@hivemagnus/magnus-server 0.0.0-stage → 1.0.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.env.example +171 -0
- package/LICENSE +202 -0
- package/README.md +169 -2
- package/agent-version.json +1 -0
- package/dist/ai/HEARTBEAT.md +56 -0
- package/dist/ai/IDENTITY.md +54 -0
- package/dist/ai/MEMORY.md +40 -0
- package/dist/ai/TOOLS.md +538 -0
- package/dist/ai/USER.md +32 -0
- package/dist/cli/args.js +62 -0
- package/dist/cli/env-catalog.js +131 -0
- package/dist/cli/preflight.js +89 -0
- package/dist/cli.js +175 -0
- package/dist/config/cors.js +16 -0
- package/dist/config/env.js +392 -0
- package/dist/config/features.js +16 -0
- package/dist/config/index.js +3 -0
- package/dist/config/listener-security.js +38 -0
- package/dist/config/version.js +31 -0
- package/dist/db/client.js +44 -0
- package/dist/db/generated/prisma/browser.js +18 -0
- package/dist/db/generated/prisma/client.js +35 -0
- package/dist/db/generated/prisma/commonInputTypes.js +11 -0
- package/dist/db/generated/prisma/enums.js +42 -0
- package/dist/db/generated/prisma/internal/class.js +50 -0
- package/dist/db/generated/prisma/internal/prismaNamespace.js +1598 -0
- package/dist/db/generated/prisma/internal/prismaNamespaceBrowser.js +1569 -0
- package/dist/db/generated/prisma/models/A2ATask.js +2 -0
- package/dist/db/generated/prisma/models/AdmissionFence.js +2 -0
- package/dist/db/generated/prisma/models/Agent.js +2 -0
- package/dist/db/generated/prisma/models/AiConversation.js +2 -0
- package/dist/db/generated/prisma/models/AiDashboard.js +2 -0
- package/dist/db/generated/prisma/models/AiMemory.js +2 -0
- package/dist/db/generated/prisma/models/AiMessage.js +2 -0
- package/dist/db/generated/prisma/models/AiScheduledOp.js +2 -0
- package/dist/db/generated/prisma/models/AlertConditionState.js +2 -0
- package/dist/db/generated/prisma/models/AlertRule.js +2 -0
- package/dist/db/generated/prisma/models/AlertViolation.js +2 -0
- package/dist/db/generated/prisma/models/AnomalyEvent.js +2 -0
- package/dist/db/generated/prisma/models/ApplicationObservation.js +2 -0
- package/dist/db/generated/prisma/models/ArdentInstallation.js +2 -0
- package/dist/db/generated/prisma/models/ArdentProfile.js +2 -0
- package/dist/db/generated/prisma/models/ArdentScopeEnrollment.js +2 -0
- package/dist/db/generated/prisma/models/CommandResourceFence.js +2 -0
- package/dist/db/generated/prisma/models/CommandResourceLease.js +2 -0
- package/dist/db/generated/prisma/models/CommandResult.js +2 -0
- package/dist/db/generated/prisma/models/CronJob.js +2 -0
- package/dist/db/generated/prisma/models/CronJobExecution.js +2 -0
- package/dist/db/generated/prisma/models/DeployJob.js +2 -0
- package/dist/db/generated/prisma/models/DriftScan.js +2 -0
- package/dist/db/generated/prisma/models/EngineeringMetricSample.js +2 -0
- package/dist/db/generated/prisma/models/EngineeringRecoveryExercise.js +2 -0
- package/dist/db/generated/prisma/models/ExecutionRun.js +2 -0
- package/dist/db/generated/prisma/models/FilePatchTask.js +2 -0
- package/dist/db/generated/prisma/models/GpuSnapshot.js +2 -0
- package/dist/db/generated/prisma/models/HealthSnapshot.js +2 -0
- package/dist/db/generated/prisma/models/HermesCoreRollout.js +2 -0
- package/dist/db/generated/prisma/models/HermesCoreRolloutHost.js +2 -0
- package/dist/db/generated/prisma/models/HermesCredentialPool.js +2 -0
- package/dist/db/generated/prisma/models/HermesCredentialPoolAttachment.js +2 -0
- package/dist/db/generated/prisma/models/HermesCronJob.js +2 -0
- package/dist/db/generated/prisma/models/HermesHook.js +2 -0
- package/dist/db/generated/prisma/models/HermesInstallation.js +2 -0
- package/dist/db/generated/prisma/models/HermesMcpServer.js +2 -0
- package/dist/db/generated/prisma/models/HermesMcpServerAttachment.js +2 -0
- package/dist/db/generated/prisma/models/HermesMessagingBinding.js +2 -0
- package/dist/db/generated/prisma/models/HermesOperationJob.js +2 -0
- package/dist/db/generated/prisma/models/HermesPluginInstall.js +2 -0
