Compare commits

..

1 Commits

Author SHA1 Message Date
c1769821d3 quality-gate: publish workspace coverage and LOC gauges 2026-04-17 05:41:01 -03:00
299 changed files with 7892 additions and 29334 deletions

374
Jenkinsfile vendored
View File

@ -7,52 +7,14 @@ pipeline {
apiVersion: v1
kind: Pod
spec:
serviceAccountName: "jenkins"
nodeSelector:
hardware: rpi5
kubernetes.io/arch: arm64
node-role.kubernetes.io/worker: "true"
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-04
- titan-06
- titan-11
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
labelSelector:
matchLabels:
jenkins/jenkins-jenkins-agent: "true"
containers:
- name: jnlp
image: jenkins/inbound-agent:3355.v388858a_47b_33-2-jdk21
resources:
requests:
cpu: "25m"
memory: "256Mi"
- name: python
image: registry.bstein.dev/bstein/python:3.12-slim
command:
- cat
tty: true
- name: quality-tools
image: registry.bstein.dev/bstein/quality-tools:sonar8.0.1-trivy0.70.0-db20260422-arm64
image: python:3.12-slim
command:
- cat
tty: true
@ -62,21 +24,9 @@ spec:
environment {
PIP_DISABLE_PIP_VERSION_CHECK = '1'
PYTHONUNBUFFERED = '1'
SUITE_NAME = 'titan_iac'
SUITE_NAME = 'titan-iac'
PUSHGATEWAY_URL = 'http://platform-quality-gateway.monitoring.svc.cluster.local:9091'
SONARQUBE_HOST_URL = 'http://sonarqube.quality.svc.cluster.local:9000'
SONARQUBE_PROJECT_KEY = 'titan_iac'
SONARQUBE_TOKEN = credentials('sonarqube-token')
VM_URL = 'http://victoria-metrics-single-server.monitoring.svc.cluster.local:8428'
QUALITY_GATE_SONARQUBE_ENFORCE = '0'
QUALITY_GATE_SONARQUBE_REPORT = 'build/sonarqube-quality-gate.json'
QUALITY_GATE_IRONBANK_ENFORCE = '1'
QUALITY_GATE_IRONBANK_REQUIRED = '0'
QUALITY_GATE_IRONBANK_REPORT = 'build/ironbank-compliance.json'
}
options {
disableConcurrentBuilds()
buildDiscarder(logRotator(daysToKeepStr: '30', numToKeepStr: '200', artifactDaysToKeepStr: '30', artifactNumToKeepStr: '120'))
}
stages {
stage('Checkout') {
@ -86,175 +36,7 @@ spec:
}
stage('Install deps') {
steps {
sh '''
set -eu
if ! command -v git >/dev/null 2>&1; then
apt-get update
apt-get install -y --no-install-recommends git ca-certificates
rm -rf /var/lib/apt/lists/*
fi
pip install --no-cache-dir -r ci/requirements.txt
'''
}
}
stage('Prepare local quality evidence') {
steps {
sh '''
set -eu
mkdir -p build
set +e
python3 -m testing.quality_gate --profile local --build-dir build
local_quality_rc=$?
set -e
printf '%s\n' "${local_quality_rc}" > build/local-quality-gate.rc
'''
}
}
stage('Collect SonarQube evidence') {
steps {
container('quality-tools') {
sh '''#!/usr/bin/env bash
set -euo pipefail
mkdir -p build
args=(
"-Dsonar.host.url=${SONARQUBE_HOST_URL}"
"-Dsonar.login=${SONARQUBE_TOKEN}"
"-Dsonar.projectKey=${SONARQUBE_PROJECT_KEY}"
"-Dsonar.projectName=${SONARQUBE_PROJECT_KEY}"
"-Dsonar.sources=."
"-Dsonar.exclusions=**/.git/**,**/build/**,**/dist/**,**/node_modules/**,**/.venv/**,**/__pycache__/**,**/coverage/**,**/test-results/**,**/playwright-report/**,services/monitoring/dashboards/**,services/monitoring/grafana-dashboard-*.yaml,services/game-stream/**"
"-Dsonar.test.inclusions=**/tests/**,**/testing/**,**/*_test.go,**/*.test.ts,**/*.test.tsx,**/*.spec.ts,**/*.spec.tsx"
)
[ -f build/coverage-unit.xml ] && args+=("-Dsonar.python.coverage.reportPaths=build/coverage-unit.xml")
set +e
sonar-scanner "${args[@]}" | tee build/sonar-scanner.log
rc=${PIPESTATUS[0]}
set -e
printf '%s\n' "${rc}" > build/sonarqube-analysis.rc
'''
}
sh '''
set -eu
mkdir -p build
python3 - <<'PY'
import base64
import json
import os
import time
import urllib.parse
import urllib.request
from pathlib import Path
host = os.getenv('SONARQUBE_HOST_URL', '').strip().rstrip('/')
project_key = os.getenv('SONARQUBE_PROJECT_KEY', '').strip()
token = os.getenv('SONARQUBE_TOKEN', '').strip()
report_path = os.getenv('QUALITY_GATE_SONARQUBE_REPORT', 'build/sonarqube-quality-gate.json')
payload = {
"status": "ERROR",
"note": "missing SONARQUBE_HOST_URL and/or SONARQUBE_PROJECT_KEY",
}
if host and project_key:
task_file = Path('.scannerwork/report-task.txt')
task_id = ''
if task_file.exists():
for line in task_file.read_text(encoding='utf-8').splitlines():
key, _, value = line.partition('=')
if key == 'ceTaskId':
task_id = value.strip()
break
if task_id:
ce_query = urllib.parse.urlencode({"id": task_id})
deadline = time.monotonic() + 180
while time.monotonic() < deadline:
ce_request = urllib.request.Request(f"{host}/api/ce/task?{ce_query}", method="GET")
if token:
encoded = base64.b64encode(f"{token}:".encode("utf-8")).decode("utf-8")
ce_request.add_header("Authorization", f"Basic {encoded}")
try:
with urllib.request.urlopen(ce_request, timeout=12) as response:
ce_payload = json.loads(response.read().decode("utf-8"))
except Exception:
time.sleep(3)
continue
status = str(ce_payload.get("task", {}).get("status", "")).upper()
if status in {"SUCCESS", "FAILED", "CANCELED"}:
break
time.sleep(3)
query = urllib.parse.urlencode({"projectKey": project_key})
request = urllib.request.Request(
f"{host}/api/qualitygates/project_status?{query}",
method="GET",
)
if token:
encoded = base64.b64encode(f"{token}:".encode("utf-8")).decode("utf-8")
request.add_header("Authorization", f"Basic {encoded}")
try:
with urllib.request.urlopen(request, timeout=12) as response:
payload = json.loads(response.read().decode("utf-8"))
except Exception as exc: # noqa: BLE001
payload = {"status": "ERROR", "error": str(exc)}
with open(report_path, "w", encoding="utf-8") as handle:
json.dump(payload, handle, indent=2, sort_keys=True)
handle.write("\\n")
PY
'''
}
}
stage('Collect IronBank evidence') {
steps {
container('quality-tools') {
sh '''#!/usr/bin/env bash
set -euo pipefail
mkdir -p build
set +e
trivy fs --cache-dir "${TRIVY_CACHE_DIR}" --skip-db-update --skip-files clusters/atlas/flux-system/gotk-components.yaml --timeout 5m --no-progress --format json --output build/trivy-fs.json --scanners vuln,secret,misconfig --severity HIGH,CRITICAL .
trivy_rc=$?
set -e
if [ ! -s build/trivy-fs.json ]; then
cat > build/ironbank-compliance.json <<EOF
{"status":"failed","compliant":false,"scanner":"trivy","scan_type":"filesystem","error":"trivy did not produce JSON output","trivy_rc":${trivy_rc}}
EOF
exit 0
fi
'''
}
sh '''
set -eu
mkdir -p build
if [ -s build/trivy-fs.json ]; then
python3 ci/scripts/supply_chain_report.py --trivy-json build/trivy-fs.json --waivers ci/titan-iac-trivy-waivers.json --output build/ironbank-compliance.json
exit 0
fi
python3 - <<'PY'
import json
import os
from pathlib import Path
report_path = Path(os.getenv('QUALITY_GATE_IRONBANK_REPORT', 'build/ironbank-compliance.json'))
if report_path.exists():
raise SystemExit(0)
status = os.getenv('IRONBANK_COMPLIANCE_STATUS', '').strip()
compliant = os.getenv('IRONBANK_COMPLIANT', '').strip().lower()
payload = {
"status": status or "unknown",
"compliant": compliant in {"1", "true", "yes", "on"} if compliant else None,
}
payload = {k: v for k, v in payload.items() if v is not None}
if "status" not in payload:
payload["status"] = "unknown"
payload["note"] = (
"Set IRONBANK_COMPLIANCE_STATUS/IRONBANK_COMPLIANT "
"or write build/ironbank-compliance.json in image-building repos."
)
report_path.parent.mkdir(parents=True, exist_ok=True)
report_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\\n", encoding="utf-8")
PY
'''
sh 'pip install --no-cache-dir -r ci/requirements.txt'
}
}
stage('Run quality gate') {
@ -284,96 +66,8 @@ PY
stage('Enforce quality gate') {
steps {
sh '''
set -euo pipefail
gate_rc="$(cat build/quality-gate.rc 2>/dev/null || echo 1)"
fail=0
if [ "${gate_rc}" -ne 0 ]; then
echo "quality gate failed with rc=${gate_rc}" >&2
fail=1
fi
enabled() {
case "$(printf '%s' "${1:-}" | tr '[:upper:]' '[:lower:]')" in
1|true|yes|on) return 0 ;;
*) return 1 ;;
esac
}
if enabled "${QUALITY_GATE_SONARQUBE_ENFORCE:-1}"; then
sonar_status="$(python3 - <<'PY'
import json
from pathlib import Path
path = Path("build/sonarqube-quality-gate.json")
if not path.exists():
print("missing")
raise SystemExit(0)
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except Exception: # noqa: BLE001
print("error")
raise SystemExit(0)
status = (payload.get("status") or payload.get("projectStatus", {}).get("status") or payload.get("qualityGate", {}).get("status") or "").strip().lower()
print(status or "missing")
PY
)"
case "${sonar_status}" in
ok|pass|passed|success) ;;
*)
echo "sonarqube gate failed: ${sonar_status}" >&2
fail=1
;;
esac
fi
ironbank_required="${QUALITY_GATE_IRONBANK_REQUIRED:-0}"
if [ "${PUBLISH_IMAGES:-false}" = "true" ]; then
ironbank_required=1
fi
if enabled "${QUALITY_GATE_IRONBANK_ENFORCE:-1}"; then
supply_status="$(python3 - <<'PY'
import json
from pathlib import Path
path = Path("build/ironbank-compliance.json")
if not path.exists():
print("missing")
raise SystemExit(0)
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except Exception: # noqa: BLE001
print("error")
raise SystemExit(0)
compliant = payload.get("compliant")
if compliant is True:
print("ok")
elif compliant is False:
print("failed")
else:
status = str(payload.get("status") or payload.get("result") or payload.get("compliance") or "").strip().lower()
print(status or "missing")
PY
)"
case "${supply_status}" in
ok|pass|passed|success|compliant) ;;
not_applicable|na|n/a)
if enabled "${ironbank_required}"; then
echo "supply chain gate required but status=${supply_status}" >&2
fail=1
fi
;;
*)
if enabled "${ironbank_required}"; then
echo "supply chain gate failed: ${supply_status}" >&2
fail=1
else
echo "supply chain gate not passing (${supply_status}) but not required for this run" >&2
fi
;;
esac
fi
exit "${fail}"
set -eu
test "$(cat build/quality-gate.rc 2>/dev/null || echo 1)" -eq 0
'''
}
}
@ -382,7 +76,7 @@ PY
script {
env.FLUX_BRANCH = sh(
returnStdout: true,
script: "grep -m1 '^\\s*branch:' clusters/atlas/flux-system/gotk-sync.yaml | sed 's/^\\s*branch:\\s*//'"
script: '''awk '/branch:/{print $2; exit}' clusters/atlas/flux-system/gotk-sync.yaml'''
).trim()
if (!env.FLUX_BRANCH) {
error('Flux branch not found in gotk-sync.yaml')
@ -399,28 +93,16 @@ PY
}
}
steps {
withCredentials([usernamePassword(credentialsId: 'gitea-pat', usernameVariable: 'GIT_USER', passwordVariable: 'GIT_TOKEN')]) {
sh '''
set -euo pipefail
if ! command -v git >/dev/null 2>&1; then
if command -v apk >/dev/null 2>&1; then
apk add --no-cache git >/dev/null
elif command -v apt-get >/dev/null 2>&1; then
apt-get update >/dev/null
apt-get install -y git >/dev/null
fi
fi
cd "${WORKSPACE:-$PWD}"
if ! git rev-parse --is-inside-work-tree >/dev/null 2>&1; then
echo "workspace is not a git checkout; skipping promote"
exit 0
fi
set +x
git config user.email "jenkins@bstein.dev"
git config user.name "jenkins"
git remote set-url origin https://${GIT_USER}:${GIT_TOKEN}@scm.bstein.dev/bstein/titan-iac.git
git push origin HEAD:${FLUX_BRANCH}
'''
container('jnlp') {
withCredentials([usernamePassword(credentialsId: 'gitea-pat', usernameVariable: 'GIT_USER', passwordVariable: 'GIT_TOKEN')]) {
sh '''
set +x
git config user.email "jenkins@bstein.dev"
git config user.name "jenkins"
git remote set-url origin https://${GIT_USER}:${GIT_TOKEN}@scm.bstein.dev/bstein/titan-iac.git
git push origin HEAD:${FLUX_BRANCH}
'''
}
}
}
}
@ -428,23 +110,15 @@ PY
post {
always {
script {
try {
if (fileExists('build/junit-unit.xml') || fileExists('build/junit-glue.xml')) {
try {
junit allowEmptyResults: true, testResults: 'build/junit-*.xml'
} catch (Throwable err) {
echo "junit step unavailable: ${err.class.simpleName}"
}
}
archiveArtifacts artifacts: 'build/**', allowEmptyArchive: true, fingerprint: true
} catch (Throwable err) {
if (err.class.simpleName == 'MissingContextVariableException') {
echo 'workspace unavailable; skipping post-build artifact collection'
} else {
throw err
if (fileExists('build/junit-unit.xml') || fileExists('build/junit-glue.xml')) {
try {
junit allowEmptyResults: true, testResults: 'build/junit-*.xml'
} catch (Throwable err) {
echo "junit step unavailable: ${err.class.simpleName}"
}
}
}
archiveArtifacts artifacts: 'build/**', allowEmptyArchive: true, fingerprint: true
}
}
}

