265 lines
7.2 KiB
YAML
265 lines
7.2 KiB
YAML
ollama:
|
|
controllers:
|
|
server-1:
|
|
type: deployment
|
|
replicas: 1
|
|
strategy: Recreate
|
|
revisionHistoryLimit: 3
|
|
pod:
|
|
labels:
|
|
ollama-type: server
|
|
affinity:
|
|
podAntiAffinity:
|
|
requiredDuringSchedulingIgnoredDuringExecution:
|
|
- labelSelector:
|
|
matchExpressions:
|
|
- key: ollama-type
|
|
operator: In
|
|
values:
|
|
- server
|
|
topologyKey: kubernetes.io/hostname
|
|
containers:
|
|
main:
|
|
image:
|
|
repository: ollama/ollama
|
|
tag: 0.9.6
|
|
pullPolicy: IfNotPresent
|
|
env:
|
|
- name: OLLAMA_KEEP_ALIVE
|
|
value: 24h
|
|
- name: OLLAMA_HOST
|
|
value: 0.0.0.0
|
|
resources:
|
|
limits:
|
|
gpu.intel.com/i915: 1
|
|
requests:
|
|
cpu: 100m
|
|
memory: 1Gi
|
|
gpu.intel.com/i915: 1
|
|
server-2:
|
|
type: deployment
|
|
replicas: 1
|
|
strategy: Recreate
|
|
revisionHistoryLimit: 3
|
|
pod:
|
|
labels:
|
|
ollama-type: server
|
|
affinity:
|
|
podAntiAffinity:
|
|
requiredDuringSchedulingIgnoredDuringExecution:
|
|
- labelSelector:
|
|
matchExpressions:
|
|
- key: ollama-type
|
|
operator: In
|
|
values:
|
|
- server
|
|
topologyKey: kubernetes.io/hostname
|
|
containers:
|
|
main:
|
|
image:
|
|
repository: ollama/ollama
|
|
tag: 0.9.6
|
|
pullPolicy: IfNotPresent
|
|
env:
|
|
- name: OLLAMA_KEEP_ALIVE
|
|
value: 24h
|
|
- name: OLLAMA_HOST
|
|
value: 0.0.0.0
|
|
resources:
|
|
limits:
|
|
gpu.intel.com/i915: 1
|
|
requests:
|
|
cpu: 100m
|
|
memory: 1Gi
|
|
gpu.intel.com/i915: 1
|
|
server-3:
|
|
type: deployment
|
|
replicas: 1
|
|
strategy: Recreate
|
|
revisionHistoryLimit: 3
|
|
pod:
|
|
labels:
|
|
ollama-type: server
|
|
affinity:
|
|
podAntiAffinity:
|
|
requiredDuringSchedulingIgnoredDuringExecution:
|
|
- labelSelector:
|
|
matchExpressions:
|
|
- key: ollama-type
|
|
operator: In
|
|
values:
|
|
- server
|
|
topologyKey: kubernetes.io/hostname
|
|
containers:
|
|
main:
|
|
image:
|
|
repository: ollama/ollama
|
|
tag: 0.9.6
|
|
pullPolicy: IfNotPresent
|
|
env:
|
|
- name: OLLAMA_KEEP_ALIVE
|
|
value: 24h
|
|
- name: OLLAMA_HOST
|
|
value: 0.0.0.0
|
|
resources:
|
|
limits:
|
|
gpu.intel.com/i915: 1
|
|
requests:
|
|
cpu: 100m
|
|
memory: 1Gi
|
|
gpu.intel.com/i915: 1
|
|
web:
|
|
type: deployment
|
|
replicas: 1
|
|
strategy: Recreate
|
|
revisionHistoryLimit: 3
|
|
containers:
|
|
main:
|
|
image:
|
|
repository: ghcr.io/open-webui/open-webui
|
|
tag: v0.6.18
|
|
pullPolicy: IfNotPresent
|
|
env:
|
|
- name: ENV
|
|
value: prod
|
|
- name: WEBUI_AUTH
|
|
value: true
|
|
- name: WEBUI_NAME
|
|
value: Ollama
|
|
- name: WEBUI_URL
|
|
value: http://ollama.alexlebens.net
|
|
- name: ENABLE_LOGIN_FORM
|
|
value: false
|
|
- name: DEFAULT_USER_ROLE
|
|
value: admin
|
|
- name: WEBUI_SECRET_KEY
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: ollama-key-secret
|
|
key: key
|
|
- name: DATABASE_URL
