alexlebens/infrastructure
1
0
Fork 0
Code Issues 1 Pull Requests 7 Actions 1 Packages Projects Releases Wiki Activity

disable extra server

Browse Source
This commit is contained in:
Alex Lebens
2025-02-28 11:50:42 -06:00
parent 6b67c2d037
commit 24a4e58054
1 changed files with 108 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files

216
clusters/cl01tl/platform/ollama/values.yaml
View File

@@ -37,80 +37,80 @@ ollama:
limits:
cpu: 4
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: 0.5.12
pullPolicy: IfNotPresent
env:
- name: OLLAMA_KEEP_ALIVE
value: 24h
- name: OLLAMA_HOST
value: 0.0.0.0
resources:
requests:
cpu: 100m
memory: 1Gi
gpu.intel.com/i915: 1
limits:
cpu: 4
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: 0.5.12
pullPolicy: IfNotPresent
env:
- name: OLLAMA_KEEP_ALIVE
value: 24h
- name: OLLAMA_HOST
value: 0.0.0.0
resources:
requests:
cpu: 100m
memory: 1Gi
gpu.intel.com/i915: 1
limits:
cpu: 4
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: 0.5.12
# pullPolicy: IfNotPresent
# env:
# - name: OLLAMA_KEEP_ALIVE
# value: 24h
# - name: OLLAMA_HOST
# value: 0.0.0.0
# resources:
# requests:
# cpu: 100m
# memory: 1Gi
# gpu.intel.com/i915: 1
# limits:
# cpu: 4
# 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: 0.5.12
# pullPolicy: IfNotPresent
# env:
# - name: OLLAMA_KEEP_ALIVE
# value: 24h
# - name: OLLAMA_HOST
# value: 0.0.0.0
# resources:
# requests:
# cpu: 100m
# memory: 1Gi
# gpu.intel.com/i915: 1
# limits:
# cpu: 4
# gpu.intel.com/i915: 1
web:
type: deployment
replicas: 1
@@ -179,20 +179,20 @@ ollama:
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
# 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:
@@ -226,26 +226,26 @@ ollama:
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
# 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