View File

@ -6,52 +6,14 @@ pipeline {
apiVersion: v1
kind: Pod
spec:
serviceAccountName: "jenkins"
nodeSelector:
hardware: rpi5
kubernetes.io/arch: arm64
node-role.kubernetes.io/worker: "true"
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-04
- titan-06
- titan-11
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
labelSelector:
matchLabels:
jenkins/jenkins-jenkins-agent: "true"
containers:
- name: jnlp
image: jenkins/inbound-agent:3355.v388858a_47b_33-2-jdk21
resources:
requests:
cpu: "25m"
memory: "256Mi"
- name: python
image: registry.bstein.dev/bstein/python:3.12-slim
command:
- cat
tty: true
- name: quality-tools
image: registry.bstein.dev/bstein/quality-tools:sonar8.0.1-trivy0.70.0-db20260422-arm64
image: python:3.12-slim
command:
- cat
tty: true
@ -61,21 +23,9 @@ spec:
environment {
PIP_DISABLE_PIP_VERSION_CHECK = '1'
PYTHONUNBUFFERED = '1'
SUITE_NAME = 'titan_iac'
SUITE_NAME = 'titan-iac'
PUSHGATEWAY_URL = 'http://platform-quality-gateway.monitoring.svc.cluster.local:9091'
SONARQUBE_HOST_URL = 'http://sonarqube.quality.svc.cluster.local:9000'
SONARQUBE_PROJECT_KEY = 'titan_iac'
SONARQUBE_TOKEN = credentials('sonarqube-token')
VM_URL = 'http://victoria-metrics-single-server.monitoring.svc.cluster.local:8428'
QUALITY_GATE_SONARQUBE_ENFORCE = '0'
QUALITY_GATE_SONARQUBE_REPORT = 'build/sonarqube-quality-gate.json'
QUALITY_GATE_IRONBANK_ENFORCE = '1'
QUALITY_GATE_IRONBANK_REQUIRED = '0'
QUALITY_GATE_IRONBANK_REPORT = 'build/ironbank-compliance.json'
}
options {
disableConcurrentBuilds()
buildDiscarder(logRotator(daysToKeepStr: '30', numToKeepStr: '200', artifactDaysToKeepStr: '30', artifactNumToKeepStr: '120'))
}
stages {
stage('Checkout') {
@ -85,175 +35,7 @@ spec:
}
stage('Install deps') {
steps {
sh '''
set -eu
if ! command -v git >/dev/null 2>&1; then
apt-get update
apt-get install -y --no-install-recommends git ca-certificates
rm -rf /var/lib/apt/lists/*
fi
pip install --no-cache-dir -r ci/requirements.txt
'''
}
}
stage('Prepare local quality evidence') {
steps {
sh '''
set -eu
mkdir -p build
set +e
python3 -m testing.quality_gate --profile local --build-dir build
local_quality_rc=$?
set -e
printf '%s\n' "${local_quality_rc}" > build/local-quality-gate.rc
'''
}
}
stage('Collect SonarQube evidence') {
steps {
container('quality-tools') {
sh '''#!/usr/bin/env bash
set -euo pipefail
mkdir -p build
args=(
"-Dsonar.host.url=${SONARQUBE_HOST_URL}"
"-Dsonar.login=${SONARQUBE_TOKEN}"
"-Dsonar.projectKey=${SONARQUBE_PROJECT_KEY}"
"-Dsonar.projectName=${SONARQUBE_PROJECT_KEY}"
"-Dsonar.sources=."
"-Dsonar.exclusions=**/.git/**,**/build/**,**/dist/**,**/node_modules/**,**/.venv/**,**/__pycache__/**,**/coverage/**,**/test-results/**,**/playwright-report/**,services/monitoring/dashboards/**,services/monitoring/grafana-dashboard-*.yaml,services/game-stream/**"
"-Dsonar.test.inclusions=**/tests/**,**/testing/**,**/*_test.go,**/*.test.ts,**/*.test.tsx,**/*.spec.ts,**/*.spec.tsx"
)
[ -f build/coverage-unit.xml ] && args+=("-Dsonar.python.coverage.reportPaths=build/coverage-unit.xml")
set +e
sonar-scanner "${args[@]}" | tee build/sonar-scanner.log
rc=${PIPESTATUS[0]}
set -e
printf '%s\n' "${rc}" > build/sonarqube-analysis.rc
'''
}
sh '''
set -eu
mkdir -p build
python3 - <<'PY'
import base64
import json
import os
import time
import urllib.parse
import urllib.request
from pathlib import Path
host = os.getenv('SONARQUBE_HOST_URL', '').strip().rstrip('/')
project_key = os.getenv('SONARQUBE_PROJECT_KEY', '').strip()
token = os.getenv('SONARQUBE_TOKEN', '').strip()
report_path = os.getenv('QUALITY_GATE_SONARQUBE_REPORT', 'build/sonarqube-quality-gate.json')
payload = {
"status": "ERROR",
"note": "missing SONARQUBE_HOST_URL and/or SONARQUBE_PROJECT_KEY",
}
if host and project_key:
task_file = Path('.scannerwork/report-task.txt')
task_id = ''
if task_file.exists():
for line in task_file.read_text(encoding='utf-8').splitlines():
key, _, value = line.partition('=')
if key == 'ceTaskId':
task_id = value.strip()
break
if task_id:
ce_query = urllib.parse.urlencode({"id": task_id})
deadline = time.monotonic() + 180
while time.monotonic() < deadline:
ce_request = urllib.request.Request(f"{host}/api/ce/task?{ce_query}", method="GET")
if token:
encoded = base64.b64encode(f"{token}:".encode("utf-8")).decode("utf-8")
ce_request.add_header("Authorization", f"Basic {encoded}")
try:
with urllib.request.urlopen(ce_request, timeout=12) as response:
ce_payload = json.loads(response.read().decode("utf-8"))
except Exception:
time.sleep(3)
continue
status = str(ce_payload.get("task", {}).get("status", "")).upper()
if status in {"SUCCESS", "FAILED", "CANCELED"}:
break
time.sleep(3)
query = urllib.parse.urlencode({"projectKey": project_key})
request = urllib.request.Request(
f"{host}/api/qualitygates/project_status?{query}",
method="GET",
)
if token:
encoded = base64.b64encode(f"{token}:".encode("utf-8")).decode("utf-8")
request.add_header("Authorization", f"Basic {encoded}")
try:
with urllib.request.urlopen(request, timeout=12) as response:
payload = json.loads(response.read().decode("utf-8"))
except Exception as exc: # noqa: BLE001
payload = {"status": "ERROR", "error": str(exc)}
with open(report_path, "w", encoding="utf-8") as handle:
json.dump(payload, handle, indent=2, sort_keys=True)
handle.write("\\n")
PY
'''
}
}
stage('Collect IronBank evidence') {
steps {
container('quality-tools') {
sh '''#!/usr/bin/env bash
set -euo pipefail
mkdir -p build
set +e
trivy fs --cache-dir "${TRIVY_CACHE_DIR}" --skip-db-update --skip-files clusters/atlas/flux-system/gotk-components.yaml --timeout 5m --no-progress --format json --output build/trivy-fs.json --scanners vuln,secret,misconfig --severity HIGH,CRITICAL .
trivy_rc=$?
set -e
if [ ! -s build/trivy-fs.json ]; then
cat > build/ironbank-compliance.json <<EOF
{"status":"failed","compliant":false,"scanner":"trivy","scan_type":"filesystem","error":"trivy did not produce JSON output","trivy_rc":${trivy_rc}}
EOF
exit 0
fi
'''
}
sh '''
set -eu
mkdir -p build
if [ -s build/trivy-fs.json ]; then
python3 ci/scripts/supply_chain_report.py --trivy-json build/trivy-fs.json --waivers ci/titan-iac-trivy-waivers.json --output build/ironbank-compliance.json
exit 0
fi
python3 - <<'PY'
import json
import os
from pathlib import Path
report_path = Path(os.getenv('QUALITY_GATE_IRONBANK_REPORT', 'build/ironbank-compliance.json'))
if report_path.exists():
raise SystemExit(0)
status = os.getenv('IRONBANK_COMPLIANCE_STATUS', '').strip()
compliant = os.getenv('IRONBANK_COMPLIANT', '').strip().lower()
payload = {
"status": status or "unknown",
"compliant": compliant in {"1", "true", "yes", "on"} if compliant else None,
}
payload = {k: v for k, v in payload.items() if v is not None}
if "status" not in payload:
payload["status"] = "unknown"
payload["note"] = (
"Set IRONBANK_COMPLIANCE_STATUS/IRONBANK_COMPLIANT "
"or write build/ironbank-compliance.json in image-building repos."
)
report_path.parent.mkdir(parents=True, exist_ok=True)
report_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\\n", encoding="utf-8")
PY
'''
sh 'pip install --no-cache-dir -r ci/requirements.txt'
}
}
stage('Run quality gate') {
@ -283,96 +65,8 @@ PY
stage('Enforce quality gate') {
steps {
sh '''
set -euo pipefail
gate_rc="$(cat build/quality-gate.rc 2>/dev/null || echo 1)"
fail=0
if [ "${gate_rc}" -ne 0 ]; then
echo "quality gate failed with rc=${gate_rc}" >&2
fail=1
fi
enabled() {
case "$(printf '%s' "${1:-}" | tr '[:upper:]' '[:lower:]')" in
1|true|yes|on) return 0 ;;
*) return 1 ;;
esac
}
if enabled "${QUALITY_GATE_SONARQUBE_ENFORCE:-1}"; then
sonar_status="$(python3 - <<'PY'
import json
from pathlib import Path
path = Path("build/sonarqube-quality-gate.json")
if not path.exists():
print("missing")
raise SystemExit(0)
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except Exception: # noqa: BLE001
print("error")
raise SystemExit(0)
status = (payload.get("status") or payload.get("projectStatus", {}).get("status") or payload.get("qualityGate", {}).get("status") or "").strip().lower()
print(status or "missing")
PY
)"
case "${sonar_status}" in
ok|pass|passed|success) ;;
*)
echo "sonarqube gate failed: ${sonar_status}" >&2
fail=1
;;
esac
fi
ironbank_required="${QUALITY_GATE_IRONBANK_REQUIRED:-0}"
if [ "${PUBLISH_IMAGES:-false}" = "true" ]; then
ironbank_required=1
fi
if enabled "${QUALITY_GATE_IRONBANK_ENFORCE:-1}"; then
supply_status="$(python3 - <<'PY'
import json
from pathlib import Path
path = Path("build/ironbank-compliance.json")
if not path.exists():
print("missing")
raise SystemExit(0)
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except Exception: # noqa: BLE001
print("error")
raise SystemExit(0)
compliant = payload.get("compliant")
if compliant is True:
print("ok")
elif compliant is False:
print("failed")
else:
status = str(payload.get("status") or payload.get("result") or payload.get("compliance") or "").strip().lower()
print(status or "missing")
PY
)"
case "${supply_status}" in
ok|pass|passed|success|compliant) ;;
not_applicable|na|n/a)
if enabled "${ironbank_required}"; then
echo "supply chain gate required but status=${supply_status}" >&2
fail=1
fi
;;
*)
if enabled "${ironbank_required}"; then
echo "supply chain gate failed: ${supply_status}" >&2
fail=1
else
echo "supply chain gate not passing (${supply_status}) but not required for this run" >&2
fi
;;
esac
fi
exit "${fail}"
set -eu
test "$(cat build/quality-gate.rc 2>/dev/null || echo 1)" -eq 0
'''
}
}
@ -381,7 +75,7 @@ PY
script {
env.FLUX_BRANCH = sh(
returnStdout: true,
script: "grep -m1 '^\\s*branch:' clusters/atlas/flux-system/gotk-sync.yaml | sed 's/^\\s*branch:\\s*//'"
script: '''awk '/branch:/{print $2; exit}' clusters/atlas/flux-system/gotk-sync.yaml'''
).trim()
if (!env.FLUX_BRANCH) {
error('Flux branch not found in gotk-sync.yaml')
@ -398,28 +92,16 @@ PY
}
}
steps {
withCredentials([usernamePassword(credentialsId: 'gitea-pat', usernameVariable: 'GIT_USER', passwordVariable: 'GIT_TOKEN')]) {
sh '''
set -euo pipefail
if ! command -v git >/dev/null 2>&1; then
if command -v apk >/dev/null 2>&1; then
apk add --no-cache git >/dev/null
elif command -v apt-get >/dev/null 2>&1; then
apt-get update >/dev/null
apt-get install -y git >/dev/null
fi
fi
cd "${WORKSPACE:-$PWD}"
if ! git rev-parse --is-inside-work-tree >/dev/null 2>&1; then
echo "workspace is not a git checkout; skipping promote"
exit 0
fi
set +x
git config user.email "jenkins@bstein.dev"
git config user.name "jenkins"
git remote set-url origin https://${GIT_USER}:${GIT_TOKEN}@scm.bstein.dev/bstein/titan-iac.git
git push origin HEAD:${FLUX_BRANCH}
'''
container('jnlp') {
withCredentials([usernamePassword(credentialsId: 'gitea-pat', usernameVariable: 'GIT_USER', passwordVariable: 'GIT_TOKEN')]) {
sh '''
set +x
git config user.email "jenkins@bstein.dev"
git config user.name "jenkins"
git remote set-url origin https://${GIT_USER}:${GIT_TOKEN}@scm.bstein.dev/bstein/titan-iac.git
git push origin HEAD:${FLUX_BRANCH}
'''
}
}
}
}
@ -427,23 +109,15 @@ PY
post {
always {
script {
try {
if (fileExists('build/junit-unit.xml') || fileExists('build/junit-glue.xml')) {
try {
junit allowEmptyResults: true, testResults: 'build/junit-*.xml'
} catch (Throwable err) {
echo "junit step unavailable: ${err.class.simpleName}"
}
}
archiveArtifacts artifacts: 'build/**', allowEmptyArchive: true, fingerprint: true
} catch (Throwable err) {
if (err.class.simpleName == 'MissingContextVariableException') {
echo 'workspace unavailable; skipping post-build artifact collection'
} else {
throw err
if (fileExists('build/junit-unit.xml') || fileExists('build/junit-glue.xml')) {
try {
junit allowEmptyResults: true, testResults: 'build/junit-*.xml'
} catch (Throwable err) {
echo "junit step unavailable: ${err.class.simpleName}"
}
}
}
archiveArtifacts artifacts: 'build/**', allowEmptyArchive: true, fingerprint: true
}
}
}

