fix: expose hwlab argo closeout diagnostics

2026-07-04 10:26:32 +00:00
parent 419ed4519b
commit 519abbdce1
2 changed files with 230 additions and 6 deletions
@@ -226,6 +226,9 @@ export function nodeRuntimeControlPlaneStatus(scoped: ReturnType<typeof parseNod
              ? codeAgentRuntimeDegradedReason ?? "code-agent-runtime-not-ready"
              : "runtime-not-ready";
  const argoReady = argo.exitCode === 0 && repoURL === spec.argoRepoUrl && targetRevision === spec.gitopsBranch && path === spec.runtimePath && syncStatus === "Synced" && health === "Healthy";
+  const argoDiagnostics = argo.exitCode === 0 && !argoReady
+    ? nodeRuntimeArgoDiagnostics(spec, probeTimeoutSeconds)
+    : null;
  const pipelineRunReady = pipelineRunProbe !== null && pipelineRunProbe.status === "True";
  const pipelineRunDegradedReason = typeof pipelineRunDiagnostics?.degradedReason === "string"
    ? pipelineRunDiagnostics.degradedReason
@@ -296,6 +299,7 @@ export function nodeRuntimeControlPlaneStatus(scoped: ReturnType<typeof parseNod
      targetGitopsRevision,
      syncStatus,
      health,
+      diagnostics: argoDiagnostics,
      result: compactRuntimeCommand(argo),
    },
    pipelineRun: pipelineRunProbe,
@@ -649,6 +653,7 @@ export function summarizeNodeRuntimeControlPlaneStatus(status: Record<string, un
      revisionObserved: typeof argo.targetGitopsRevision === "string" && argo.syncRevision === argo.targetGitopsRevision,
      syncStatus: argo.syncStatus ?? null,
      health: argo.health ?? null,
+      diagnostics: compactArgoDiagnostics(record(argo.diagnostics)),
    },
    runtime: {
      namespace: runtime.namespace ?? null,
@@ -716,6 +721,120 @@ export function summarizeNodeRuntimeControlPlaneStatus(status: Record<string, un
  };
 }

+function nodeRuntimeArgoDiagnostics(spec: HwlabRuntimeLaneSpec, timeoutSeconds: number): Record<string, unknown> {
+  const resourceTemplate = `{{range .status.resources}}{{.kind}}{{"\\t"}}{{.namespace}}{{"\\t"}}{{.name}}{{"\\t"}}{{.status}}{{"\\t"}}{{with .health}}{{.status}}{{"\\t"}}{{printf "%.500s" .message}}{{else}}{{"\\t"}}{{end}}{{"\\n"}}{{end}}`;
+  const operationTemplate = `{{with .status.operationState}}{{.phase}}{{"\\t"}}{{printf "%.500s" .message}}{{"\\t"}}{{.startedAt}}{{"\\t"}}{{.finishedAt}}{{"\\t"}}{{with .syncResult}}{{.revision}}{{"\\t"}}{{with .source}}{{.repoURL}}{{end}}{{end}}{{"\\n"}}{{with .syncResult}}{{range .resources}}{{.group}}{{"\\t"}}{{.kind}}{{"\\t"}}{{.namespace}}{{"\\t"}}{{.name}}{{"\\t"}}{{.status}}{{"\\t"}}{{printf "%.500s" .message}}{{"\\t"}}{{.hookPhase}}{{"\\t"}}{{.syncPhase}}{{"\\n"}}{{end}}{{end}}{{end}}`;
+  const conditionTemplate = `{{range .status.conditions}}{{.type}}{{"\\t"}}{{printf "%.500s" .message}}{{"\\t"}}{{.lastTransitionTime}}{{"\\n"}}{{end}}`;
+  const eventTemplate = `{{range .items}}{{.type}}{{"\\t"}}{{.reason}}{{"\\t"}}{{printf "%.500s" .message}}{{"\\t"}}{{.count}}{{"\\t"}}{{.lastTimestamp}}{{"\\n"}}{{end}}`;
+  const resourceResult = runNodeK3sArgs(spec, ["kubectl", "-n", "argocd", "get", "application", spec.app, "-o", `go-template=${resourceTemplate}`], timeoutSeconds);
+  const operationResult = runNodeK3sArgs(spec, ["kubectl", "-n", "argocd", "get", "application", spec.app, "-o", `go-template=${operationTemplate}`], timeoutSeconds);
+  const conditionResult = runNodeK3sArgs(spec, ["kubectl", "-n", "argocd", "get", "application", spec.app, "-o", `go-template=${conditionTemplate}`], timeoutSeconds);
+  const eventsResult = runNodeK3sArgs(spec, ["kubectl", "-n", "argocd", "get", "events", "--field-selector", `involvedObject.name=${spec.app}`, "--sort-by=.lastTimestamp", "-o", `go-template=${eventTemplate}`], timeoutSeconds);
+  const resources = argoResourceRows(resourceResult.stdout);
+  const problemResources = resources.filter((item) => {
+    const sync = typeof item.status === "string" ? item.status : null;
+    const health = typeof item.healthStatus === "string" ? item.healthStatus : null;
+    return (sync !== null && sync !== "Synced") || (health !== null && health !== "Healthy");
+  }).slice(0, 12);
+  const operation = argoOperationRows(operationResult.stdout);
+  const conditions = argoConditionRows(conditionResult.stdout).slice(-8);