- package/dist/db/generated/prisma/models/HermesProfile.js +2 -0
- package/dist/db/generated/prisma/models/HermesProviderConfig.js +2 -0
- package/dist/db/generated/prisma/models/HermesRegistryClient.js +2 -0
- package/dist/db/generated/prisma/models/HermesRegistryInventory.js +2 -0
- package/dist/db/generated/prisma/models/HermesRegistryReceipt.js +2 -0
- package/dist/db/generated/prisma/models/HermesRegistryRevision.js +2 -0
- package/dist/db/generated/prisma/models/HermesRegistryTarget.js +2 -0
- package/dist/db/generated/prisma/models/HermesRole.js +2 -0
- package/dist/db/generated/prisma/models/HermesRoleApplication.js +2 -0
- package/dist/db/generated/prisma/models/HermesSkillAssignment.js +2 -0
- package/dist/db/generated/prisma/models/HermesToolsetPolicy.js +2 -0
- package/dist/db/generated/prisma/models/IncidentLesson.js +2 -0
- package/dist/db/generated/prisma/models/LibraryCategory.js +2 -0
- package/dist/db/generated/prisma/models/LibraryFile.js +2 -0
- package/dist/db/generated/prisma/models/LogEntry.js +2 -0
- package/dist/db/generated/prisma/models/MagnusReleaseCanary.js +2 -0
- package/dist/db/generated/prisma/models/MaintenanceWindow.js +2 -0
- package/dist/db/generated/prisma/models/ManagedCronDefinition.js +2 -0
- package/dist/db/generated/prisma/models/ManagedCronOccurrence.js +2 -0
- package/dist/db/generated/prisma/models/Node.js +2 -0
- package/dist/db/generated/prisma/models/Notification.js +2 -0
- package/dist/db/generated/prisma/models/NotificationPreference.js +2 -0
- package/dist/db/generated/prisma/models/RemediationExecution.js +2 -0
- package/dist/db/generated/prisma/models/RemediationRule.js +2 -0
- package/dist/db/generated/prisma/models/RemoteCommand.js +2 -0
- package/dist/db/generated/prisma/models/ResourceDrainParticipant.js +2 -0
- package/dist/db/generated/prisma/models/ResourceEnrollment.js +2 -0
- package/dist/db/generated/prisma/models/ResourceParticipantWork.js +2 -0
- package/dist/db/generated/prisma/models/Role.js +2 -0
- package/dist/db/generated/prisma/models/RoleApplication.js +2 -0
- package/dist/db/generated/prisma/models/RolloutTarget.js +2 -0
- package/dist/db/generated/prisma/models/SecurityExportOutbox.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionAction.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionActionApproval.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCase.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCaseActivity.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCaseEntity.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCaseEvent.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCaseRelationship.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCollector.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionCommandOutbox.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionDeletionReceipt.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionDelivery.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionEntity.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionEntityAlias.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionEvent.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionEventEntity.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionEvidence.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionInvestigation.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionNonce.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionReceiptChainHead.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionRuleVersion.js +2 -0
- package/dist/db/generated/prisma/models/SecurityFusionSourceState.js +2 -0
- package/dist/db/generated/prisma/models/SecurityServiceState.js +2 -0
- package/dist/db/generated/prisma/models/SharedResourceFence.js +2 -0
- package/dist/db/generated/prisma/models/SharedResourceLease.js +2 -0
- package/dist/db/generated/prisma/models/SharedResourceLeaseResource.js +2 -0
- package/dist/db/generated/prisma/models/SkillInstallation.js +2 -0
- package/dist/db/generated/prisma/models/SkillRegistryEntry.js +2 -0
