Skip to content

Commit 034aba9

Browse files
committed
feat: reorg builds/deploys
1 parent f6d3623 commit 034aba9

File tree

18 files changed

+76
-117
lines changed

18 files changed

+76
-117
lines changed

.ko.yaml

Lines changed: 5 additions & 73 deletions
Original file line numberDiff line numberDiff line change
@@ -154,87 +154,19 @@ builds:
154154
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
155155
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"
156156

157-
- id: janitor-provider-aws
158-
dir: providers/aws
157+
- id: janitor-provider
158+
dir: janitor-provider
159159
main: .
160160
ldflags:
161161
- "-s -w"
162162
- "-X main.version={{.Env.VERSION}} -X main.commit={{.Env.GIT_COMMIT}} -X main.date={{.Env.BUILD_DATE}}"
163163
annotations:
164-
org.opencontainers.image.description: "AWS provider for the NVSentinel Janitor"
164+
org.opencontainers.image.description: "Example provider for NVSentinel Janitor"
165165
labels:
166166
org.opencontainers.image.source: "https://github.com/nvidia/nvsentinel"
167167
org.opencontainers.image.licenses: "Apache-2.0"
168-
org.opencontainers.image.title: "janitor-provider-aws"
169-
org.opencontainers.image.description: "AWS implementation of the CSP provider interface for Janitor"
170-
org.opencontainers.image.version: "{{.Env.VERSION}}"
171-
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
172-
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"
173-
174-
- id: janitor-provider-azure
175-
dir: providers/azure
176-
main: .
177-
ldflags:
178-
- "-s -w"
179-
- "-X main.version={{.Env.VERSION}} -X main.commit={{.Env.GIT_COMMIT}} -X main.date={{.Env.BUILD_DATE}}"
180-
annotations:
181-
org.opencontainers.image.description: "Azure provider for the NVSentinel Janitor"
182-
labels:
183-
org.opencontainers.image.source: "https://github.com/nvidia/nvsentinel"
184-
org.opencontainers.image.licenses: "Apache-2.0"
185-
org.opencontainers.image.title: "janitor-provider-azure"
186-
org.opencontainers.image.description: "Azure implementation of the CSP provider interface for Janitor"
187-
org.opencontainers.image.version: "{{.Env.VERSION}}"
188-
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
189-
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"
190-
191-
- id: janitor-provider-gcp
192-
dir: providers/gcp
193-
main: .
194-
ldflags:
195-
- "-s -w"
196-
- "-X main.version={{.Env.VERSION}} -X main.commit={{.Env.GIT_COMMIT}} -X main.date={{.Env.BUILD_DATE}}"
197-
annotations:
198-
org.opencontainers.image.description: "GCP provider for the NVSentinel Janitor"
199-
labels:
200-
org.opencontainers.image.source: "https://github.com/nvidia/nvsentinel"
201-
org.opencontainers.image.licenses: "Apache-2.0"
202-
org.opencontainers.image.title: "janitor-provider-gcp"
203-
org.opencontainers.image.description: "GCP implementation of the CSP provider interface for Janitor"
204-
org.opencontainers.image.version: "{{.Env.VERSION}}"
205-
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
206-
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"
207-
208-
- id: janitor-provider-kwok
209-
dir: providers/kwok
210-
main: .
211-
ldflags:
212-
- "-s -w"
213-
- "-X main.version={{.Env.VERSION}} -X main.commit={{.Env.GIT_COMMIT}} -X main.date={{.Env.BUILD_DATE}}"
214-
annotations:
215-
org.opencontainers.image.description: "KWOK provider for the NVSentinel Janitor"
216-
labels:
217-
org.opencontainers.image.source: "https://github.com/nvidia/nvsentinel"
218-
org.opencontainers.image.licenses: "Apache-2.0"
219-
org.opencontainers.image.title: "janitor-provider-kwok"
220-
org.opencontainers.image.description: "KWOK implementation of the CSP provider interface for Janitor"
221-
org.opencontainers.image.version: "{{.Env.VERSION}}"
222-
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
223-
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"
224-
225-
- id: janitor-provider-oci
226-
dir: providers/oci
227-
main: .
228-
ldflags:
229-
- "-s -w"
230-
- "-X main.version={{.Env.VERSION}} -X main.commit={{.Env.GIT_COMMIT}} -X main.date={{.Env.BUILD_DATE}}"
231-
annotations:
232-
org.opencontainers.image.description: "OCI provider for the NVSentinel Janitor"
233-
labels:
234-
org.opencontainers.image.source: "https://github.com/nvidia/nvsentinel"
235-
org.opencontainers.image.licenses: "Apache-2.0"
236-
org.opencontainers.image.title: "janitor-provider-oci"
237-
org.opencontainers.image.description: "OCI implementation of the CSP provider interface for Janitor"
168+
org.opencontainers.image.title: "NVSentinel Janitor-Provider"
169+
org.opencontainers.image.description: "Example implementation of the CSP provider interface for Janitor"
238170
org.opencontainers.image.version: "{{.Env.VERSION}}"
239171
org.opencontainers.image.revision: "{{.Env.GIT_COMMIT}}"
240172
org.opencontainers.image.created: "{{.Env.BUILD_DATE}}"

distros/kubernetes/nvsentinel/Chart.yaml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -55,9 +55,9 @@ dependencies:
5555
- name: janitor
5656
version: "0.1.0"
5757
condition: global.janitor.enabled
58-
- name: janitor-provider-kwok
58+
- name: janitor-provider
5959
version: "0.1.0"
60-
condition: global.janitor.enabled && global.janitor.provider == "kwok"
60+
condition: global.janitorProvider.enabled
6161
- name: metadata-collector
6262
version: "0.1.0"
6363
condition: global.metadataCollector.enabled

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/Chart.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/Chart.yaml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@
1313
# limitations under the License.
1414

1515
apiVersion: v2
16-
name: janitor-provider-kwok
17-
description: KWOK provider for NVSentinel Janitor
16+
name: janitor-provider
17+
description: Example provider for NVSentinel Janitor
1818
type: application
1919
version: 0.1.0

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/templates/_helpers.tpl renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/templates/_helpers.tpl

File renamed without changes.

