4.2. VNF Resiliency¶
The VNF is responsible for meeting its resiliency goals and must factor in expected availability of the targeted virtualization environment. This is likely to be much lower than found in a traditional data center. Resiliency is defined as the ability of the VNF to respond to error conditions and continue to provide the service intended. A number of software resiliency dimensions have been identified as areas that should be addressed to increase resiliency. As VNFs are deployed into the Network Cloud, resiliency must be designed into the VNF software to provide high availability versus relying on the Network Cloud to achieve that end.
Section 4.2 Resiliency in VNF Guidelines describes the overall guidelines for designing VNFs to meet resiliency goals. Below are more detailed resiliency requirements for VNFs.
4.2.1. All Layer Redundancy¶
Design the VNF to be resilient to the failures of the underlying virtualized infrastructure (Network Cloud). VNF design considerations would include techniques such as multiple vLANs, multiple local and geographic instances, multiple local and geographic data replication, and virtualized services such as Load Balancers.
All Layer Redundancy Requirements
The VNF MUST meet their own resiliency goals and not rely on the Network Cloud.
The VNF MUST design resiliency into a VNF such that the resiliency deployment model (e.g., active-active) can be chosen at run-time.
The VNF MUST survive any single points of failure within the Network Cloud (e.g., virtual NIC, VM, disk failure).
The VNF MUST survive any single points of software failure internal to the VNF (e.g., in memory structures, JMS message queues).
The VNF MUST be designed, built and packaged to enable deployment across multiple fault zones (e.g., VNFCs deployed in different servers, racks, OpenStack regions, geographies) so that in the event of a planned/unplanned downtime of a fault zone, the overall operation/throughput of the VNF is maintained.
The VNF MUST support the ability to failover a VNFC automatically to other geographically redundant sites if not deployed active-active to increase the overall resiliency of the VNF.
The VNF MUST support the ability of the VNFC to be deployable in multi-zoned cloud sites to allow for site support in the event of cloud zone failure or upgrades.
The VNF MUST NOT impact the ability of the VNF to provide service/function due to a single container restart.
The VNF SHOULD support container snapshots if not for rebuild and evacuate for rollback or back out mechanism.
4.2.2. Minimize Cross Data-Center Traffic¶
Avoid performance-sapping data center-to-data center replication delay by applying techniques such as caching and persistent transaction paths - Eliminate replication delay impact between data centers by using a concept of stickiness (i.e., once a client is routed to data center “A”, the client will stay with Data center “A” until the entire session is completed).
Minimize Cross Data-Center Traffic Requirements
The VNF SHOULD minimize the propagation of state information across multiple data centers to avoid cross data center traffic.
4.2.3. Application Resilient Error Handling¶
Ensure an application communicating with a downstream peer is equipped to intelligently handle all error conditions. Make sure code can handle exceptions seamlessly - implement smart retry logic and implement multi-point entry (multiple data centers) for back-end system applications.
Application Resilient Error Handling Requirements
The VNF MUST detect communication failure for inter VNFC instance and intra/inter VNF and re-establish communication automatically to maintain the VNF without manual intervention to provide service continuity.
The VNF MUST handle the restart of a single VNFC instance without requiring all VNFC instances to be restarted.
The VNF MUST handle the start or restart of VNFC instances in any order with each VNFC instance establishing or re-establishing required connections or relationships with other VNFC instances and/or VNFs required to perform the VNF function/role without requiring VNFC instance(s) to be started/restarted in a particular order.
The VNF MUST handle errors and exceptions so that they do not interrupt processing of incoming VNF requests to maintain service continuity (where the error is not directly impacting the software handling the incoming request).
The VNF MUST provide the ability to modify the number of retries, the time between retries and the behavior/action taken after the retries have been exhausted for exception handling to allow the NCSP to control that behavior, where the interface and/or functional specification allows for altering behaviour.
The VNF MUST fully exploit exception handling to the extent that resources (e.g., threads and memory) are released when no longer needed regardless of programming language.
The VNF MUST handle replication race conditions both locally and geo-located in the event of a data base instance failure to maintain service continuity.
The VNF MUST automatically retry/resubmit failed requests made by the software to its downstream system to increase the success rate.
The VNF MUST NOT require any manual steps to get it ready for service after a container rebuild.
The VNF MUST provide a mechanism and tool to start VNF containers (VMs) without impacting service or service quality assuming another VNF in same or other geographical location is processing service requests.
The VNF MUST provide a mechanism and tool to perform a graceful shutdown of all the containers (VMs) in the VNF without impacting service or service quality assuming another VNF in same or other geographical location can take over traffic and process service requests.