- package/dist/db/generated/prisma/models/SkillTag.js +2 -0
- package/dist/db/generated/prisma/models/SkillVersion.js +2 -0
- package/dist/db/generated/prisma/models/TeamApprovalDecision.js +2 -0
- package/dist/db/generated/prisma/models/User.js +2 -0
- package/dist/db/generated/prisma/models/WebhookDelivery.js +2 -0
- package/dist/db/generated/prisma/models/WebhookEndpoint.js +2 -0
- package/dist/db/generated/prisma/models.js +2 -0
- package/dist/db/index.js +3 -0
- package/dist/db/read-retry.js +35 -0
- package/dist/db/seed.js +348 -0
- package/dist/index.js +162 -0
- package/dist/middleware/auth.js +361 -0
- package/dist/middleware/body-limit.js +14 -0
- package/dist/middleware/correlation-id.js +13 -0
- package/dist/middleware/error-handler.js +28 -0
- package/dist/middleware/iam-token.js +21 -0
- package/dist/middleware/rate-limit.js +78 -0
- package/dist/middleware/request-logger.js +85 -0
- package/dist/middleware/security-headers.js +23 -0
- package/dist/middleware/tracing.js +120 -0
- package/dist/middleware/user-status.js +33 -0
- package/dist/openapi/spec.js +2656 -0
- package/dist/queue/redis.js +20 -0
- package/dist/queue/requeue.js +27 -0
- package/dist/queue/workers.js +342 -0
- package/dist/routes/ai-schedule.js +73 -0
- package/dist/routes/alerts.js +79 -0
- package/dist/routes/anomalies.js +101 -0
- package/dist/routes/ardent.js +82 -0
- package/dist/routes/assistant.js +417 -0
- package/dist/routes/audit.js +138 -0
- package/dist/routes/codex.js +65 -0
- package/dist/routes/commands.js +226 -0
- package/dist/routes/credentials.js +58 -0
- package/dist/routes/cron.js +292 -0
- package/dist/routes/dependencies.js +49 -0
- package/dist/routes/deploy.js +18 -0
- package/dist/routes/discovery.js +168 -0
- package/dist/routes/docs.js +24 -0
- package/dist/routes/drift.js +126 -0
- package/dist/routes/events-stream.js +87 -0
- package/dist/routes/files.js +214 -0
- package/dist/routes/fleet-checks.js +21 -0
- package/dist/routes/health.js +133 -0
- package/dist/routes/heartbeat.js +545 -0
- package/dist/routes/hermes-registry.js +278 -0
- package/dist/routes/hermes.js +61 -0
- package/dist/routes/library.js +402 -0
- package/dist/routes/logs.js +172 -0
- package/dist/routes/managed-cron.js +23 -0
- package/dist/routes/mcp.js +39 -0
- package/dist/routes/metrics.js +187 -0
- package/dist/routes/nodes.js +951 -0
- package/dist/routes/notifications.js +130 -0
- package/dist/routes/ops.js +258 -0
- package/dist/routes/orchestra.js +430 -0
- package/dist/routes/predictions.js +23 -0
- package/dist/routes/profiling.js +68 -0
- package/dist/routes/quick-command.js +72 -0
- package/dist/routes/remediation.js +111 -0
- package/dist/routes/resource-leases.js +297 -0
- package/dist/routes/roles.js +370 -0
- package/dist/routes/rollout.js +124 -0
- package/dist/routes/scan.js +116 -0
- package/dist/routes/security-backup.js +58 -0
- package/dist/routes/security-fusion.js +605 -0
- package/dist/routes/security-integration.js +71 -0
- package/dist/routes/self-update.js +129 -0
- package/dist/routes/shared-resource-leases.js +78 -0
- package/dist/routes/skills.js +485 -0
- package/dist/routes/stats.js +96 -0
- package/dist/routes/team-approvals.js +23 -0
- package/dist/routes/terminal-ticket.js +15 -0
- package/dist/routes/terminal.js +17 -0
- package/dist/routes/topology.js +37 -0
- package/dist/routes/users.js +219 -0
- package/dist/routes/webhooks.js +137 -0
- package/dist/schemas/command-reconciliation.js +14 -0
- package/dist/schemas/commands.js +51 -0
- package/dist/schemas/cron.js +33 -0
- package/dist/schemas/deploy.js +20 -0
- package/dist/schemas/files.js +14 -0
- package/dist/schemas/hermes.js +69 -0
- package/dist/schemas/index.js +11 -0
- package/dist/schemas/library.js +48 -0
- package/dist/schemas/logs.js +35 -0
- package/dist/schemas/managed-cron.js +127 -0
- package/dist/schemas/node-responses.js +87 -0
- package/dist/schemas/node-session-receipt.js +84 -0
- package/dist/schemas/nodes.js +36 -0