View File

@ -6,50 +6,30 @@ from __future__ import annotations
import json
import os
from glob import glob
from pathlib import Path
import sys
import urllib.error
import urllib.request
import xml.etree.ElementTree as ET
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from ci.scripts import publish_test_metrics_quality as _quality_helpers
CANONICAL_CHECKS = _quality_helpers.CANONICAL_CHECKS
_build_check_statuses = _quality_helpers._build_check_statuses
_combine_statuses = _quality_helpers._combine_statuses
_infer_sonarqube_status = _quality_helpers._infer_sonarqube_status
_infer_source_lines_over_500 = _quality_helpers._infer_source_lines_over_500
_infer_supply_chain_status = _quality_helpers._infer_supply_chain_status
_infer_workspace_coverage_percent = _quality_helpers._infer_workspace_coverage_percent
_load_optional_json = _quality_helpers._load_optional_json
_normalize_result_status = _quality_helpers._normalize_result_status
def _escape_label(value: str) -> str:
"""Escape a Prometheus label value without changing its content."""
return value.replace("\\", "\\\\").replace("\n", "\\n").replace('"', '\\"')
def _label_str(labels: dict[str, str]) -> str:
"""Render a stable Prometheus label set from a mapping."""
parts = [f'{key}="{_escape_label(val)}"' for key, val in labels.items() if val]
return "{" + ",".join(parts) + "}" if parts else ""
def _read_text(url: str) -> str:
"""Fetch a plain-text response body from the given URL."""
with urllib.request.urlopen(url, timeout=10) as response:
return response.read().decode("utf-8")
def _post_text(url: str, payload: str) -> None:
"""PUT a plain-text payload and fail on any 4xx/5xx response."""
request = urllib.request.Request(
url,
data=payload.encode("utf-8"),
method="PUT",
method="POST",
headers={"Content-Type": "text/plain"},
)
with urllib.request.urlopen(request, timeout=10) as response:
@ -58,7 +38,6 @@ def _post_text(url: str, payload: str) -> None:
def _parse_junit(path: str) -> dict[str, int]:
"""Parse a JUnit XML file into aggregate test counters."""
if not os.path.exists(path):
return {"tests": 0, "failures": 0, "errors": 0, "skipped": 0}
@ -85,7 +64,6 @@ def _parse_junit(path: str) -> dict[str, int]:
def _collect_junit_totals(pattern: str) -> dict[str, int]:
"""Sum JUnit counters across every XML file matching the pattern."""
totals = {"tests": 0, "failures": 0, "errors": 0, "skipped": 0}
for path in sorted(glob(pattern)):
parsed = _parse_junit(path)
@ -94,38 +72,7 @@ def _collect_junit_totals(pattern: str) -> dict[str, int]:
return totals
def _collect_junit_cases(pattern: str) -> list[tuple[str, str]]:
"""Collect individual JUnit test-case statuses for flaky-test trend panels."""
cases: list[tuple[str, str]] = []
for path in sorted(glob(pattern)):
if not os.path.exists(path):
continue
root = ET.parse(path).getroot()
suites: list[ET.Element]
if root.tag == "testsuite":
suites = [root]
elif root.tag == "testsuites":
suites = [elem for elem in root if elem.tag == "testsuite"]
else:
suites = []
for suite in suites:
for test_case in suite.findall("testcase"):
case_name = test_case.attrib.get("name", "").strip()
class_name = test_case.attrib.get("classname", "").strip()
if not case_name:
continue
full_name = f"{class_name}.{case_name}" if class_name else case_name
status = "passed"
if test_case.find("failure") is not None or test_case.find("error") is not None:
status = "failed"
elif test_case.find("skipped") is not None:
status = "skipped"
cases.append((full_name, status))
return cases
def _read_exit_code(path: str) -> int:
"""Read the quality-gate exit code, defaulting to failure if missing."""
try:
with open(path, "r", encoding="utf-8") as handle:
return int(handle.read().strip())
@ -134,7 +81,6 @@ def _read_exit_code(path: str) -> int:
def _load_summary(path: str) -> dict:
"""Load the JSON quality-gate summary, returning an empty mapping on error."""
try:
with open(path, "r", encoding="utf-8") as handle:
return json.load(handle)
@ -143,7 +89,6 @@ def _load_summary(path: str) -> dict:
def _summary_float(summary: dict, key: str) -> float:
"""Extract a float-like value from the summary, defaulting to 0.0."""
value = summary.get(key)
if isinstance(value, (int, float)):
return float(value)
@ -151,7 +96,6 @@ def _summary_float(summary: dict, key: str) -> float:
def _summary_int(summary: dict, key: str) -> int:
"""Extract an int-like value from the summary, defaulting to 0."""
value = summary.get(key)
if isinstance(value, int):
return value
@ -161,7 +105,6 @@ def _summary_int(summary: dict, key: str) -> int:
def _fetch_existing_counter(pushgateway_url: str, metric: str, labels: dict[str, str]) -> float:
"""Return the current counter value for a labeled metric if present."""
text = _read_text(f"{pushgateway_url.rstrip('/')}/metrics")
for line in text.splitlines():
if not line.startswith(metric + "{"):
@ -182,34 +125,22 @@ def _build_payload(
suite: str,
status: str,
tests: dict[str, int],
test_cases: list[tuple[str, str]],
ok_count: int,
failed_count: int,
branch: str,
build_number: str,
jenkins_job: str,
summary: dict | None = None,
workspace_line_coverage_percent: float = 0.0,
source_files_total: int = 0,
source_lines_over_500: int = 0,
check_statuses: dict[str, str] | None = None,
) -> str:
"""Build the Pushgateway payload for the current suite run."""
passed = max(tests["tests"] - tests["failures"] - tests["errors"] - tests["skipped"], 0)
build_labels = _label_str(
{
"suite": suite,
"branch": branch or "unknown",
"build_number": build_number or "unknown",
"jenkins_job": jenkins_job or suite,
}
)
test_case_base_labels = {
"suite": suite,
"branch": branch or "unknown",
"build_number": build_number or "unknown",
"jenkins_job": jenkins_job or suite,
}
lines = [
"# TYPE platform_quality_gate_runs_total counter",
f'platform_quality_gate_runs_total{{suite="{suite}",status="ok"}} {ok_count}',
@ -222,85 +153,43 @@ def _build_payload(
"# TYPE titan_iac_quality_gate_run_status gauge",
f'titan_iac_quality_gate_run_status{{suite="{suite}",status="ok"}} {1 if status == "ok" else 0}',
f'titan_iac_quality_gate_run_status{{suite="{suite}",status="failed"}} {1 if status == "failed" else 0}',
"# TYPE platform_quality_gate_build_info gauge",
f"platform_quality_gate_build_info{build_labels} 1",
"# TYPE titan_iac_quality_gate_build_info gauge",
f"titan_iac_quality_gate_build_info{build_labels} 1",
"# TYPE platform_quality_gate_workspace_line_coverage_percent gauge",
f'platform_quality_gate_workspace_line_coverage_percent{{suite="{suite}"}} {workspace_line_coverage_percent:.3f}',
"# TYPE platform_quality_gate_source_files_total gauge",
f'platform_quality_gate_source_files_total{{suite="{suite}"}} {source_files_total}',
"# TYPE platform_quality_gate_source_lines_over_500_total gauge",
f'platform_quality_gate_source_lines_over_500_total{{suite="{suite}"}} {source_lines_over_500}',
]
if check_statuses:
results = summary.get("results", []) if isinstance(summary, dict) else []
if results:
lines.append("# TYPE titan_iac_quality_gate_checks_total gauge")
for check_name in CANONICAL_CHECKS:
check_status = check_statuses.get(check_name, "not_applicable")
for result in results:
check_name = result.get("name")
check_status = result.get("status")
if not check_name or not check_status:
continue
lines.append(
f'titan_iac_quality_gate_checks_total{{suite="{suite}",check="{_escape_label(check_name)}",result="{_escape_label(check_status)}"}} 1'
f'titan_iac_quality_gate_checks_total{{suite="{suite}",check="{_escape_label(str(check_name))}",result="{_escape_label(str(check_status))}"}} 1'
)
lines.append("# TYPE platform_quality_gate_test_case_result gauge")
if test_cases:
for test_name, test_status in test_cases:
labels = {
**test_case_base_labels,
"test": test_name,
"status": test_status,
}
lines.append(
f"platform_quality_gate_test_case_result{_label_str(labels)} 1"
)
else:
labels = {**test_case_base_labels, "test": "__no_test_cases__", "status": "skipped"}
lines.append(
f"platform_quality_gate_test_case_result{_label_str(labels)} 1"
)
return "\n".join(lines) + "\n"
def main() -> int:
"""Publish the quality-gate metrics and print a compact run summary."""
suite = os.getenv("SUITE_NAME", "titan_iac")
suite = os.getenv("SUITE_NAME", "titan-iac")
pushgateway_url = os.getenv("PUSHGATEWAY_URL", "http://platform-quality-gateway.monitoring.svc.cluster.local:9091")
job_name = os.getenv("QUALITY_GATE_JOB_NAME", "platform-quality-ci")
junit_glob = os.getenv("JUNIT_GLOB", os.getenv("JUNIT_PATH", "build/junit-*.xml"))
exit_code_path = os.getenv("QUALITY_GATE_EXIT_CODE_PATH", os.getenv("GLUE_EXIT_CODE_PATH", "build/quality-gate.rc"))
summary_path = os.getenv("QUALITY_GATE_SUMMARY_PATH", "build/quality-gate-summary.json")
branch = os.getenv("BRANCH_NAME") or os.getenv("GIT_BRANCH") or "unknown"
if branch.startswith("origin/"):
branch = branch[len("origin/") :]
branch = os.getenv("BRANCH_NAME", os.getenv("GIT_BRANCH", ""))
build_number = os.getenv("BUILD_NUMBER", "")
jenkins_job = os.getenv("JOB_NAME", "titan-iac")
tests = _collect_junit_totals(junit_glob)
test_cases = _collect_junit_cases(junit_glob)
exit_code = _read_exit_code(exit_code_path)
status = "ok" if exit_code == 0 else "failed"
summary = _load_summary(summary_path)
workspace_line_coverage_percent = _summary_float(summary, "workspace_line_coverage_percent")
if workspace_line_coverage_percent <= 0:
workspace_line_coverage_percent = _infer_workspace_coverage_percent(summary, "build/coverage-unit.xml")
source_files_total = _summary_int(summary, "source_files_total")
source_lines_over_500 = _summary_int(summary, "source_lines_over_500")
if source_lines_over_500 <= 0:
source_lines_over_500 = _infer_source_lines_over_500(summary)
sonarqube_report = _load_optional_json(os.getenv("QUALITY_GATE_SONARQUBE_REPORT", "build/sonarqube-quality-gate.json"))
supply_chain_report = _load_optional_json(os.getenv("QUALITY_GATE_IRONBANK_REPORT", "build/ironbank-compliance.json"))
truthy = {"1", "true", "yes", "on"}
supply_chain_required = (
os.getenv("QUALITY_GATE_IRONBANK_REQUIRED", "0").strip().lower() in truthy
or os.getenv("PUBLISH_IMAGES", "false").strip().lower() in truthy
)
check_statuses = _build_check_statuses(
summary=summary,
tests=tests,
workspace_line_coverage_percent=workspace_line_coverage_percent,
source_lines_over_500=source_lines_over_500,
sonarqube_report=sonarqube_report,
supply_chain_report=supply_chain_report,
supply_chain_required=supply_chain_required,
)
ok_count = int(
_fetch_existing_counter(
@ -325,17 +214,13 @@ def main() -> int:
suite=suite,
status=status,
tests=tests,
test_cases=test_cases,
ok_count=ok_count,
failed_count=failed_count,
branch=branch,
build_number=build_number,
jenkins_job=jenkins_job,
summary=summary,
workspace_line_coverage_percent=workspace_line_coverage_percent,
source_files_total=source_files_total,
source_lines_over_500=source_lines_over_500,
check_statuses=check_statuses,
)
push_url = f"{pushgateway_url.rstrip('/')}/metrics/job/{job_name}/suite/{suite}"
_post_text(push_url, payload)
@ -349,14 +234,13 @@ def main() -> int:
"tests_skipped": tests["skipped"],
"ok_count": ok_count,
"failed_count": failed_count,
"checks_recorded": len(check_statuses),
"checks_recorded": len(summary.get("results", [])) if isinstance(summary, dict) else 0,
"workspace_line_coverage_percent": workspace_line_coverage_percent,
"source_files_total": source_files_total,
"source_lines_over_500": source_lines_over_500,
}
print(json.dumps(summary, sort_keys=True))
return 0
if __name__ == "__main__": # pragma: no cover
if __name__ == "__main__":
raise SystemExit(main())

View File

@ -1,204 +0,0 @@
#!/usr/bin/env python3
"""Quality/status helpers for publish_test_metrics."""
from __future__ import annotations
import json
from pathlib import Path
import xml.etree.ElementTree as ET
SUCCESS_STATUSES = {"ok", "pass", "passed", "success", "compliant"}
NOT_APPLICABLE_STATUSES = {"not_applicable", "n/a", "na", "none", "skipped"}
FAILED_STATUSES = {"failed", "fail", "error", "errors", "warn", "warning", "red"}
CANONICAL_CHECKS = [
"tests",
"coverage",
"loc",
"docs_naming",
"gate_glue",
"sonarqube",
"supply_chain",
]
def _infer_workspace_coverage_percent(summary: dict, default_xml: str) -> float:
"""Infer workspace line coverage from quality summary coverage XML metadata."""
results = summary.get("results", []) if isinstance(summary, dict) else []
coverage_xml = default_xml
for result in results:
if not isinstance(result, dict):
continue
if str(result.get("name") or "").strip().lower() != "coverage":
continue
candidate = str(result.get("coverage_xml") or "").strip()
if candidate:
coverage_xml = candidate
break
xml_path = Path(coverage_xml)
if not xml_path.exists():
return 0.0
try:
root = ET.parse(xml_path).getroot()
line_rate = root.attrib.get("line-rate")
if line_rate is None:
return 0.0
return float(line_rate) * 100.0
except (ET.ParseError, OSError, ValueError):
return 0.0
def _infer_source_lines_over_500(summary: dict) -> int:
"""Infer over-limit source file count from hygiene issue payloads."""
results = summary.get("results", []) if isinstance(summary, dict) else []
for result in results:
if not isinstance(result, dict):
continue
if str(result.get("name") or "").strip().lower() not in {"hygiene", "loc", "smell"}:
continue
issues = result.get("issues")
if not isinstance(issues, list):
continue
return sum(1 for item in issues if isinstance(item, str) and item.startswith("file exceeds"))
return 0
def _normalize_result_status(value: str | None, default: str = "failed") -> str:
"""Map arbitrary check status text into canonical check result buckets."""
if not value:
return default
normalized = value.strip().lower()
if normalized in SUCCESS_STATUSES:
return "ok"
if normalized in NOT_APPLICABLE_STATUSES:
return "not_applicable"
if normalized in FAILED_STATUSES:
return "failed"
return default
def _load_optional_json(path: str | None) -> dict:
"""Load an optional JSON report file, returning an empty object when absent."""
if not path:
return {}
candidate = Path(path)
if not candidate.exists():
return {}
try:
return json.loads(candidate.read_text(encoding="utf-8"))
except json.JSONDecodeError:
return {}
def _combine_statuses(statuses: list[str]) -> str:
"""Roll up many check statuses into one canonical result."""
if not statuses:
return "not_applicable"
if any(status == "failed" for status in statuses):
return "failed"
if all(status == "not_applicable" for status in statuses):
return "not_applicable"
if all(status in {"ok", "not_applicable"} for status in statuses):
return "ok"
return "failed"
def _infer_sonarqube_status(report: dict) -> str:
"""Infer canonical SonarQube check status from its JSON report payload."""
if not report:
return "not_applicable"
status = (
report.get("projectStatus", {}).get("status")
or report.get("qualityGate", {}).get("status")
or report.get("status")
)
return _normalize_result_status(str(status) if status is not None else None, default="failed")
def _infer_supply_chain_status(report: dict, required: bool) -> str:
"""Infer canonical supply-chain status from IronBank/artifact report payload."""
if not report:
return "failed" if required else "not_applicable"
compliant = report.get("compliant")
if isinstance(compliant, bool):
if compliant:
return "ok"
return "failed" if required else "not_applicable"
status = report.get("status")
if status is None:
return "failed" if required else "not_applicable"
normalized = _normalize_result_status(str(status), default="failed")
if normalized == "failed" and not required:
return "not_applicable"
if normalized == "not_applicable" and required:
return "failed"
return normalized
def _build_check_statuses(
summary: dict | None,
tests: dict[str, int],
workspace_line_coverage_percent: float,
source_lines_over_500: int,
sonarqube_report: dict,
supply_chain_report: dict,
supply_chain_required: bool,
) -> dict[str, str]:
"""Generate the canonical quality-check status map for dashboarding."""
raw_results = summary.get("results", []) if isinstance(summary, dict) else []
status_by_name: dict[str, str] = {}
for result in raw_results:
if not isinstance(result, dict):
continue
check_name = str(result.get("name") or "").strip().lower()
if not check_name:
continue
status_by_name[check_name] = _normalize_result_status(result.get("status"), default="failed")
tests_status = status_by_name.get("tests")
if not tests_status:
candidate_keys = ["unit", "integration", "e2e", "pytest", "test", "tests"]
candidates = [status_by_name[key] for key in candidate_keys if key in status_by_name]
if candidates:
tests_status = _combine_statuses(candidates)
elif tests["tests"] > 0:
tests_status = "ok" if (tests["failures"] + tests["errors"]) == 0 else "failed"
else:
tests_status = "not_applicable"
coverage_status = status_by_name.get("coverage")
if not coverage_status:
if workspace_line_coverage_percent > 0:
coverage_status = "ok" if workspace_line_coverage_percent >= 95.0 else "failed"
else:
coverage_status = "not_applicable"
loc_status = status_by_name.get("loc")
if not loc_status:
loc_status = "ok" if source_lines_over_500 == 0 else "failed"
docs_naming_status = status_by_name.get("docs_naming")
if not docs_naming_status:
candidates = [status_by_name[key] for key in ["docs", "hygiene", "smell", "lint", "naming"] if key in status_by_name]
docs_naming_status = _combine_statuses(candidates) if candidates else "not_applicable"
gate_glue_status = status_by_name.get("gate_glue")
if not gate_glue_status:
candidates = [status_by_name[key] for key in ["gate_glue", "glue", "gate"] if key in status_by_name]
gate_glue_status = _combine_statuses(candidates) if candidates else "not_applicable"
sonarqube_status = status_by_name.get("sonarqube") or _infer_sonarqube_status(sonarqube_report)
supply_chain_status = status_by_name.get("supply_chain") or _infer_supply_chain_status(
supply_chain_report,
required=supply_chain_required,
)
return {
"tests": tests_status,
"coverage": coverage_status,
"loc": loc_status,
"docs_naming": docs_naming_status,
"gate_glue": gate_glue_status,
"sonarqube": sonarqube_status,
"supply_chain": supply_chain_status,
}

