaml
@@ -5,7 +5,7 @@
 apiVersion: helm.toolkit.fluxcd.io/v2
 kind: HelmRelease
 metadata:
-  name: &app ollama
+  name: &app mmseqs2
 spec:
   interval: 30m
   chart:
@@ -26,7 +26,7 @@ spec:
       retries: 3
   values:
     controllers:
-      ollama:
+      mmseqs2:
         type: deployment
         annotations:
           reloader.stakater.com/auto: "true"
@@ -41,11 +41,11 @@ spec:
               requests:
                 cpu: 200m
                 memory: 4Gi
-                # gpu.intel.com/i915: "1"
+                nvidia.com/gpu: 2
               limits:
-                cpu: 8000m
-                memory: 8Gi
-                # gpu.intel.com/i915: "1"
+                cpu: 32000m
+                memory: 64Gi
+                nvidia.com/gpu: 4
     service:
       app:
         controller: *app
@@ -74,6 +74,7 @@ spec:
           - secretName: mmseqs2-tls
             hosts: [*host]
     persistence:
+      # TODO: Replace with existing PVC
       data:
         storageClass: local-nvme
         accessMode: ReadWriteMany

diff --git a/kubernetes/arc1/apps/machine-learning/qlora/app/helmrelease.yaml b/kubernetes/arc1/apps/machine-learning/qlora/app/helmrelease.yaml
@@ -0,0 +1,102 @@
+---
+# yaml-language-server: $schema=https://raw.githubusercontent.com/bjw-s/helm-charts/main/charts/other/app-template/schemas/helmrelease-helm-v2.schema.json
+# TODO: Finish this
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: &app qlora
+spec:
+  interval: 30m
+  chart:
+    spec:
+      chart: app-template
+      version: 3.5.1
+      sourceRef:
+        kind: HelmRepository
+        name: bjw-s
+        namespace: flux-system
+  install:
+    remediation:
+      retries: 3
+  upgrade:
+    cleanupOnFail: true
+    remediation:
+      strategy: rollback
+      retries: 3
+  values:
+    controllers:
+      qlora:
+        type: deployment
+        annotations:
+          reloader.stakater.com/auto: "true"
+        pod:
+          runtimeClassName: nvidia
+          terminationGracePeriodSeconds: 1
+          affinity:
+            nodeAffinity:
+              requiredDuringSchedulingIgnoredDuringExecution:
+                nodeSelectorTerms:
+                  - matchExpressions:
+                      - key: nvidia.com/gpu.present
+                        operator: In
+                        values:
+                          - "true"
+        containers:
+          app:
+            image:
+              repository: ghcr.io/rarecompute/qlora-docker
+              tag: main@sha256:e56350596e17af5198bfc848b0de3a5a11cb98d97e1e02dbb322467269342541
+            env:
+              TZ: ${TIMEZONE}
+              # GITHUB_REPO: https://github.com/RareCompute/example-models
+              NVIDIA_VISIBLE_DEVICES: all
+              NVIDIA_DRIVER_CAPABILITIES: all
+            securityContext:
+              capabilities.drop: ["ALL"]
+            resources:
+              requests:
+                cpu: 200m
+                memory: 8Gi
+              limits:
+                cpu: 16
+                memory: 32Gi
+                nvidia.com/gpu: 4
+            #            probes:
+            #              liveness:
+            #                enabled: true
+            #              readiness:
+            #                enabled: true
+            #              startup:
+            #                enabled: false
+            #                spec:
+            #                  failureThreshold: 30
+            #                  periodSeconds: 5
+    service:
+      app:
+        controller: *app
+        annotations:
+          teleport.dev/name: *app
+        labels:
+          teleport: enabled
+        ports:
+          http:
+            port: &port 80
+    persistence:
+      app:
+        storageClass: local-nvme
+        accessMode: ReadWriteOnce
+        size: 2Gi
+        globalMounts:
+          - path: /app
+      workspace:
+        storageClass: local-nvme
+        # TODO: OpenEBS only support ReadWriteOnce
+        accessMode: ReadWriteOnce
+        size: 2048Gi
+        retain: true
+        globalMounts:
+          - path: /workspace
+      tmp:
+        type: emptyDir
+        globalMounts:
+          - path: /tmp
diff --git a/kubernetes/arc1/apps/machine-learning/qlora/app/kustomization.yaml b/kubernetes/arc1/apps/machine-learning/qlora/app/kustomization.yaml
@@ -0,0 +1,6 @@
+---
+# yaml-language-server: $schema=https://json.schemastore.org/kustomization
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+resources:
+  - ./helmrelease.yaml
diff --git a/kubernetes/arc1/apps/machine-learning/qlora/ks.yaml b/kubernetes/arc1/apps/machine-learning/qlora/ks.yaml
@@ -0,0 +1,24 @@
+---
+# yaml-language-server: $schema=https://raw.githubusercontent.com/fluxcd-community/flux2-schemas/main/kustomization-kustomize-v1.json
+apiVersion: kustomize.toolkit.fluxcd.io/v1
+kind: Kustomization
+metadata:
+  name: &app qlora
+  namespace: flux-system
+spec:
+  targetNamespace: machine-learning
+  commonMetadata:
+    labels:
+      app.kubernetes.io/name: *app
+  path: ./kubernetes/arc1/apps/machine-learning/qlora/app
+  prune: true
+  sourceRef:
+    kind: GitRepository
+    name: k8s-gitops
+  wait: false
+  interval: 30m
+  retryInterval: 1m
+  timeout: 5m
+  postBuild:
+    substitute:
+      APP: *app
diff --git a/kubernetes/arc1/apps/network-system/external-dns/app/helmrelease.yaml b/kubernetes/arc1/apps/network-system/external-dns/app/helmrelease.yaml
@@ -0,0 +1,48 @@
+---
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: &app external-dns
+spec:
+  interval: 30m
+  chart:
+    spec:
+      chart: external-dns
+      version: 1.15.0
+      sourceRef:
+        kind: HelmRepository
+        name: external-dns
+        namespace: flux-system
+  install:
+    crds: CreateReplace
+    remediation:
+      retries: 3
+  upgrade:
+    cleanupOnFail: true
+    crds: CreateReplace
+    remediation:
+      strategy: rollback
+      retries: 3
+  values:
+    fullnameOverride: *app
+    provider: cloudflare
+    env:
+      - name: CF_API_TOKEN
+        valueFrom:
+          secretKeyRef:
+            name: external-dns-secret
+            key: api-token
+    extraArgs:
+      - --ingress-class=external
+      - --cloudflare-proxied
+      - --crd-source-apiversion=externaldns.k8s.io/v1alpha1
+      - --crd-source-kind=DNSEndpoint
+    policy: sync
+    sources: ["crd", "ingress"]
+    txtPrefix: k8s.