- package/dist/schemas/orchestra.js +118 -0
- package/dist/schemas/resource-participant-reconciliation.js +99 -0
- package/dist/schemas/roles.js +54 -0
- package/dist/schemas/shared-resource-lease.js +132 -0
- package/dist/schemas/skills.js +44 -0
- package/dist/schemas/users.js +28 -0
- package/dist/scripts/run-disposable-recovery-exercise.js +39 -0
- package/dist/sdk/index.js +2 -0
- package/dist/sdk/node-api-client.js +474 -0
- package/dist/security/approval-assertion.js +66 -0
- package/dist/security/command-completion-registry.js +35 -0
- package/dist/security/command-request.js +12 -0
- package/dist/security/command-result-signature.js +34 -0
- package/dist/security/managed-http.js +45 -0
- package/dist/security/native-completion-registry.js +44 -0
- package/dist/server.js +178 -0
- package/dist/services/admission-fence.js +71 -0
- package/dist/services/ai-assistant-service.js +2523 -0
- package/dist/services/ai-correlation-service.js +225 -0
- package/dist/services/ai-export-service.js +141 -0
- package/dist/services/ai-insights-service.js +219 -0
- package/dist/services/ai-schedule-service.js +145 -0
- package/dist/services/ai-scratchpad-service.js +222 -0
- package/dist/services/ai-tools.js +1343 -0
- package/dist/services/alert-service.js +313 -0
- package/dist/services/anomaly-detector.js +419 -0
- package/dist/services/api-local-work.js +79 -0
- package/dist/services/application-probes.js +309 -0
- package/dist/services/ardent-scope-enrollment.js +123 -0
- package/dist/services/ardent-service.js +86 -0
- package/dist/services/audit-service.js +308 -0
- package/dist/services/codex-service.js +209 -0
- package/dist/services/command-lifecycle.js +60 -0
- package/dist/services/command-resource-leases.js +305 -0
- package/dist/services/control-host-node.js +242 -0
- package/dist/services/credential-rotation-service.js +294 -0
- package/dist/services/dependency-service.js +145 -0
- package/dist/services/disposable-recovery-runner.js +595 -0
- package/dist/services/docker-group-health.js +121 -0
- package/dist/services/engineering-assurance.js +492 -0
- package/dist/services/engineering-evidence-ledger.js +94 -0
- package/dist/services/engineering-metric-evidence.js +26 -0
- package/dist/services/event-bus.js +227 -0
- package/dist/services/fleet-check-log.js +102 -0
- package/dist/services/forecast-service.js +154 -0
- package/dist/services/health-retention.js +175 -0
- package/dist/services/heartbeat-stream.js +63 -0
- package/dist/services/hermes-catalog-import.js +62 -0
- package/dist/services/hermes-cli-parser.js +671 -0
- package/dist/services/hermes-core-rollout.js +90 -0
- package/dist/services/hermes-dashboard-endpoint.js +13 -0
- package/dist/services/hermes-plugin-source.js +149 -0
- package/dist/services/hermes-registry-service.js +258 -0
- package/dist/services/hermes-service.js +566 -0
- package/dist/services/hivemem-client.js +85 -0
- package/dist/services/host-local-writer.js +82 -0
- package/dist/services/incident-evidence.js +110 -0
- package/dist/services/incident-lessons.js +186 -0
- package/dist/services/llm-providers/anthropic.js +143 -0
- package/dist/services/llm-providers/index.js +7 -0
- package/dist/services/llm-providers/llama-cpp.js +237 -0
- package/dist/services/llm-providers/ollama.js +292 -0
- package/dist/services/llm-providers/openai-compatible.js +200 -0
- package/dist/services/llm-providers/provider-error.js +36 -0
- package/dist/services/llm-providers/registry.js +45 -0
- package/dist/services/llm-providers/types.js +2 -0
- package/dist/services/log-emitter.js +19 -0
- package/dist/services/log-retention.js +66 -0
- package/dist/services/maintenance-service.js +29 -0
- package/dist/services/managed-cron.js +493 -0
- package/dist/services/managed-ssh-admission.js +68 -0
- package/dist/services/monitor-pulse.js +10 -0
- package/dist/services/node-issue-monitor.js +71 -0
- package/dist/services/node-service.js +35 -0
- package/dist/services/notification-retention.js +51 -0