View File

@ -1,173 +0,0 @@
"""Build a titan-iac supply-chain compliance report from Trivy evidence."""
from __future__ import annotations
import argparse
import datetime as dt
import json
from pathlib import Path
from typing import Any
FAIL_SEVERITIES = {"HIGH", "CRITICAL"}
def _read_json(path: Path) -> dict[str, Any]:
"""Read a JSON object from disk for use as pipeline evidence."""
payload = json.loads(path.read_text(encoding="utf-8"))
if not isinstance(payload, dict):
raise ValueError(f"{path} must contain a JSON object")
return payload
def _parse_day(raw: str | None) -> dt.date | None:
"""Parse an ISO day while letting optional waiver dates stay optional."""
if not raw:
return None
return dt.date.fromisoformat(raw)
def _today(override: str | None = None) -> dt.date:
"""Return the policy day so tests can pin expiry behavior."""
return _parse_day(override) or dt.date.today()
def _load_waiver_pairs(path: Path | None, policy_day: dt.date) -> tuple[set[tuple[str, str]], int]:
"""Return active ``(misconfiguration id, target)`` waivers and expired count."""
if path is None or not path.exists():
return set(), 0
payload = _read_json(path)
default_expires_at = payload.get("default_expires_at")
active: set[tuple[str, str]] = set()
expired = 0
for entry in payload.get("misconfigurations", []):
if not isinstance(entry, dict):
continue
misconfiguration_id = str(entry.get("id") or "").strip()
if not misconfiguration_id:
continue
expires_at = _parse_day(str(entry.get("expires_at") or default_expires_at or ""))
targets = entry.get("targets", [])
if not isinstance(targets, list):
continue
if expires_at and expires_at < policy_day:
expired += len(targets)
continue
# Waivers are target-specific so a new unsafe manifest fails until it is
# either fixed or deliberately accepted with a fresh expiration.
for target in targets:
if isinstance(target, str) and target:
active.add((misconfiguration_id, target))
return active, expired
def _iter_failed_misconfigurations(payload: dict[str, Any]):
"""Yield failed high/critical Trivy misconfiguration records."""
for result in payload.get("Results", []):
if not isinstance(result, dict):
continue
target = str(result.get("Target") or "")
for item in result.get("Misconfigurations") or []:
if not isinstance(item, dict):
continue
if item.get("Status") != "FAIL":
continue
if str(item.get("Severity") or "").upper() not in FAIL_SEVERITIES:
continue
yield target, item
def _count_vulnerabilities(payload: dict[str, Any], severity: str) -> int:
"""Count Trivy vulnerabilities at a specific severity."""
count = 0
for result in payload.get("Results", []):
if not isinstance(result, dict):
continue
for item in result.get("Vulnerabilities") or []:
if isinstance(item, dict) and str(item.get("Severity") or "").upper() == severity:
count += 1
return count
def _count_secrets(payload: dict[str, Any]) -> int:
"""Count detected secrets in the Trivy filesystem report."""
count = 0
for result in payload.get("Results", []):
if isinstance(result, dict):
count += len(result.get("Secrets") or [])
return count
def build_report(
trivy_payload: dict[str, Any],
waiver_path: Path | None = None,
today_override: str | None = None,
) -> dict[str, Any]:
"""Build the compliance summary consumed by the quality gate."""
policy_day = _today(today_override)
active_waivers, expired_waivers = _load_waiver_pairs(waiver_path, policy_day)
open_misconfigs: list[dict[str, str]] = []
waived_misconfigs = 0
for target, item in _iter_failed_misconfigurations(trivy_payload):
misconfiguration_id = str(item.get("ID") or "")
if (misconfiguration_id, target) in active_waivers:
waived_misconfigs += 1
continue
open_misconfigs.append(
{
"id": misconfiguration_id,
"target": target,
"severity": str(item.get("Severity") or ""),
"title": str(item.get("Title") or ""),
}
)
critical = _count_vulnerabilities(trivy_payload, "CRITICAL")
high = _count_vulnerabilities(trivy_payload, "HIGH")
secrets = _count_secrets(trivy_payload)
status = "ok" if critical == 0 and secrets == 0 and not open_misconfigs else "failed"
return {
"status": status,
"compliant": status == "ok",
"category": "artifact_security",
"scan_type": "filesystem",
"scanner": "trivy",
"critical_vulnerabilities": critical,
"high_vulnerabilities": high,
"high_vulnerability_policy": "observe",
"secrets": secrets,
"high_or_critical_misconfigurations": len(open_misconfigs),
"waived_misconfigurations": waived_misconfigs,
"expired_waivers": expired_waivers,
"waiver_file": str(waiver_path) if waiver_path else "",
"open_misconfiguration_examples": open_misconfigs[:20],
}
def main(argv: list[str] | None = None) -> int:
"""CLI entrypoint used by Jenkins after the Trivy scan completes."""
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--trivy-json", required=True)
parser.add_argument("--waivers")
parser.add_argument("--output", required=True)
parser.add_argument("--today")
args = parser.parse_args(argv)
trivy_payload = _read_json(Path(args.trivy_json))
waiver_path = Path(args.waivers) if args.waivers else None
report = build_report(trivy_payload, waiver_path=waiver_path, today_override=args.today)
output_path = Path(args.output)
output_path.parent.mkdir(parents=True, exist_ok=True)
output_path.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n", encoding="utf-8")
return 0
if __name__ == "__main__": # pragma: no cover
raise SystemExit(main())

View File

@ -1,108 +0,0 @@
"""Glue checks for Ariadne schedules exported to VictoriaMetrics."""
from __future__ import annotations
import os
from datetime import datetime, timezone
from pathlib import Path
import requests
import yaml
CONFIG_PATH = Path(__file__).with_name("config.yaml")
def _load_config() -> dict:
with CONFIG_PATH.open("r", encoding="utf-8") as handle:
return yaml.safe_load(handle) or {}
def _query(promql: str) -> list[dict]:
vm_url = os.environ.get("VM_URL", "http://victoria-metrics-single-server:8428").rstrip("/")
response = requests.get(f"{vm_url}/api/v1/query", params={"query": promql}, timeout=10)
response.raise_for_status()
payload = response.json()
return payload.get("data", {}).get("result", [])
def _expected_tasks() -> list[dict]:
cfg = _load_config()
tasks = [
_normalize_task(item, cfg)
for item in cfg.get("ariadne_schedule_tasks", [])
]
assert tasks, "No Ariadne schedule tasks configured"
return tasks
def _normalize_task(item: object, cfg: dict) -> dict:
if isinstance(item, str):
return {
"task": item,
"check_last_success": True,
"max_success_age_hours": cfg.get("max_success_age_hours", 48),
}
if isinstance(item, dict):
normalized = dict(item)
normalized.setdefault("check_last_success", True)
normalized.setdefault("max_success_age_hours", cfg.get("max_success_age_hours", 48))
return normalized
raise TypeError(f"Unsupported Ariadne schedule task config entry: {item!r}")
def _tracked_tasks(tasks: list[dict]) -> list[dict]:
tracked = [item for item in tasks if item.get("check_last_success")]
assert tracked, "No Ariadne schedule tasks are marked for success tracking"
return tracked
def _task_regex(tasks: list[dict]) -> str:
return "|".join(item["task"] for item in tasks)
def test_ariadne_schedule_series_exist():
tasks = _expected_tasks()
selector = _task_regex(tasks)
series = _query(f'ariadne_schedule_next_run_timestamp_seconds{{task=~"{selector}"}}')
seen = {item.get("metric", {}).get("task") for item in series}
missing = [item["task"] for item in tasks if item["task"] not in seen]
assert not missing, f"Missing next-run metrics for: {', '.join(missing)}"
def test_ariadne_schedule_recent_success():
tasks = _tracked_tasks(_expected_tasks())
selector = _task_regex(tasks)
series = _query(f'ariadne_schedule_last_success_timestamp_seconds{{task=~"{selector}"}}')
seen = {item.get("metric", {}).get("task") for item in series}
missing = [item["task"] for item in tasks if item["task"] not in seen]
assert not missing, f"Missing last-success metrics for: {', '.join(missing)}"
now = datetime.now(timezone.utc)
age_by_task = {
item.get("metric", {}).get("task"): (now - datetime.fromtimestamp(float(item["value"][1]), tz=timezone.utc)).total_seconds() / 3600
for item in series
}
too_old = [
f"{task} ({age_by_task[task]:.1f}h > {item['max_success_age_hours']}h)"
for item in tasks
if (task := item["task"]) in age_by_task and age_by_task[task] > float(item["max_success_age_hours"])
]
assert not too_old, "Ariadne schedules are stale: " + ", ".join(too_old)
def test_ariadne_schedule_last_status_present_and_boolean():
tasks = _tracked_tasks(_expected_tasks())
selector = _task_regex(tasks)
series = _query(f'ariadne_schedule_last_status{{task=~"{selector}"}}')
seen = {item.get("metric", {}).get("task") for item in series}
missing = [item["task"] for item in tasks if item["task"] not in seen]
assert not missing, f"Missing last-status metrics for: {', '.join(missing)}"
invalid = []
for item in series:
task = item.get("metric", {}).get("task")
value = float(item["value"][1])
if value not in (0.0, 1.0):
invalid.append(f"{task}={value}")
assert not invalid, f"Unexpected Ariadne last-status values: {', '.join(invalid)}"

View File

@ -1,5 +1,3 @@
"""Glue checks for the metrics the quality-gate publishes."""
from __future__ import annotations
import os
@ -25,63 +23,26 @@ def _query(promql: str) -> list[dict]:
return payload.get("data", {}).get("result", [])
def _expected_tasks() -> list[dict]:
cfg = _load_config()
tasks = [
_normalize_task(item, cfg)
for item in cfg.get("ariadne_schedule_tasks", [])
]
assert tasks, "No Ariadne schedule tasks configured"
return tasks
def test_glue_metrics_present():
series = _query('kube_cronjob_labels{label_atlas_bstein_dev_glue="true"}')
assert series, "No glue cronjob label series found"
def _normalize_task(item: object, cfg: dict) -> dict:
if isinstance(item, str):
return {
"task": item,
"check_last_success": True,
"max_success_age_hours": cfg.get("max_success_age_hours", 48),
}
if isinstance(item, dict):
normalized = dict(item)
normalized.setdefault("check_last_success", True)
normalized.setdefault("max_success_age_hours", cfg.get("max_success_age_hours", 48))
return normalized
raise TypeError(f"Unsupported Ariadne schedule task config entry: {item!r}")
def _tracked_tasks(tasks: list[dict]) -> list[dict]:
tracked = [item for item in tasks if item.get("check_last_success")]
assert tracked, "No Ariadne schedule tasks are marked for success tracking"
return tracked
def _task_regex(tasks: list[dict]) -> str:
return "|".join(item["task"] for item in tasks)
def test_glue_metrics_success_join():
query = (
"kube_cronjob_status_last_successful_time "
'and on(namespace,cronjob) kube_cronjob_labels{label_atlas_bstein_dev_glue="true"}'
)
series = _query(query)
assert series, "No glue cronjob last success series found"
def test_ariadne_schedule_metrics_present():
tasks = _expected_tasks()
selector = _task_regex(tasks)
series = _query(f'ariadne_schedule_next_run_timestamp_seconds{{task=~"{selector}"}}')
seen = {item.get("metric", {}).get("task") for item in series}
missing = [item["task"] for item in tasks if item["task"] not in seen]
cfg = _load_config()
expected = cfg.get("ariadne_schedule_tasks", [])
if not expected:
return
series = _query("ariadne_schedule_next_run_timestamp_seconds")
tasks = {item.get("metric", {}).get("task") for item in series}
missing = [task for task in expected if task not in tasks]
assert not missing, f"Missing Ariadne schedule metrics for: {', '.join(missing)}"
def test_ariadne_schedule_success_and_status_metrics_present():
tasks = _tracked_tasks(_expected_tasks())
selector = _task_regex(tasks)
success = _query(f'ariadne_schedule_last_success_timestamp_seconds{{task=~"{selector}"}}')
status = _query(f'ariadne_schedule_last_status{{task=~"{selector}"}}')
success_tasks = {item.get("metric", {}).get("task") for item in success}
status_tasks = {item.get("metric", {}).get("task") for item in status}
expected = {item["task"] for item in tasks}
missing_success = sorted(expected - success_tasks)
missing_status = sorted(expected - status_tasks)
assert not missing_success, f"Missing Ariadne success metrics for: {', '.join(missing_success)}"
assert not missing_status, f"Missing Ariadne status metrics for: {', '.join(missing_status)}"

