NSO consists of a number of modules and executable components. These executable components will be referred to by their command-line name, e.g. ncs, ncs-netsim, ncs_cli, etc. ncs is used to refer to the executable, the running daemon.

Starting NSO

When NSO is started, it reads its configuration file and starts all subsystems configured to start (such as NETCONF, CLI, etc.).

By default, NSO starts in the background without an associated terminal. It is recommended to use a when installing NSO for production deployment. This will create an init script that starts NSO when the system boots, and makes NSO start the service manager.

Licensing NSO

NSO is licensed using Cisco Smart Licensing. To register your NSO instance, you need to enter a token from your Cisco Smart Software Manager account. For more information on this topic, see .

Configuring NSO

NSO is configured in the following two ways:

• Through its configuration file, ncs.conf.

• Through whatever data is configured at run-time over any northbound, for example, turning on trace using the CLI.

ncs.conf File

The configuration file ncs.conf is read at startup and can be reloaded. Below is an example of the most common settings. It is included here as an example and should be self-explanatory. See for more information. Important configuration settings are:

• load-path: where NSO should look for compiled YANG files, such as data models for NEDs or Services.

• db-dir: the directory on disk that CDB uses for its storage and any temporary files being used. It is also the directory where CDB searches for initialization files. This should be a local disk and not NFS mounted for performance reasons.

• Various log settings.

The ncs.conf file is described in the . There is a large number of configuration items in ncs.conf, most of them have sane default values. The ncs.conf file is an XML file that must adhere to the tailf-ncs-config.yang model. If we start the NSO daemon directly, we must provide the path to the NCS configuration file as in:

However, in a System Install, systemd is typically used to start NSO, and it will pass the appropriate options to the ncs command. Thus, NSO is started with the command:

It is possible to edit the ncs.conf file, and then tell NSO to reload the edited file without restarting the daemon as in:

This command also tells NSO to close and reopen all log files, which makes it suitable to use from a system like logrotate.

In this section, some of the important configuration settings will be described and discussed.

Exposed Interfaces

NSO allows access through a number of different interfaces, depending on the use case. In the default configuration, clients can access the system locally through an unauthenticated IPC socket (with the ncs* family of commands, port 4569) and plain (non-HTTPS) HTTP web server (port 8080). Additionally, the system enables remote access through SSH-secured NETCONF and CLI (ports 2022 and 2024).

We strongly encourage you to review and customize the exposed interfaces to your needs in the ncs.conf configuration file. In particular, set:

• /ncs-config/webui/match-host-name to true.

• /ncs-config/webui/server-name to the hostname of the server.

If you decide to allow remote access to the web server, also make sure you use TLS-secured HTTPS instead of HTTP. Not doing so exposes you to security risks.

To additionally secure IPC access, refer to .

For more details on individual interfaces and their use, see .

Dynamic Configuration

Let's look at all the settings that can be manipulated through the NSO northbound interfaces. NSO itself has a number of built-in YANG modules. These YANG modules describe the structure that is stored in CDB. Whenever we change anything under, say /devices/device, it will change the CDB, but it will also change the configuration of NSO. We call this dynamic configuration since it can be changed at will through all northbound APIs.

We summarize the most relevant parts below:

tailf-ncs.yang Module

This is the most important YANG module that is used to control and configure NSO. The module can be found at: $NCS_DIR/src/ncs/yang/tailf-ncs.yang in the release. Everything in that module is available through the northbound APIs. The YANG module has descriptions for everything that can be configured.

tailf-common-monitoring2.yang and tailf-ncs-monitoring2.yang are two modules that are relevant to monitoring NSO.

Built-in or External SSH Server

NSO has a built-in SSH server which makes it possible to SSH directly into the NSO daemon. Both the NSO northbound NETCONF agent and the CLI need SSH. To configure the built-in SSH server we need a directory with server SSH keys - it is specified via /ncs-config/aaa/ssh-server-key-dir in ncs.conf. We also need to enable /ncs-config/netconf-north-bound/transport/ssh and /ncs-config/cli/ssh in ncs.conf. In a System Install, ncs.conf is installed in the "config directory", by default /etc/ncs, with the SSH server keys in /etc/ncs/ssh.

Run-time Configuration

There are also configuration parameters that are more related to how NSO behaves when talking to the devices. These reside in devices global-settings.

User Management

Users are configured at the path aaa authentication users.

Access control, including group memberships, is managed using the NACM model (RFC 6536).

Adding a User

Adding a user includes the following steps:

1. Create the user: admin@ncs(config)# aaa authentication users user <user-name>.

2. Add the user to a NACM group: admin@ncs(config)# nacm groups <group-name> admin user-name <user-name>.

3. Verify/change access rules.

It is likely that the new user also needs access to work with device configuration. The mapping from NSO users and corresponding device authentication is configured in authgroups. So, the user needs to be added there as well.

If the last step is forgotten, you will see the following error:

Monitoring NSO

This section describes how to monitor NSO. See also .

Use the command ncs --status to get runtime information on NSO.

NSO Status

Checking the overall status of NSO can be done using the shell:

Or, in the CLI:

For details on the output see $NCS_DIR/src/yang/tailf-common-monitoring2.yang.

Below is an overview of the output:

It is also important to look at the packages that are loaded. This can be done in the CLI with:

Monitoring the NSO Daemon

NSO runs the following processes:

• The daemon: ncs.smp: this is the NCS process running in the Erlang VM.

• Java VM: com.tailf.ncs.NcsJVMLauncher: service applications implemented in Java run in this VM. There are several options on how to start the Java VM, it can be monitored and started/restarted by NSO or by an external monitor. See the Manual Page and the java-vm settings in the CLI.

• Python VMs: NSO packages can be implemented in Python. The individual packages can be configured to run a VM each or share a Python VM. Use the

Logging

NSO has extensive logging functionality. Log settings are typically very different for a production system compared to a development system. Furthermore, the logging of the NSO daemon and the NSO Java VM/Python VM is controlled by different mechanisms. During development, we typically want to turn on the developer-log. The sample ncs.conf that comes with the NSO release has log settings suitable for development, while the ncs.conf created by a System Install are suitable for production deployment.

NSO logs in /logs in your running directory, (depends on your settings in ncs.conf). You might want the log files to be stored somewhere else. See man ncs.conf for details on how to configure the various logs. Below is a list of the most useful log files:

• ncs.log : NCS daemon log. See . Can be configured to Syslog.

• ncserr.log.1, ncserr.log.idx, ncserr.log.siz: if the NSO daemon has a problem. this contains debug information relevant to support. The content can be displayed with ncs --printlog ncserr.log.

Syslog

NSO can syslog to a local Syslog. See man ncs.conf how to configure the Syslog settings. All Syslog messages are documented in Log Messages. The ncs.conf also lets you decide which of the logs should go into Syslog: ncs.log, devel.log, netconf.log, snmp.log, audit.log, WebUI access log. There is also a possibility to integrate with rsyslog to log the NCS, developer, audit, netconf, SNMP, and WebUI access logs to syslog with the facility set to daemon in ncs.conf. For reference, see the upgrade-l2 example, located in examples.ncs/development-guide/high-availability/hcc .

Below is an example of Syslog configuration:

Log messages are described on the link below:

NSO Alarms

NSO generates alarms for serious problems that must be remedied. Alarms are available over all the northbound interfaces and exist at the path /alarms. NSO alarms are managed as any other alarms by the general NSO Alarm Manager, see the specific section on the alarm manager in order to understand the general alarm mechanisms.

The NSO alarm manager also presents a northbound SNMP view, alarms can be retrieved as an alarm table, and alarm state changes are reported as SNMP Notifications. See the "NSO Northbound" documentation on how to configure the SNMP Agent.

This is also documented in the example /examples.ncs/getting-started/using-ncs/5-snmp-alarm-northbound.

Alarms are described on the link below:

Trace ID

NSO can issue a unique Trace ID per northbound request, visible in logs and trace headers. This Trace ID can be used to follow the request from service invocation to configuration changes pushed to any device affected by the change. The Trace ID may either be passed in from an external client or generated by NSO.

Trace ID is enabled by default, and can be turned off by adding the following snippet to NSO.conf:

Trace ID is propagated downwards in LSA setups and is fully integrated with commit queues.

Trace ID can be passed to NSO over NETCONF, RESTCONF, JSON-RPC, or CLI as a commit parameter.

If Trace ID is not given as a commit parameter, NSO will generate one. The generated Trace ID is an array of 16 random bytes, encoded as a 32-character hexadecimal string, in accordance with .

For RESTCONF requests, this generated Trace ID will be communicated back to the requesting client as an HTTP header called X-Cisco-NSO-Trace-ID. The trace-id query parameter can also be used with RPCs and actions to relay a trace-id from northbound requests.

For NETCONF, the Trace ID will be returned as an attribute called trace-id.

Trace ID will appear in relevant log entries and trace file headers on the form trace-id=....

Disaster Management

This section describes a number of disaster scenarios and recommends various actions to take in the different disaster variants.

NSO Fails to Start

CDB keeps its data in four files A.cdb, C.cdb, O.cdb and S.cdb. If NSO is stopped, these four files can be copied, and the copy is then a full backup of CDB.

Furthermore, if neither files exist in the configured CDB directory, CDB will attempt to initialize from all files in the CDB directory with the suffix .xml.

Thus, there exist two different ways to re-initiate CDB from a previously known good state, either from .xml files or from a CDB backup. The .xml files would typically be used to reinstall factory defaults whereas a CDB backup could be used in more complex scenarios.

If the S.cdb file has become inconsistent or has been removed, all commit queue items will be removed, and devices not yet processed out of sync. For such an event, appropriate alarms will be raised on the devices and any service instance that has unprocessed device changes will be set in the failed state.

When NSO starts and fails to initialize, the following exit codes can occur:

• Exit codes 1 and 19 mean that an internal error has occurred. A text message should be in the logs, or if the error occurred at startup before logging had been activated, on standard error (standard output if NSO was started with --foreground --verbose). Generally, the message will only be meaningful to the NSO developers, and an internal error should always be reported to support.

• Exit codes 2 and 3 are only used for the NCS control commands (see the section COMMUNICATING WITH NCS in the in Manual Pages manual page) and mean that the command failed due to timeout. Code 2 is used when the initial connect to NSO didn't succeed within 5 seconds (or the TryTime if given), while code 3 means that the NSO daemon did not complete the command within the time given by the --timeout option.

If the NSO daemon starts normally, the exit code is 0.

If the AAA database is broken, NSO will start but with no authorization rules loaded. This means that all write access to the configuration is denied. The NSO CLI can be started with a flag ncs_cli --noaaa that will allow full unauthorized access to the configuration.

NSO Failure After Startup

NSO attempts to handle all runtime problems without terminating, e.g., by restarting specific components. However, there are some cases where this is not possible, described below. When NSO is started the default way, i.e. as a daemon, the exit codes will of course not be available, but see the --foreground option in the Manual Page.

• Out of memory: If NSO is unable to allocate memory, it will exit by calling abort(3). This will generate an exit code, as for reception of the SIGABRT signal - e.g. if NSO is started from a shell script, it will see 134, as the exit code (128 + the signal number).

• Out of file descriptors for accept(2): If NSO fails to accept a TCP connection due to lack of file descriptors, it will log this and then exit with code 25. To avoid this problem, make sure that the process and system-wide file descriptor limits are set high enough, and if needed configure session limits in ncs.conf. The out-of-file descriptors issue may also manifest itself in that applications are no longer able to open new file descriptors. In many Linux systems, the default limit is 1024, but if we, for example, assume that there are four northbound interface ports, CLI, RESTCONF, SNMP, WebUI/JSON-RPC, or similar, plus a few hundred IPC ports, x 1024 == 5120. But one might as well use the next power of two, 8192, to be on the safe side.

  Several application issues can contribute to consuming extra ports. In the scope of an NSO application that could, for example, be a script application that invokes CLI command or a callback daemon application that does not close the connection socket as it should.

Transaction Commit Failure

When the system is updated, NSO executes a two-phase commit protocol towards the different participating databases including CDB. If a participant fails in the commit() phase although the participant succeeded in the preparation phase, the configuration is possibly in an inconsistent state.