4.2.4. System Resource Optimization¶
Ensure an application is using appropriate system resources for the task at hand; for example, do not use network or IO operations inside critical sections, which could end up blocking other threads or processes or eating memory if they are unable to complete. Critical sections should only contain memory operation, and should not contain any network or IO operation.
System Resource Optimization Requirements
The VNF MUST NOT execute long running tasks (e.g., IO, database, network operations, service calls) in a critical section of code, so as to minimize blocking of other operations and increase concurrent throughput.
The VNF MUST automatically advertise newly scaled components so there is no manual intervention required.
The VNF MUST utilize FQDNs (and not IP address) for both Service Chaining and scaling.
The VNF MUST deliver any and all functionality from any VNFC in the pool (where pooling is the most suitable solution). The VNFC pool member should be transparent to the client. Upstream and downstream clients should only recognize the function being performed, not the member performing it.
The VNF SHOULD automatically enable/disable added/removed sub-components or component so there is no manual intervention required.
The VNF SHOULD support the ability to scale down a VNFC pool without jeopardizing active sessions. Ideally, an active session should not be tied to any particular VNFC instance.
The VNF SHOULD support load balancing and discovery mechanisms in resource pools containing VNFC instances.
The VNF SHOULD utilize resource pooling (threads, connections, etc.) within the VNF application so that resources are not being created and destroyed resulting in resource management overhead.
The VNF SHOULD use techniques such as “lazy loading” when initialization includes loading catalogues and/or lists which can grow over time, so that the VNF startup time does not grow at a rate proportional to that of the list.
The VNF SHOULD release and clear all shared assets (memory, database operations, connections, locks, etc.) as soon as possible, especially before long running sync and asynchronous operations, so as to not prevent use of these assets by other entities.
4.2.5. Application Configuration Management¶
Leverage configuration management audit capability to drive conformity to develop gold configurations for technologies like Java, Python, etc.
Application Configuration Management Requirements
The VNF MUST allow configurations and configuration parameters to be managed under version control to ensure consistent configuration deployment, traceability and rollback.
The VNF MUST allow configurations and configuration parameters to be managed under version control to ensure the ability to rollback to a known valid configuration.
The VNF MUST allow changes of configuration parameters to be consumed by the VNF without requiring the VNF or its sub-components to be bounced so that the VNF availability is not effected.
4.2.6. Intelligent Transaction Distribution & Management¶
Leverage Intelligent Load Balancing and redundant components (hardware and modules) for all transactions, such that at any point in the transaction: front end, middleware, back end – a failure in any one component does not result in a failure of the application or system; i.e., transactions will continue to flow, albeit at a possibly reduced capacity until the failed component restores itself. Create redundancy in all layers (software and hardware) at local and remote data centers; minimizing interdependencies of components (i.e. data replication, deploying non-related elements in the same container).
Intelligent Transaction Distribution & Management Requirements
The VNF SHOULD use intelligent routing by having knowledge of multiple downstream/upstream endpoints that are exposed to it, to ensure there is no dependency on external services (such as load balancers) to switch to alternate endpoints.
The VNF SHOULD use redundant connection pooling to connect to any backend data source that can be switched between pools in an automated/scripted fashion to ensure high availability of the connection to the data source.
The VNF SHOULD include control loop mechanisms to notify the consumer of the VNF of their exceeding SLA thresholds so the consumer is able to control its load against the VNF.
4.2.7. Deployment Optimization¶
Reduce opportunity for failure, by human or by machine, through smarter deployment practices and automation. This can include rolling code deployments, additional testing strategies, and smarter deployment automation (remove the human from the mix).
Deployment Optimization Requirements
The VNF MUST support at least two major versions of the VNF software and/or sub-components to co-exist within production environments at any time so that upgrades can be applied across multiple systems in a staggered manner.
The VNF MUST support the existence of multiple major/minor versions of the VNF software and/or sub-components and interfaces that support both forward and backward compatibility to be transparent to the Service Provider usage.
The VNF MUST support hitless staggered/rolling deployments between its redundant instances to allow “soak-time/burn in/slow roll” which can enable the support of low traffic loads to validate the deployment prior to supporting full traffic loads.
The VNF MUST support the ability of a requestor of the service to determine the version (and therefore capabilities) of the service so that Network Cloud Service Provider can understand the capabilities of the service.
The VNF MUST test for adherence to the defined performance budgets at each layer, during each delivery cycle with delivered results, so that the performance budget is measured and the code is adjusted to meet performance budget.