View File

@ -1,407 +0,0 @@
{
"version": 1,
"generated_from": "Jenkins titan-iac build 225 Trivy filesystem scan",
"default_expires_at": "2026-05-22",
"ticket": "atlas-quality-wave-k8s-hardening",
"default_reason": "Existing Kubernetes manifest hardening baseline accepted only for the first quality-gate rollout; fix or renew explicitly before expiry.",
"misconfigurations": [
{
"id": "DS-0002",
"targets": [
"dockerfiles/Dockerfile.ananke-node-helper"
]
},
{
"id": "KSV-0009",
"targets": [
"services/mailu/vip-controller.yaml",
"services/maintenance/k3s-agent-restart-daemonset.yaml"
]
},
{
"id": "KSV-0010",
"targets": [
"services/maintenance/k3s-agent-restart-daemonset.yaml",
"services/maintenance/metis-sentinel-amd64-daemonset.yaml",
"services/maintenance/metis-sentinel-arm64-daemonset.yaml",
"services/monitoring/jetson-tegrastats-exporter.yaml"
]
},
{
"id": "KSV-0014",
"targets": [
"infrastructure/cert-manager/cleanup/cert-manager-cleanup-job.yaml",
"infrastructure/core/node-prefer-noschedule-cronjob.yaml",
"infrastructure/core/ntp-sync-daemonset.yaml",
"infrastructure/longhorn/adopt/longhorn-helm-adopt-job.yaml",
"infrastructure/longhorn/core/longhorn-disk-tags-ensure-job.yaml",
"infrastructure/longhorn/core/longhorn-settings-ensure-job.yaml",
"infrastructure/longhorn/core/vault-sync-deployment.yaml",
"infrastructure/longhorn/ui-ingress/oauth2-proxy-longhorn.yaml",
"infrastructure/modules/profiles/components/device-plugin-jetson/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-minipc/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-tethys/daemonset.yaml",
"infrastructure/postgres/statefulset.yaml",
"infrastructure/vault-csi/vault-csi-provider.yaml",
"services/ai-llm/deployment.yaml",
"services/bstein-dev-home/backend-deployment.yaml",
"services/bstein-dev-home/chat-ai-gateway-deployment.yaml",
"services/bstein-dev-home/frontend-deployment.yaml",
"services/bstein-dev-home/oneoffs/migrations/portal-migrate-job.yaml",
"services/bstein-dev-home/oneoffs/portal-onboarding-e2e-test-job.yaml",
"services/bstein-dev-home/vault-sync-deployment.yaml",
"services/bstein-dev-home/vaultwarden-cred-sync-cronjob.yaml",
"services/comms/atlasbot-deployment.yaml",
"services/comms/coturn.yaml",
"services/comms/element-call-deployment.yaml",
"services/comms/guest-name-job.yaml",
"services/comms/guest-register-deployment.yaml",
"services/comms/livekit-token-deployment.yaml",
"services/comms/livekit.yaml",
"services/comms/mas-deployment.yaml",
"services/comms/oneoffs/bstein-force-leave-job.yaml",
"services/comms/oneoffs/comms-secrets-ensure-job.yaml",
"services/comms/oneoffs/mas-admin-client-secret-ensure-job.yaml",
"services/comms/oneoffs/mas-db-ensure-job.yaml",
"services/comms/oneoffs/mas-local-users-ensure-job.yaml",
"services/comms/oneoffs/othrys-kick-numeric-job.yaml",
"services/comms/oneoffs/synapse-admin-ensure-job.yaml",
"services/comms/oneoffs/synapse-seeder-admin-ensure-job.yaml",
"services/comms/oneoffs/synapse-signingkey-ensure-job.yaml",
"services/comms/oneoffs/synapse-user-seed-job.yaml",
"services/comms/pin-othrys-job.yaml",
"services/comms/reset-othrys-room-job.yaml",
"services/comms/seed-othrys-room.yaml",
"services/comms/vault-sync-deployment.yaml",
"services/comms/wellknown.yaml",
"services/crypto/monerod/deployment.yaml",
"services/crypto/wallet-monero-temp/deployment.yaml",
"services/crypto/xmr-miner/deployment.yaml",
"services/crypto/xmr-miner/vault-sync-deployment.yaml",
"services/crypto/xmr-miner/xmrig-daemonset.yaml",
"services/finance/actual-budget-deployment.yaml",
"services/finance/firefly-cronjob.yaml",
"services/finance/firefly-deployment.yaml",
"services/finance/firefly-user-sync-cronjob.yaml",
"services/finance/oneoffs/finance-secrets-ensure-job.yaml",
"services/gitea/deployment.yaml",
"services/harbor/vault-sync-deployment.yaml",
"services/health/wger-admin-ensure-cronjob.yaml",
"services/health/wger-deployment.yaml",
"services/health/wger-user-sync-cronjob.yaml",
"services/jellyfin/deployment.yaml",
"services/jellyfin/loader.yaml",
"services/jenkins/deployment.yaml",
"services/jenkins/vault-sync-deployment.yaml",
"services/keycloak/deployment.yaml",
"services/keycloak/oneoffs/actual-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/harbor-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/ldap-federation-job.yaml",
"services/keycloak/oneoffs/logs-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/mas-secrets-ensure-job.yaml",
"services/keycloak/oneoffs/metis-node-passwords-secret-ensure-job.yaml",
"services/keycloak/oneoffs/metis-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/metis-ssh-keys-secret-ensure-job.yaml",
"services/keycloak/oneoffs/portal-admin-client-secret-ensure-job.yaml",
"services/keycloak/oneoffs/portal-e2e-client-job.yaml",
"services/keycloak/oneoffs/portal-e2e-execute-actions-email-test-job.yaml",
"services/keycloak/oneoffs/portal-e2e-target-client-job.yaml",
"services/keycloak/oneoffs/portal-e2e-token-exchange-permissions-job.yaml",
"services/keycloak/oneoffs/portal-e2e-token-exchange-test-job.yaml",
"services/keycloak/oneoffs/quality-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/realm-settings-job.yaml",
"services/keycloak/oneoffs/soteria-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/synapse-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/user-overrides-job.yaml",
"services/keycloak/oneoffs/vault-oidc-secret-ensure-job.yaml",
"services/keycloak/vault-sync-deployment.yaml",
"services/logging/node-image-gc-rpi4-daemonset.yaml",
"services/logging/node-image-prune-rpi5-daemonset.yaml",
"services/logging/node-log-rotation-daemonset.yaml",
"services/logging/oauth2-proxy.yaml",
"services/logging/oneoffs/opensearch-dashboards-setup-job.yaml",
"services/logging/oneoffs/opensearch-ism-job.yaml",
"services/logging/oneoffs/opensearch-observability-setup-job.yaml",
"services/logging/opensearch-prune-cronjob.yaml",
"services/logging/vault-sync-deployment.yaml",
"services/mailu/mailu-sync-cronjob.yaml",
"services/mailu/mailu-sync-listener.yaml",
"services/mailu/oneoffs/mailu-sync-job.yaml",
"services/mailu/vault-sync-deployment.yaml",
"services/mailu/vip-controller.yaml",
"services/maintenance/ariadne-deployment.yaml",
"services/maintenance/disable-k3s-traefik-daemonset.yaml",
"services/maintenance/image-sweeper-cronjob.yaml",
"services/maintenance/k3s-agent-restart-daemonset.yaml",
"services/maintenance/metis-deployment.yaml",
"services/maintenance/metis-k3s-token-sync-cronjob.yaml",
"services/maintenance/metis-sentinel-amd64-daemonset.yaml",
"services/maintenance/metis-sentinel-arm64-daemonset.yaml",
"services/maintenance/node-image-sweeper-daemonset.yaml",
"services/maintenance/node-nofile-daemonset.yaml",
"services/maintenance/oauth2-proxy-metis.yaml",
"services/maintenance/oauth2-proxy-soteria.yaml",
"services/maintenance/oneoffs/ariadne-migrate-job.yaml",
"services/maintenance/oneoffs/k3s-traefik-cleanup-job.yaml",
"services/maintenance/oneoffs/titan-24-rootfs-sweep-job.yaml",
"services/maintenance/pod-cleaner-cronjob.yaml",
"services/maintenance/soteria-deployment.yaml",
"services/maintenance/vault-sync-deployment.yaml",
"services/monitoring/dcgm-exporter.yaml",
"services/monitoring/jetson-tegrastats-exporter.yaml",
"services/monitoring/oneoffs/grafana-org-bootstrap.yaml",
"services/monitoring/oneoffs/grafana-user-dedupe-job.yaml",
"services/monitoring/platform-quality-gateway-deployment.yaml",
"services/monitoring/platform-quality-suite-probe-cronjob.yaml",
"services/monitoring/postmark-exporter-deployment.yaml",
"services/monitoring/vmalert-atlas-availability.yaml",
"services/monitoring/vault-sync-deployment.yaml",
"services/nextcloud-mail-sync/cronjob.yaml",
"services/nextcloud/collabora.yaml",
"services/nextcloud/cronjob.yaml",
"services/nextcloud/deployment.yaml",
"services/nextcloud/maintenance-cronjob.yaml",
"services/oauth2-proxy/deployment.yaml",
"services/openldap/statefulset.yaml",
"services/outline/deployment.yaml",
"services/outline/redis-deployment.yaml",
"services/pegasus/deployment.yaml",
"services/pegasus/vault-sync-deployment.yaml",
"services/planka/deployment.yaml",
"services/quality/oauth2-proxy-sonarqube.yaml",
"services/quality/sonarqube-deployment.yaml",
"services/quality/sonarqube-exporter-deployment.yaml",
"services/sui-metrics/base/deployment.yaml",
"services/typhon/vault-sync-deployment.yaml",
"services/vault/k8s-auth-config-cronjob.yaml",
"services/vault/oidc-config-cronjob.yaml",
"services/vault/statefulset.yaml",
"services/vaultwarden/deployment.yaml"
]
},
{
"id": "KSV-0017",
"targets": [
"infrastructure/modules/profiles/components/device-plugin-jetson/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-minipc/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-tethys/daemonset.yaml",
"services/logging/node-image-gc-rpi4-daemonset.yaml",
"services/logging/node-image-prune-rpi5-daemonset.yaml",
"services/logging/node-log-rotation-daemonset.yaml",
"services/maintenance/disable-k3s-traefik-daemonset.yaml",
"services/maintenance/image-sweeper-cronjob.yaml",
"services/maintenance/k3s-agent-restart-daemonset.yaml",
"services/maintenance/metis-deployment.yaml",
"services/maintenance/metis-sentinel-amd64-daemonset.yaml",
"services/maintenance/metis-sentinel-arm64-daemonset.yaml",
"services/maintenance/node-image-sweeper-daemonset.yaml",
"services/maintenance/node-nofile-daemonset.yaml",
"services/maintenance/oneoffs/titan-24-rootfs-sweep-job.yaml",
"services/monitoring/dcgm-exporter.yaml",
"services/monitoring/jetson-tegrastats-exporter.yaml"
]
},
{
"id": "KSV-0041",
"targets": [
"infrastructure/cert-manager/cleanup/cert-manager-cleanup-rbac.yaml",
"infrastructure/longhorn/adopt/longhorn-adopt-rbac.yaml",
"infrastructure/traefik/clusterrole.yaml",
"services/bstein-dev-home/rbac.yaml",
"services/comms/comms-secrets-ensure-rbac.yaml",
"services/comms/mas-db-ensure-rbac.yaml",
"services/comms/mas-secrets-ensure-rbac.yaml",
"services/maintenance/soteria-rbac.yaml"
]
},
{
"id": "KSV-0047",
"targets": [
"services/monitoring/rbac.yaml"
]
},
{
"id": "KSV-0053",
"targets": [
"services/comms/comms-secrets-ensure-rbac.yaml",
"services/comms/mas-db-ensure-rbac.yaml",
"services/jenkins/serviceaccount.yaml",
"services/maintenance/ariadne-rbac.yaml"
]
},
{
"id": "KSV-0056",
"targets": [
"infrastructure/cert-manager/cleanup/cert-manager-cleanup-rbac.yaml",
"infrastructure/longhorn/adopt/longhorn-adopt-rbac.yaml",
"services/jenkins/serviceaccount.yaml",
"services/maintenance/disable-k3s-traefik-rbac.yaml",
"services/maintenance/k3s-traefik-cleanup-rbac.yaml"
]
},
{
"id": "KSV-0114",
"targets": [
"infrastructure/cert-manager/cleanup/cert-manager-cleanup-rbac.yaml"
]
},
{
"id": "KSV-0118",
"targets": [
"infrastructure/cert-manager/cleanup/cert-manager-cleanup-job.yaml",
"infrastructure/core/coredns-deployment.yaml",
"infrastructure/core/node-prefer-noschedule-cronjob.yaml",
"infrastructure/core/ntp-sync-daemonset.yaml",
"infrastructure/longhorn/adopt/longhorn-helm-adopt-job.yaml",
"infrastructure/longhorn/core/longhorn-disk-tags-ensure-job.yaml",
"infrastructure/longhorn/core/longhorn-settings-ensure-job.yaml",
"infrastructure/longhorn/core/vault-sync-deployment.yaml",
"infrastructure/longhorn/ui-ingress/oauth2-proxy-longhorn.yaml",
"infrastructure/modules/profiles/components/device-plugin-jetson/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-minipc/daemonset.yaml",
"infrastructure/modules/profiles/components/device-plugin-tethys/daemonset.yaml",
"infrastructure/postgres/statefulset.yaml",
"infrastructure/vault-csi/vault-csi-provider.yaml",
"services/ai-llm/deployment.yaml",
"services/bstein-dev-home/backend-deployment.yaml",
"services/bstein-dev-home/chat-ai-gateway-deployment.yaml",
"services/bstein-dev-home/frontend-deployment.yaml",
"services/bstein-dev-home/oneoffs/migrations/portal-migrate-job.yaml",
"services/bstein-dev-home/oneoffs/portal-onboarding-e2e-test-job.yaml",
"services/bstein-dev-home/vault-sync-deployment.yaml",
"services/bstein-dev-home/vaultwarden-cred-sync-cronjob.yaml",
"services/comms/atlasbot-deployment.yaml",
"services/comms/coturn.yaml",
"services/comms/element-call-deployment.yaml",
"services/comms/guest-name-job.yaml",
"services/comms/livekit-token-deployment.yaml",
"services/comms/livekit.yaml",
"services/comms/mas-deployment.yaml",
"services/comms/oneoffs/bstein-force-leave-job.yaml",
"services/comms/oneoffs/comms-secrets-ensure-job.yaml",
"services/comms/oneoffs/mas-admin-client-secret-ensure-job.yaml",
"services/comms/oneoffs/mas-db-ensure-job.yaml",
"services/comms/oneoffs/mas-local-users-ensure-job.yaml",
"services/comms/oneoffs/othrys-kick-numeric-job.yaml",
"services/comms/oneoffs/synapse-admin-ensure-job.yaml",
"services/comms/oneoffs/synapse-seeder-admin-ensure-job.yaml",
"services/comms/oneoffs/synapse-signingkey-ensure-job.yaml",
"services/comms/oneoffs/synapse-user-seed-job.yaml",
"services/comms/pin-othrys-job.yaml",
"services/comms/reset-othrys-room-job.yaml",
"services/comms/seed-othrys-room.yaml",
"services/comms/vault-sync-deployment.yaml",
"services/comms/wellknown.yaml",
"services/crypto/monerod/deployment.yaml",
"services/crypto/wallet-monero-temp/deployment.yaml",
"services/crypto/xmr-miner/deployment.yaml",
"services/crypto/xmr-miner/vault-sync-deployment.yaml",
"services/crypto/xmr-miner/xmrig-daemonset.yaml",
"services/finance/firefly-cronjob.yaml",
"services/finance/firefly-deployment.yaml",
"services/finance/firefly-user-sync-cronjob.yaml",
"services/finance/oneoffs/finance-secrets-ensure-job.yaml",
"services/gitea/deployment.yaml",
"services/harbor/vault-sync-deployment.yaml",
"services/health/wger-admin-ensure-cronjob.yaml",
"services/health/wger-deployment.yaml",
"services/health/wger-user-sync-cronjob.yaml",
"services/jellyfin/loader.yaml",
"services/jenkins/deployment.yaml",
"services/jenkins/vault-sync-deployment.yaml",
"services/keycloak/oneoffs/actual-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/harbor-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/ldap-federation-job.yaml",
"services/keycloak/oneoffs/logs-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/mas-secrets-ensure-job.yaml",
"services/keycloak/oneoffs/metis-node-passwords-secret-ensure-job.yaml",
"services/keycloak/oneoffs/metis-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/metis-ssh-keys-secret-ensure-job.yaml",
"services/keycloak/oneoffs/portal-admin-client-secret-ensure-job.yaml",
"services/keycloak/oneoffs/portal-e2e-client-job.yaml",
"services/keycloak/oneoffs/portal-e2e-execute-actions-email-test-job.yaml",
"services/keycloak/oneoffs/portal-e2e-target-client-job.yaml",
"services/keycloak/oneoffs/portal-e2e-token-exchange-permissions-job.yaml",
"services/keycloak/oneoffs/portal-e2e-token-exchange-test-job.yaml",
"services/keycloak/oneoffs/quality-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/realm-settings-job.yaml",
"services/keycloak/oneoffs/soteria-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/synapse-oidc-secret-ensure-job.yaml",
"services/keycloak/oneoffs/user-overrides-job.yaml",
"services/keycloak/oneoffs/vault-oidc-secret-ensure-job.yaml",
"services/keycloak/vault-sync-deployment.yaml",
"services/logging/node-image-gc-rpi4-daemonset.yaml",
"services/logging/node-image-prune-rpi5-daemonset.yaml",
"services/logging/node-log-rotation-daemonset.yaml",
"services/logging/oauth2-proxy.yaml",
"services/logging/oneoffs/opensearch-dashboards-setup-job.yaml",
"services/logging/oneoffs/opensearch-ism-job.yaml",
"services/logging/oneoffs/opensearch-observability-setup-job.yaml",
"services/logging/opensearch-prune-cronjob.yaml",
"services/logging/vault-sync-deployment.yaml",
"services/mailu/mailu-sync-cronjob.yaml",
"services/mailu/mailu-sync-listener.yaml",
"services/mailu/oneoffs/mailu-sync-job.yaml",
"services/mailu/vault-sync-deployment.yaml",
"services/mailu/vip-controller.yaml",
"services/maintenance/ariadne-deployment.yaml",
"services/maintenance/disable-k3s-traefik-daemonset.yaml",
"services/maintenance/image-sweeper-cronjob.yaml",
"services/maintenance/k3s-agent-restart-daemonset.yaml",
"services/maintenance/metis-deployment.yaml",
"services/maintenance/metis-k3s-token-sync-cronjob.yaml",
"services/maintenance/metis-sentinel-amd64-daemonset.yaml",
"services/maintenance/metis-sentinel-arm64-daemonset.yaml",
"services/maintenance/node-image-sweeper-daemonset.yaml",
"services/maintenance/node-nofile-daemonset.yaml",
"services/maintenance/oauth2-proxy-metis.yaml",
"services/maintenance/oauth2-proxy-soteria.yaml",
"services/maintenance/oneoffs/ariadne-migrate-job.yaml",
"services/maintenance/oneoffs/k3s-traefik-cleanup-job.yaml",
"services/maintenance/oneoffs/titan-24-rootfs-sweep-job.yaml",
"services/maintenance/pod-cleaner-cronjob.yaml",
"services/maintenance/soteria-deployment.yaml",
"services/maintenance/vault-sync-deployment.yaml",
"services/monitoring/dcgm-exporter.yaml",
"services/monitoring/jetson-tegrastats-exporter.yaml",
"services/monitoring/oneoffs/grafana-org-bootstrap.yaml",
"services/monitoring/oneoffs/grafana-user-dedupe-job.yaml",
"services/monitoring/platform-quality-gateway-deployment.yaml",
"services/monitoring/platform-quality-suite-probe-cronjob.yaml",
"services/monitoring/postmark-exporter-deployment.yaml",
"services/monitoring/vmalert-atlas-availability.yaml",
"services/monitoring/vault-sync-deployment.yaml",
"services/nextcloud/collabora.yaml",
"services/oauth2-proxy/deployment.yaml",
"services/openldap/statefulset.yaml",
"services/outline/deployment.yaml",
"services/outline/redis-deployment.yaml",
"services/pegasus/vault-sync-deployment.yaml",
"services/quality/oauth2-proxy-sonarqube.yaml",
"services/quality/sonarqube-deployment.yaml",
"services/quality/sonarqube-exporter-deployment.yaml",
"services/sui-metrics/base/deployment.yaml",
"services/sui-metrics/overlays/atlas/patch-node-selector.yaml",
"services/typhon/deployment.yaml",
"services/typhon/vault-sync-deployment.yaml",
"services/vault/k8s-auth-config-cronjob.yaml",
"services/vault/oidc-config-cronjob.yaml",
"services/vaultwarden/deployment.yaml"
]
},
{
"id": "KSV-0121",
"targets": [
"services/logging/node-image-gc-rpi4-daemonset.yaml",
"services/logging/node-image-prune-rpi5-daemonset.yaml",
"services/logging/node-log-rotation-daemonset.yaml",
"services/maintenance/disable-k3s-traefik-daemonset.yaml",
"services/maintenance/image-sweeper-cronjob.yaml",
"services/maintenance/metis-deployment.yaml",
"services/maintenance/node-image-sweeper-daemonset.yaml",
"services/maintenance/node-nofile-daemonset.yaml",
"services/maintenance/oneoffs/titan-24-rootfs-sweep-job.yaml"
]
}
]
}