- package/dist/services/notification-service.js +94 -0
- package/dist/services/operational-observations.js +112 -0
- package/dist/services/operations-config.js +67 -0
- package/dist/services/ops-service.js +491 -0
- package/dist/services/orchestra-extras-service.js +370 -0
- package/dist/services/orchestra-resources-service.js +530 -0
- package/dist/services/os-scheduler-writer.js +73 -0
- package/dist/services/password-service.js +22 -0
- package/dist/services/prediction-service.js +284 -0
- package/dist/services/recovery-exercises.js +682 -0
- package/dist/services/remediation-service.js +325 -0
- package/dist/services/resource-enrollment.js +433 -0
- package/dist/services/resource-node-lock.js +18 -0
- package/dist/services/resource-participant-work.js +302 -0
- package/dist/services/role-service.js +388 -0
- package/dist/services/rollout-service.js +440 -0
- package/dist/services/security-export.js +185 -0
- package/dist/services/security-fusion-actions.js +22 -0
- package/dist/services/security-fusion-approvals.js +83 -0
- package/dist/services/security-fusion-evidence-integrity.js +144 -0
- package/dist/services/security-fusion-ingest.js +101 -0
- package/dist/services/security-fusion-outbox.js +185 -0
- package/dist/services/security-fusion-receipt-chain.js +134 -0
- package/dist/services/security-fusion-service.js +751 -0
- package/dist/services/security-fusion-verification.js +50 -0
- package/dist/services/self-update-service.js +200 -0
- package/dist/services/shared-resource-leases.js +413 -0
- package/dist/services/skill-publication-bundle.js +90 -0
- package/dist/services/skill-publication-settlement.js +36 -0
- package/dist/services/skill-publication-state.js +40 -0
- package/dist/services/skill-registry-service.js +327 -0
- package/dist/services/skill-service.js +148 -0
- package/dist/services/ssh-exec-service.js +139 -0
- package/dist/services/ssh-host-verification.js +27 -0
- package/dist/services/tailscale-node-key-service.js +75 -0
- package/dist/services/team-approvals.js +118 -0
- package/dist/services/topology-service.js +138 -0
- package/dist/services/transfer-authority-snapshot.js +151 -0
- package/dist/services/webhook-credential-migration.js +97 -0
- package/dist/services/webhook-security.js +242 -0
- package/dist/services/webhook-service.js +494 -0
- package/dist/types/auth.js +3 -0
- package/dist/types/index.js +3 -0
- package/dist/types/library.js +3 -0
- package/dist/types/notifications.js +3 -0
- package/dist/types/roles.js +3 -0
- package/dist/types/shared-contracts.js +3 -0
- package/dist/utils/env.js +141 -0
- package/dist/utils/heartbeat-peer.js +19 -0
- package/dist/utils/idempotency.js +45 -0
- package/dist/utils/ip.js +25 -0
- package/dist/utils/json-fields.js +94 -0
- package/dist/utils/local-tls.js +39 -0
- package/dist/utils/magnus-home.js +15 -0
- package/dist/utils/node-url.js +59 -0
- package/dist/utils/redact.js +26 -0
- package/dist/utils/safe-equal.js +13 -0
- package/dist/utils/sentinel-nodes.js +16 -0
- package/dist/utils/validate.js +41 -0
- package/package.json +119 -4
- package/prisma/migrations/20261031000000_baseline/migration.sql +2935 -0
- package/prisma/migrations/migration_lock.toml +3 -0
- package/prisma/schema.prisma +2185 -0
- package/prisma.config.ts +14 -0
|
@@ -0,0 +1,313 @@
|
|
|
1
|
+
import { nanoid } from 'nanoid';
|
|
2
|
+
import { getDb } from '../db/index.js';
|
|
3
|
+
import { createNotificationInTransaction } from './notification-service.js';
|
|
4
|
+
import { matchingMaintenance } from './maintenance-service.js';
|
|
5
|
+
import { eventBus, EventChannels } from './event-bus.js';
|
|
6
|
+
const MAX_OBSERVATION_GAP_MS = 5 * 60 * 1000;
|
|
7
|
+
/**
|
|
8
|
+
* Evaluate all active alert rules against a node's health snapshot.
|
|
9
|
+
* Called from the heartbeat route after writing the snapshot.
|
|
10
|
+
*/
|
|
11
|
+
export async function evaluateHealth(nodeId, nodeName, metrics) {
|
|
12
|
+
const db = getDb();
|
|
13
|
+
const rules = await db.alertRule.findMany({ where: { active: true } });