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/templates/deployment.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/templates/deployment.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -37,7 +37,7 @@ spec:
3737
{{- end }}
3838
serviceAccountName: {{ include "provider.serviceAccountName" . }}
3939
containers:
40-
- name: janitor-provider-kwok
40+
- name: janitor-provider
4141
image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default ((.Values.global).image).tag | default .Chart.AppVersion }}"
4242
imagePullPolicy: {{ .Values.image.pullPolicy }}
4343
resources:

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/templates/networkpolicy.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/templates/networkpolicy.yaml

File renamed without changes.

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/templates/service.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/templates/service.yaml

File renamed without changes.

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/templates/serviceaccount.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/templates/serviceaccount.yaml

File renamed without changes.

distros/kubernetes/nvsentinel/charts/janitor-provider-kwok/values.yaml renamed to distros/kubernetes/nvsentinel/charts/janitor-provider/values.yaml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -16,12 +16,12 @@ replicaCount: 1
1616

1717
# Image configuration
1818
image:
19-
repository: ghcr.io/nvidia/nvsentinel/janitor-provider-kwok
19+
repository: ghcr.io/nvidia/nvsentinel/janitor-provider
2020
pullPolicy: IfNotPresent
2121
tag: ""
2222

2323
nameOverride: ""
24-
fullnameOverride: "janitor-provider-kwok"
24+
fullnameOverride: "janitor-provider"
2525

2626
serviceAccount:
2727
# Specifies whether a service account should be created

distros/kubernetes/nvsentinel/values-tilt.yaml

Lines changed: 23 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -76,7 +76,9 @@ global:
7676

7777
janitor:
7878
enabled: true
79-
provider: "kwok"
79+
80+
janitorProvider:
81+
enabled: true
8082

8183
mongodbStore:
8284
enabled: true
@@ -281,7 +283,7 @@ health-events-analyzer:
281283

282284
janitor:
283285
config:
284-
cspProviderHost: "janitor-provider-kwok.nvsentinel.svc.cluster.local:50051"
286+
cspProviderHost: "janitor-provider.nvsentinel.svc.cluster.local:50051"
285287
webhook:
286288
certIssuer: "janitor-selfsigned-issuer"
287289

@@ -293,7 +295,7 @@ labeler:
293295

294296
csp-health-monitor:
295297
logLevel: debug
296-
cspName: "gcp" # this will be updated to aws as well dynamically during tests
298+
cspName: "gcp" # this will be updated to aws as well dynamically during tests
297299
quarantineTriggerEngine:
298300
logLevel: debug
299301
configToml:
@@ -304,17 +306,17 @@ csp-health-monitor:
304306
endpointOverride: "csp-api-mock.nvsentinel.svc.cluster.local:50051"
305307
aws:
306308
endpointOverride: "http://csp-api-mock.nvsentinel.svc.cluster.local:8080/aws/health"
307-
309+
308310
affinity:
309311
podAntiAffinity:
310312
requiredDuringSchedulingIgnoredDuringExecution:
311-
- labelSelector:
312-
matchLabels:
313-
app.kubernetes.io/name: kwok
314-
namespaceSelector:
315-
matchLabels:
316-
kubernetes.io/metadata.name: kube-system
317-
topologyKey: kubernetes.io/hostname
313+
- labelSelector:
314+
matchLabels:
315+
app.kubernetes.io/name: kwok
316+
namespaceSelector:
317+
matchLabels:
318+
kubernetes.io/metadata.name: kube-system
319+
topologyKey: kubernetes.io/hostname
318320

319321
# Enable node metadata enrichment for testing
320322
platformConnector:
@@ -323,32 +325,32 @@ platformConnector:
323325
maxNodeConditionMessageLength: 1024
324326
qps: 5.0
325327
burst: 10
326-
328+
327329
pipeline:
328330
- name: MetadataAugmentor
329331
enabled: true
330332
config: /etc/config/metadata.toml
331-
333+
332334
- name: OverrideTransformer
333335
enabled: true
334336
config: /etc/config/overrides.toml
335-
337+
336338
transformers:
337339
MetadataAugmentor:
338340
cacheSize: 50
339341
cacheTTLSeconds: 3600
340342
allowedLabels:
341343
- "topology.kubernetes.io/zone"
342344
- "topology.kubernetes.io/region"
343-
345+
344346
OverrideTransformer:
345347
rules:
346-
- name: "escalate-xid-94"
347-
when: 'event.agent == "syslog-health-monitor" && "94" in event.errorCode'
348-
override:
349-
isFatal: true
350-
recommendedAction: "CONTACT_SUPPORT"
351-
348+
- name: "escalate-xid-94"
349+
when: 'event.agent == "syslog-health-monitor" && "94" in event.errorCode'
350+
override:
351+
isFatal: true
352+
recommendedAction: "CONTACT_SUPPORT"
353+
352354
# - name: "zone-specific-override"
353355
# when: 'event.metadata["topology.kubernetes.io/zone"] == "us-west1-a" && event.componentClass == "GPU"'
354356
# override:

0 commit comments

Comments
 (0)