Skip to content

Hyper-V health and alert architecture

Health answers what is currently unhealthy; alerts answer what an operator should act on. They are related but not identical. Unit monitors establish object health, aggregate monitors organize the four SCOM health dimensions, dependency monitors propagate service impact, and alerts originate at the most actionable layer. Rules collect diagnostic history or alert on discrete events that do not have a durable state model.

Health construction

Rendering diagram…

Aggregate and dependency monitors normally do not generate duplicate alerts. The leaf monitor that has the evidence and remediation context alerts; its state then rolls up for service impact.

Standard health dimensions

DimensionHyper-V examplesTypical evidence
AvailabilityHost role, VM service, cluster/node/resource, CSV online, Replica availabilityService state, cluster state, authoritative events, heartbeat
ConfigurationVersion drift, unsupported topology, integration configuration, network authority driftRegistry, CIM, PowerShell, configuration provider
PerformanceSustained CPU pressure, memory pressure, latency, queue, packet lossPerformance counters and calculated property bags
SecurityExplicit security posture signals that are both supported and actionableSecurity/configuration provider; disabled unless evidence supports default health

Security is not a catch-all for general configuration. The first release includes Security health only for signals with a documented source, owner, remediation, and support boundary.

Stateful threshold pattern

Rendering diagram…

SCOM may render a missing or uninitialized state as unmonitored rather than a custom fourth state; the implementation must map that platform behavior explicitly. It must not substitute Healthy for missing evidence.

Threshold contract

Every numeric monitor defines the complete time behavior, not only a percentage:

Rendering diagram…

The default host-memory design does not alert at 75% utilization alone. It combines available or reserved host memory, Hyper-V pressure, paging, sustained duration, topology, and lab evidence. The same evidence contract applies to CPU, storage, network, and VM pressure.

Alert decision flow

Rendering diagram…

Alert contract

Every enabled alert-generating workflow must define:

  • source object and monitor/rule ID;
  • condition, operational impact, and evidence captured in alert parameters;
  • severity and priority with a consistent mapping;
  • whether the alert auto-resolves, and the exact healthy/reset evidence;
  • suppression key and repeat-count behavior for event storms;
  • maintenance, migration, backup, checkpoint, drain, and failover behavior;
  • probable causes, validation commands or views, remediation, escalation, and recovery verification;
  • related performance, event, state, and task views; and
  • DA branch and parent impact.

Dependency and symptom suppression

Rendering diagram…

Suppression must be implemented only where targeting and correlation are deterministic. When SCOM cannot safely suppress a symptom, the MP should delay the child condition or provide correlation knowledge instead of hiding a potentially independent fault.

Population-aware VM health

Rendering diagram…

An intentionally powered-off VM must not make the DA unhealthy by default. Expected state is an explicit discovered or configured policy, not a guess based on one sample.

Rollup defaults

ScopeProposed defaultReason
Unit monitor to dimensionWorst state within that dimensionPreserve the most severe supported condition
Dimension to objectStandard SCOM object health behaviorKeep Health Explorer predictable
Critical infrastructure child to branchWorst stateOne failed quorum, CSV, or required host dependency can be service-critical
Redundant host populationTopology-aware or percentage rollupOne drained node may not equal cluster outage
VM populationExpected-state plus absolute and percentage policyAvoid one intentionally stopped VM poisoning a large service
Monitoring pipelineWorst state with explicit freshness deadlinesPrevent false confidence when telemetry is absent
Branch to DA rootImpact-weighted dependency monitorsAvailability-critical branches may affect root differently from advisory configuration

Monitoring coverage sets

Coverage setIntended behavior
CoreLow-noise availability, data-integrity, and monitoring-pipeline health enabled
BalancedCore plus validated predictive performance and configuration monitoring
Deep diagnosticHigh-cardinality collection and disabled-by-default monitors enabled selectively

Coverage sets describe which validated workflows are selected. Lab, Standard, and Strict tuning templates describe environment posture and the values applied to selected workflows. These are orthogonal decisions, not separate sealed runtime products. Discovery changes remain in the customer Discovery Overrides MP; monitor, rule, alert, and collection changes remain in the customer Monitoring Overrides MP. See Override and tuning architecture.

Decision gate

ADR 0029 cannot be accepted until threshold and lab research provides threshold evidence, fault/recovery results, noise assessment, VM expected-state policy, and final Must/Should/Could/Collect/Diagnostic classification.

Released under the MIT License.