|
|
14
|
+
if (rules.length === 0)
|
|
15
|
+
return;
|
|
16
|
+
for (const rule of rules) {
|
|
17
|
+
// Check node filter
|
|
18
|
+
if (!matchesNodeFilter(rule.nodeFilter, nodeId, nodeName))
|
|
19
|
+
continue;
|
|
20
|
+
if (rule.metric === 'gpu_fan' && !metrics.gpuSampleIsNew)
|
|
21
|
+
continue;
|
|
22
|
+
const value = extractMetric(rule.metric, metrics);
|
|
23
|
+
if (value === null) {
|
|
24
|
+
if (rule.durationSeconds > 0 && rule.metric !== 'gpu_fan') {
|
|
25
|
+
await clearConditionState(rule.id, nodeId);
|
|
26
|
+
}
|
|
27
|
+
continue;
|
|
28
|
+
}
|
|
29
|
+
const violated = evaluateOperator(value, rule.operator, rule.threshold);
|
|
30
|
+
if (violated) {
|
|
31
|
+
const observedAt = rule.metric === 'gpu_fan'
|
|
32
|
+
? metrics.gpuSampledAt ?? new Date()
|
|
33
|
+
: new Date();
|
|
34
|
+
if (await hasMetDuration(rule, nodeId, value, observedAt, metrics)) {
|
|
35
|
+
await handleViolation(rule, nodeId, nodeName, value);
|
|
36
|
+
}
|
|
37
|
+
}
|
|
38
|
+
else {
|
|
39
|
+
if (rule.durationSeconds > 0 && rule.metric !== 'gpu_fan') {
|
|
40
|
+
await clearConditionState(rule.id, nodeId);
|
|
41
|
+
}
|
|
42
|
+
await resolveViolation(rule.id, nodeId);
|
|
43
|
+
}
|
|
44
|
+
}
|
|
45
|
+
}
|
|
46
|
+
function matchesNodeFilter(filter, nodeId, nodeName) {
|
|
47
|
+
if (filter === '*')
|
|
48
|
+
return true;
|
|
49
|
+
const parts = filter.split(',').map((s) => s.trim()).filter(Boolean);
|
|
50
|
+
return parts.some((part) => part === nodeId || part === nodeName || nodeName.includes(part));
|
|
51
|
+
}
|
|
52
|
+
function extractMetric(metric, metrics) {
|
|
53
|
+
switch (metric) {
|
|
54
|
+
case 'cpu':
|
|
55
|
+
return metrics.cpuUsagePercent ?? null;
|
|
56
|
+
case 'memory':
|
|
57
|
+
return metrics.memoryPercent ?? null;
|
|
58
|
+
case 'swap': {
|
|
59
|
+
const total = Number(metrics.swapTotalBytes ?? 0);
|
|
60
|
+
const used = Number(metrics.swapUsedBytes ?? 0);
|
|
61
|
+
if (total === 0)
|
|
62
|
+
return 0;
|
|
63
|
+
return (used / total) * 100;
|
|
64
|
+
}
|
|
65
|
+
case 'disk': {
|
|
66
|
+
// Use the highest disk usage across all mountpoints
|
|
67
|
+
const disks = metrics.disks ?? [];
|
|
68
|
+
if (disks.length === 0)
|
|
69
|
+
return null;
|
|
70
|
+
return Math.max(...disks.map((d) => d.usagePercent ?? 0));
|
|
71
|
+
}
|
|
72
|
+
case 'gpu_fan':
|
|
73
|
+
return metrics.gpuFanPercent ?? null;
|
|
74
|
+
default:
|
|
75
|
+
return null;
|
|
76
|
+
}
|
|
77
|
+
}
|
|
78
|
+
async function clearConditionState(ruleId, nodeId) {
|
|
79
|
+
const db = getDb();
|
|
80
|
+
await db.alertConditionState.deleteMany({ where: { ruleId, nodeId } });
|
|
81
|
+
}
|
|
82
|
+
async function hasMetDuration(rule, nodeId, value, observedAt, metrics) {
|
|
83
|
+
if (rule.durationSeconds <= 0)
|
|
84
|
+
return true;
|
|
85
|
+
if (rule.metric === 'gpu_fan') {
|
|
86
|
+
return hasGpuFanMetDuration(rule, nodeId, observedAt, metrics.gpuDevices ?? []);
|
|
87
|
+
}
|
|
88
|
+
const db = getDb();
|
|
89
|
+
const existing = await db.alertConditionState.findUnique({
|
|
90
|
+
where: { ruleId_nodeId: { ruleId: rule.id, nodeId } },
|
|
91
|
+
});
|
|
92
|
+
const duplicateOrOlderSample = existing && observedAt.getTime() <= existing.lastObservedAt.getTime();
|
|
93
|
+
if (duplicateOrOlderSample) {
|
|
94
|
+
return existing.lastObservedAt.getTime() - existing.breachedAt.getTime() > rule.durationSeconds * 1000;
|
|
95
|
+
}
|
|
96
|
+
const hasGap = existing
|
|
97
|
+
&& observedAt.getTime() - existing.lastObservedAt.getTime() > MAX_OBSERVATION_GAP_MS;
|
|
98
|
+
const breachedAt = !existing || hasGap ? observedAt : existing.breachedAt;
|
|
99
|
+
await db.alertConditionState.upsert({
|
|
100
|
+
where: { ruleId_nodeId: { ruleId: rule.id, nodeId } },
|
|
101
|
+
update: { breachedAt, lastObservedAt: observedAt, lastValue: value },
|
|
102
|
+
create: { ruleId: rule.id, nodeId, breachedAt, lastObservedAt: observedAt, lastValue: value },
|
|
103
|
+
});
|
|
104
|
+