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: ai-llm
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/ai-llm

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: bstein-dev-home-migrations
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/bstein-dev-home/oneoffs/migrations

View File

@ -13,14 +13,14 @@ spec:
git:
checkout:
ref:
branch: main
branch: feature/ariadne
commit:
author:
email: ops@bstein.dev
name: flux-bot
messageTemplate: "chore(bstein-dev-home): automated image update"
push:
branch: main
branch: feature/ariadne
update:
strategy: Setters
path: services/bstein-dev-home

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: bstein-dev-home
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/bstein-dev-home

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: comms
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
prune: true
@ -15,3 +13,5 @@ spec:
path: ./services/comms
targetNamespace: comms
timeout: 2m
dependsOn:
- name: traefik

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: crypto
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/crypto

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: finance
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/finance

View File

@ -1,29 +0,0 @@
# clusters/atlas/flux-system/applications/game-stream/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: game-stream
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/game-stream
targetNamespace: game-stream
prune: true
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
dependsOn:
- name: cert-manager
- name: keycloak
- name: traefik
- name: vault
healthChecks:
- apiVersion: apps/v1
kind: Deployment
name: oauth2-proxy-wolf
namespace: game-stream
wait: false
timeout: 10m

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: gitea
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/gitea

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: harbor
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/harbor

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: health
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/health
@ -17,6 +15,7 @@ spec:
dependsOn:
- name: keycloak
- name: postgres
- name: traefik
- name: vault
healthChecks:
- apiVersion: apps/v1

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: jellyfin
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/jellyfin

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: jenkins
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/jenkins
@ -16,6 +14,7 @@ spec:
targetNamespace: jenkins
dependsOn:
- name: helm
- name: traefik
healthChecks:
- apiVersion: apps/v1
kind: Deployment

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: keycloak
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
prune: true

View File

@ -21,14 +21,10 @@ resources:
- sui-metrics/kustomization.yaml
- openldap/kustomization.yaml
- keycloak/kustomization.yaml
- quality/kustomization.yaml
- oauth2-proxy/kustomization.yaml
- mailu/kustomization.yaml
- jenkins/kustomization.yaml
- ai-llm/kustomization.yaml
- openclaw/kustomization.yaml
- game-stream/kustomization.yaml
- veles/kustomization.yaml
- typhon/kustomization.yaml
- nextcloud/kustomization.yaml
- nextcloud-mail-sync/kustomization.yaml

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: mailu
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: monerod
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/crypto/monerod

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: nextcloud-mail-sync
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
prune: true

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: nextcloud
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/nextcloud

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: oauth2-proxy
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
prune: true

View File

@ -1,34 +0,0 @@
# clusters/atlas/flux-system/applications/openclaw/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: openclaw
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/openclaw
targetNamespace: openclaw
prune: true
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
wait: true
timeout: 30m
healthChecks:
- apiVersion: apps/v1
kind: Deployment
name: openclaw-ollama
namespace: openclaw
- apiVersion: apps/v1
kind: Deployment
name: openclaw
namespace: openclaw
dependsOn:
- name: cert-manager
- name: core
- name: longhorn
- name: traefik

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: openldap
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
prune: true

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: outline
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/outline
@ -17,6 +15,7 @@ spec:
dependsOn:
- name: keycloak
- name: mailu
- name: traefik
healthChecks:
- apiVersion: apps/v1
kind: Deployment

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: pegasus
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/pegasus

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: planka
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/planka
@ -17,6 +15,7 @@ spec:
dependsOn:
- name: keycloak
- name: mailu
- name: traefik
healthChecks:
- apiVersion: apps/v1
kind: Deployment

View File

@ -1,36 +0,0 @@
# clusters/atlas/flux-system/applications/quality/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: quality
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/quality
prune: true
sourceRef:
kind: GitRepository
name: flux-system
targetNamespace: quality
dependsOn:
- name: cert-manager
- name: keycloak
- name: vault
- name: postgres
healthChecks:
- apiVersion: apps/v1
kind: Deployment
name: sonarqube
namespace: quality
- apiVersion: apps/v1
kind: Deployment
name: sonarqube-exporter
namespace: quality
- apiVersion: apps/v1
kind: Deployment
name: oauth2-proxy-sonarqube
namespace: quality
wait: false
timeout: 20m

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: sui-metrics
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/sui-metrics/overlays/atlas

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: typhon
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/typhon

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: vault
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: vaultwarden
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
suspend: false
@ -19,3 +17,4 @@ spec:
wait: true
dependsOn:
- name: helm
- name: traefik

View File

@ -1,29 +0,0 @@
# clusters/atlas/flux-system/applications/veles/image-automation.yaml
# Staged for the first Veles image rollout. Add this file to the parent
# applications kustomization after the namespace exists and the Harbor repos
# have initial tags.
apiVersion: image.toolkit.fluxcd.io/v1
kind: ImageUpdateAutomation
metadata:
name: veles
namespace: veles
spec:
interval: 1m0s
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
git:
checkout:
ref:
branch: main
commit:
author:
email: ops@bstein.dev
name: flux-bot
messageTemplate: "chore(veles): automated image update"
push:
branch: main
update:
strategy: Setters
path: services/veles

View File

@ -1,28 +0,0 @@
# clusters/atlas/flux-system/applications/veles/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: veles
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/veles
targetNamespace: veles
prune: true
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
dependsOn:
- name: cert-manager
- name: core
- name: keycloak
- name: longhorn
- name: traefik
- name: vault
- name: vault-csi
- name: vault-injector
wait: false
timeout: 20m

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: wallet-monero-temp
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/crypto/wallet-monero-temp

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: xmr-miner
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/crypto/xmr-miner

View File

@ -5966,9 +5966,6 @@ spec:
- args:
- --events-addr=http://notification-controller.$(RUNTIME_NAMESPACE).svc.cluster.local./
- --watch-all-namespaces=true
- --concurrent=1
- --requeue-dependency=5s
- --interval-jitter-percentage=30
- --log-level=info
- --log-encoding=json
- --enable-leader-election

View File

@ -7,7 +7,7 @@ metadata:
name: flux-system
namespace: flux-system
spec:
interval: 15m0s
interval: 1m0s
ref:
branch: main
secretRef:
@ -20,7 +20,7 @@ metadata:
name: flux-system
namespace: flux-system
spec:
interval: 1h0m0s
interval: 10m0s
path: ./clusters/atlas/flux-system
prune: true
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: cert-manager-cleanup
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/cert-manager/cleanup

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: cert-manager
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/cert-manager

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: core
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./infrastructure/core

View File

@ -1,21 +0,0 @@
# clusters/atlas/flux-system/platform/descheduler/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: descheduler
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/descheduler
prune: true
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
targetNamespace: kube-system
dependsOn:
- name: helm
- name: core
wait: true

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: gitops-ui
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
timeout: 10m
@ -18,4 +16,5 @@ spec:
targetNamespace: flux-system
dependsOn:
- name: helm
- name: traefik
wait: true

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: helm
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
resources:
- core/kustomization.yaml
- helm/kustomization.yaml
- descheduler/kustomization.yaml
- resource-guardrails/kustomization.yaml
- cert-manager/kustomization.yaml
- metallb/kustomization.yaml
- traefik/kustomization.yaml

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: logging
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/logging

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: longhorn-adopt
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/longhorn/adopt

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: longhorn-ui
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./infrastructure/longhorn/ui-ingress

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: longhorn
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/longhorn/core

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: maintenance
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/maintenance

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: metallb
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: monitoring
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./services/monitoring

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: postgres
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./infrastructure/postgres

View File

@ -1,19 +0,0 @@
# clusters/atlas/flux-system/platform/resource-guardrails/kustomization.yaml
apiVersion: kustomize.toolkit.fluxcd.io/v1
kind: Kustomization
metadata:
name: resource-guardrails
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./infrastructure/resource-guardrails
prune: true
sourceRef:
kind: GitRepository
name: flux-system
namespace: flux-system
dependsOn:
- name: core
wait: true

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: traefik
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 10m
path: ./infrastructure/traefik

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: vault-csi
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
sourceRef:

View File

@ -4,8 +4,6 @@ kind: Kustomization
metadata:
name: vault-injector
namespace: flux-system
annotations:
kustomize.toolkit.fluxcd.io/ssa: IfNotPresent
spec:
interval: 30m
path: ./infrastructure/vault-injector

View File

@ -2,8 +2,4 @@ FROM python:3.11-slim
ENV PIP_DISABLE_PIP_VERSION_CHECK=1
RUN pip install --no-cache-dir requests psycopg2-binary \
&& groupadd --system guest-tools \
&& useradd --system --uid 65532 --gid guest-tools --home-dir /nonexistent --shell /usr/sbin/nologin guest-tools
USER guest-tools
RUN pip install --no-cache-dir requests psycopg2-binary

View File

@ -1,12 +1,15 @@
# Use the mirrored Harbor artifact so CI does not depend on Docker Hub egress.
FROM registry.bstein.dev/streaming/data-prepper@sha256:32ac6ad42e0f12da08bebee307e290b17d127b30def9b06eeaffbcbbc5033e83
FROM --platform=$BUILDPLATFORM opensearchproject/data-prepper:2.8.0 AS source
FROM --platform=$TARGETPLATFORM eclipse-temurin:17-jre
ENV DATA_PREPPER_PATH=/usr/share/data-prepper
USER root
RUN apt-get update \
&& apt-get install -y --no-install-recommends bc \
&& rm -rf /var/lib/apt/lists/*
RUN useradd -u 10001 -M -U -d / -s /usr/sbin/nologin data_prepper \
&& mkdir -p /var/log/data-prepper
COPY --from=source /usr/share/data-prepper /usr/share/data-prepper
RUN chown -R 10001:10001 /usr/share/data-prepper /var/log/data-prepper
USER 10001
WORKDIR /usr/share/data-prepper

View File

@ -1,13 +1,10 @@
FROM ghcr.io/element-hq/lk-jwt-service:0.3.0 AS base
FROM alpine:3.20
RUN apk add --no-cache ca-certificates \
&& addgroup -S livekit-token \
&& adduser -S -D -H -u 65532 -G livekit-token livekit-token
RUN apk add --no-cache ca-certificates
COPY --from=base /lk-jwt-service /lk-jwt-service
COPY dockerfiles/vault-entrypoint.sh /entrypoint.sh
RUN chmod 0755 /entrypoint.sh
USER livekit-token
ENTRYPOINT ["/entrypoint.sh"]
CMD ["/lk-jwt-service"]

View File

@ -29,12 +29,10 @@ FROM ${DEBIAN_IMAGE}
RUN set -eux; \
apt-get update; \
apt-get install -y --no-install-recommends ca-certificates; \
update-ca-certificates; rm -rf /var/lib/apt/lists/*; \
groupadd --system p2pool; \
useradd --system --uid 65532 --gid p2pool --home-dir /nonexistent --shell /usr/sbin/nologin p2pool
update-ca-certificates; rm -rf /var/lib/apt/lists/*
COPY --from=fetch /out/p2pool /usr/local/bin/p2pool
RUN /usr/local/bin/p2pool --version || true
EXPOSE 3333
USER p2pool
ENTRYPOINT ["/usr/local/bin/p2pool"]

View File

@ -26,12 +26,9 @@ RUN set -eux; \
curl -fsSL "$URL" -o /opt/monero/monero.tar.bz2; \
tar -xjf /opt/monero/monero.tar.bz2 -C /opt/monero --strip-components=1; \
install -m 0755 /opt/monero/monero-wallet-rpc /usr/local/bin/monero-wallet-rpc; \
rm -f /opt/monero/monero.tar.bz2; \
groupadd --system monero; \
useradd --system --uid 1000 --gid monero --home-dir /nonexistent --shell /usr/sbin/nologin monero
rm -f /opt/monero/monero.tar.bz2
ENV PATH="/usr/local/bin:/usr/bin:/bin"
RUN /usr/local/bin/monero-wallet-rpc --version || true
EXPOSE 18083
USER monero

View File

@ -23,14 +23,10 @@ RUN set -eux; \
mkdir -p /opt/monero; \
tar -xjf /tmp/monero.tar.bz2 -C /opt/monero --strip-components=1; \
rm -f /tmp/monero.tar.bz2; \
groupadd --system monero; \
useradd --system --uid 1000 --gid monero --home-dir /nonexistent --shell /usr/sbin/nologin monero; \
mkdir -p /data; \
chown monero:monero /data; \
chmod 0770 /data
ENV LD_LIBRARY_PATH=/opt/monero:/opt/monero/lib \
PATH="/opt/monero:${PATH}"
USER monero
CMD ["/opt/monero/monerod", "--version"]

View File

@ -1,13 +1,10 @@
FROM quay.io/oauth2-proxy/oauth2-proxy:v7.6.0 AS base
FROM alpine:3.20
RUN apk add --no-cache ca-certificates \
&& addgroup -S oauth2-proxy \
&& adduser -S -D -H -u 65532 -G oauth2-proxy oauth2-proxy
RUN apk add --no-cache ca-certificates
COPY --from=base /bin/oauth2-proxy /bin/oauth2-proxy
COPY dockerfiles/vault-entrypoint.sh /entrypoint.sh
RUN chmod 0755 /entrypoint.sh
USER oauth2-proxy
ENTRYPOINT ["/entrypoint.sh"]
CMD ["/bin/oauth2-proxy"]

View File

@ -1,13 +1,10 @@
FROM registry.bstein.dev/streaming/pegasus:1.2.32 AS base
FROM alpine:3.20
RUN apk add --no-cache ca-certificates \
&& addgroup -S pegasus \
&& adduser -S -D -H -u 65532 -G pegasus pegasus
RUN apk add --no-cache ca-certificates
COPY --from=base /pegasus /pegasus
COPY dockerfiles/vault-entrypoint.sh /entrypoint.sh
RUN chmod 0755 /entrypoint.sh
USER pegasus
ENTRYPOINT ["/entrypoint.sh"]
CMD ["/pegasus"]