When NSO considers the configuration to be in an inconsistent state, operations will continue. It is still possible to use NETCONF, the CLI, and all other northbound management agents. The CLI has a different prompt which reflects that the system is considered to be in an inconsistent state and also the Web UI shows this:

The MAAPI API has two interface functions that can be used to set and retrieve the consistency status, those are maapi_set_running_db_status() and maapi_get_running_db_status() corresponding. This API can thus be used to manually reset the consistency state. The only alternative to reset the state to a consistent state is by reloading the entire configuration.

Backup and Restore

All parts of the NSO installation can be backed up and restored with standard file system backup procedures.

The most convenient way to do backup and restore is to use the ncs-backup command. In that case, the following procedure is used.

Take a Backup

NSO Backup backs up the database (CDB) files, state files, config files, and rollback files from the installation directory. To take a complete backup (for disaster recovery), use:

The backup will be stored in the "run directory", by default /var/opt/ncs, as /var/opt/ncs/backups/[email protected].

For more information on backup, refer to the in Manual Pages.

Restore a Backup

NSO Restore is performed if you would like to switch back to a previous good state or restore a backup.

It is always advisable to stop NSO before performing a restore.

1. First stop NSO if NSO is not stopped yet.

2. Restore the backup.

   Select the backup to be restored from the available list of backups. The configuration and database with run-time state files are restored in /etc/ncs and /var/opt/ncs.

3. Start NSO.

Rollbacks

NSO supports creating rollback files during the commit of a transaction that allows for rolling back the introduced changes. Rollbacks do not come without a cost and should be disabled if the functionality is not going to be used. Enabling rollbacks impacts both the time it takes to commit a change and requires sufficient storage on disk.

Rollback files contain a set of headers and the data required to restore the changes that were made when the rollback was created. One of the header fields includes a unique rollback ID that can be used to address the rollback file independent of the rollback numbering format.

The use of rollbacks from the supported APIs and the CLI is documented in the documentation for the given API.

ncs.conf Config for Rollback

As described , NSO is configured through the configuration file, ncs.conf. In that file, we have the following items related to rollbacks:

• /ncs-config/rollback/enabled: If set to true, then a rollback file will be created whenever the running configuration is modified.

• /ncs-config/rollback/directory: Location where rollback files will be created.

• /ncs-config/rollback/history-size: The number of old rollback files to save.

Troubleshooting

New users can face problems when they start to use NSO. If you face an issue, reach out to our support team regardless if your problem is listed here or not.

Some noteworthy issues are covered here.

General Troubleshooting Strategies

If you have trouble starting or running NSO, examples, or the clients you write, here are some troubleshooting tips.

Cisco Smart Licensing is a cloud-based approach to licensing and it simplifies the purchase, deployment, and management of Cisco software assets. Entitlements are purchased through a Cisco account via Cisco Commerce Workspace (CCW) and are immediately deposited into a Smart Account for usage. This eliminates the need to install license files on every device. Products that are smart-enabled communicate directly to Cisco to report consumption.

Cisco Smart Software Manager (CSSM) enables the management of software licenses and Smart Account from a single portal. The interface allows you to activate your product, manage entitlements, and renew and upgrade software.

A functioning Smart Account is required to complete the registration process. For detailed information about CSSM, see Cisco Smart Software Manager.

Smart Accounts and Virtual Accounts

A Virtual Account exists as a sub-account within the Smart Account. Virtual Accounts are a customer-defined structure based on organizational layout, business function, geography, or any defined hierarchy. They are created and maintained by the Smart Account administrator(s).

Visit to learn about how to create and manage Smart Accounts.

Request a Smart Account

The creation of a new Smart Account is a one-time event and subsequent management of users is a capability provided through the tool. To request a Smart Account, visit and take the following steps:

1. After logging in select Request a Smart Account in the Administration section.
2. Select the type of Smart Account to create. There are two options: (a) Individual Smart Account requiring agreement to represent your company. By creating this Smart Account you agree to authorization to create and manage product and service entitlements, users, and roles on behalf of your organization. (b) Create the account on behalf of someone else.
3. Provide the required domain identifier and the preferred account name.

Adding Users to a Smart Account

Smart Account user management is available in the Administration section of . Take the following steps to add a new user to a Smart Account:

1. After logging in Select Manage Smart Account in the Administration section.
2. Choose the Users tab.
3. Select New User and follow the instructions in the wizard to add a new user.

Create a License Registration Token

1. To create a new token, log into CSSM and select the appropriate Virtual Account.
2. Click on the Smart Licenses link to enter CSSM.
3. In CSSM click on New Token.

Notes on Configuring Smart Licensing

• If ncs.conf contains configuration for any of java-executable, java-options, override-url/url, or proxy/url under the configure path /ncs-config/smart-license/smart-agent/ any corresponding configuration done via the CLI is ignored.

• The smart licensing component of NSO runs its own Java virtual machine. Usually, the default Java options are sufficient:

  If you for some reason need to modify the Java options, remember to include the default values as found in the YANG model.

Validation and Troubleshooting

Available show and debug Commands

• show license all: Displays all information.

• show license status: Displays status information.

• show license summary: Displays summary.

All user code that needs to run in NSO must be part of a package. A package is basically a directory of files with a fixed file structure or a tar archive with the same directory layout. A package consists of code, YANG modules, etc., that are needed to add an application or function to NSO. Packages are a controlled way to manage loading and versions of custom applications.

Network Element Drivers (NEDs) are also packages. Each NED allows NSO to manage a network device of a specific type. Except for third-party YANG NED packages which do not contain a YANG device model by default (and must be downloaded and fixed before adding to the package), a NED typically contains a device YANG model and the code, specifying how NSO should connect to the device. For NETCONF devices, NSO includes built-in tools to help you build a NED, as described in , that you can use if needed. Otherwise, a third-party YANG NED, if available, should be used instead. Vendors, in some cases, provide the required YANG device models but not the entire NED. In practice, all NSO instances use at least one NED. The set of used NED packages depends on the number of different device types the NSO manages.

When NSO starts, it searches for packages to load. The ncs.conf parameter /ncs-config/load-path defines a list of directories. At initial startup, NSO searches these directories for packages and copies the packages to a private directory tree in the directory defined by the /ncs-config/state-dir parameter in ncs.conf

Alarm Type Descriptions

This section provides necessary information on NED (Network Element Driver) administration with a focus on Cisco-provided NEDs. If you're planning to use NEDs not provided by Cisco, refer to the NED Development to build your own NED packages.

NED represents a key NSO component that makes it possible for the NSO core system to communicate southbound with network devices in most deployments. NSO has a built-in client that can be used to communicate southbound with NETCONF-enabled devices. Many network devices are, however, not NETCONF-enabled, and there exist a wide variety of methods and protocols for configuring network devices, ranging from simple CLI to HTTP/REST-enabled devices. For such cases, it is necessary to use a NED to allow NSO communicate southbound with the network device.

Even for NETCONF-enabled devices, it is possible that the NSO's built-in NETCONF client cannot be used, for instance, if the devices do not strictly follow the specification for the NETCONF protocol. In such cases, one must also use a NED to seamlessly communicate with the device. See Managing Cisco-provided third Party YANG NEDs for more information on third-party YANG NEDs.

Types of NED Packages

A NED package is a package that NSO uses to manage a particular type of device. A NED is a piece of code that enables communication with a particular type of managed device. You add NEDs to NSO as a special kind of package, called NED packages.

A NED package must provide a device YANG model as well as define means (protocol) to communicate with the device. The latter can either leverage the NSO built-in NETCONF and SNMP support or use a custom implementation. When a package provides custom protocol implementation, typically written in Java, it is called a CLI NED or a Generic NED.

Cisco provides and supports a number of such NEDs. With these Cisco-provided NEDs, a major category are CLI NEDs which communicate with a device through its CLI instead of a dedicated API.

CLI NED

This NED category is targeted at devices that use CLI as a configuration interface. Cisco-provided CLI NEDs are available for various network devices from different vendors. Many different CLI syntaxes are supported.

The driver element in a CLI NED implemented by the Cisco NSO NED team typically consists of the following three parts:

• The protocol client, responsible for connecting to and interacting with the device. The protocols supported are SSH and Telnet.

• A fast and versatile CLI parser (+ emitter), usually referred to as the turbo parser.

• Various transform engines capable of converting data between NSO and device formats.

The YANG models in a CLI NED are developed and maintained by the Cisco NSO NED team. Usually, the models for a CLI NED are structured to mimic the CLI command hierarchy on the device.

Generic NED

A generic NED is typically used to communicate with non-CLI devices, such as devices using protocols like REST, TL1, Corba, SOAP, RESTCONF, or gNMI as a configuration interface. Even NETCONF-enabled devices in many cases require a generic NED to function properly with NSO.

The driver element in a Generic NED implemented by the Cisco NED team typically consists of the following parts:

• The protocol client, responsible for interacting with the device.

• Various transform engines capable of converting data between NSO and the device formats, usually JSON and/or XML transformers.

There are two types of Generic NEDs maintained by the Cisco NSO NED team:

• NEDs with Cisco-owned YANG models. These NEDs have models developed and maintained by the Cisco NSO NED team.

• NEDs targeted at YANG models from third-party vendors, also known as, third-party YANG NEDs.

Generic Cisco-provided NEDs with Cisco-owned YANG Models

Generic NEDs belonging to the first category typically handle devices that are model-driven. For instance, devices using proprietary protocols based on REST, SOAP, Corba, etc. The YANG models for such NEDs are usually structured to mimic the messages used by the proprietary protocol of the device.

Third-party YANG NEDs

As the name implies, this NED category is used for cases where the device YANG models are not implemented, maintained, or owned by the Cisco NSO NED team. Instead, the YANG models are typically provided by the device vendor itself, or by organizations like IETF, IEEE, ONF, or OpenConfig.

This category of NEDs has some special characteristics that set them apart from all other NEDs developed by the Cisco NSO NED team:

• Targeted for devices supporting model-driven protocols like NETCONF, RESTCONF, and gNMI.

• Delivered from the software.cisco.com portal without any device YANG models included. There are several reasons for this, such as legal restrictions that prevent Cisco from re-distributing YANG models from other vendors, or the availability of several different version bundles for open-source YANG, like OpenConfig. The version used by the NED must match the version used by the targeted device.

• The NEDs can be bundled with various fixes to solve shortcomings in the YANG models, the download sources, and/or in the device. These fixes are referred to as recipes.

Since the third-party NEDs are delivered without any device YANG models, there are additional steps required to make this category of NEDs operational:

1. The device models need to be downloaded and copied into the NED package source tree. This can be done by using a special (optional) downloader tool bundled with each third-party YANG NED, or in any custom way.

2. The NED must be rebuilt with the downloaded YANG models.

This procedure is thoroughly described in .

Recipes

A third-party YANG NED can be bundled with up to three types of recipe modules. These recipes are used by the NED to solve various types of issues related to:

• The source of the YANG files.

• The YANG files.

• The device itself.

The recipes represent the characteristics and the real value of a third-party YANG NED. Recipes are typically adapted for a certain bundle of YANG models and/or certain device types. This is why there exist many different third-party YANG NEDs, each one adapted for a specific protocol, a specific model package, and/or a specific device.

Download Recipes

When downloading the YANG files, it is first of all important to know which source to use. In some cases, the source is the device itself. For instance, if the device is enabled for NETCONF and sometimes for RESTCONF (in rare cases).

In other cases, the device does not support model download. This applies to all gNMI-enabled devices and most RESTCONF devices too. In this case, the source can be a public Git repository or an archive file provided by the device vendor.

Another important question is what YANG models and what versions to download. To make this task easier, third-party NEDs can be bundled with the download recipes. These are presets to be used with the downloader tool bundled with the NED. There can be several profiles, each representing a preset that has been verified to work by the Cisco NSO NED team. A profile can point out a certain source to download from. It can also limit the scope of the download so that only certain YANG files are selected.