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: ollama-web-postgresql-17-cluster-app
|
|
key: uri
|
|
- name: OLLAMA_BASE_URL
|
|
value: http://ollama-server-1.ollama:11434
|
|
- name: ENABLE_OAUTH_SIGNUP
|
|
value: true
|
|
- name: OAUTH_USERNAME_CLAIM
|
|
value: preferred_username
|
|
- name: OAUTH_CLIENT_SECRET
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: ollama-oidc-secret
|
|
key: secret
|
|
- name: OAUTH_CLIENT_ID
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: ollama-oidc-secret
|
|
key: client
|
|
- name: OAUTH_PROVIDER_NAME
|
|
value: Authentik
|
|
- name: OPENID_PROVIDER_URL
|
|
value: https://authentik.alexlebens.net/application/o/ollama/.well-known/openid-configuration
|
|
resources:
|
|
requests:
|
|
cpu: 10m
|
|
memory: 1Gi
|
|
service:
|
|
server-1:
|
|
controller: server-1
|
|
ports:
|
|
http:
|
|
port: 11434
|
|
targetPort: 11434
|
|
protocol: HTTP
|
|
server-2:
|
|
controller: server-2
|
|
ports:
|
|
http:
|
|
port: 11434
|
|
targetPort: 11434
|
|
protocol: HTTP
|
|
server-3:
|
|
controller: server-3
|
|
ports:
|
|
http:
|
|
port: 11434
|
|
targetPort: 11434
|
|
protocol: HTTP
|
|
web:
|
|
controller: web
|
|
ports:
|
|
http:
|
|
port: 80
|
|
targetPort: 8080
|
|
protocol: HTTP
|
|
persistence:
|
|
server-1:
|
|
storageClass: ceph-block
|
|
accessMode: ReadWriteOnce
|
|
size: 40Gi
|
|
retain: true
|
|
advancedMounts:
|
|
server-1:
|
|
main:
|
|
- path: /root/.ollama
|
|
readOnly: false
|
|
server-2:
|
|
storageClass: ceph-block
|
|
accessMode: ReadWriteOnce
|
|
size: 40Gi
|
|
retain: true
|
|
advancedMounts:
|
|
server-2:
|
|
main:
|
|
- path: /root/.ollama
|
|
readOnly: false
|
|
server-3:
|
|
storageClass: ceph-block
|
|
accessMode: ReadWriteOnce
|
|
size: 40Gi
|
|
retain: true
|
|
advancedMounts:
|
|
server-3:
|
|
main:
|
|
- path: /root/.ollama
|
|
readOnly: false
|
|
web-data:
|
|
storageClass: ceph-block
|
|
accessMode: ReadWriteOnce
|
|
size: 5Gi
|
|
retain: true
|
|
advancedMounts:
|
|
web:
|
|
main:
|
|
- path: /app/backend/data
|
|
readOnly: false
|
|
postgres-17-cluster:
|
|
nameOverride: ollama-web-postgresql-17
|
|
mode: standalone
|
|
cluster:
|
|
storage:
|
|
storageClass: local-path
|
|
walStorage:
|
|
storageClass: local-path
|
|
monitoring:
|
|
enabled: true
|
|
prometheusRule:
|
|
enabled: true
|
|
recovery:
|
|
method: objectStore
|
|
objectStore:
|
|
endpointURL: https://nyc3.digitaloceanspaces.com
|
|
destinationPath: s3://postgres-backups-ce540ddf106d186bbddca68a/cl01tl/ollama/ollama-web-postgresql-17-cluster
|
|
endpointCredentials: ollama-web-postgresql-17-cluster-backup-secret
|
|
recoveryIndex: 2
|
|
backup:
|
|
enabled: true
|
|
endpointURL: https://nyc3.digitaloceanspaces.com
|
|
destinationPath: s3://postgres-backups-ce540ddf106d186bbddca68a/cl01tl/ollama/ollama-web-postgresql-17-cluster
|
|
endpointCredentials: ollama-web-postgresql-17-cluster-backup-secret
|
|
backupIndex: 2
|
|
retentionPolicy: "7d"
|