High Availability for Enterprise Imaging: Active-Active, Active-Passive, and Automated Failover

High availability keeps imaging services accessible when a server, interface, network path, or site component fails. Depending on the design, traffic may be distributed across active nodes or redirected to standby infrastructure. A complete HA design must also preserve state, monitor dependencies, prevent duplicate processing, and prove failover through scheduled testing.
Enterprise imaging workflows depend on more than one application server. DICOM routing, HL7 and FHIR interfaces, patient and study metadata, worklists, queues, databases, storage, identity services, network paths, certificates, PACS, VNA, viewers, and reporting systems all contribute to availability.
Adding a second server does not create high availability by itself. The architecture must detect the correct failures, isolate unhealthy resources, redirect traffic, preserve transaction state, maintain sufficient remaining capacity, and confirm that downstream clinical systems continue to receive accurate data.
What Does High Availability Mean for Enterprise Imaging?
High availability, or HA, is an architectural and operational approach that reduces service interruption when a component fails or requires maintenance.
In enterprise imaging, HA may protect:
- DICOM study ingestion and routing
- DICOM query and retrieve services
- HL7 order, result, and status interfaces
- FHIR and other API connections
- DICOMweb services
- Modality worklists
- Metadata transformation and normalization
- AI routing and result delivery
- PACS and VNA connectivity
- Web-based imaging access
- Workflow queues and operational databases
A complete design must define which services remain available, which failures the architecture can tolerate, how quickly traffic can be rerouted to another node, and what happens to transactions active when the failure occurs.
HA usually targets localized failures. A data center outage, regional cloud outage, ransomware incident, or widespread data corruption may require a separate enterprise imaging disaster recovery plan.
Failure Scenarios an Imaging HA Design Should Address
The required response depends on what failed and whether other components remain healthy.
| Failure scenario | Potential imaging impact | Required HA control |
|---|---|---|
| Application process failure | DICOM, HL7, FHIR, or web services stop responding even though the server remains online | Application-level health check, service restart, or node failover |
| Server or virtual-machine failure | All services on the node become unavailable | Redundant node, state synchronization, and automated or controlled failover |
| Network-interface failure | The node runs, but clients cannot reach it | Redundant interfaces, separate network paths, and endpoint health monitoring |
| Switch, firewall, or routing failure | Multiple nodes may become unreachable through the same path | Network redundancy, independent paths, and external service checks |
| Virtual IP or DNS failure | Healthy nodes remain available, but clients cannot locate the service | Redundant endpoint control and tested traffic-redirection procedures |
| Database, queue, or shared-state failure | Nodes run, but transactions cannot be processed consistently | Replicated or highly available state services and transaction-integrity controls |
| Shared-storage failure | Multiple application nodes lose access to images, configuration, or logs | Redundant storage or an alternate data path outside the same failure domain |
| Downstream PACS or VNA failure | The router remains available, but delivery stops | Destination health checks, queuing, alternate routing, and controlled retry |
| Capacity loss after node failure | Remaining nodes stay online but cannot handle the full workload | Reserved capacity and failover testing under production-like load |
| Split-brain | More than one node believes it owns the same active workload | Quorum, fencing, or another mechanism that prevents conflicting active ownership |
| Site or regional failure | Local nodes and shared services become unavailable | Cross-site architecture or disaster recovery |
Health checks should test the service that matters. A successful ping only proves that the host responded. It does not prove that the DICOM listener can accept associations, that the database can commit state, that the routing queue can process studies, or that a downstream archive can receive them.
Active-Active, Active-Passive, Load Balancing, and Disaster Recovery
Microsoft’s failover clustering guidance describes active-active and active-passive as distinct availability configurations. Each design has different requirements for traffic distribution, state, capacity, and failure handling.
| Architecture | Normal operation | Response to failure | Key requirements and limitations |
|---|---|---|---|
| Active-active HA | Two or more nodes process traffic simultaneously | New work moves to the remaining active nodes | Requires load distribution, consistent configuration, coordinated state, duplicate prevention, and enough remaining capacity |
| Active-passive HA | One node processes traffic while one or more synchronized nodes remain on standby | A standby node becomes active and assumes the service endpoint | Requires reliable health detection, state synchronization, promotion, Virtual IP or endpoint transfer, and protection against split brain |
| Load balancing | Traffic is distributed across multiple services or destinations | An unhealthy destination may be removed from rotation | Improves distribution and capacity, but does not automatically synchronize application state, recover a database, or provide full disaster recovery |
| Disaster recovery | Production operates from the primary environment | Systems and data are restored or activated in another environment after a major event | Requires protected recovery copies, alternate infrastructure, documented recovery order, RTO and RPO targets, and restoration testing |
Active-Active High Availability
In an active-active design, every participating node handles production work. A load balancer or routing layer distributes new transactions across healthy nodes.