return observedAt.getTime() - breachedAt.getTime() > rule.durationSeconds * 1000;
|
|
105
|
+
}
|
|
106
|
+
function gpuDeviceKey(device) {
|
|
107
|
+
return device.uuid || `${device.vendor}:${device.index}`;
|
|
108
|
+
}
|
|
109
|
+
async function hasGpuFanMetDuration(rule, nodeId, observedAt, currentDevices) {
|
|
110
|
+
const candidates = currentDevices.filter((device) => device.fanPercent !== null && evaluateOperator(device.fanPercent, rule.operator, rule.threshold));
|
|
111
|
+
if (candidates.length === 0)
|
|
112
|
+
return false;
|
|
113
|
+
const windowStart = new Date(observedAt.getTime() - rule.durationSeconds * 1000 - MAX_OBSERVATION_GAP_MS);
|
|
114
|
+
const snapshots = await getDb().gpuSnapshot.findMany({
|
|
115
|
+
where: { nodeId, sampledAt: { gte: windowStart, lte: observedAt } },
|
|
116
|
+
orderBy: { sampledAt: 'desc' },
|
|
117
|
+
select: { sampledAt: true, devices: true },
|
|
118
|
+
});
|
|
119
|
+
return candidates.some((candidate) => {
|
|
120
|
+
const key = gpuDeviceKey(candidate);
|
|
121
|
+
let newestEvidenceAt = observedAt;
|
|
122
|
+
for (const snapshot of snapshots) {
|
|
123
|
+
if (newestEvidenceAt.getTime() - snapshot.sampledAt.getTime() > MAX_OBSERVATION_GAP_MS)
|
|
124
|
+
return false;
|
|
125
|
+
let devices = [];
|
|
126
|
+
try {
|
|
127
|
+
const parsed = JSON.parse(snapshot.devices);
|
|
128
|
+
if (Array.isArray(parsed))
|
|
129
|
+
devices = parsed;
|
|
130
|
+
}
|
|
131
|
+
catch {
|
|
132
|
+
return false;
|
|
133
|
+
}
|
|
134
|
+
const matching = devices.find((device) => gpuDeviceKey(device) === key);
|
|
135
|
+
if (!matching || matching.fanPercent === null
|
|
136
|
+
|| !evaluateOperator(matching.fanPercent, rule.operator, rule.threshold))
|
|
137
|
+
return false;
|
|
138
|
+
if (observedAt.getTime() - snapshot.sampledAt.getTime() > rule.durationSeconds * 1000)
|
|
139
|
+
return true;
|
|
140
|
+
newestEvidenceAt = snapshot.sampledAt;
|
|
141
|
+
}
|
|
142
|
+
return false;
|
|
143
|
+
});
|
|
144
|
+
}
|
|
145
|
+
function metricLabel(metric) {
|
|
146
|
+
return metric.replaceAll('_', ' ').toUpperCase();
|
|
147
|
+
}
|
|
148
|
+
function formatDuration(seconds) {
|
|
149
|
+
if (seconds <= 0)
|
|
150
|
+
return '';
|
|
151
|
+
if (seconds % 3600 === 0)
|
|
152
|
+
return ` for ${seconds / 3600}h`;
|
|
153
|
+
if (seconds % 60 === 0)
|
|
154
|
+
return ` for ${seconds / 60}m`;
|
|
155
|
+
return ` for ${seconds}s`;
|
|
156
|
+
}
|
|
157
|
+
function evaluateOperator(value, operator, threshold) {
|
|
158
|
+
switch (operator) {
|
|
159
|
+
case 'gt':
|
|
160
|
+
return value > threshold;
|
|
161
|
+
case 'gte':
|
|
162
|
+
return value >= threshold;
|
|
163
|
+
case 'lt':
|
|
164
|
+
return value < threshold;
|
|
165
|
+
case 'lte':
|
|
166
|
+
return value <= threshold;
|
|
167
|
+
case 'eq':
|
|
168
|
+
return value === threshold;
|
|
169
|
+
default:
|
|
170
|
+
return false;
|
|
171
|
+
}
|
|
172
|
+
}
|
|
173
|
+
async function handleViolation(rule, nodeId, nodeName, value) {
|
|
174
|
+
const db = getDb();
|
|
175
|
+
const maintenance = await matchingMaintenance(nodeId);
|
|
176
|
+
const created = await db.$transaction(async (tx) => {
|
|
177
|
+
await tx.$executeRaw `SELECT pg_advisory_xact_lock(hashtextextended(${`alert:${rule.id}:${nodeId}`}, 0))`;
|
|
178
|
+
// Check if there's already an active violation for this rule + node.
|
|
179
|
+
const existing = await tx.alertViolation.findFirst({
|
|
180
|
+
where: {
|
|
181
|
+
ruleId: rule.id,
|
|
182
|
+
nodeId,
|
|
183
|
+
status: 'active',
|
|
184
|
+
},
|
|
185
|
+
orderBy: { createdAt: 'desc' },
|
|
186
|
+
});
|
|
187
|
+
if (existing) {
|
|
188
|
+
// Check cooldown — don't create a new notification within the cooldown period.
|
|
189
|
+
const cooldownMs = rule.cooldownMinutes * 60 * 1000;
|
|
190
|
+
if (Date.now() - existing.createdAt.getTime() < cooldownMs) {
|
|
191
|
+
return false; // Within cooldown
|
|
192
|
+
}
|
|
193
|
+
}
|
|
194
|
+
// Create new violation and its notification outbox in one transaction.