The VNF MUST test for adherence to the defined performance budget at each layer, during each delivery cycle so that the performance budget is measured and feedback is provided where the performance budget is not met.
The VNF SHOULD test for adherence to the defined resiliency rating recommendation at each layer, during each delivery cycle with delivered results, so that the resiliency rating is measured and the code is adjusted to meet software resiliency requirements.
The VNF SHOULD test for adherence to the defined resiliency rating recommendation at each layer, during each delivery cycle so that the resiliency rating is measured and feedback is provided where software resiliency requirements are not met.
4.2.8. Monitoring & Dashboard¶
Promote dashboarding as a tool to monitor and support the general operational health of a system. It is critical to the support of the implementation of many resiliency patterns essential to the maintenance of the system. It can help identify unusual conditions that might indicate failure or the potential for failure. This would contribute to improve Mean Time to Identify (MTTI), Mean Time to Repair (MTTR), and post-incident diagnostics.
Monitoring & Dashboard Requirements
The VNF MUST provide a method of metrics gathering for each layer’s performance to identify/document variances in the allocations so they can be addressed.
The VNF MUST provide unique traceability of a transaction through its life cycle to ensure quick and efficient troubleshooting.
The VNF MUST provide a method of metrics gathering and analysis to evaluate the resiliency of the software from both a granular as well as a holistic standpoint. This includes, but is not limited to thread utilization, errors, timeouts, and retries.
The VNF MUST provide operational instrumentation such as logging, so as to facilitate quick resolution of issues with the VNF to provide service continuity.
The VNF MUST monitor for and alert on (both sender and receiver) errant, running longer than expected and missing file transfers, so as to minimize the impact due to file transfer errors.
The VNF SHOULD use an appropriately configured logging level that can be changed dynamically, so as to not cause performance degradation of the VNF due to excessive logging.
The VNF SHOULD utilize Cloud health checks, when available from the Network Cloud, from inside the application through APIs to check the network connectivity, dropped packets rate, injection, and auto failover to alternate sites if needed.
The VNF SHOULD conduct a resiliency impact assessment for all inter/intra-connectivity points in the VNF to provide an overall resiliency rating for the VNF to be incorporated into the software design and development of the VNF.
4.2.9. Virtual Function - Container Recovery Requirements¶
As part of life cycle management, for Cloud environment, VNFs need to support a set of basic recovery capabilities to maintain the health and extend the life of the VNF, eliminating and reducing the frequency that an entire VNF needs to be rebuilt or re-instantiated to recover one or more of its containers. For instance, a VNF in an Openstack environment is composed of one or more containers called VMs (Virtual Machines). During the life of a VNF it is expected that Cloud infrastructure hardware will fail or they would need to be taken down for maintenance or hardware and software upgrades (e.g. firmware upgrades, HostOS (Hypervisor), power maintenance, power outages, etc.) To deal with such life cycle events without having to rebuild entire VNFs or even entire sites these basic recovery capabilities of individual containers, Virtual Machines or other, must be supported.
Evacuate(VM): The Controller client is requesting moving a specified VM from its current AIC host to another (when the host is down). Moving from a specified Host will be supported at in a later release (Openstack).
Migrate (VM): The Controller client is requesting migrating a running target VM from its current AIC host to another. Migrating a running target VM from a specified Host will be supported at in a later release (Openstack).
Reboot(VM): The Controller client is requesting to reboot the VM. Options are soft (graceful) or hard (Openstack).
Rebuild (VM): The Controller client is recreating a target VM instance to a known (good) state (Openstack).
Restart (VM): The Controller client is requesting to restart the VM (Openstack).
Snapshot (VM): The Controller client is requesting to create a snapshot of a VNF or VM and store it (Openstack).
Start (VM): The Controller client is requesting to start the VM (Openstack).
Stop (VM): The Controller client is requesting to stop the VM (Openstack).
The VNF MUST support ONAP Controller’s Restart (stop/start or reboot) command.
The VNF MUST support ONAP Controller’s Migrate command that moves container (VM) from a live Physical Server / Compute Node to another live Physical Server / Compute Node.
Note: Container migrations MUST be transparent to the VNF and no more intrusive than a stop, followed by some down time for the migration to be performed from one Compute Node / Physical Server to another, followed by a start of the same VM with same configuration on the new Compute Node / Physical Server.
The VNF MUST support ONAP Controller’s Rebuild command.
The VNF MUST support a container rebuild mechanism based on existing image (e.g. Glance image in Openstack environment) or a snapshot.
The VNF MUST support ONAP Controller’s Evacuate command.
The VNF MUST support ONAP Controller’s Snapshot command.