+  const events = argoEventRows(eventsResult.stdout).slice(-8);
+  return {
+    ok: resourceResult.exitCode === 0 && operationResult.exitCode === 0 && conditionResult.exitCode === 0,
+    application: spec.app,
+    resourceCount: resources.length,
+    problemResourceCount: problemResources.length,
+    problemResources,
+    operationState: operation.state,
+    operationResources: operation.resources.slice(0, 12),
+    conditions,
+    events,
+    result: compactRuntimeCommand(operationResult),
+    resourcesResult: compactRuntimeCommand(resourceResult),
+    conditionsResult: compactRuntimeCommand(conditionResult),
+    eventsResult: compactRuntimeCommand(eventsResult),
+    valuesPrinted: false,
+  };
+}
+
+function compactArgoDiagnostics(diagnostics: Record<string, unknown>): Record<string, unknown> | null {
+  if (Object.keys(diagnostics).length === 0) return null;
+  return {
+    ok: diagnostics.ok === true,
+    problemResourceCount: diagnostics.problemResourceCount ?? null,
+    problemResources: Array.isArray(diagnostics.problemResources) ? diagnostics.problemResources.slice(0, 8) : [],
+    operationState: diagnostics.operationState ?? null,
+    operationResources: Array.isArray(diagnostics.operationResources) ? diagnostics.operationResources.slice(0, 8) : [],
+    conditions: Array.isArray(diagnostics.conditions) ? diagnostics.conditions.slice(0, 6) : [],
+    events: Array.isArray(diagnostics.events) ? diagnostics.events.slice(0, 6) : [],
+    valuesPrinted: false,
+  };
+}
+
+function shortDiagnosticText(value: unknown): string | null {
+  if (typeof value !== "string" || value.length === 0) return null;
+  return webObserveShort(value.replace(/\s+/gu, " ").trim(), 500);
+}
+
+function argoResourceRows(text: string): Record<string, unknown>[] {
+  return text.split(/\r?\n/u).map((line) => {
+    const [kind = "", namespace = "", name = "", status = "", healthStatus = "", healthMessage = ""] = line.split("\t");
+    if (kind.length === 0 && name.length === 0) return null;
+    return {
+      kind: kind || null,
+      namespace: namespace || null,
+      name: name || null,
+      status: status || null,
+      healthStatus: healthStatus || null,
+      healthMessage: shortDiagnosticText(healthMessage),
+    };
+  }).filter((item): item is Record<string, unknown> => item !== null);
+}
+
+function argoOperationRows(text: string): { state: Record<string, unknown>; resources: Record<string, unknown>[] } {
+  const lines = text.split(/\r?\n/u).filter((line) => line.length > 0);
+  const [phase = "", message = "", startedAt = "", finishedAt = "", syncResultRevision = "", syncResultSource = ""] = (lines.shift() ?? "").split("\t");
+  return {
+    state: {
+      phase: phase || null,
+      message: shortDiagnosticText(message),
+      startedAt: startedAt || null,
+      finishedAt: finishedAt || null,
+      syncResultRevision: syncResultRevision || null,
+      syncResultSource: syncResultSource || null,
+    },
+    resources: lines.map((line) => {
+      const [group = "", kind = "", namespace = "", name = "", status = "", resourceMessage = "", hookPhase = "", syncPhase = ""] = line.split("\t");
+      return {
+        group: group || null,
+        kind: kind || null,
+        namespace: namespace || null,
+        name: name || null,
+        status: status || null,
+        message: shortDiagnosticText(resourceMessage),
+        hookPhase: hookPhase || null,
+        syncPhase: syncPhase || null,
+      };
+    }).filter((item) => item.status !== "Synced" || item.message !== null || item.hookPhase !== null),
+  };
+}
+
+function argoConditionRows(text: string): Record<string, unknown>[] {
+  return text.split(/\r?\n/u).map((line) => {
+    const [type = "", message = "", lastTransitionTime = ""] = line.split("\t");
+    if (type.length === 0) return null;
+    return { type, message: shortDiagnosticText(message), lastTransitionTime: lastTransitionTime || null };
+  }).filter((item): item is Record<string, unknown> => item !== null);
+}
+
+function argoEventRows(text: string): Record<string, unknown>[] {
+  return text.split(/\r?\n/u).map((line) => {
+    const [type = "", reason = "", message = "", count = "", lastTimestamp = ""] = line.split("\t");