YANG Recipes (YR)

Third-party YANG files can often contain various types of errors, ranging from real bugs that cause compilation errors to certain YANG constructs that are known to cause runtime issues in NSO. To ensure that the files can be built correctly, the third-party NEDs can be bundled with YANG recipes. These recipes patch the downloaded YANG files before they are built by the NSO compiler. This procedure is performed automatically by the make system when the NED is rebuilt after downloading the device YANG files. For more information, refer to .

Runtime Recipes (RR)

Many devices enabled for NETCONF, RESTCONF, or gNMI sometimes deviate in their runtime behavior. This can make it impossible to interact properly with NSO. These deviations can be on any level in the runtime behavior, such as:

• The configuration protocol is not properly implemented, i.e., the device lacks support for mandatory parts of, for instance, the RESTCONF RFC.

• The device returns "dirty" configuration dumps, for instance, JSON or XML containing invalid elements.

• Special quirks are required when applying new configuration on a device. May also require additional transforms of the payload before it is relayed by the NED.

• The device has aliasing issues, possibly caused by overlapping YANG models. If leaf X in model A is modified, the device will automatically modify leaf Y in model B as well.

A third-party YANG NED can be bundled with runtime recipes to solve these kinds of issues, if necessary. How this is implemented varies from NED to NED. In some cases, a NED has a fixed set of recipes that are always used. Alternatively, a NED can support several different recipes, which can be configured through a NED setting, referred to as a runtime profile. For example, a multi-vendor third-party YANG NED might have one runtime profile for each device type supported:

NED Settings

NED settings are YANG models augmented as configurations in NSO and control the behavior of the NED. These settings are augmented under:

• /devices/global-settings/ned-settings

• /devices/profiles/ned-settings

• /devices/device/ned-settings

Most NEDs are instrumented with a large number of NED settings that can be used to customize the device instance configured in NSO. The README file in the respective NED contains more information on these.

Purpose of NED ID

Each managed device in NSO has a device type that informs NSO how to communicate with the device. When managing NEDs, the device type is either cli or generic. The other two device types, netconf and snmp, are used in NETCONF and SNMP packages and are further described in this guide.

In addition, a special NED ID identifier is needed. Simply put, this identifier is a handle in NSO pointing to the NED package. NSO uses the identifier when it is about to invoke the driver in a NED package. The identifier ensures that the driver of the correct NED package is called for a given device instance. For more information on how to set up a new device instance, see .

Each NED package has a NED ID, which is mandatory. The NED ID is a simple string that can have any format. For NEDs developed by the Cisco NSO NED team, the NED ID is formatted as <NED NAME>-<gen | cli>-<NED VERSION MAJOR>.<NED VERSION MINOR>.

Examples

• onf-tapi_rc-gen-2.0

• cisco-iosxr-cli-7.43

The NED ID for a certain NED package stays the same from one version to another, as long as no backward incompatible changes have been done to the YANG models. Upgrading a NED from one version to another, where the NED ID is the same, is simple as it only requires replacing the old NED package with the new one in NSO and then reloading all packages.

Upgrading a NED package from one version to another, where the NED ID is not the same (typically indicated by a change of major or minor number in the NED version), requires additional steps. The new NED package first needs to be installed side-by-side with the old one. Then, a NED migration needs to be performed. This procedure is thoroughly described in .

The Cisco NSO NED team ensures that our CLI NEDs, as well as Generic NEDs with Cisco-owned models, have version numbers and NED ID that indicate any possible backward incompatible YANG model changes. When a NED with such an incompatible change is released, the minor digit in the version is always incremented. The case is a bit different for our third-party YANG NEDs since it is up to the end user to select the NED ID to be used. This is further described in .

NED Versioning Scheme

A NED is assigned a version number consisting of a sequence of numbers separated by dots. The first two numbers represent the major and minor version, and the third number represents the maintenance version.

For example, the number 5.8.1 indicates a maintenance release (1) for the minor release 5.8. Incompatible YANG model changes require either the major or minor version number to be changed. This means that any version within the 5.8.x series is backward compatible with the previous versions.

When a newer maintenance release with the same major/minor version replaces a NED release, NSO can perform a simple data model upgrade to handle stored instance data in the CDB (Configuration Database). This type of upgrade does not pose a risk of data loss.

However, when a NED is replaced by a new major/minor release, it becomes a NED migration. These migrations are complex because the YANG model changes can potentially result in the loss of instance data if not handled correctly.

NED Installation in NSO

This section describes the NED installation in NSO for Local and System installs. Consult the README.md supplied with the NED for the most up-to-date installation description.

Local Install of NED in NSO

This section describes how to install a NED package on a locally installed NSO. See for more information.

Follow the instructions below to install a NED package:

1. Download the latest production-grade version of the NED from software.cisco.com using the URLs provided on your NED license certificates. All NED packages are files with the .signed.bin extension named using the following rule: ncs-<NSO VERSION>-<NED NAME>-<NED VERSION>.signed.bin. The NED package ncs-6.0-cisco-iosxr-7.43.signed.bin will be used in the example below. It is assumed the NED package has been downloaded into the directory named /tmp/ned-package-store. The environment variable NSO_RUNDIR needs to be configured to point to the NSO runtime directory. Example:

2. Unpack the NED package and verify its signature.

Alternatively, the tar.gz file can be installed directly into NSO. In this case, skip steps 3 and 4, and do as below instead:

System Install of Cisco-provided NED in NSO

This section describes how to install a NED package on a system-installed NSO. See for more information.

1. Download the latest production-grade version of the NED from software.cisco.com using the URLs provided on your NED license certificates. All NED packages are files with the .signed.bin extension named using the following rule: ncs-<NSO_VERSION>-<NED NAME>-<NED VERSION>.signed.bin. The NED package ncs-6.0-cisco-iosxr-7.43.signed.bin will be used in the example below. It is assumed that the package has been downloaded into the directory named /tmp/ned-package-store.

2. Unpack the NED package and verify its signature.

   In case the signature cannot be verified (for instance, if access to internet is down), do as below instead.

Configuring a device with the new Cisco-provided NED

The basic steps for configuring a device instance using the newly installed NED package are described in this section. Only the most basic configuration steps are covered here.

Many NEDs require additional custom configuration to be operational. This applies in particular to Generic NEDs. Information about such additional configuration can be found in the files README.md and README-ned-settings.md bundled with the NED package.

The following info is necessary to proceed with the basic setup of a device instance in NSO:

• NED ID of the new NED.

• Connection information for the device to connect to (address and port).

• Authentication information to the device (username and password).

CLI NED Setup

For CLI NEDs, it is mandatory to specify the protocol to be used, either SSH or Telnet.

The following values will be used for this example:

• NED ID: cisco-iosxr-cli-7.43

• Address: 10.10.1.1

• Port: 22

Do the CLI NED setup as below:

1. Start an NSO CLI session.

2. Enter the configuration mode.

3. Configure a new authentication group to be used for this device.

4. Configure the new device instance.

Cisco-provided Generic NED Setup

This example shows a simple setup of a generic NED.

The following values will be used for this example:

• NED ID: onf-tapi_rc-gen-2.0

• Address: 10.10.1.2

• Port: 443

Do the Generic NED setup as below:

1. Start an NSO CLI session.

2. Enter the configuration mode.

3. Configure a new authentication group to be used for this device.

4. Configure the new device instance.

Managing Cisco-provided third Party YANG NEDs

The third-party YANG NED type is a special category of the generic NED type targeted for devices supporting protocols like NETCONF, RESTCONF, and gNMI. As the name implies, this NED category is used for cases where the device YANG models are not implemented or maintained by the Cisco NSO NED Team. Instead, the YANG models are typically provided by the device vendor itself or by organizations like IETF, IEEE, ONF, or OpenConfig.

A third-party YANG NED package is delivered from the software.cisco.com portal without any device YANG models included. It is required that the models are first downloaded, followed by a rebuild and reload of the package, before the NED can become fully operational. This task needs to be performed by the NED user.

Downloading with the NED Built-in Download Tool

This section gives a brief instruction on how to download the device YANG models using the special downloader tool that is bundled with each third-party YANG NED. Each specific NED can contain specific requirements regarding downloading/rebuilding. Before proceeding, check the file README-rebuild.md bundled with the NED package. Furthermore, it is recommended to use a non-production NSO environment for this task.

1. Download and install the third-party YANG NED package into NSO, see .

2. Configure a device instance using as usual. See for more information. The device name dev-1 will be used in this example.

3. Open an NCS CLI session (non-configure mode).

4. The installed NED is now basically empty. It contains no YANG models except some used by the NED internally. This can be verified with the following CLI commands:

Rebuilding NED with Downloaded YANG Files

The NED must be rebuilt when the device YANG models have been downloaded and stored properly. Compiling third-party YANG files is often combined with various types of issues caused by bad or odd YANG constructs. Such issues typically cause compiler errors or unwanted runtime errors in NSO. A third-party YANG NED is configured to take care of all currently known build issues. It will automatically patch the problematic files such that they build properly for NSO. This is done using a set of YANG build recipes bundled with the NED package.

Before rebuilding the NED, it is important to know the path to the target directory used for the downloaded YANG files. This is the same as the local directory if the built-in NED downloader tool was used, see .

This example uses the environment variable NED_YANG_TARGET_DIR to represent the target directory.

To rebuild the NED with the downloaded YANG file:

1. Enter the NED build directory, which is the parent directory to the target directory.

2. Run the make clean all command. The output from the make command can be massive, depending on the number of YANG files, etc. After this step, the NED is rebuilt with the device YANG models included. Lines like below indicate that the NED has applied a number of YANG recipes (patches) to solve known issues with the YANG files:

Reloading the NED Package into NSO

This is the final step to make a third-party YANG NED operational. If the NED built-in YANG downloader tool was used together with no local-dir argument specified (i.e., the default), the only thing required is a package reload in NSO, which you can do by running the packages reload or the packages add command.

If another target directory was used for the YANG file download, it is necessary to first do a proper re-install of the NED package. See .

Rebuilding the NED with a Unique NED ID

A common use case is to have many different versions of a certain device type in the network. All devices can be managed by the same third-party YANG NED. However, each device will likely have its unique set of YANG files (or versions) which this NED has to be rebuilt for.

To set up NSO for this kind of scenario, some additional steps need to be taken:

• Each flavor of the NED needs to be built in a separate source directory, i.e., untar the third-party YANG NED package at multiple locations.

• Each flavor of the re-built NED must have its own unique NED-ID. This will make NSO allow multiple versions of the same NED package to co-exist.

The default NED ID for a third-party YANG NED typically looks like this: <NED NAME>-gen-<NED VERSION MAJOR DIGIT>.<NED VERSION MINOR DIGIT>

The NED build system allows for a customized NED ID by setting one or several of three make variables in any combination when rebuilding the NED:

• NED_ID_SUFFIX

• NED_ID_MAJOR

• NED_ID_MINOR

Do as follows to build each flavor of the third-party YANG NED. Do it in iterations, one at a time:

1. Unpack the empty NED package as described in .

2. Unpack the NED package again in a separate location. Rename the NED directory to something unique.

3. Configure a device instance using the installed NED, as described in . Configure it to connect to the first variant of the device.

4. Follow the instructions in to download the YANG files. Configure local-dir

Upgrading a Cisco-provided Third Party YANG NED to a Newer Version

The NSO procedure to upgrade a NED package to a newer version uses the following approach:

• If there are no backward incompatible changes in the schemas (YANG models) of respective NEDs, simply replace the old NED with the new one and reload all packages in NSO.

• In case there are backwards incompatible changes present in the schemas, some administration is required: the new NED needs to be installed side-by-side with the old NED, after which a NED migration must be performed to properly update the data in CDB using the new schemas. More information about NED migration is available in .

Whether or not there are backward incompatible differences present between two versions of the same NED, is determined by the NED ID. If the versions have the same NED ID, they are fully compatible; otherwise, the NED IDs will differ, typically indicated by the major and/or minor number in the NED ID.

The third-party YANG NEDs add some extra complexity to the NED migration feature. This is because the device YANG models are not included in the NED package. It is up to the end user to select the YANG model versions to use and also to configure the NED ID. If the same NED, at a later stage, needs to be upgraded and rebuilt with newer versions of the YANG model, a decision has to be made regarding the NED ID: Is it safe to use the same NED ID, or should a new one be used?