This approach can improve resource utilization and capacity by keeping standby hardware active. If one node fails, the other nodes can continue processing.
Active-active designs also introduce additional state-management requirements. The architecture must answer:
- Which node owns each transaction?
- Can two nodes process the same study or message?
- How do nodes share routing configuration and destination status?
- Where do queues and audit records reside?
- How does the system maintain study-level routing consistency?
- Can the remaining nodes handle peak traffic after one node fails?
- How are partially completed transactions reconciled?
For DICOM workflows, an active-active design should keep the same study together when downstream behavior requires it. It should also prevent multiple nodes from unnecessarily delivering the same SOP Instance.
Active-Passive High Availability
In an active-passive design, one node provides the production service, while the other remains synchronized and ready to take over.
When the active node fails, the cluster must:
- Detect that the service or node is unhealthy.
- Confirm that the failed node no longer owns the active workload.
- Promote a standby node.
- Transfer the service endpoint, such as a Virtual IP.
- Resume processing from the available state.
- Validate connections, queues, and downstream delivery.
Active-passive configurations often simplify ownership because only one node processes the protected workload at a time. The tradeoff is that standby capacity remains unused during normal operation, and failover depends on reliable promotion and endpoint transfer.
DICOM Load Balancing
DICOM load balancing distributes imaging traffic across multiple destinations based on availability, capacity, DICOM metadata, or a routing policy.
Common methods include:
- Round robin
- Weighted round robin
- Destination health and availability
- DICOM tag-based routing
- Thread or connection limits
- Modality, facility, priority, or workflow rules
Load balancing can support an active-active design when multiple available services can safely process the same workload. It can also operate in front of downstream PACS or VNA ingestion points.
Load balancing alone does not guarantee HA. The architecture must still protect routing configuration, queues, transaction state, databases, network paths, and the load balancer itself.
What Happens When a DICOM Router Fails?
When a DICOM Router fails, the exact sequence depends on whether the architecture uses active-active or active-passive nodes. A controlled failure sequence should address the following steps.
1. The Sending System Connects to a Stable Endpoint
Modalities, PACS, VNAs, and other DICOM systems connect to a defined Application Entity Title, IP address, and port. An HA design should avoid requiring every sending system to change its endpoint during a node failure.
An active-passive cluster may use a Virtual IP. An active-active design may use an application-aware load balancer or routing endpoint.
2. Health Monitoring Detects the Failure
The monitoring service detects that the active router, application process, network path, or required dependency is unhealthy.
Useful checks may include:
- Node and operating-system availability
- DICOM association negotiation
- DICOM Verification, or C-ECHO
- Application-process state
- Database and queue access
- Storage availability
- Network reachability
- Certificate validity
- Downstream destination availability
A C-ECHO response verifies that a DICOM application responds, but it does not prove that the entire routing workflow can store and deliver a study.
3. The Failed Node Is Isolated
Before another node assumes ownership, the cluster should confirm that the failed node can no longer process the same workload. Quorum or fencing controls help prevent a split-brain condition.
4. Traffic Moves to a Healthy Node
In an active-passive configuration, the standby node becomes active and assumes the shared endpoint.
In an active-active configuration, the load-balancing layer removes the failed node from rotation and sends new associations to the remaining nodes.
5. In-Flight Associations Are Resolved
An association that was active when the node failed may terminate before the sender receives a final response. The sending system should handle the transaction in accordance with its documented retry behavior.
The DICOM Storage Service Class requires the sending application to recognize successful and failed C-STORE responses and take the implementation-defined action documented in its conformance statement.
If the connection fails before the sender receives a successful C-STORE response, the sender may retry the object. The receiving workflow should be prepared to recognize the SOP Instance UID and avoid unsafe or unnecessary duplicate downstream processing.
Where explicit confirmation of safekeeping is required, DICOM Storage Commitment may form part of the workflow.
6. Queues and Workflow State Are Recovered
The available node must determine:
- Which studies were accepted
- Which instances remain queued
- Which destinations already received each object
- Which transactions require retry
- Which jobs were incomplete
- Whether configuration and routing state remain consistent
HA must preserve more than data. It must preserve enough transaction state to resume work without losing or incorrectly repeating clinical messages.
7. Downstream Delivery Is Validated
Monitoring should confirm that studies and messages reach the intended PACS, VNA, EHR, RIS, viewer, or AI application.