+    txtOwnerId: default
+    domainFilters: ["${SECRET_EXTERNAL_DOMAIN}"]
+    serviceMonitor:
+      enabled: true
+    podAnnotations:
+      secret.reloader.stakater.com/reload: external-dns-secret
diff --git a/kubernetes/arc1/apps/network-system/external-dns/app/kustomization.yaml b/kubernetes/arc1/apps/network-system/external-dns/app/kustomization.yaml
@@ -0,0 +1,6 @@
+---
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+resources:
+  - ./secret.sops.yaml
+  - ./helmrelease.yaml
diff --git a/kubernetes/arc1/apps/network-system/external-dns/app/secret.sops.yaml b/kubernetes/arc1/apps/network-system/external-dns/app/secret.sops.yaml
@@ -0,0 +1,26 @@
+apiVersion: v1
+kind: Secret
+metadata:
+  name: external-dns-secret
+stringData:
+  api-token: ENC[AES256_GCM,data:JTphCAIBIt4+al2z37lUck1Zq7hlW2W5V9/KASVQae5jwkxI03OVmQ==,iv:ZxU2FjPgVtVpP8/xmxOun3jjoxeJ3gp2vIF3ml+A04c=,tag:GjT20j6rAp0sDhaq2WAkfw==,type:str]
+sops:
+  kms: []
+  gcp_kms: []
+  azure_kv: []
+  hc_vault: []
+  age:
+    - recipient: age1ey3reuxyffqynll464r4q3tlhq5v73nxesyktr44lfez8jzxm94s0644n7
+      enc: |
+        -----BEGIN AGE ENCRYPTED FILE-----
+        YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBrNVdPRFFxekJvQlprM0ps
+        QXBNenFZR2k3bUt3dFloRkNUbFFDWldYUlc0CnpOMVg3cytnblBIOUptdHRsTGFa
+        Qk4vay9lWjAyR2gwTmUxQ0U4QVVZc1kKLS0tIHRUWGYxYSttaTMrZGFUeloxNDh5
+        T2pIZXdHNEt5azJPU2lpcVpydjlIMVUKQm2uTdBw+T1RMWS1XTZ1TACEujS7nr8e
+        VWTpDUY+LRU62sJx30evTAUitn+qTxF2jhMudLC7YZBwrQD+pPTISA==
+        -----END AGE ENCRYPTED FILE-----
+  lastmodified: "2024-11-19T02:05:15Z"
+  mac: ENC[AES256_GCM,data:xTYv3RcaPMbV9gwhrtlYXVQ++y5ozSlzorHnloOc+40kI7dc5ue9cmQufthIhsoL6ebcbDz6/cN1/rjMcBrlPxBop0nGkvyyjL/PybEEwrHwa82/fEuGkdJcJKZF76FYdOSXeJY41m7QOJlh31xOIkRmZUAZg+2N94VCeyPrJ5U=,iv:wLBRew0er+yA2EWqnnipPdUa297Ph2O4WBchuizuOVM=,tag:8Ni9BfQvcoSQ1PwJrlOlBw==,type:str]
+  pgp: []
+  encrypted_regex: ^(data|stringData)$
+  version: 3.9.1
diff --git a/kubernetes/arc1/apps/network-system/external-dns/ks.yaml b/kubernetes/arc1/apps/network-system/external-dns/ks.yaml
@@ -0,0 +1,20 @@
+---
+apiVersion: kustomize.toolkit.fluxcd.io/v1
+kind: Kustomization
+metadata:
+  name: &app external-dns
+  namespace: flux-system
+spec:
+  targetNamespace: network-system
+  commonMetadata:
+    labels:
+      app.kubernetes.io/name: *app
+  path: ./kubernetes/arc1/apps/network-system/external-dns/app
+  prune: true
+  sourceRef:
+    kind: GitRepository
+    name: k8s-gitops
+  wait: true
+  interval: 30m
+  retryInterval: 1m
+  timeout: 5m
diff --git a/kubernetes/arc1/apps/network-system/kustomization.yaml b/kubernetes/arc1/apps/network-system/kustomization.yaml
@@ -5,4 +5,5 @@ kind: Kustomization
 resources:
   - ./namespace.yaml
   - ./echo-server/ks.yaml
+  - ./external-dns/ks.yaml
   - ./k8tz/ks.yaml