Using a unique NED ID for each NED package is always the safe option. It minimizes the risk of data loss during package upgrade, etc. However, in some cases, it might be beneficial to use the same NED ID when upgrading a NED package since it minimizes the administration in NSO, i.e., simply replace the old NED package with the new one without any need of NED migration.

This kind of use case can occur when the firmware is upgraded on an NSO-controlled device. For example, assume that we have an optical device that supports the TAPI YANG models from the Open Networking Foundation. Current firmware supports version 2.1.3 of the TAPI bundle. The third-party YANG NED onf-tapi_rc has been rebuilt accordingly with TAPI version 2.1.3 and the default NED ID onf-tapi_rc-gen-2.0. This NED package is installed in NSO and a device instance named dev-1 is configured using it. Next, the optical device is upgraded with the new firmware that supports the TAPI bundle version 2.3.1 instead. The onf-tapi_rc NED needs to be upgraded accordingly. The question is what NED ID to use?

To upgrade a Cisco-provided third-party YANG NED to a newer version:

1. Unpack a fresh copy of the onf-tapi_rc NED package.

2. Download the TAPI models v2.3.1 from the TAPI public Git repository.

3. Rebuild the NED package with a temporary unique NED ID for this rebuild. Any unique NED ID works for this.

   This will generate the NED ID: onf-tapi_rc-gen-2.3.1.

NED Migration

If you upgrade a managed device (such as installing a new firmware), the device data model can change in a significant way. If this is the case, you usually need to use a different and newer NED with an updated YANG model.

When the changes in the NED are not backward compatible, the NED is assigned a new ned-id to avoid breaking existing code. On the plus side, this allows you to use both versions of the NED at the same time, so some devices can use the new version and some can use the old one. As a result, there is no need to upgrade all devices at the same time. The downside is, NSO doesn't know the two NEDs are related and will not perform any upgrade on its own due to different ned-ids. Instead, you must manually change the NED of a managed device through a NED migration.

Migration is required when upgrading a NED and the NED-ID changes, which is signified by a change in either the first or the second number in the NED package version. For example, if you're upgrading the existing router-nc-1.0.1 NED to router-nc-1.2.0 or router-nc-2.0.2, you must perform NED migration. On the other hand, upgrading to router-nc-1.0.2 or router-nc-1.0.3 retains the same ned-id and you can upgrade the router-1.0.1 package in place, directly replacing it with the new one. However, note that some third-party, non-Cisco packages may not adhere to this standard versioning convention. In that case, you must check the ned-id values to see whether migration is needed.

A potential issue with a new NED is that it can break an existing service or other packages that rely on it. To help service developers and operators verify or upgrade the service code, NSO provides additional options of migration tooling for identifying the paths and service instances that may be impacted. Therefore, ensure that all the other packages are compatible with the new NED before you start migrating devices.

To prepare for the NED migration process, first, load the new NED package into NSO with either packages reload or packages add command. Then, use the show packages command to verify that both NEDs, the new and the old, are present. Finally, you may perform the migration of devices either one by one or multiple at a time.

Depending on your operational policies, this may be done during normal operations and does not strictly require a maintenance window, as the migration only reads from and doesn't write to a network device. Still, it is recommended that you create an NSO backup before proceeding.

Note that changing a ned-id also affects device templates if you use them. To make existing device templates compatible with the new ned-id, you can use the copy action. It will copy the configuration used for one ned-id to another, as long as the schema nodes used haven't changed between the versions. The following example demonstrates the copy action usage:

For individual devices, use the /devices/device/migrate action, with the new-ned-id parameter. Without additional options, the command will read and update the device configuration in NSO. As part of this process, NSO migrates all the configuration and service meta-data. Use the dry-run option to see what the command would do and verbose to list all impacted service instances.

You may also use the no-networking option to prevent NSO from generating any southbound traffic towards the device. In this case, only the device configuration in the CDB is used for the migration but then NSO can't know if the device is in sync. Afterward, you must use the compare-config or the sync-from action to remedy this.

For migrating multiple devices, use the /devices/migrate action, which takes the same options. However, with this action, you must also specify the old-ned-id, which limits the migration to devices using the old NED. You can further restrict the action with the device parameter, selecting only specific devices.

It is possible for a NED migration to fail if the new NED is not entirely backward compatible with the old one and the device has an active configuration that is incompatible with the new NED version. In such cases, NSO will produce an error with the YANG constraint that is not satisfied. Here, you must first manually adjust the device configuration to make it compatible with the new NED, and then you can perform the migration as usual.

Depending on what changes are introduced by the migration and how these impact the services, it might be good to re-deploy the affected services before removing the old NED package. It is especially recommended in the following cases:

• When the service touches a list key that has changed. As long as the old schema is loaded, NSO is able to perform an upgrade.

• When a namespace that was used by the service has been removed. The service diffset, that is, the recorded configuration changes created by the service, will no longer be valid. The diffset is needed for the correct get-modifications output, deep-check-sync, and similar operations.

Revision Merge Functionality

The YANG modeling language supports the notion of a module revision. It allows users to distinguish between different versions of a module, so the module can evolve over time. If you wish to use a new revision of a module for a managed device, for example, to access new features, you generally need to create a new NED.

When a model evolves quickly and you have many devices that require the use of a lot of different revisions, you will need to maintain a high number of NEDs, which are mostly the same. This can become especially burdensome during NSO version upgrades, when all NEDs may need to be recompiled.

When a YANG module is only updated in a backward-compatible way (following the upgrade rules in RFC6020 or RFC7950), the NSO compiler, ncsc, allows you to pack multiple module revisions into the same package. This way, a single NED with multiple device model revisions can be used, instead of multiple NEDs. Based on the capabilities exchange, NSO will then use the correct revision for communication with each device.

However, there is a major downside to this approach. While the exact revision is known for each communication session with the managed device, the device model in NSO does not have that information. For that reason, the device model always uses the latest revision. When pushing configuration to a device that only supports an older revision, NSO silently drops the unsupported parts. This may have surprising results, as the NSO copy can contain configuration that is not really supported on the device. Use the no-revision-drop commit parameter when you want to make sure you are not committing config that is not supported by a device.

If you still wish to use this functionality, you can create a NED package with the ncs-make-package --netconf-ned command as you would otherwise. However, the supplied source YANG directory should contain YANG modules with different revisions. The files should follow the module-or-submodule-name@revision-date.yang naming convention, as specified in the RFC6020. Some versions of the compiler require you to use the --no-fail-on-warnings option with the ncs-make-package command or the build process may fail.

The examples.ncs/development-guide/ned-upgrade/yang-revision example shows how you can perform a YANG model upgrade. The original, 1.0 version of the router NED uses the [email protected] YANG model. First, it is updated to the version 1.0.1 [email protected] using a revision merge approach. This is possible because the changes are backward-compatible.

In the second part of the example, the updates in [email protected] introduce breaking changes, therefore the version is increased to 1.1 and a different NED-ID is assigned to the NED. In this case, you can't use revision merge and the usual NED migration procedure is required.

As a single NSO node can fail or lose network connectivity, you can configure multiple nodes in a highly available (HA) setup, which replicates the CDB configuration and operational data across participating nodes. It allows the system to continue functioning even when some nodes are inoperable.

The replication architecture is that of one active primary and a number of secondaries. This means all configuration write operations must occur on the primary, which distributes the updates to the secondaries.

Operational data in the CDB may be replicated or not based on the tailf:persistent statement in the data model. If replicated, operational data writes can only be performed on the primary, whereas non-replicated operational data can also be written on the secondaries.

Replication is supported in several different architectural setups. For example, two-node active/standby designs as well as multi-node clusters with runtime software upgrade.

This feature is independent of but compatible with the Layered Service Architecture (LSA), which also configures multiple NSO nodes to provide additional scalability. When the following text simply refers to a cluster, it identifies the set of NSO nodes participating in the same HA group, not an LSA cluster, which is a separate concept.

NSO supports the following options for implementing an HA setup to cater to the widest possible range of use cases (only one can be used at a time):

• : Using a modern, consensus-based algorithm, it offers a robust, hands-off solution that works best in the majority of cases.

• : A less sophisticated solution that allows you to influence the primary selection but may require occasional manual operator action.

• : NSO only provides data replication; all other functions, such as primary selection and group membership management, are performed by an external application, using the HA framework (HAFW).

In addition to data replication, having a fixed address to connect to the current primary in an HA group greatly simplifies access for operators, users, and other systems alike. Use or an to manage it.

NSO HA Raft

is a consensus algorithm that reliably distributes a set of changes to a group of nodes and robustly handles network and node failure. It can operate in the face of multiple, subsequent failures, while also allowing a previously failed or disconnected node to automatically rejoin the cluster without risk of data conflicts.

Compared to traditional fail-over HA solutions, Raft relies on the consensus of the participating nodes, which addresses the so-called “split-brain” problem, where multiple nodes assume a primary role. This problem is especially characteristic of two-node systems, where it is impossible for a single node on its own to distinguish between losing network connectivity itself versus the other node malfunctioning. For this reason, Raft requires at least three nodes in the cluster.

Raft achieves robustness by requiring at least three nodes in the HA cluster. Three is the recommended cluster size, allowing the cluster to operate in the face of a single node failure. In case you need to tolerate two nodes failing simultaneously, you can add two additional nodes, for a 5-node cluster. However, permanently having more than five nodes in a single cluster is currently not recommended since Raft requires the majority of the currently configured nodes in the cluster to reach consensus. Without the consensus, the cluster cannot function.

You can start a sample HA Raft cluster using the examples.ncs/high-availability/raft-cluster example to test it out. The scripts in the example show various aspects of cluster setup and operation, which are further described in the rest of this section.

Optionally, examples using separate containers for each HA Raft cluster member with NSO system installations are available and referenced in the examples.ncs/development-guide/high-availability/hcc example in the NSO example set.

Overview of Raft Operation

The Raft algorithm works with the concept of (election) terms. In each term, nodes in the cluster vote for a leader. The leader is elected when it receives the majority of the votes. Since each node only votes for a single leader in a given term, there can only be one leader in the cluster for this term.

Once elected, the leader becomes responsible for distributing the changes and ensuring consensus in the cluster for that term. Consensus means that the majority of the participating nodes must confirm a change before it is accepted. This is required for the system to ensure no changes ever get overwritten and provide reliability guarantees. On the other hand, it also means more than half of the nodes must be available for normal operation.

Changes can only be performed on the leader, that will accept the change after the majority of the cluster nodes confirm it. This is the reason a typical Raft cluster has an odd number of nodes; exactly half of the nodes agreeing on a change is not sufficient. It also makes a two-node cluster (or any even number of nodes in a cluster) impractical; the system as a whole is no more available than it is with one fewer node.

If the connection to the leader is broken, such as during a network partition, the nodes start a new term and a new election. Another node can become a leader if it gets the majority of the votes of all nodes initially in the cluster. While gathering votes, the node has the status of a candidate. In case multiple nodes assume candidate status, a split-vote scenario may occur, which is resolved by starting a fresh election until a candidate secures the majority vote.

If it happens that there aren't enough reachable nodes to obtain a majority, a candidate can stay in the candidate state for an indefinite time. Otherwise, when a node votes for a candidate, it becomes a follower and stays a follower in this term, regardless if the candidate is elected or not.

Additionally, the NSO node can also be in the stalled state, if HA Raft is enabled but the node has not joined a cluster.

Node Names and Certificates

Each node in an HA Raft cluster needs a unique name. Names are usually in the ADDRESS format, where ADDRESS identifies a network host where the NSO process is running, such as a fully qualified domain name (FQDN) or an IPv4 address.

Other nodes in the cluster must be able to resolve and reach the ADDRESS, which creates a dependency on the DNS if you use domain names instead of IP addresses.

Limitations of the underlying platform place a constraint on the format of ADDRESS, which can't be a simple short name (without a dot), even if the system is able to resolve such a name using hosts file or a similar mechanism.