View File

@ -1,48 +0,0 @@
# dockerfiles/Dockerfile.quality-tools
FROM debian:bookworm-slim
ARG SONAR_SCANNER_VERSION=8.0.1.6346
ARG TRIVY_VERSION=0.70.0
ENV TRIVY_CACHE_DIR=/opt/trivy-cache
SHELL ["/bin/bash", "-o", "pipefail", "-c"]
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
bash \
ca-certificates \
curl \
git \
jq \
unzip \
&& rm -rf /var/lib/apt/lists/* \
&& groupadd --system quality-tools \
&& useradd --system --uid 65532 --gid quality-tools --home-dir /nonexistent --shell /usr/sbin/nologin quality-tools
RUN set -eux; \
scanner_zip="sonar-scanner-cli-${SONAR_SCANNER_VERSION}-linux-aarch64.zip"; \
base_url="https://binaries.sonarsource.com/Distribution/sonar-scanner-cli"; \
curl -fsSL "${base_url}/${scanner_zip}" -o "/tmp/${scanner_zip}"; \
curl -fsSL "${base_url}/${scanner_zip}.sha256" -o "/tmp/${scanner_zip}.sha256"; \
printf '%s %s\n' "$(cat "/tmp/${scanner_zip}.sha256")" "/tmp/${scanner_zip}" | sha256sum -c -; \
unzip -q "/tmp/${scanner_zip}" -d /opt; \
ln -s "/opt/sonar-scanner-${SONAR_SCANNER_VERSION}-linux-aarch64/bin/sonar-scanner" /usr/local/bin/sonar-scanner; \
rm -f "/tmp/${scanner_zip}" "/tmp/${scanner_zip}.sha256"
RUN set -eux; \
trivy_tgz="trivy_${TRIVY_VERSION}_Linux-ARM64.tar.gz"; \
curl -fsSL "https://github.com/aquasecurity/trivy/releases/download/v${TRIVY_VERSION}/${trivy_tgz}" -o "/tmp/${trivy_tgz}"; \
tar -C /usr/local/bin -xzf "/tmp/${trivy_tgz}" trivy; \
rm -f "/tmp/${trivy_tgz}"; \
trivy --version; \
sonar-scanner -v
RUN set -eux; \
mkdir -p "${TRIVY_CACHE_DIR}"; \
trivy image --download-db-only --cache-dir "${TRIVY_CACHE_DIR}"; \
chmod -R a+rX "${TRIVY_CACHE_DIR}"; \
mkdir -p /workspace; \
chown quality-tools:quality-tools /workspace
WORKDIR /workspace
USER quality-tools

View File

@ -27,53 +27,12 @@ spec:
timeout: 10m
values:
installCRDs: true
replicaCount: 2
podDisruptionBudget:
enabled: true
minAvailable: 1
extraArgs:
- --acme-http01-solver-nameservers=1.1.1.1:53,8.8.8.8:53
resources:
requests:
cpu: 50m
memory: 128Mi
limits:
cpu: 500m
memory: 512Mi
nodeSelector:
node-role.kubernetes.io/worker: "true"
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: atlas.bstein.dev/spillover
operator: DoesNotExist
- weight: 95
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
- weight: 90
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi5
- weight: 50
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi4
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
@ -83,63 +42,10 @@ spec:
- rpi5
- rpi4
webhook:
replicaCount: 2
podDisruptionBudget:
enabled: true
minAvailable: 1
resources:
requests:
cpu: 100m
memory: 128Mi
limits:
cpu: 500m
memory: 512Mi
livenessProbe:
failureThreshold: 8
initialDelaySeconds: 90
periodSeconds: 10
successThreshold: 1
timeoutSeconds: 5
readinessProbe:
failureThreshold: 8
initialDelaySeconds: 10
periodSeconds: 5
successThreshold: 1
timeoutSeconds: 5
nodeSelector:
node-role.kubernetes.io/worker: "true"
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: atlas.bstein.dev/spillover
operator: DoesNotExist
- weight: 95
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
- weight: 90
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi5
- weight: 50
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi4
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
@ -149,51 +55,10 @@ spec:
- rpi5
- rpi4
cainjector:
replicaCount: 2
podDisruptionBudget:
enabled: true
minAvailable: 1
resources:
requests:
cpu: 50m
memory: 128Mi
limits:
cpu: 500m
memory: 512Mi
nodeSelector:
node-role.kubernetes.io/worker: "true"
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: atlas.bstein.dev/spillover
operator: DoesNotExist
- weight: 95
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
- weight: 90
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi5
- weight: 50
preference:
matchExpressions:
- key: hardware
operator: In
values:
- rpi4
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:

View File

@ -10,7 +10,6 @@ data:
errors
cache 30
hosts {
192.168.22.9 agent.bstein.dev
192.168.22.9 alerts.bstein.dev
192.168.22.9 auth.bstein.dev
192.168.22.9 bstein.dev
@ -29,7 +28,6 @@ data:
192.168.22.9 matrix.live.bstein.dev
192.168.22.9 metrics.bstein.dev
192.168.22.9 monero.bstein.dev
192.168.22.9 moonlight.bstein.dev
10.43.6.87 money.bstein.dev
192.168.22.9 notes.bstein.dev
192.168.22.9 office.bstein.dev
@ -42,7 +40,6 @@ data:
192.168.22.9 secret.bstein.dev
192.168.22.9 sso.bstein.dev
192.168.22.9 stream.bstein.dev
192.168.22.9 wolf.bstein.dev
192.168.22.9 tasks.bstein.dev
192.168.22.9 vault.bstein.dev
fallthrough

View File

@ -4,12 +4,8 @@ kind: Kustomization
resources:
- ../modules/base
- ../modules/profiles/atlas-ha
- node-prefer-noschedule-serviceaccount.yaml
- node-prefer-noschedule-rbac.yaml
- node-prefer-noschedule-cronjob.yaml
- coredns-custom.yaml
- coredns-deployment.yaml
- ntp-sync-daemonset.yaml
- workload-profiles.yaml
- ../sources/cert-manager/letsencrypt.yaml
- ../sources/cert-manager/letsencrypt-prod.yaml

View File

@ -1,80 +0,0 @@
# infrastructure/core/node-prefer-noschedule-cronjob.yaml
apiVersion: batch/v1
kind: CronJob
metadata:
name: node-prefer-noschedule
namespace: kube-system
spec:
schedule: "* * * * *"
concurrencyPolicy: Replace
successfulJobsHistoryLimit: 1
failedJobsHistoryLimit: 3
jobTemplate:
spec:
backoffLimit: 0
template:
spec:
serviceAccountName: node-prefer-noschedule
restartPolicy: Never
containers:
- name: taint
image: bitnami/kubectl@sha256:554ab88b1858e8424c55de37ad417b16f2a0e65d1607aa0f3fe3ce9b9f10b131
command:
- /usr/bin/env
- bash
- -ceu
- |
k() {
kubectl --request-timeout=10s "$@"
}
clear_worker() {
local node="${1}"
local hardware="${2}"
if k get node "${node}" >/dev/null 2>&1; then
k label node "${node}" node-role.kubernetes.io/worker=true "hardware=${hardware}" --overwrite=true || true
k label node "${node}" atlas.bstein.dev/spillover- || true
k taint node "${node}" node.kubernetes.io/unschedulable:NoSchedule- || true
k uncordon "${node}" || true
else
echo "skipping missing node ${node}"
fi
}
clear_worker titan-04 rpi5
clear_worker titan-05 rpi5
clear_worker titan-07 rpi5
clear_worker titan-08 rpi5
clear_worker titan-11 rpi5
clear_worker titan-12 rpi4
clear_worker titan-14 rpi4
clear_worker titan-18 rpi4
clear_worker titan-22 amd64
if k get node titan-22 >/dev/null 2>&1; then
k label node titan-22 atlas.bstein.dev/general-compute=last-resort --overwrite=true || true
fi
if k get node titan-23 >/dev/null 2>&1; then
k label node titan-23 \
veles.bstein.dev/simulation=true \
veles.bstein.dev/node-pool=oceanus \
node-role.kubernetes.io/veles-sim=true \
longhorn-host=true \
hardware=oceanus \
--overwrite=true || true
k label node titan-23 node-role.kubernetes.io/worker- || true
k taint node titan-23 veles.bstein.dev/simulation=true:NoSchedule --overwrite=true || true
else
echo "skipping missing node titan-23"
fi
for node in titan-13 titan-15 titan-17 titan-19; do
if k get node "${node}" >/dev/null 2>&1; then
k label node "${node}" atlas.bstein.dev/spillover=true longhorn-host=true --overwrite=true || true
k taint node "${node}" longhorn=true:PreferNoSchedule --overwrite=true || true
k taint node "${node}" atlas.bstein.dev/spillover=true:PreferNoSchedule --overwrite=true || true
else
echo "skipping missing node ${node}"
fi
done

View File

@ -1,22 +0,0 @@
# infrastructure/core/node-prefer-noschedule-rbac.yaml
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
name: node-prefer-noschedule
rules:
- apiGroups: [""]
resources: ["nodes"]
verbs: ["get", "list", "patch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: node-prefer-noschedule
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: node-prefer-noschedule
subjects:
- kind: ServiceAccount
name: node-prefer-noschedule
namespace: kube-system

View File

@ -1,6 +0,0 @@
# infrastructure/core/node-prefer-noschedule-serviceaccount.yaml
apiVersion: v1
kind: ServiceAccount
metadata:
name: node-prefer-noschedule
namespace: kube-system

View File

@ -1,27 +0,0 @@
# infrastructure/core/workload-profiles.yaml
apiVersion: v1
kind: ConfigMap
metadata:
name: atlas-workload-profiles
namespace: kube-system
data:
profiles.yaml: |
profiles:
tiny:
request: { cpu: 25m, memory: 64Mi }
limit: { cpu: 200m, memory: 256Mi }
light:
request: { cpu: 50m, memory: 128Mi }
limit: { cpu: 500m, memory: 512Mi }
standard:
request: { cpu: 250m, memory: 512Mi }
limit: { cpu: "1", memory: 1Gi }
heavy:
request: { cpu: 500m, memory: 1Gi }
limit: { cpu: 1500m, memory: 3Gi }
ci:
request: { cpu: 512m, memory: 512Mi }
limit: { cpu: 1500m, memory: 2Gi }
scavenger:
request: { cpu: 10m, memory: 32Mi }
limit: { cpu: 250m, memory: 256Mi }

View File

@ -1,97 +0,0 @@
# infrastructure/descheduler/helmrelease.yaml
apiVersion: helm.toolkit.fluxcd.io/v2
kind: HelmRelease
metadata:
name: descheduler
namespace: kube-system
spec:
interval: 30m
install:
remediation:
retries: 3
upgrade:
remediation:
retries: 3
chart:
spec:
chart: descheduler
version: 0.33.0
sourceRef:
kind: HelmRepository
name: descheduler
namespace: flux-system
values:
kind: CronJob
schedule: "*/20 * * * *"
successfulJobsHistoryLimit: 1
failedJobsHistoryLimit: 3
resources:
requests:
cpu: 50m
memory: 96Mi
limits:
cpu: 200m
memory: 256Mi
deschedulerPolicyAPIVersion: descheduler/v1alpha2
deschedulerPolicy:
maxNoOfPodsToEvictPerNode: 2
maxNoOfPodsToEvictPerNamespace: 2
profiles:
- name: atlas-rpi-balance
pluginConfig:
- name: DefaultEvictor
args:
nodeFit: true
minPodAge: 10m
ignorePvcPods: true
evictLocalStoragePods: false
- name: RemovePodsHavingTooManyRestarts
args:
podRestartThreshold: 12
includingInitContainers: true
- name: RemovePodsViolatingNodeAffinity
args:
nodeAffinityType:
- requiredDuringSchedulingIgnoredDuringExecution
- name: RemovePodsViolatingTopologySpreadConstraint
- name: RemovePodsViolatingNodeTaints
- name: LowNodeUtilization
args:
thresholds:
cpu: 45
memory: 45
pods: 45
targetThresholds:
cpu: 75
memory: 75
pods: 75
plugins:
balance:
enabled:
- RemovePodsViolatingTopologySpreadConstraint
- LowNodeUtilization
deschedule:
enabled:
- RemovePodsHavingTooManyRestarts
- RemovePodsViolatingNodeTaints
- RemovePodsViolatingNodeAffinity
priorityClassName: system-cluster-critical
nodeSelector:
node-role.kubernetes.io/control-plane: "true"
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-0a
tolerations:
- key: node-role.kubernetes.io/control-plane
operator: Exists
effect: NoSchedule
- key: node-role.kubernetes.io/master
operator: Exists
effect: NoSchedule

View File

@ -1,5 +0,0 @@
# infrastructure/descheduler/kustomization.yaml
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- helmrelease.yaml

View File

@ -26,9 +26,6 @@ spec:
cleanupOnFail: true
timeout: 15m
values:
global:
nodeSelector:
longhorn-host: "true"
service:
ui:
type: NodePort
@ -81,23 +78,3 @@ spec:
tag: v2.16.0
defaultSettings:
systemManagedPodsImagePullPolicy: Always
taintToleration: veles.bstein.dev/simulation=true:NoSchedule
longhornManager:
tolerations:
- key: veles.bstein.dev/simulation
operator: Equal
value: "true"
effect: NoSchedule
nodeSelector:
longhorn-host: "true"
longhornDriver:
tolerations:
- key: veles.bstein.dev/simulation
operator: Equal
value: "true"
effect: NoSchedule
nodeSelector:
longhorn-host: "true"
longhornUI:
nodeSelector:
longhorn-host: "true"

View File

@ -7,9 +7,7 @@ resources:
- secretproviderclass.yaml
- vault-sync-deployment.yaml
- helmrelease.yaml
- veles-recurring-jobs.yaml
- longhorn-settings-ensure-job.yaml
- longhorn-csi-toleration-ensure-job.yaml
- longhorn-disk-tags-ensure-job.yaml
configMapGenerator:

View File

@ -1,106 +0,0 @@
# infrastructure/longhorn/core/longhorn-csi-toleration-ensure-job.yaml
apiVersion: batch/v1
kind: Job
metadata:
name: longhorn-csi-toleration-ensure-4
namespace: longhorn-system
spec:
backoffLimit: 0
activeDeadlineSeconds: 240
ttlSecondsAfterFinished: 3600
template:
spec:
serviceAccountName: longhorn-service-account
restartPolicy: Never
nodeSelector:
kubernetes.io/hostname: titan-11
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/arch
operator: In
values: ["arm64"]
- key: node-role.kubernetes.io/worker
operator: Exists
containers:
- name: patch
image: bitnami/kubectl@sha256:554ab88b1858e8424c55de37ad417b16f2a0e65d1607aa0f3fe3ce9b9f10b131
command: ["/bin/sh", "-c"]
args:
- |
set -euo pipefail
ns="longhorn-system"
ds="longhorn-csi-plugin"
key="veles.bstein.dev/simulation"
value="true"
effect="NoSchedule"
patch_daemonset() {
target="$1"
current="$(kubectl -n "${ns}" get daemonset "${target}" -o json)"
if echo "${current}" | jq -e \
--arg key "${key}" \
--arg value "${value}" \
--arg effect "${effect}" \
'.spec.template.spec.tolerations[]? | select(.key == $key and .value == $value and .effect == $effect)' >/dev/null; then
echo "${target} already tolerates ${key}=${value}:${effect}"
return 0
fi
patch="$(echo "${current}" | jq -c \
--arg key "${key}" \
--arg value "${value}" \
--arg effect "${effect}" \
'{
spec: {
template: {
spec: {
tolerations: ((.spec.template.spec.tolerations // []) + [
{key: $key, operator: "Equal", value: $value, effect: $effect}
])
}
}
}
}')"
kubectl -n "${ns}" patch daemonset "${target}" --type=merge -p "${patch}"
}
patch_daemonset "${ds}"
engine_daemonsets="$(kubectl -n "${ns}" get daemonset -l longhorn.io/component=engine-image -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')"
for engine_ds in ${engine_daemonsets}; do
patch_daemonset "${engine_ds}"
done
csi_ready="false"
for attempt in $(seq 1 90); do
if kubectl get csinode titan-23 -o json | jq -e '.spec.drivers[]? | select(.name == "driver.longhorn.io")' >/dev/null; then
echo "driver.longhorn.io registered on titan-23"
csi_ready="true"
break
fi
sleep 2
done
if [ "${csi_ready}" != "true" ]; then
echo "driver.longhorn.io did not register on titan-23 before timeout" >&2
exit 1
fi
for engine_ds in ${engine_daemonsets}; do
for attempt in $(seq 1 90); do
if kubectl -n "${ns}" get pods -o json | jq -e \
--arg engine_ds "${engine_ds}" \
'.items[] | select(.spec.nodeName == "titan-23") | select(.metadata.ownerReferences[]?.name == $engine_ds) | select([.status.containerStatuses[]?.ready] | all)' >/dev/null; then
echo "${engine_ds} ready on titan-23"
break
fi
if [ "${attempt}" = "90" ]; then
echo "${engine_ds} did not become ready on titan-23 before timeout" >&2
exit 1
fi
sleep 2
done
done