A router can be healthy while a destination remains unavailable. The workflow should queue, redirect, retry, or escalate according to policy.
8. The Failed Node Is Repaired and Returned Carefully
The recovered node should not automatically rejoin production until its configuration, state, software version, certificates, network settings, and health checks match the active environment.
Failback or rebalancing should occur as a controlled change.
Enterprise Imaging HA Dependency Checklist
Every protected service should have a documented dependency map.
Compute and Platform
- Physical or virtual nodes
- Hypervisor or container platform
- CPU, memory, and storage capacity
- Operating-system services
- Software packages and versions
- Licensing services
- Time synchronization
Network and Endpoints
- Virtual IP addresses
- DNS
- DICOM AE Titles, IP addresses, and ports
- HL7 and FHIR endpoints
- DICOMweb and REST endpoints
- Firewalls and network-access rules
- Switches, routers, and network paths
- Load balancers
- TLS certificates and trust stores
Application State
- DICOM routing configuration
- HL7 and FHIR interfaces
- Destination definitions
- Tag-transformation rules
- Message and study queues
- Job status
- Audit history
- Logs
- Database state
- Cache contents
- Study and SOP Instance tracking
Security and Administration
- Service accounts
- Privileged administrative access
- Authentication and directory services
- Encryption keys
- Certificate-renewal processes
- Monitoring credentials
- Configuration backups
Connected Clinical Systems
- Modalities
- PACS
- VNA
- EHR
- RIS
- Reporting systems
- Clinical viewers
- AI applications
- Cloud services
- External imaging exchanges
Operational Requirements
- Peak and sustained traffic
- Capacity after losing one node
- Queue-growth limits
- Retry timing
- Escalation thresholds
- Maintenance procedures
- Failover ownership
- Failback approval
- Downtime communications
- Disaster-recovery escalation
A design remains only as available as its unprotected dependencies. Two application nodes connected to one database, switch, storage array, certificate authority, or authentication service can still share a single point of failure.
Testing Automated Failover
Automated failover should be tested under controlled, production-like conditions. Restarting a standby node does not prove that the complete workflow can survive a failure.
The AWS Well-Architected Reliability guidance recommends monitoring workload components, failing over to healthy resources, and testing recovery processes. The same principles apply to imaging HA whether the nodes run on-premises, in the cloud, or in a hybrid environment.
Failure Tests
| Test | What it proves |
|---|---|
| Stop the protected application process | Health monitoring detects an application failure, not only a server failure |
| Power off or isolate the active node | Another node can assume the workload |
| Turn off the production network interface | Endpoint and network-path redundancy work correctly |
| Interrupt the heartbeat path | Quorum behavior prevents split brain |
| Make the database or queue unavailable | The system detects a dependency failure and does not process inconsistent state |
| Make a downstream PACS or VNA unavailable | The router queues, redirects, retries, or escalates according to policy |
| Fail a node during active DICOM transfers | The sender and receiver handle interrupted associations and retries correctly |
| Run the remaining node at expected peak load | The reduced cluster retains sufficient capacity |
| Expire or replace a certificate in a test environment | TLS-dependent services continue or alert correctly |
| Restore the failed node | The node rejoins without stale configuration or duplicate processing |
| Perform planned failback | Ownership and traffic return without losing state or transactions |
Workflow Validation
Each test should verify:
- Time from failure to detection
- Time from detection to traffic redirection
- Time to the first successful DICOM, HL7, FHIR, or web transaction
- Number of interrupted associations
- Number of failed or retried studies and messages
- Duplicate study or message processing
- Queue growth and drain time
- Study and message ordering where required
- Destination accuracy
- Remaining-node CPU, memory, storage, and network capacity
- Alert generation and operator notification
- Audit-log continuity
- Clinical-user access
- Failback behavior
Testing Frequency
The test schedule should reflect clinical risk, transaction volume, architecture changes, and organizational policy.
A practical program may include:
- Continuous node, application, queue, and destination monitoring
- Quarterly component or node-failure testing for critical routing services
- Semiannual dependency and reduced-capacity testing
- Annual end-to-end continuity and disaster-recovery exercises
- Targeted failover testing after upgrades, network changes, certificate changes, or configuration revisions
Organizations should document expected behavior before each test and compare it with the actual result. Every failed check should produce an owner and corrective action.
Does High Availability Replace Disaster Recovery?
No. HA and disaster recovery address different failure scopes.