You specify the node address in the ncs.conf file as the value for node-address, under the listen container. You can also use the full node name (with the “@” character), however, that is usually unnecessary as the system prepends ncsd@ as-needed.

Another aspect in which ADDRESS plays a role is authentication. The HA system uses mutual TLS to secure communication between cluster nodes. This requires you to configure a trusted Certificate Authority (CA) and a key/certificate pair for each node. When nodes connect, they check that the certificate of the peer validates against the CA and matches the ADDRESS of the peer.

In most cases, this means the ADDRESS must appear in the node certificate's Subject Alternative Name (SAN) extension, as dNSName (see ).

Create and use a self-signed CA to secure the NSO HA Raft cluster. A self-signed CA is the only secure option. The CA should only be used to sign the certificates of the member nodes in one NSO HA Raft cluster. It is critical for security that the CA is not used to sign any other certificates. Any certificate signed by the CA can be used to gain complete control of the NSO HA Raft cluster.

See the examples.ncs/high-availability/raft-cluster example for one way to set up a self-signed CA and provision individual node certificates. The example uses a shell script gen_tls_certs.sh that invokes the openssl command. Consult the section for using it independently of the example.

Examples using separate containers for each HA Raft cluster member with NSO system installations that use a variant of the gen_tls_certs.sh script are available and referenced in the examples.ncs/development-guide/high-availability/hcc example in the NSO example set.

The following is a HA Raft configuration snippet for ncs.conf that includes certificate settings and a sample ADDRESS:

Recipe for a Self-signed CA

HA Raft uses a standard TLS protocol with public key cryptography for securing cross-node communication, where each node requires a separate public/private key pair and a corresponding certificate. Key and certificate management is a broad topic and is critical to the overall security of the system.

The following text provides a recipe for generating certificates using a self-signed CA. It uses strong cryptography and algorithms that are deemed suitable for production use. However, it makes a few assumptions that may not be appropriate for all environments. Always consider how they affect your own deployment and consult a security professional if in doubt.

The recipe makes the following assumptions:

• You use a secured workstation or server to run these commands and handle the generated keys with care. In particular, you must copy the generated keys to NSO nodes in a secure fashion, such as using scp.

• The CA is used solely for a single NSO HA Raft cluster, with certificates valid for 10 years, and provides no CRL. If a single key or host is compromised, a new CA and all key/certificate pairs must be recreated and reprovisioned in the cluster.

• Keys and signatures based on ecdsa-with-sha384/P-384 are sufficiently secure for the vast majority of environments. However, if your organization has specific requirements, be sure to follow those.

To use this recipe, first, prepare a working environment on a secure host by creating a new directory and copying the gen_tls_certs.sh script from $NCS_DIR/examples.ncs/high-availability/raft-cluster into it. Additionally, ensure that the openssl command, version 1.1 or later, is available and the system time is set correctly. Supposing that you have a cluster named lower-west, you might run:

The recipe relies on the gen_tls_certs.sh script to generate individual certificates. For clusters using FQDN node addresses, invoke the script with full hostnames of all the participating nodes. For example:

If your HA cluster is using IP addresses instead, add the -a option to the command and list the IPs:

The script outputs the location of the relevant files and you should securely transfer each set of files to the corresponding NSO node. For each node, transfer only the three files: ca.crt, host.crt, and host.key.

Once certificates are deployed, you can check their validity with the openssl verify command:

This command takes into account the current time and can be used during troubleshooting. It can also display information contained in the certificate if you use the openssl x509 -text -in ssl/certs/node1.example.org.crt -noout variant. The latter form allows you to inspect the incorporated hostname/IP address and certificate validity dates.

Actions

NSO HA Raft can be controlled through several actions. All actions are found under /ha-raft/. In the best-case scenario, you will only need the create-cluster action to initialize the cluster and the read-only and create-cluster actions when upgrading the NSO version. The available actions are listed below:

Network and ncs.conf Prerequisites

In addition to the network connectivity required for the normal operation of a standalone NSO node, nodes in the HA Raft cluster must be able to initiate TCP connections from a random ephemeral client port to the following ports on other nodes:

• Port 4369

• Ports in the range 4370-4399 (configurable)

You can change the ports in the second listed range from the default of 4370-4399. Use the min-port and max-port settings of the ha-raft/listen container.

The Raft implementation does not impose any other hard limits on the network but you should keep in mind that consensus requires communication with other nodes in the cluster. A high round-trip latency between cluster nodes is likely to negatively impact the transaction throughput of the system.

The HA Raft cluster also requires compatible ncs.conf files among the member nodes. In particular, /ncs-config/cdb/operational/enabled and /ncs-config/rollback/enabled values affect replication behavior and must match. Likewise, each member must have the same set of encryption keys and the keys cannot be changed while the cluster is in operation.

To update the ncs.conf configuration, you must manually update the copy on each member node, making sure the new versions contain compatible values. Then perform the reload on the leader and the follower members will automatically reload their copies of the configuration file as well.

If a node is a cluster member but has been configured with a new, incompatible ncs.conf file, it gets automatically disabled. See the /ha-raft/status/disabled-reason for reason. You can re-enable the node with the ha-raft reset command, once you have reconciled the incompatibilities.

Connected Nodes and Node Discovery

Raft has a notion of cluster configuration, in particular, how many and which members the cluster has. You define member nodes when you first initialize the cluster with the create-cluster command or use the adjust-membership command. The member nodes allow the cluster to know how many nodes are needed for consensus and similar.

However, not all cluster members may be reachable or alive all the time. Raft implementation in NSO uses TCP connections between nodes to transport data. The TCP connections are authenticated and encrypted using TLS by default (see ). A working connection between nodes is essential for the cluster to function but a number of factors, such as firewall rules or expired/invalid certificates, can prevent the connection from establishing.

Therefore, NSO distinguishes between configured member nodes and nodes to which it has established a working transport connection. The latter are called connected nodes. In a normal, fully working, and properly configured cluster, the connected nodes will be the same as member nodes (except for the current node).

To help troubleshoot connectivity issues without affecting cluster operation, connected nodes will show even nodes that are not actively participating in the cluster but have established a transport connection to nodes in the cluster. The optional discovery mechanism, described next, relies on this functionality.

NSO includes a mechanism that simplifies the initial cluster setup by enumerating known nodes. This mechanism uses a set of seed nodes to discover all connectable nodes, which can then be used with the create-cluster command to form a Raft cluster.

When you specify one or more nodes with the /ha-raft/seed-nodes/seed-node setting in the ncs.conf file, the current node tries to establish a connection to these seed nodes, in order to discover the list of all nodes potentially participating in the cluster. For the discovery to work properly, all other nodes must also use seed nodes and the set of seed nodes must overlap. The recommended practice is to use the same set of seed nodes on every participating node.

Along with providing an autocompletion list for the create-cluster command, this feature streamlines the discovery of node names when using NSO in containerized or other dynamic environments, where node addresses are not known in advance.

Initial Cluster Setup

Creating a new HA cluster consists of two parts: configuring the individual nodes and running the create-cluster action.

First, you must update the ncs.conf configuration file for each node. All HA Raft configuration comes under the /ncs-config/ha-raft element.

As part of the configuration, you must:

• Enable HA Raft functionality through the enabled leaf.

• Set node-address and the corresponding TLS parameters (see ).

• Identify the cluster this node belongs to with cluster-name.

The cluster name is simply a character string that uniquely identifies this HA cluster. The nodes in the cluster must use the same cluster name or they will refuse to establish a connection. This setting helps prevent mistakenly adding a node to the wrong cluster when multiple clusters are in operation, such as in an LSA setup.

With all the nodes configured and running, connect to the node that you would like to become the initial leader and invoke the ha-raft create-cluster action. The action takes a list of nodes identified by their names. If you have configured seed-nodes, you will get auto-completion support, otherwise, you have to type in the names of the nodes yourself.

This action makes the current node a cluster leader and joins the other specified nodes to the newly created cluster. For example:

You can use the show ha-raft command on any node to inspect the status of the HA Raft cluster. The output includes the current cluster leader and members according to this node, as well as information about the local node, such as node name (local-node) and role. The status/connected-node list contains the names of the nodes with which this node has active network connections.

In case you get an error, such as the Error: NSO can't reach member node 'ncsd@ADDRESS'., please verify all of the following:

• The node at the ADDRESS is reachable. You can use the ping ADDRESS command, for example.

• The problematic node has the correct ncs.conf configuration, especially cluster-name and node-address. The latter should match the ADDRESS and should contain at least one dot.

In addition to the above, you may also examine the logs/raft.log file for detailed information on the error message and overall operation of the Raft algorithm. The amount of information in the file is controlled by the /ncs-config/logs/raft-log configuration in the ncs.conf.

Cluster Management

After the initial cluster setup, you can add new nodes or remove existing nodes from the cluster with the help of the ha-raft adjust-membership action. For example:

When removing nodes using the ha-raft adjust-membership remove-node command, the removed node is not made aware that it is removed from the cluster and continues signaling the other nodes. This is a limitation in the algorithm, as it must also handle situations, where the removed node is down or unreachable. To prevent further communication with the cluster, it is important you ensure the removed node is shut down. You should shut down the to-be-removed node prior to removal from the cluster, or immediately after it. The former is recommended but the latter is required if there are only two nodes left in the cluster and shutting down prior to removal would prevent the cluster from reaching consensus.

Additionally, you can force an existing follower node to perform a full re-sync from the leader by invoking the ha-raft reset action with the force option. Using this action on the leader will make the node give up the leader role and perform a sync with the newly elected leader.

As leader selection during the Raft election is not deterministic, NSO provides the ha-raft handover action, which allows you to either trigger a new election if called with no arguments or transfer leadership to a specific node. The latter is especially useful when, for example, one of the nodes resides in a different location and more traffic between locations may incur extra costs or additional latency, so you prefer this node is not the leader under normal conditions.

Migrating From Existing Rule-based HA

If you have an existing HA cluster using the rule-based built-in HA, you can migrate it to use HA Raft instead. This procedure is performed in four distinct high-level steps:

• Ensuring the existing cluster meets migration prerequisites.

• Preparing the required HA Raft configuration files.

• Switching to HA Raft.

• Adding additional nodes to the cluster.

The procedure does not perform an NSO version upgrade, so the cluster remains on the same version. It also does not perform any schema upgrades, it only changes the type of the HA cluster.

The migration procedure is in place, that is, the existing nodes are disconnected from the old cluster and connected to the new one. This results in a temporary disruption of the service, so it should be performed during a service window.

First, you should ensure the cluster meets migration prerequisites. The cluster must use:

• NSO 6.1.2 or later

• tailf-hcc 6.0 or later (if used)

In case these prerequisites are not met, follow the standard upgrade procedures to upgrade the existing cluster to supported versions first.

Additionally, ensure that all used packages are compatible with HA Raft, as NSO uses some new or updated notifications about HA state changes. Also, verify the network supports the new cluster communications (see ).

Secondly, prepare all the ncs.conf and related files for each node, such as certificates and keys. Create a copy of all the ncs.conf files and disable or remove the existing >ha< section in the copies. Then add the required configuration items to the copies, as described in and . Do not update the ncs.conf files used by the nodes yet.

It is recommended but not necessary that you set the seed nodes in ncs.conf to the designated primary and fail-over primary. Do this for all ncs.conf files for all nodes.

Procedure 1. Migration to HA Raft

1. With the new configurations at hand and verified, start the switch to HA Raft. The cluster nodes should be in their nominal, designated roles. If not, perform a failover first.

2. On the designated (actual) primary, called node1, enable read-only mode.

3. Then take a backup of all nodes.

4. Once the backup successfully completes, stop the designated fail-over primary (actual secondary) NSO process, update its

Security Considerations

Communication between the NSO nodes in an HA Raft cluster takes place over Distributed Erlang, an RPC protocol transported over TLS (unless explicitly disabled by setting /ncs-config/ha-raft/ssl/enabled to 'false').