View File

@ -2,7 +2,7 @@
apiVersion: batch/v1
kind: Job
metadata:
name: longhorn-disk-tags-ensure-3
name: longhorn-disk-tags-ensure-1
namespace: longhorn-system
spec:
backoffLimit: 0

View File

@ -2,18 +2,15 @@
apiVersion: batch/v1
kind: Job
metadata:
name: longhorn-settings-ensure-10
name: longhorn-settings-ensure-4
namespace: longhorn-system
spec:
backoffLimit: 0
activeDeadlineSeconds: 240
ttlSecondsAfterFinished: 3600
template:
spec:
serviceAccountName: longhorn-service-account
restartPolicy: Never
nodeSelector:
kubernetes.io/hostname: titan-11
volumes:
- name: longhorn-settings-ensure-script
configMap:

View File

@ -17,28 +17,10 @@ import urllib.request
LONGHORN_NS = "longhorn-system"
LONGHORN_API = "/apis/longhorn.io/v1beta2/namespaces/{namespace}/nodes"
DESIRED_DISK_TAGS = {
"/mnt/astreae": ["astreae"],
"/mnt/asteria": ["asteria"],
"/mnt/veles": ["veles-oceanus", "veles-db", "veles-artifacts"],
"/mnt/veles-db": ["veles-oceanus", "veles-db"],
"/mnt/veles-artifacts": ["veles-oceanus", "veles-artifacts"],
DESIRED_TAGS = {
"/mnt/astreae": "astreae",
"/mnt/asteria": "asteria",
}
DESIRED_NODE_TAGS = {
"titan-23": ["veles-oceanus"],
}
DESIRED_NODE_DISKS = {
"titan-23": {
"veles-oceanus": {
"path": "/mnt/veles",
"allowScheduling": True,
"evictionRequested": False,
"storageReserved": 0,
"tags": ["veles-oceanus", "veles-db", "veles-artifacts"],
}
}
}
DISABLE_DEFAULT_DISK_NODES = {"titan-23"}
def api_base() -> str:
@ -81,30 +63,8 @@ def list_nodes() -> list[dict]:
return data.get("items", [])
def merged_tags(current_tags: list[str], desired_tags: list[str]) -> list[str]:
return sorted(dict.fromkeys([*current_tags, *desired_tags]))
def patch_node_tags(node_name: str, desired_tags: list[str]) -> None:
body = {"spec": {"tags": desired_tags}}
request_json(
"PATCH",
f"{LONGHORN_API.format(namespace=LONGHORN_NS)}/{node_name}",
body=body,
)
def patch_disk_tags(node_name: str, disk_name: str, desired_tags: list[str]) -> None:
body = {"spec": {"disks": {disk_name: {"tags": desired_tags}}}}
request_json(
"PATCH",
f"{LONGHORN_API.format(namespace=LONGHORN_NS)}/{node_name}",
body=body,
)
def patch_disks(node_name: str, disks: dict) -> None:
body = {"spec": {"disks": disks}}
def patch_disk_tags(node_name: str, disk_name: str, desired_tag: str) -> None:
body = {"spec": {"disks": {disk_name: {"tags": [desired_tag]}}}}
request_json(
"PATCH",
f"{LONGHORN_API.format(namespace=LONGHORN_NS)}/{node_name}",
@ -118,52 +78,18 @@ def main() -> int:
for node in list_nodes():
name = node.get("metadata", {}).get("name", "")
desired_node_tags = DESIRED_NODE_TAGS.get(name)
if desired_node_tags:
current_node_tags = node.get("spec", {}).get("tags") or []
next_node_tags = merged_tags(current_node_tags, desired_node_tags)
if current_node_tags != next_node_tags:
print(f"patching {name} node tags={current_node_tags!r} -> {next_node_tags!r}")
patch_node_tags(name, next_node_tags)
changed += 1
else:
skipped += 1
spec_disks = node.get("spec", {}).get("disks", {}) or {}
desired_disks = DESIRED_NODE_DISKS.get(name, {})
missing_disks = {
disk_name: disk_spec
for disk_name, disk_spec in desired_disks.items()
if disk_name not in spec_disks
}
if missing_disks:
print(f"adding {name} disks={sorted(missing_disks)}")
patch_disks(name, missing_disks)
changed += len(missing_disks)
spec_disks = {**spec_disks, **missing_disks}
if name in DISABLE_DEFAULT_DISK_NODES:
disable_patch = {}
for disk_name, disk in spec_disks.items():
disk_path = (disk.get("path") or "").rstrip("/")
if disk_path == "/var/lib/longhorn" and disk.get("allowScheduling", True):
disable_patch[disk_name] = {"allowScheduling": False}
if disable_patch:
print(f"disabling default Longhorn scheduling on {name} disks={sorted(disable_patch)}")
patch_disks(name, disable_patch)
changed += len(disable_patch)
for disk_name, disk in spec_disks.items():
disk_path = disk.get("path")
desired_disk_tags = DESIRED_DISK_TAGS.get(disk_path)
if not desired_disk_tags:
desired_tag = DESIRED_TAGS.get(disk_path)
if not desired_tag:
continue
current_tags = disk.get("tags") or []
if current_tags == desired_disk_tags:
if current_tags == [desired_tag]:
skipped += 1
continue
print(f"patching {name}:{disk_name} path={disk_path} tags={current_tags!r} -> {desired_disk_tags!r}")
patch_disk_tags(name, disk_name, desired_disk_tags)
print(f"patching {name}:{disk_name} path={disk_path} tags={current_tags!r} -> {[desired_tag]!r}")
patch_disk_tags(name, disk_name, desired_tag)
changed += 1
print(f"done: changed={changed} skipped={skipped}")

View File

@ -4,12 +4,11 @@ set -eu
# Longhorn blocks direct CR patches for some settings; use the internal API instead.
api_base="http://longhorn-backend.longhorn-system.svc:9500/v1/settings"
curl_opts="-fsS --connect-timeout 3 --max-time 15"
wait_for_api() {
attempts=30
while [ "${attempts}" -gt 0 ]; do
if curl ${curl_opts} "${api_base}" >/dev/null 2>&1; then
if curl -fsS "${api_base}" >/dev/null 2>&1; then
return 0
fi
attempts=$((attempts - 1))
@ -23,32 +22,17 @@ update_setting() {
name="$1"
value="$2"
current="$(curl ${curl_opts} "${api_base}/${name}" || true)"
current="$(curl -fsS "${api_base}/${name}" || true)"
if echo "${current}" | grep -Fq "\"value\":\"${value}\""; then
echo "Setting ${name} already set."
return 0
fi
echo "Setting ${name} -> ${value}"
out="$(mktemp)"
if curl ${curl_opts} -o "${out}" -X PUT \
curl -fsS -X PUT \
-H "Content-Type: application/json" \
-d "{\"value\":\"${value}\"}" \
"${api_base}/${name}"; then
rm -f "${out}"
return 0
fi
current="$(curl ${curl_opts} "${api_base}/${name}" || true)"
if echo "${current}" | grep -Fq "\"value\":\"${value}\""; then
echo "Setting ${name} stored; Longhorn will apply it when current state allows."
rm -f "${out}"
return 0
fi
cat "${out}" >&2 || true
rm -f "${out}"
return 1
"${api_base}/${name}" >/dev/null
}
wait_for_api
@ -56,8 +40,3 @@ update_setting default-engine-image "registry.bstein.dev/infra/longhorn-engine:v
update_setting default-instance-manager-image "registry.bstein.dev/infra/longhorn-instance-manager:v1.8.2"
update_setting default-backing-image-manager-image "registry.bstein.dev/infra/longhorn-backing-image-manager:v1.8.2"
update_setting support-bundle-manager-image "registry.bstein.dev/infra/longhorn-support-bundle-kit:v0.0.56"
update_setting taint-toleration "veles.bstein.dev/simulation=true:NoSchedule"
# Keep storage-heavy nodes from getting hammered by rebuild storms and skew.
update_setting replica-auto-balance "best-effort"
update_setting concurrent-replica-rebuild-per-node-limit "2"
update_setting node-down-pod-deletion-policy "delete-both-statefulset-and-deployment-pod"

View File

@ -26,16 +26,6 @@ spec:
- key: hardware
operator: In
values: ["rpi5", "rpi4"]
- weight: 90
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- titan-13
- titan-15
- titan-17
- titan-19
containers:
- name: sync
image: alpine:3.20

View File

@ -1,60 +0,0 @@
# infrastructure/longhorn/core/veles-recurring-jobs.yaml
apiVersion: longhorn.io/v1beta2
kind: RecurringJob
metadata:
name: veles-postgres-backup
namespace: longhorn-system
spec:
name: veles-postgres-backup
cron: "30 5 * * *"
task: backup
groups:
- veles
- veles-postgres
retain: 7
concurrency: 1
---
apiVersion: longhorn.io/v1beta2
kind: RecurringJob
metadata:
name: veles-postgres-snapshot
namespace: longhorn-system
spec:
name: veles-postgres-snapshot
cron: "*/30 * * * *"
task: snapshot
groups:
- veles
- veles-postgres
retain: 8
concurrency: 1
---
apiVersion: longhorn.io/v1beta2
kind: RecurringJob
metadata:
name: veles-artifacts-backup
namespace: longhorn-system
spec:
name: veles-artifacts-backup
cron: "45 5 * * *"
task: backup
groups:
- veles
- veles-artifacts
retain: 7
concurrency: 1
---
apiVersion: longhorn.io/v1beta2
kind: RecurringJob
metadata:
name: veles-artifacts-snapshot
namespace: longhorn-system
spec:
name: veles-artifacts-snapshot
cron: "15 */6 * * *"
task: snapshot
groups:
- veles
- veles-artifacts
retain: 8
concurrency: 1

View File

@ -3,4 +3,3 @@ apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- scavenger.yaml
- veles.yaml

View File

@ -1,17 +0,0 @@
# infrastructure/modules/base/priorityclass/veles.yaml
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
name: veles-core
value: 500
globalDefault: false
description: "For Veles core database, API, and controller workloads"
---
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
name: veles-sim
value: 50
globalDefault: false
preemptionPolicy: Never
description: "For Veles simulation jobs; lower than core and non-preempting"

View File

@ -5,6 +5,3 @@ resources:
- asteria.yaml
- asteria-encrypted.yaml
- astreae.yaml
- veles-oceanus-db.yaml
- veles-oceanus-artifacts.yaml
- veles-oceanus-policy.yaml

View File

@ -1,21 +0,0 @@
# infrastructure/modules/base/storageclass/veles-oceanus-artifacts.yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: veles-oceanus-artifacts
annotations:
veles.bstein.dev/allowed-namespace: veles
provisioner: driver.longhorn.io
parameters:
nodeSelector: veles-oceanus
diskSelector: veles-oceanus,veles-artifacts
fromBackup: ""
numberOfReplicas: "1"
staleReplicaTimeout: "30"
fsType: ext4
replicaAutoBalance: disabled
dataLocality: strict-local
recurringJobSelector: '[{"name":"veles-artifacts-backup","isGroup":false},{"name":"veles-artifacts-snapshot","isGroup":false}]'
reclaimPolicy: Retain
allowVolumeExpansion: true
volumeBindingMode: WaitForFirstConsumer

View File

@ -1,21 +0,0 @@
# infrastructure/modules/base/storageclass/veles-oceanus-db.yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: veles-oceanus-db
annotations:
veles.bstein.dev/allowed-namespace: veles
provisioner: driver.longhorn.io
parameters:
nodeSelector: veles-oceanus
diskSelector: veles-oceanus,veles-db
fromBackup: ""
numberOfReplicas: "1"
staleReplicaTimeout: "30"
fsType: ext4
replicaAutoBalance: disabled
dataLocality: strict-local
recurringJobSelector: '[{"name":"veles-postgres-backup","isGroup":false},{"name":"veles-postgres-snapshot","isGroup":false}]'
reclaimPolicy: Retain
allowVolumeExpansion: true
volumeBindingMode: WaitForFirstConsumer

View File

@ -1,25 +0,0 @@
# infrastructure/modules/base/storageclass/veles-oceanus-policy.yaml
apiVersion: admissionregistration.k8s.io/v1
kind: ValidatingAdmissionPolicy
metadata:
name: veles-oceanus-storage-namespace
spec:
failurePolicy: Fail
matchConstraints:
resourceRules:
- apiGroups: [""]
apiVersions: ["v1"]
operations: ["CREATE", "UPDATE"]
resources: ["persistentvolumeclaims"]
validations:
- expression: "!has(object.spec.storageClassName) || !(object.spec.storageClassName in ['veles-oceanus-db', 'veles-oceanus-artifacts']) || object.metadata.namespace == 'veles'"
message: "Veles Oceanus storage classes are reserved for namespace veles"
---
apiVersion: admissionregistration.k8s.io/v1
kind: ValidatingAdmissionPolicyBinding
metadata:
name: veles-oceanus-storage-namespace
spec:
policyName: veles-oceanus-storage-namespace
validationActions:
- Deny

View File

@ -25,7 +25,6 @@ spec:
serviceAccountName: postgres-vault
nodeSelector:
node-role.kubernetes.io/worker: "true"
hardware: rpi5
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
@ -36,17 +35,7 @@ spec:
values: ["true"]
- key: hardware
operator: In
values: ["rpi5"]
- key: kubernetes.io/hostname
operator: NotIn
values: ["titan-06"]
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
preference:
matchExpressions:
- key: kubernetes.io/hostname
operator: In
values: ["titan-05", "titan-07", "titan-08", "titan-11"]
values: ["rpi4", "rpi5"]
containers:
- name: postgres
image: postgres:15

View File

@ -1,5 +0,0 @@
# infrastructure/resource-guardrails/kustomization.yaml
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- limitranges.yaml

View File

@ -1,188 +0,0 @@
# infrastructure/resource-guardrails/limitranges.yaml
apiVersion: v1
kind: List
items:
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: ai
spec: &defaultCompute
limits:
- type: Container
defaultRequest:
cpu: 50m
memory: 96Mi
default:
cpu: 500m
memory: 512Mi
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: bstein-dev-home
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: cert-manager
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: climate
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: comms
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: crypto
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: finance
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: gitea
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: harbor
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: health
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: jellyfin
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: jenkins
spec:
limits:
- type: Container
defaultRequest:
cpu: 25m
memory: 256Mi
default:
cpu: 1500m
memory: 2Gi
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: logging
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: mailu-mailserver
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: maintenance
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: metallb-system
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: monitoring
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: nextcloud
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: outline
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: openclaw
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: planka
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: postgres
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: quality
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: sso
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: sui-metrics
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: traefik
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: vault
spec: *defaultCompute
- apiVersion: v1
kind: LimitRange
metadata:
name: atlas-default-compute
namespace: vaultwarden
spec: *defaultCompute

View File

@ -1,9 +0,0 @@
# infrastructure/sources/helm/descheduler.yaml
apiVersion: source.toolkit.fluxcd.io/v1
kind: HelmRepository
metadata:
name: descheduler
namespace: flux-system
spec:
interval: 1h
url: https://kubernetes-sigs.github.io/descheduler/

Some files were not shown because too many files have changed in this diff Show More