|
|
195
|
+
await tx.alertViolation.create({
|
|
196
|
+
data: {
|
|
197
|
+
id: nanoid(),
|
|
198
|
+
ruleId: rule.id,
|
|
199
|
+
nodeId,
|
|
200
|
+
nodeName,
|
|
201
|
+
metric: rule.metric,
|
|
202
|
+
value,
|
|
203
|
+
threshold: rule.threshold,
|
|
204
|
+
severity: rule.severity,
|
|
205
|
+
},
|
|
206
|
+
});
|
|
207
|
+
const label = metricLabel(rule.metric);
|
|
208
|
+
const duration = formatDuration(rule.durationSeconds);
|
|
209
|
+
const severity = rule.severity;
|
|
210
|
+
await createNotificationInTransaction(tx, {
|
|
211
|
+
type: `alert.${rule.metric}`,
|
|
212
|
+
severity,
|
|
213
|
+
title: `${rule.name}: ${label} ${value.toFixed(1)}% on ${nodeName}`,
|
|
214
|
+
message: `Alert rule "${rule.name}" triggered: ${label} is ${value.toFixed(1)}% (threshold: ${rule.threshold}%${duration}) on node ${nodeName}.`,
|
|
215
|
+
nodeId,
|
|
216
|
+
nodeName,
|
|
217
|
+
sourceRef: rule.id,
|
|
218
|
+
metadata: {
|
|
219
|
+
alertRuleId: rule.id,
|
|
220
|
+
metric: rule.metric,
|
|
221
|
+
value,
|
|
222
|
+
threshold: rule.threshold,
|
|
223
|
+
severity: rule.severity,
|
|
224
|
+
},
|
|
225
|
+
}, maintenance?.id ?? null);
|
|
226
|
+
return true;
|
|
227
|
+
});
|
|
228
|
+
if (!created)
|
|
229
|
+
return;
|
|
230
|
+
void eventBus.publish(EventChannels.ALERT_FIRED, {
|
|
231
|
+
type: 'health.alert', nodeId, nodeName, metric: rule.metric,
|
|
232
|
+
value, threshold: rule.threshold, severity: rule.severity,
|
|
233
|
+
timestamp: new Date().toISOString(), source: 'control-api',
|
|
234
|
+
});
|
|
235
|
+
}
|
|
236
|
+
async function resolveViolation(ruleId, nodeId) {
|
|
237
|
+
const db = getDb();
|
|
238
|
+
await db.alertViolation.updateMany({
|
|
239
|
+
where: { ruleId, nodeId, status: 'active' },
|
|
240
|
+
data: { status: 'resolved', resolvedAt: new Date() },
|
|
241
|
+
});
|
|
242
|
+
}
|
|
243
|
+
// ─── CRUD helpers ───
|
|
244
|
+
export async function listAlertRules() {
|
|
245
|
+
const db = getDb();
|
|
246
|
+
return db.alertRule.findMany({
|
|
247
|
+
orderBy: { createdAt: 'desc' },
|
|
248
|
+
include: { _count: { select: { violations: true } } },
|
|
249
|
+
});
|
|
250
|
+
}
|
|
251
|
+
export async function getAlertRule(id) {
|
|
252
|
+
const db = getDb();
|
|
253
|
+
return db.alertRule.findUnique({
|
|
254
|
+
where: { id },
|
|
255
|
+
include: {
|
|
256
|
+
_count: { select: { violations: true } },
|
|
257
|
+
violations: {
|
|
258
|
+
orderBy: { createdAt: 'desc' },
|
|
259
|
+
take: 20,
|
|
260
|
+
},
|
|
261
|
+
},
|
|
262
|
+
});
|
|
263
|
+
}
|
|
264
|
+
export async function createAlertRule(data) {
|
|
265
|
+
const db = getDb();
|
|
266
|
+
return db.alertRule.create({
|
|
267
|
+
data: {
|
|
268
|
+
id: nanoid(),
|
|
269
|
+
name: data.name,
|
|
270
|
+
metric: data.metric,
|
|
271
|
+
operator: data.operator,
|
|
272
|
+
threshold: data.threshold,
|
|
273
|
+
durationSeconds: data.durationSeconds ?? 0,
|
|
274
|
+
severity: data.severity ?? 'warning',
|
|
275
|
+
nodeFilter: data.nodeFilter ?? '*',
|
|
276
|
+
cooldownMinutes: data.cooldownMinutes ?? 15,
|
|
277
|
+
createdBy: data.createdBy ?? null,
|
|
278
|
+
},
|
|
279
|
+
});
|
|
280
|
+
}
|
|
281
|
+
export async function updateAlertRule(id, data) {
|
|
282
|
+
const db = getDb();
|
|
283
|
+
return db.alertRule.update({ where: { id }, data });
|
|
284
|
+
}
|
|
285
|
+
export async function deleteAlertRule(id) {
|
|
286
|
+
const db = getDb();
|
|
287
|
+
await db.alertRule.delete({ where: { id } });
|
|
288
|
+
}
|
|
289
|
+
export async function listAlertViolations(params) {
|
|
290
|
+
const db = getDb();
|
|
291
|
+
const page = params.page ?? 1;
|
|
292
|
+
const pageSize = Math.min(params.pageSize ?? 50, 100);
|
|
293
|
+
const skip = (page - 1) * pageSize;
|
|
294
|
+
const where = {};
|
|
295
|
+
if (params.ruleId)
|
|
296
|
+
where['ruleId'] = params.ruleId;
|
|
297
|
+
if (params.nodeId)
|
|
298
|
+
where['nodeId'] = params.nodeId;
|
|
299
|
+
if (params.status)
|
|
300
|
+
where['status'] = params.status;
|
|
301
|
+
const [violations, total] = await Promise.all([
|
|
302
|
+
db.alertViolation.findMany({
|
|
303
|
+
where,
|
|
304
|
+
orderBy: { createdAt: 'desc' },
|
|
305
|
+
skip,
|
|
306
|
+
take: pageSize,
|
|
307
|
+
include: { rule: { select: { name: true } } },
|
|
308
|
+
}),
|
|
309
|
+
db.alertViolation.count({ where }),
|
|
310
|
+
]);
|
|
311
|
+
return { violations, total };
|
|
312
|
+
}
|
|
313
|
+
//# sourceMappingURL=alert-service.js.map
|