TLS (Transport Layer Security) provides Authentication and Privacy by only allowing NSO nodes to connect using certificates and keys issued from the same Certificate Authority (CA). Distributed Erlang is transported over TLS 1.2. Access to a host can be revoked by the CA through the means of a CRL (Certificate Revocation List). To enforce certificate revocation within an HA Raft cluster, invoke the action /ha-raft/disconnect to terminate the pre-existing connection. A connection to the node can re-establish once the node's certificate is valid.

Please ensure the CA key is kept in a safe place since it can be used to generate new certificates and key pairs for peers.

Distributed Erlang supports for multiple NSO nodes to run on the same host and the node addresses are resolved by the epmd () service. Once resolved, the NSO nodes communicate directly.

The ports epmd and the NSO nodes listen to can be found in . epmd binds the wildcard IPv4 address 0.0.0.0 and the IPv6 address ::.

In case epmd is exposed to a DoS attack, the HA Raft members may be unable to resolve addresses and communication could be disrupted. Please ensure traffic on these ports are only accepted between the HA Raft members by using firewall rules or other means.

Two NSO nodes can only establish a connection if a shared secret "cookie" matches. The cookie is optionally configured from /ncs-config/ha-raft/cluster-name. Please note the cookie is not a security feature but a way to isolate HA Raft clusters and to avoid accidental misuse.

Packages Upgrades in Raft Cluster

NSO contains a mechanism for distributing packages to nodes in a Raft cluster, greatly simplifying package management in a highly-available setup.

You perform all package management operations on the current leader node. To identify the leader node, you can use the show ha-raft status leader command on a running cluster.

Invoking the packages reload command makes the leader node update its currently loaded packages, identical to a non-HA, single-node setup. At the same time, the leader also distributes these packages to the followers to load. However, the load paths on the follower nodes, such as /var/opt/ncs/packages/, are not updated. This means, that if a leader election took place, a different leader was elected, and you performed another packages reload, the system would try to load the versions of the packages on this other leader, which may be out of date or not even present.

The recommended approach is, therefore, to use the packages ha sync and-reload command instead, unless a load path is shared between NSO nodes, such as the same network drive. This command distributes and updates packages in the load paths on the follower nodes, as well as loading them.

For the full procedure, first, ensure all cluster nodes are up and operational, then follow these steps on the leader node:

• Perform a full backup of the NSO instance, such as running ncs-backup.

• Add, replace, or remove packages on the filesystem. The exact location depends on the type of NSO deployment, for example /var/opt/ncs/packages/.

• Invoke the packages ha sync and-reload or packages ha sync and-add command to start the upgrade process.

Note that while the upgrade is in progress, writes to the CDB are not allowed and will be rejected.

For a packages ha sync and-reload example see the raft-upgrade-l2 NSO system installation-based example referenced by the examples.ncs/development-guide/high-availability/hcc example in the NSO example set.

For more details, troubleshooting, and general upgrade recommendations, see and .

Version Upgrade of Cluster Nodes

Currently, the only supported and safe way of upgrading the Raft HA cluster NSO version requires that the cluster be taken offline since the nodes must, at all times, run the same software version.

Do not attempt an upgrade unless all cluster member nodes are up and actively participating in the cluster. Verify the current cluster state with the show ha-raft status command. All member nodes must also be present in the connected-node list.

The procedure differentiates between the current leader node versus followers. To identify the leader, you can use the show ha-raft status leader command on a running cluster.

Procedure 2. Cluster version upgrade

1. On the leader, first enable read-only mode using the ha-raft read-only mode true command and then verify that all cluster nodes are in sync with the show ha-raft status log replications state command.

2. Before embarking on the upgrade procedure, it's imperative to backup each node. This ensures that you have a safety net in case of any unforeseen issues. For example, you can use the $NCS_DIR/bin/ncs-backup command.

3. Delete the $NCS_RUN_DIR/cdb/compact.lock

For a standard System Install, the single-node procedure is described in , but in general depends on the NSO deployment type. For example, it will be different for containerized environments. For specifics, please refer to the documentation for the deployment type.

For an example see the raft-upgrade-l2 NSO system installation-based example referenced by the examples.ncs/development-guide/high-availability/hcc example in the NSO example set.

If the upgrade fails before or during the upgrade of the original leader, start up the original followers to restore service and then restore the original leader, using backup as necessary.

However, if the upgrade fails after the original leader was successfully upgraded, you should still be able to complete the cluster upgrade. If you are unable to upgrade a follower node, you may provision a (fresh) replacement and the data and packages in use will be copied from the leader.

NSO Rule-based HA

NSO can manage the HA groups based on a set of predefined rules. This functionality was added in NSO 5.4 and is sometimes referred to simply as the built-in HA. However, since NSO 6.1, HA Raft (which is also built-in) is available as well, and is likely a better choice in most situations.

Rule-based HA allows administrators to:

• Configure HA group members with IP addresses and default roles

• Configure failover behavior

• Configure start-up behavior

• Configure HA group members with IP addresses and default roles

NSO rule-based HA is defined in tailf-ncs-high-availability.yang, with data residing under the /high-availability/ container.

NSO rule-based HA does not manage any virtual IP addresses, or advertise any BGP routes or similar. This must be handled by an external package. Tail-f HCC 5.x and greater has this functionality compatible with NSO rule-based HA. You can read more about the HCC package in the .

Prerequisites

To use NSO rule-based HA, HA must first be enabled in ncs.conf - See .

HA Member Configuration

All HA group members are defined under /high-availability/ha-node. Each configured node must have a unique IP address configured and a unique HA ID. Additionally, nominal roles and fail-over settings may be configured on a per-node basis.

The HA Node ID is a unique identifier used to identify NSO instances in an HA group. The HA ID of the local node - relevant amongst others when an action is called - is determined by matching configured HA node IP addresses against IP addresses assigned to the host machine of the NSO instance. As the HA ID is crucial to NSO HA, NSO rule-based HA will not function if the local node cannot be identified.

To join a HA group, a shared secret must be configured on the active primary and any prospective secondary. This is used for a CHAP-2-like authentication and is specified under /high-availability/token/.

The token configured on the secondary node is overwritten with the encrypted token of type aes-256-cfb-128-encrypted-string from the primary node when the secondary node connects to the primary. If there is a mismatch between the encrypted-string configuration on the nodes, NSO will not decrypt the HA token to match the token presented. As a result, the primary node denies the secondary node access the next time the HA connection needs to reestablish with a "Token mismatch, secondary is not allowed" error.

See the upgrade-l2 example, referenced from examples.ncs/development-guide/high-availability/hcc, for an example setup and the for a description of the example.

Also, note that the ncs.crypto_keys file is highly sensitive. The file contains the encryption keys for all CDB data that is encrypted on disk. Besides the HA token, this often includes passwords for various entities, such as login credentials to managed devices.

HA Roles

NSO can assume HA roles primary, secondary and none. Roles can be assigned directly through actions, or at startup or failover. See for the definition of these roles.

NSO rule-based HA distinguishes between the concepts of nominal role and assigned role. Nominal-role is configuration data that applies when an NSO instance starts up and at failover. The assigned role is the role that the NSO instance has been ordered to assume either by an action or as a result of startup or failover.

Failover

Failover may occur when a secondary node loses the connection to the primary node. A secondary may then take over the primary role. Failover behavior is configurable and controlled by the parameters:

• /high-availability/ha-node{id}/failover-primary

• /high-availability/settings/enable-failover

For automatic failover to function, /high-availability/settings/enable-failover must be se to true. It is then possible to enable at most one node with a nominal role secondary as failover-primary, by setting the parameter /high-availability/ha-node{id}/failover-primary. The failover works in both directions; if a nominal primary is currently connected to the failover-primary as a secondary and loses the connection, then it will attempt to take over as a primary.

Before failover happens, a failover-primary-enabled secondary node may attempt to reconnect to the previous primary before assuming the primary role. This behavior is configured by the parameters denoting how many reconnect attempts will be made, and with which interval, respectively.

• /high-availability/settings/reconnect-attempts

• /high-availability/settings/reconnect-interval

HA Members that are assigned as secondaries, but are neither failover-primaries nor set with a nominal-role primary, may attempt to rejoin the HA group after losing connection to the primary.

This is controlled by /high-availability/settings/reconnect-secondaries. If this is true, secondary nodes will query the nodes configured under /high-availability/ha-node for an NSO instance that currently has the primary role. Any configured nominal roles will not be considered. If no primary node is found, subsequent attempts to rejoin the HA setup will be issued with an interval defined by /high-availability/settings/reconnect-interval.

In case a net-split provokes a failover it is possible to end up in a situation with two primaries, both nodes accepting writes. The primaries are then not synchronized and will end up in a split brain. Once one of the primaries joins the other as a secondary, the HA cluster is once again consistent but any out-of-sync changes will be overwritten.

To prevent split-brain from occurring, NSO 5.7 or later comes with a rule-based algorithm. The algorithm is enabled by default, it can be disabled or changed from the parameters:

• /high-availability/settings/consensus/enabled [true]

• /high-availability/settings/consensus/algorithm [ncs:rule-based]

The rule-based algorithm can be used in either of the two HA constellations:

• Two nodes: one nominal primary and one nominal secondary configured as failover-primary.

• Three nodes: one nominal primary, one nominal secondary configured as failover-primary, and one perpetual secondary.

On failover:

• Failover-primary: become primary but enable read-only mode. Once the secondary joins, disable read-only.

• Nominal primary: on loss of all secondaries, change role to none. If one secondary node is connected, stay primary.

To restore the HA cluster one may need to manually invoke the /high-availability/be-secondary-to action.

The read-write mode can manually be enabled from the /high-availability/read-only action with the parameter mode passed with value false.

When any node loses connection, this can also be observed in high-availability alarms as either a ha-primary-down or a ha-secondary-down alarm.

Startup

Startup behavior is defined by a combination of the parameters /high-availability/settings/start-up/assume-nominal-role and /high-availability/settings/start-up/join-ha as well as the node's nominal role:

Actions

NSO rule-based HA can be controlled through several actions. All actions are found under /high-availability/. The available actions are listed below:

Status Check

The current state of NSO rule-based HA can be monitored by observing /high-availability/status/. Information can be found about the current active HA mode and the current assigned role. For nodes with active mode primary, a list of connected nodes and their source IP addresses is shown. For nodes with assigned role secondary the latest result of the be-secondary operation is listed. All NSO rule-based HA status information is non-replicated operational data - the result here will differ between nodes connected in an HA setup.

Tail-f HCC Package

The Tail-f HCC package extends the built-in HA functionality by providing virtual IP addresses (VIPs) that can be used to connect to the NSO HA group primary node. HCC ensures that the VIP addresses are always bound by the HA group primary and never bound by a secondary. Each time a node transitions between primary and secondary states HCC reacts by binding (primary) or unbinding (secondary) the VIP addresses.

HCC manages IP addresses at the link layer (OSI layer 2) for Ethernet interfaces, and optionally, also at the network layer (OSI layer 3) using BGP router advertisements. The layer-2 and layer-3 functions are mostly independent and this document describes the details of each one separately. However, the layer-3 function builds on top of the layer-2 function. The layer-2 function is always necessary, otherwise, the Linux kernel on the primary node would not recognize the VIP address or accept traffic directed to it.

Dependencies

Both the HCC layer-2 VIP and layer-3 BGP functionality depend on iproute2 utilities and awk. An optional dependency is arping (either from iputils or Thomas Habets arping implementation) which allows HCC to announce the VIP to MAC mapping to all nodes in the network by sending gratuitous ARP requests.

The HCC layer-3 BGP functionality depends on the daemon version 2.x being installed on each NSO host that is configured to run HCC in BGP mode.

GoBGP is open-source software originally developed by NTT Communications and released under the Apache License 2.0. GoBGP can be obtained directly from https://osrg.github.io/gobgp/ and is also packaged for mainstream Linux distributions.

The HCC layer-3 DNS Update functionality depends on the command line utility nsupdate.

Tools Dependencies are listed below:

Same as with built-in HA functionality, all NSO instances must be configured to run in HA mode. See the on how to enable HA on NSO instances.

Running the HCC Package with NSO as a Non-Root User

GoBGP uses TCP port 179 for its communications and binds to it at startup. As port 179 is considered a privileged port it is normally required to run gobgpd as root.