HA can reduce downtime when a protected node, service, interface, or network component fails. It may not protect against:
- Loss of the entire data center or cloud region
- Ransomware affecting all connected nodes
- Corruption replicated across the cluster
- Compromise of shared identity or administrative systems
- Loss of shared storage
- A configuration error deployed to every node
- Prolonged failure of a critical third-party service
Enterprise imaging disaster recovery restores applications and data from an alternate or known-good environment after a major event. It requires protected recovery copies, restoration priorities, RTO and RPO targets, testing, reconciliation, and failback.
An HA configuration may form one part of a disaster-recovery architecture, but it does not eliminate the need for backups and recovery planning.
Frequently Asked Questions
What does high availability mean for enterprise imaging?
High availability uses redundant components, health monitoring, state management, and failover to keep DICOM, HL7, FHIR, DICOMweb, and related imaging services accessible during localized failures. The design must also protect dependencies and provide enough capacity after a component fails.
What is active-active availability?
Active-active availability uses two or more nodes that process production traffic simultaneously. When one node fails, the remaining nodes continue processing, provided they have enough capacity, and the system can maintain consistent state and prevent duplicate work.
What happens when a DICOM router fails?
Health monitoring detects the failure, isolates the unhealthy node, and redirects new traffic or promotes a standby node. Interrupted DICOM associations may need to be retried, while the routing system reconciles accepted studies, queued work, destination status, and duplicate risk.
How is automated failover tested?
Teams deliberately stop application processes, nodes, network paths, or dependencies and measure detection, traffic redirection, transaction recovery, queue behavior, remaining capacity, alerts, and failback. Testing should include active DICOM and interface traffic, not only idle servers.
Does HA replace disaster recovery?
No. HA reduces interruption during localized component failures. Disaster recovery restores systems and data after major events such as site loss, ransomware, regional outages, or widespread corruption.
How Dicom Systems Supports High Availability for Enterprise Imaging
The Dicom Systems Unifier platform provides vendor-neutral workflow and interoperability services across DICOM, DICOMweb, HL7, FHIR, PACS, VNA, EHR, RIS, modalities, viewers, AI applications, and cloud environments.
Health systems choose the systems and workflows they need to protect. Dicom Systems configures the Unifier architecture, routing, endpoints, monitoring, and failover behavior to meet those requirements.
Unifier Active-Passive High Availability
Unifier High Availability uses two or more Unifier appliances in an active-passive configuration.

Unifier High Availability during normal operations, with the active node processing workflow traffic and the passive node remaining synchronized.

Unifier High Availability after failover, with the passive node promoted and workflow traffic redirected through the Virtual IP.
The architecture includes:
- An active Unifier node that processes DICOM, HL7, and web-based workflow traffic
- One or more synchronized passive nodes
- A front-facing Virtual IP that provides a stable endpoint
- Automatic promotion of a passive node when the active node becomes unavailable
- Virtual IP redirection to the newly active node
- Dedicated heartbeat connectivity for node health and coordination
- Cluster messaging for membership, synchronization, and resource coordination
- Support for virtualized, on-premises, cloud, and hybrid environments
Modalities, PACS, VNAs, EHRs, and other systems connect to the Virtual IP rather than to an individual node. This allows the cluster to redirect supported traffic without requiring every connected system to change its endpoint.
Unifier DICOM Load Balancing
Unifier DICOM Load Balancing distributes imaging traffic across multiple downstream services or PACS ingestion points.
Available routing methods include:
- Round robin
- Weighted round robin
- Configurable thread limits
- DICOM tag-based routing
- Application-level DICOM traffic control
- Destination availability and workflow rules
- Study-level destination consistency
Load balancing can operate independently or alongside Unifier High Availability. The appropriate configuration depends on whether the organization needs traffic distribution, node failover, downstream capacity management, or a combination of these capabilities.
High Availability and Disaster Recovery Together
Unifier High Availability addresses node and service failures within the protected architecture. Broader continuity may also require alternate sites, protected data copies, backup clinical viewing, archive recovery, and documented disaster-recovery procedures.
Dicom Systems can help health systems:
- Identify the DICOM, DICOMweb, HL7, FHIR, and web services that require HA
- Choose active-passive, load-balanced, or combined workflow patterns
- Map application, network, state, storage, security, and downstream dependencies
- Configure stable endpoints, health monitoring, routing, and failover
- Define queue, retry, and duplicate-handling behavior
- Validate reduced-capacity performance after a node failure
- Test node, service, network, and destination failures
- Coordinate HA with enterprise imaging disaster recovery
Meet with a Dicom Systems workflow expert to review the imaging systems, interfaces, nodes, destinations, and failure domains you need to protect and define the load balancing, active-passive failover, state coordination, capacity, testing, and disaster-recovery requirements.