+    if (type.length === 0 && reason.length === 0) return null;
+    return { type: type || null, reason: reason || null, message: shortDiagnosticText(message), count: numericField(count), lastTimestamp: lastTimestamp || null };
+  }).filter((item): item is Record<string, unknown> => item !== null);
+}
+
 export function nodeRuntimeStatusNextAction(status: Record<string, unknown>, scoped: ReturnType<typeof parseNodeScopedDelegatedOptions>): string {
  const reason = typeof status.degradedReason === "string" ? status.degradedReason : null;
  if (reason === null) return `${nodeRuntimeStatusCommand(scoped)} --full`;
@@ -1521,6 +1640,15 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "  }",
    "  return changed;",
    "}",
+    "function isPrometheusOperatorResource(item) { return item && item.apiVersion && String(item.apiVersion).startsWith('monitoring.coreos.com/') && ['ServiceMonitor', 'PrometheusRule', 'PodMonitor', 'Probe'].includes(item.kind); }",
+    "function stripPrometheusOperatorResources(doc) {",
+    "  if (!(overlay.observability && overlay.observability.prometheusOperator === false)) return { docs: [doc], changed: false };",
+    "  if (doc && doc.kind === 'List' && Array.isArray(doc.items)) {",
+    "    const items = doc.items.filter((item) => !isPrometheusOperatorResource(item));",
+    "    return { docs: items.length > 0 ? [{ ...doc, items }] : [], changed: items.length !== doc.items.length };",
+    "  }",
+    "  return isPrometheusOperatorResource(doc) ? { docs: [], changed: true } : { docs: [doc], changed: false };",
+    "}",
    "function containerHasVolumeMount(container, name) { return isObject(container) && Array.isArray(container.volumeMounts) && container.volumeMounts.some((mount) => mount && mount.name === name); }",
    "function removeMetricsSidecar(podSpec) {",
    "  if (!isObject(podSpec)) return false;",
@@ -1675,8 +1803,14 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "    if (path.basename(file) === 'kustomization.yaml') continue;",
    "    const docs = readYamlDocuments(file);",
    "    let changed = false;",
+    "    const nextDocs = [];",
    "    for (const doc of docs) {",
-    "      for (const item of listItems(doc).filter(Boolean)) {",
+    "      const stripped = stripPrometheusOperatorResources(doc);",
+    "      changed = stripped.changed || changed;",
+    "      observabilityChanged = observabilityChanged || stripped.changed;",
+    "      for (const nextDoc of stripped.docs) {",
+    "        nextDocs.push(nextDoc);",
+    "        for (const item of listItems(nextDoc).filter(Boolean)) {",
    "        if (!isObject(item)) continue;",
    "        if (overlay.observability && overlay.observability.prometheusOperator === false) {",
    "          const metadataChanged = stripMonitoringMetadata(item.metadata);",
@@ -1702,8 +1836,12 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "        codeAgentRuntimeChanged = codeAgentRuntimeChanged || envChanged.codeAgentRuntimeChanged;",
    "        cloudWebRuntimeChanged = cloudWebRuntimeChanged || envChanged.cloudWebRuntimeChanged;",
    "      }",
+    "      }",
+    "    }",
+    "    if (changed) {",
+    "      if (nextDocs.length === 0) fs.rmSync(file, { force: true });",
+    "      else writeYamlDocuments(file, nextDocs);",
    "    }",
-    "    if (changed) writeYamlDocuments(file, docs);",
    "  }",
    "  return { observabilityChanged, startupProbeChanged, imageRewriteChanged, gitReadUrlChanged, publicEndpointChanged, dbSslModeChanged, codeAgentRuntimeChanged, cloudWebRuntimeChanged };",
    "}",
@@ -1712,7 +1850,12 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "  if (!fs.existsSync(file)) return false;",
    "  const doc = readYaml(file) || {};",
    "  const resources = Array.isArray(doc.resources) ? doc.resources : [];",
-    "  const next = resources.filter((item) => !(overlay.observability && overlay.observability.prometheusOperator === false && item === 'observability.yaml'));",
+    "  const next = resources.filter((item) => {",
+    "    if (!(overlay.observability && overlay.observability.prometheusOperator === false)) return true;",
+    "    const resource = String(item);",
+    "    if (resource === 'observability.yaml') return false;",