When NSO is running as a non-root user the gobgpd command will be executed as the same user as NSO and will prevent gobgpd from binding to port 179.

There a multiple ways of handling this and two are listed here.

1. Set capability CAP_NET_BIND_SERVICE on the gobgpd file. May not be supported by all Linux distributions.

2. Set the owner to root and the setuid bit of the gobgpd file. Works on all Linux distributions.

3. The

Tail-f HCC Compared with HCC Version 4.x and Older

HA Group Management Decisions

Tail-f HCC 5.x or later does not participate in decisions on which NSO node is primary or secondary. These decisions are taken by NSO's built-in HA and then pushed as notifications to HCC. The NSO built-in HA functionality is available in NSO starting with version 5.4, where older NSO versions are not compatible with the HCC 5.x or later.

Embedded BGP Daemon

HCC 5.x or later operates a GoBGP daemon as a subprocess completely managed by NSO. The old HCC function pack interacted with an external Quagga BGP daemon using a NED interface.

Automatic Interface Assignment

HCC 5.x or later automatically associates VIP addresses with Linux network interfaces using the ip utility from the iproute2 package. VIP addresses are also treated as /32 without defining a new subnet. The old HCC function pack used explicit configuration to associate VIPs with existing addresses on each NSO host and define IP subnets for VIP addresses.

Upgrading

Since version 5.0, HCC relies on the NSO built-in HA for cluster management and only performs address or route management in reaction to cluster changes. Therefore, no special measures are necessary if using HCC when performing an NSO version upgrade or a package upgrade. Instead, you should follow the standard best practice HA upgrade procedure from .

A reference to upgrade examples can be found in the NSO example set under examples.ncs/development-guide/high-availability/hcc/README.

Layer-2

The purpose of the HCC layer-2 functionality is to ensure that the configured VIP addresses are bound in the Linux kernel of the NSO primary node only. This ensures that the primary node (and only the primary node) will accept traffic directed toward the VIP addresses.

HCC also notifies the local layer-2 network when VIP addresses are bound by sending Gratuitous ARP (GARP) packets. Upon receiving the Gratuitous ARP, all the nodes in the network update their ARP tables with the new mapping so they can continue to send traffic to the non-failed, now primary node.

Operational Details

HCC binds the VIP addresses as additional (alias) addresses on existing Linux network interfaces (e.g. eth0). The network interface for each VIP is chosen automatically by performing a kernel routing lookup on the VIP address. That is, the VIP will automatically be associated with the same network interface that the Linux kernel chooses to send traffic to the VIP.

This means that you can map each VIP onto a particular interface by defining a route for a subnet that includes the VIP. If no such specific route exists the VIP will automatically be mapped onto the interface of the default gateway.

Configuration

The layer-2 functionality is configured by providing a list of IPv4 and/or IPv6 VIP addresses and enabling HCC. The VIP configuration parameters are found under /hcc:hcc.

Global Layer-2 Configuration:

Example Configuration

Layer-3 BGP

The purpose of the HCC layer-3 BGP functionality is to operate a BGP daemon on each NSO node and to ensure that routes for the VIP addresses are advertised by the BGP daemon on the primary node only.

The layer-3 functionality is an optional add-on to the layer-2 functionality. When enabled, the set of BGP neighbors must be configured separately for each NSO node. Each NSO node operates an embedded BGP daemon and maintains connections to peers but only the primary node announces the VIP addresses.

The layer-3 functionality relies on the layer-2 functionality to assign the virtual IP addresses to one of the host's interfaces. One notable difference in assigning virtual IP addresses when operating in Layer-3 mode is that the virtual IP addresses are assigned to the loopback interface lo rather than to a specific physical interface.

Operational Details

HCC operates a subprocess as an embedded BGP daemon. The BGP daemon is started, configured, and monitored by HCC. The HCC YANG model includes basic BGP configuration data and state data.

Operational data in the YANG model includes the state of the BGP daemon subprocess and the state of each BGP neighbor connection. The BGP daemon writes log messages directly to NSO where the HCC module extracts updated operational data and then repeats the BGP daemon log messages into the HCC log verbatim. You can find these log messages in the developer log (devel.log).

Configuration

The layer-3 BGP functionality is configured as a list of BGP configurations with one list entry per node. Configurations are separate because each NSO node usually has different BGP neighbors with their own IP addresses, authentication parameters, etc.

The BGP configuration parameters are found under /hcc:hcc/bgp/node{id}.

Per-Node Layer-3 Configuration:

Each NSO node can connect to a different set of BGP neighbors. For each node, the BGP neighbor list configuration parameters are found under /hcc:hcc/bgp/node{id}/neighbor{address}.

Per-Neighbor BGP Configuration:

Example

Layer-3 DNS Update

The purpose of the HCC layer-3 DNS Update functionality is to notify a DNS server of the IP address change of the active primary NSO server, allowing the DNS server to update the DNS record for the given domain name.

Geographically redundant NSO setup typically relies on DNS support. To enable this use case, tailf-hcc can dynamically update DNS with the nsupdate utility on HA status change notification.

The DNS server used should support updates through nsupdate command (RFC 2136).

Operational Details

HCC listens on the underlying NSO HA notifications stream. When HCC receives a notification about an NSO node being Primary, it updates the DNS Server with the IP address of the Primary NSO for the given hostname. The HCC YANG model includes basic DNS configuration data and operational status data.

Operational data in the YANG model includes the result of the latest DNS update operation.

If the DNS Update is unsuccessful, an error message will be populated in operational data, for example:

Configuration

The layer-3 DNS Update functionality needs DNS-related information like DNS server IP address, port, zone, etc, and information about NSO nodes involved in HA - node, ip, and location.

The DNS configuration parameters are found under /hcc:hcc/dns.

Layer-3 DNS Configuration:

Each NSO node can be placed in a separate Location/Site/Availability-Zone. This is configured as a list member configuration, with one list entry per node ID. The member list configuration parameters are found under /hcc:hcc/dns/member{node-id}.

Example

Here is an example configuration for a setup of two dual-stack NSO nodes, node-1 and node-2, that have an IPv4 and an IPv6 address configured. The configuration also sets up an update signing with the specified key.

Usage

This section describes basic deployment scenarios for HCC. Layer-2 mode is demonstrated first and then the layer-3 BGP functionality is configured in addition:

A reference to container-based examples for the layer-2 and layer-3 deployment scenarios described here can be found in the NSO example set under examples.ncs/development-guide/high-availability/hcc.

Both scenarios consist of two test nodes: london and paris with a single IPv4 VIP address. For the layer-2 scenario, the nodes are on the same network. The layer-3 scenario also involves a BGP-enabled router node as the london and paris nodes are on two different networks.

Layer-2 Deployment

The layer-2 operation is configured by simply defining the VIP addresses and enabling HCC. The HCC configuration on both nodes should match, otherwise, the primary node's configuration will overwrite the secondary node configuration when the secondary connects to the primary node.

Addresses:

Configuring VIPs:

Verifying VIP Availability:

Once enabled, HCC on the HA group primary node will automatically assign the VIP addresses to corresponding Linux network interfaces.

On the secondary node, HCC will not configure these addresses.

Layer-2 Example Implementation:

A reference to a container-based example of the layer-2 scenario can be found in the NSO example set. See the examples.ncs/development-guide/high-availability/hcc/README

Enabling Layer-3 BGP

Layer-3 operation is configured for each NSO HA group node separately. The HCC configuration on both nodes should match, otherwise, the primary node's configuration will overwrite the configuration on the secondary node.

Addresses:

Configuring BGP for Paris Node:

Configuring BGP for London Node:

Check BGP Neighbor Connectivity:

Check neighbor connectivity on the paris primary node. Note that its connection to neighbor 192.168.31.2 (router) is ESTABLISHED.

Check neighbor connectivity on the london secondary node. Note that the primary node also has an ESTABLISHED connection to its neighbor 192.168.30.2 (router). The primary and secondary nodes both maintain their BGP neighbor connections at all times when BGP is enabled, but only the primary node announces routes for the VIPs.

Check Advertised BGP Routes Neighbors:

Check the BGP routes received by the router.

The VIP subnet is routed to the paris host, which is the primary node.

Layer-3 BGP Example Implementation:

A reference to a container-based example of the combined layer-2 and layer-3 BGP scenario can be found in the NSO example set. See the examples.ncs/development-guide/high-availability/hcc/README

Enabling Layer-3 DNS

If enabled prior to the HA being established, HCC will update the DNS server with the IP address of the Primary node once a primary is selected.

If an HA is already operational, and Layer-3 DNS is enabled and configured afterward, HCC will not update the DNS server automatically. An automatic DNS server update will only happen if a HA switchover happens. HCC exposes an update action to manually trigger an update to the DNS server with the IP address of the primary node.

DNS Update Action:

The user can explicitly update DNS from the specific NSO node by running the update action.

Check the result of invoking the DNS update utility using the operational data in /hcc/dns:

One way to verify DNS server updates is through the nslookup program. However, be mindful of the DNS caching mechanism, which may cache the old value for the amount of time controlled by the TTL setting.

DNS get-node-location Action:

/hcc/dns/member holds the information about all members involved in HA. The get-node-location action provides information on the location of an NSO node.

Data Model

The HCC data model can be found in the HCC package (tailf-hcc.yang).

Setup with an External Load Balancer

As an alternative to the HCC package, NSO built-in HA, either rule-based or HA Raft, can also be used in conjunction with a load balancer device in a reverse proxy configuration. Instead of managing the virtual IP address directly as HCC does, this setup relies on an external load balancer to route traffic to the currently active primary node.

The load balancer uses HTTP health checks to determine which node is currently the active primary. The example, found in the examples.ncs/development-guide/high-availability/load-balancer directory uses HTTP status codes on the health check endpoint to easily distinguish whether the node is currently primary or not.

In the example, freely available HAProxy software is used as a load balancer to demonstrate the functionality. It is configured to steer connections on localhost to either of the TCP port 2024 (SSH CLI) and TCP port 8080 (web UI and RESTCONF) to the active node in a 2-node HA cluster. The HAProxy software is required if you wish to run this example yourself.

You can start all the components in the example by running the make build start command. At the beginning, the first node n1 is the active primary. Connecting to the localhost port 2024 will establish a connection to this node:

Then, you can disable the high availability subsystem on n1 to simulate a node failure.

Disconnect and wait a few seconds for the built-in HA to perform the failover to node n2. The time depends on the high-availability/settings/reconnect-interval and is set quite aggressively in this example to make the failover in about 6 seconds. Reconnect with the SSH client and observe the connection is now made to the fail-over node which has become the active primary:

Finally, shut down the example with the make stop clean command.

NB Listens to Addresses on HA Primary for Load Balancers

NSO can be configured for the HA primary to listen on additional ports for the northbound interfaces NETCONF, RESTCONF, the web server (including JSON-RPC), and the CLI over SSH. Once a different node transitions to role primary the configured listen addresses are brought up on that node instead.

When the following configuration is added to ncs.conf, then the primary HA node will listen(2) and bind(2) port 1830 on the wildcard IPv4 and IPv6 addresses.

A similar configuration can be added for other NB interfaces, see the ha-primary-listen list under /ncs-config/{restconf,webui,cli}.

HA Framework Requirements

If an external HAFW is used, NSO only replicates the CDB data. NSO must be told by the HAFW which node should be primary and which nodes should be secondaries.

The HA framework must also detect when nodes fail and instruct NSO accordingly. If the primary node fails, the HAFW must elect one of the remaining secondaries and appoint it the new primary. The remaining secondaries must also be informed by the HAFW about the new primary situation.

Mode of Operation

NSO must be instructed through the ncs.conf configuration file that it should run in HA mode. The following configuration snippet enables HA mode:

Make sure to restart the ncs process for the changes to take effect.

The IP address and the port above indicate which IP and which port should be used for the communication between the HA nodes. extra-listen is an optional list of ip:port pairs that a HA primary also listens to for secondary connections. For IPv6 addresses, the syntax [ip]:port may be used. If the :port is omitted, port is used. The tick-timeout is a duration indicating how often each secondary must send a tick message to the primary indicating liveness. If the primary has not received a tick from a secondary within 3 times the configured tick time, the secondary is considered to be dead. Similarly, the primary sends tick messages to all the secondaries. If a secondary has not received any tick messages from the primary within the 3 times the timeout, the secondary will consider the primary dead and report accordingly.