+    "    return !(/\\.ya?ml$/u.test(resource) && !fs.existsSync(path.join(runtimePath, resource)));",
+    "  });",
    "  let changed = false;",
    "  if (next.length !== resources.length) { doc.resources = next; writeYaml(file, doc); changed = true; }",
    "  const observabilityFile = path.join(runtimePath, 'observability.yaml');",
@@ -1877,8 +2020,8 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "  console.error(JSON.stringify({ event: 'unidesk-public-exposure-postprocess', ok: true, applied: true, changed, filePath: file, hostname: exposure.hostname, serverAddr: exposure.serverAddr, serverPort: exposure.serverPort, webProxy: exposure.webProxy.name, apiProxy: exposure.apiProxy.name, extraProxyCount: Array.isArray(exposure.extraProxies) ? exposure.extraProxies.length : 0, configSha256: tomlSha256 }));",
    "  return { configured: true, changed, foundConfigMap, foundDeployment };",
    "}",
-    "const kustomizationChanged = patchKustomization();",
    "const runtimeWorkloadsChanged = patchRuntimeWorkloads();",
+    "const kustomizationChanged = patchKustomization();",
    "const externalPostgresChanged = patchExternalPostgres();",
    "const healthContractChanged = patchHealthContract();",
    "const publicExposureChanged = patchPublicExposure();",
@@ -1958,8 +2101,10 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "  if (!exposure || !Array.isArray(exposure.extraProxies)) return [];",
    "  return exposure.extraProxies.filter((proxy) => proxy && proxy.cloudWebEnvName && proxy.publicBaseUrl).map((proxy) => ({ name: String(proxy.cloudWebEnvName), value: String(proxy.publicBaseUrl) }));",
    "}",
+    "function isPrometheusOperatorResource(item) { return item && item.apiVersion && String(item.apiVersion).startsWith('monitoring.coreos.com/') && ['ServiceMonitor', 'PrometheusRule', 'PodMonitor', 'Probe'].includes(item.kind); }",
    "function workloadRef(item, file, container) { return { file, kind: item && item.kind, name: item && item.metadata && item.metadata.name, container: container && container.name }; }",
    "function workloadChecks() {",
+    "  const monitoringResources = [];",
    "  const metricsRefs = [];",
    "  const missingStartupProbes = [];",
    "  const publicRuntimeImages = [];",
@@ -1987,6 +2132,7 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "    if (path.basename(file) === 'kustomization.yaml') continue;",
    "    for (const doc of readYamlDocuments(file)) {",
    "      for (const item of listItems(doc).filter(Boolean)) {",
+    "        if (isPrometheusOperatorResource(item)) monitoringResources.push(workloadRef(item, file, null));",
    "        const podSpec = podSpecFor(item);",
    "        if (!isObject(podSpec)) continue;",
    "        for (const container of Array.isArray(podSpec.containers) ? podSpec.containers : []) {",
@@ -2026,7 +2172,7 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "      }",
    "    }",
    "  }",
-    "  return { metricsRefs, missingStartupProbes, publicRuntimeImages, staleGitReadUrls, wrongPublicEndpoints, wrongDbSslModes, wrongCodeAgentRuntimeEnvs, wrongCloudWebRuntimeEnvs };",
+    "  return { monitoringResources, metricsRefs, missingStartupProbes, publicRuntimeImages, staleGitReadUrls, wrongPublicEndpoints, wrongDbSslModes, wrongCodeAgentRuntimeEnvs, wrongCloudWebRuntimeEnvs };",
    "}",
    "const checks = [];",
    "const workloadCheck = workloadChecks();",
@@ -2035,6 +2181,7 @@ export function nodeRuntimePipelinePostprocessScript(): string[] {
    "  if (!fs.existsSync(kustomizationPath)) fail('kustomization-missing');",
    "  const resources = readYaml(kustomizationPath).resources || [];",
    "  if (resources.includes('observability.yaml')) fail('observability-resource-still-rendered', { file: kustomizationPath });",
+    "  if (workloadCheck.monitoringResources.length > 0) fail('prometheus-operator-resource-still-rendered', { refs: workloadCheck.monitoringResources.slice(0, 12), count: workloadCheck.monitoringResources.length });",
    "  if (workloadCheck.metricsRefs.length > 0) fail('observability-sidecar-still-rendered', { refs: workloadCheck.metricsRefs.slice(0, 12), count: workloadCheck.metricsRefs.length });",
    "  checks.push('observability-disabled');",
    "}",