A HA node can be in one of three states: NONE, SECONDARY or PRIMARY. Initially, a node is in the NONE state. This implies that the node will read its configuration from CDB, stored locally on file. Once the HA framework has decided whether the node should be a secondary or a primary the HAFW must invoke either the methods Ha.beSecondary(primary) or Ha.bePrimary()

When an NSO HA node starts, it always starts up in mode NONE. At this point, there are no other nodes connected. Each NSO node reads its configuration data from the locally stored CDB and applications on or off the node may connect to NSO and read the data they need. Although write operations are allowed in the NONE state it is highly discouraged to initiate southbound communication unless necessary. A node in NONE state should only be used to configure NSO itself or to do maintenance such as upgrades. When in NONE state, some features are disabled, including but not limited to:

• commit queue

• NSO scheduler

• nano-service side effect queue

This is to avoid situations where multiple NSO nodes are trying to perform the same southbound operation simultaneously.

At some point, the HAFW will command some nodes to become secondary nodes of a named primary node. When this happens, each secondary node tracks changes and (logically or physically) copies all the data from the primary. Previous data at the secondary node is overwritten.

Note that the HAFW, by using NSO's start phases, can make sure that NSO does not start its northbound interfaces (NETCONF, CLI, ...) until the HAFW has decided what type of node it is. Furthermore once a node has been set to the SECONDARY state, it is not possible to initiate new write transactions towards the node. It is thus never possible for an agent to write directly into a secondary node. Once a node is returned either to the NONE state or to the PRIMARY state, write transactions can once again be initiated towards the node.

The HAFW may command a secondary node to become primary at any time. The secondary node already has up-to-date data, so it simply stops receiving updates from the previous primary. Presumably, the HAFW also commands the primary node to become a secondary node or takes it down, or handles the situation somehow. If it has crashed, the HAFW tells the secondary to become primary, restarts the necessary services on the previous primary node, and gives it an appropriate role, such as secondary. This is outside the scope of NSO.

Each of the primary and secondary nodes has the same set of all callpoints and validation points locally on each node. The start sequence has to make sure the corresponding daemons are started before the HAFW starts directing secondary nodes to the primary, and before replication starts. The associated callbacks will however only be executed at the primary. If e.g. the validation executing at the primary needs to read data that is not stored in the configuration and only available on another node, the validation code must perform any needed RPC calls.

If the order from the HAFW is to become primary, the node will start to listen for incoming secondaries at the ip:port configured under /ncs-config/ha. The secondaries TCP connect to the primary and this socket is used by NSO to distribute the replicated data.

If the order is to be a secondary, the node will contact the primary and possibly copy the entire configuration from the primary. This copy is not performed if the primary and secondary decide that they have the same version of the CDB database loaded, in which case nothing needs to be copied. This mechanism is implemented by use of a unique token, the transaction id - it contains the node id of the node that generated it and a time stamp, but is effectively "opaque".

This transaction ID is generated by the cluster primary each time a configuration change is committed, and all nodes write the same transaction ID into their copy of the committed configuration. If the primary dies and one of the remaining secondaries is appointed the new primary, the other secondaries must be told to connect to the new primary. They will compare their last transaction ID to the one from the newly appointed primary. If they are the same, no CDB copy occurs. This will be the case unless a configuration change has sneaked in since both the new primary and the remaining secondaries will still have the last transaction ID generated by the old primary - the new primary will not generate a new transaction ID until a new configuration change is committed. The same mechanism works if a secondary node is simply restarted. No cluster reconfiguration will lead to a CDB copy unless the configuration has been changed in between.

Northbound agents should run on the primary, an agent can't commit write operations at a secondary node.

When an agent commits its CDB data, CDB will stream the committed data out to all registered secondaries. If a secondary dies during the commit, nothing will happen, the commit will succeed anyway. When and if the secondary reconnects to the cluster, the secondary will have to copy the entire configuration. All data on the HA sockets between NSO nodes only go in the direction from the primary to the secondaries. A secondary that isn't reading its data will eventually lead to a situation with full TCP buffers at the primary. In principle, it is the responsibility of HAFW to discover this situation and notify the primary NSO about the hanging secondary. However, if 3 times the tick timeout is exceeded, NSO will itself consider the node dead and notify the HAFW. The default value for tick timeout is 20 seconds.

The primary node holds the active copy of the entire configuration data in CDB. All configuration data has to be stored in CDB for replication to work. At a secondary node, any request to read will be serviced while write requests will be refused. Thus, the CDB subscription code works the same regardless of whether the CDB client is running at the primary or at any of the secondaries. Once a secondary has received the updates associated to a commit at the primary, all CDB subscribers at the secondary will be duly notified about any changes using the normal CDB subscription mechanism.

If the system has been set up to subscribe for NETCONF notifications, the secondaries will have all subscriptions as configured in the system, but the subscription will be idle. All NETCONF notifications are handled by the primary, and once the notifications get written into stable storage (CDB) at the primary, the list of received notifications will be replicated to all secondaries.

Security Aspects

We specify in ncs.conf which IP address the primary should bind for incoming secondaries. If we choose the default value 0.0.0.0 it is the responsibility of the application to ensure that connection requests only arrive from acceptable trusted sources through some means of firewalling.

A cluster is also protected by a token, a secret string only known to the application. The Ha.connect() method must be given the token. A secondary node that connects to a primary node negotiates with the primary using a CHAP-2-like protocol, thus both the primary and the secondary are ensured that the other end has the same token without ever revealing their own token. The token is never sent in clear text over the network. This mechanism ensures that a connection from an NSO secondary to a primary can only succeed if they both have the same token.

It is indeed possible to store the token itself in CDB, thus an application can initially read the token from the local CDB data, and then use that token in . the constructor for the Ha class. In this case, it may very well be a good idea to have the token stored in CDB be of type tailf:aes-256-cfb-128-encrypted-string.

If the actual CDB data that is sent on the wire between cluster nodes is sensitive, and the network is untrusted, the recommendation is to use IPSec between the nodes. An alternative option is to decide exactly which configuration data is sensitive and then use the tailf:aes-256-cfb-128-encrypted-string type for that data. If the configuration data is of type tailf:aes-256-cfb-128-encrypted-string the encrypted data will be sent on the wire in update messages from the primary to the secondaries.

API

There are two APIs used by the HA framework to control the replication aspects of NSO. First, there exists a synchronous API used to tell NSO what to do, secondly, the application may create a notifications socket and subscribe to HA-related events where NSO notifies the application on certain HA-related events such as the loss of the primary, etc. The HA-related notifications sent by NSO are crucial to how to program the HA framework.

The HA-related classes reside in the com.tailf.ha package. See Javadocs for reference. The HA notifications-related classes reside in the com.tailf.notif package, See Javadocs for reference.

Ticks

The configuration parameter /ncs-cfg/ha/tick-timeout is by default set to 20 seconds. This means that every 20 seconds each secondary will send a tick message on the socket leading to the primary. Similarly, the primary will send a tick message every 20 seconds on every secondary socket.

This aliveness detection mechanism is necessary for NSO. If a socket gets closed all is well, NSO will clean up and notify the application accordingly using the notifications API. However, if a remote node freezes, the socket will not get properly closed at the other end. NSO will distribute update data from the primary to the secondaries. If a remote node is not reading the data, TCP buffer will get full and NSO will have to start to buffer the data. NSO will buffer data for at most tickTime times 3 time units. If a tick has not been received from a remote node within that time, the node will be considered dead. NSO will report accordingly over the notifications socket and either remove the hanging secondary or, if it is a secondary that loses contact with the primary, go into the initial NONE state.

If the HAFW can be really trusted, it is possible to set this timeout to PT0S, i.e zero, in which case the entire dead-node-detection mechanism in NSO is disabled.

Relay Secondaries

The normal setup of an NSO HA cluster is to have all secondaries connected directly to the primary. This is a configuration that is both conceptually simple and reasonably straightforward to manage for the HAFW. In some scenarios, in particular a cluster with multiple secondaries at a location that is network-wise distant from the primary, it can however be sub-optimal, since the replicated data will be sent to each remote secondary individually over a potentially low-bandwidth network connection.

To make this case more efficient, we can instruct a secondary to be a relay for other secondaries, by invoking the Ha.beRelay() method. This will make the secondary start listening on the IP address and port configured for HA in ncs.conf, and handle connections from other secondaries in the same manner as the cluster primary does. The initial CDB copy (if needed) to a new secondary will be done from the relay secondary, and when the relay secondary receives CDB data for replication from its primary, it will distribute the data to all its connected secondaries in addition to updating its own CDB copy.

To instruct a node to become a secondary connected to a relay secondary, we use the Ha.beSecondary() method as usual, but pass the node information for the relay secondary instead of the node information for the primary. I.e. the "sub-secondary" will in effect consider the relay secondary as its primary. To instruct a relay secondary to stop being a relay, we can invoke the Ha.beSecondary() method with the same parameters as in the original call. This is a no-op for a "normal" secondary, but it will cause a relay secondary to stop listening for secondary connections, and disconnect any already connected "sub-secondaries".

This setup requires special consideration by the HAFW. Instead of just telling each secondary to connect to the primary independently, it must set up the secondaries that are intended to be relays, and tell them to become relays, before telling the "sub-secondaries" to connect to the relay secondaries. Consider the case of a primary M and a secondary S0 in one location, and two secondaries S1 and S2 in a remote location, where we want S1 to act as relay for S2. The setup of the cluster then needs to follow this procedure:

1. Tell M to be primary.

2. Tell S0 and S1 to be secondary with M as primary.

3. Tell S1 to be relay.

4. Tell S2 to be secondary with S1 as primary.

Conversely, the handling of network outages and node failures must also take the relay secondary setup into account. For example, if a relay secondary loses contact with its primary, it will transition to the NONE state just like any other secondary, and it will then disconnect its sub-secondaries which will cause those to transition to NONE too, since they lost contact with "their" primary. Or if a relay secondary dies in a way that is detected by its sub-secondaries, they will also transition to NONE. Thus in the example above, S1 and S2 needs to be handled differently. E.g. if S2 dies, the HAFW probably won't take any action, but if S1 dies, it makes sense to instruct S2 to be a secondary of M instead (and when S1 comes back, perhaps tell S2 to be a relay and S1 to be a secondary of S2).

Besides the use of Ha.beRelay(), the API is mostly unchanged when using relay secondaries. The HA event notifications reporting the arrival or the death of a secondary are still generated only by the "real" cluster primary. If the Ha.HaStatus() method is used towards a relay secondary, it will report the node state as SECONDARY_RELAY rather than just SECONDARY, and the array of nodes will have its primary as the first element (same as for a "normal" secondary), followed by its "sub-secondaries" (if any).

CDB Replication

When HA is enabled in ncs.conf, CDB automatically replicates data written on the primary to the connected secondary nodes. Replication is done on a per-transaction basis to all the secondaries in parallel and is synchronous. When NSO is in secondary mode the northbound APIs are in read-only mode, that is the configuration can not be changed on a secondary other than through replication updates from the primary. It is still possible to read from for example NETCONF or CLI (if they are enabled) on a secondary. CDB subscriptions work as usual. When NSO is in the NONE state CDB is unlocked and it behaves as when NSO is not in HA mode at all.

Unlike configuration data, operational data is replicated only if it is defined as persistent in the data model (using the tailf:persistent extension).

The Problem

Users log into NSO through the CLI, NETCONF, RESTCONF, SNMP, or via the Web UI. In either case, users need to be authenticated. That is, a user needs to present credentials, such as a password or a public key to gain access. As an alternative, for RESTCONF, users can be authenticated via token validation.

Once a user is authenticated, all operations performed by that user need to be authorized. That is, certain users may be allowed to perform certain tasks, whereas others are not. This is called authorization. We differentiate between the authorization of commands and the authorization of data access.