NSO Device Manager
Learn the concepts of NSO device management.
Last updated
Learn the concepts of NSO device management.
Last updated
© Copyright 2024 Cisco Systems, Inc. | This site is part of the official Cisco Crosswork NSO documentation set.
The NSO device manager is the center of NSO. The device manager maintains a flat list of all managed devices. NSO keeps the primary copy of the configuration for each managed device in CDB. Whenever a configuration change is done to the list of device configuration primary copies, the device manager will partition this network configuration change into the corresponding changes for the managed devices. The device manager passes on the required changes to the NEDs (Network Element Drivers). A NED needs to be installed for every type of device OS, like Cisco IOS NED, Cisco XR NED, Juniper JUNOS NED, etc. The NEDs communicate through the native device protocol southbound.
The NEDs fall into the following categories:
NETCONF-capable device: The Device Manager will produce NETCONF edit-config RPC operations for each participating device.
SNMP device: The Device Manager translates the changes made to the configuration into the corresponding SNMP SET PDUs.
Device with Cisco CLI: The device has a CLI with the same structure as Cisco IOS or XR routers. The Device Manager and a CLI NED are used to produce the correct sequence of CLI commands which reflects the changes made to the configuration.
Other devices: For devices that do not fit into any of the above-mentioned categories, a corresponding Generic NED is invoked. Generic NEDs are used for proprietary protocols like REST and for CLI flavors that do not resemble IOS or XR. The Device Manager will inform the Generic NED about the made changes and the NED will translate these to the appropriate operations toward the device.
NSO orchestrates an atomic transaction that has the very desirable characteristic of either the transaction as a whole ending up on all participating devices and in the NSO primary copy, or alternatively, the whole transaction getting aborted and resultingly, all changes getting automatically rolled back.
The architecture of the NETCONF protocol is the enabling technology making it possible to push out configuration changes to managed devices and then in the case of other errors, roll back changes. Devices that do not support NETCONF, i.e., devices that do not have transactional capabilities can also participate, however depending on the device, error recovery may not be as good as it is for a proper NETCONF-enabled device.
To understand the main idea behind the NSO device manager it is necessary to understand the NSO data model and how NSO incorporates the YANG data models from the different managed devices.
The NEDs will publish YANG data models even for non-NETCONF devices. In the case of SNMP the YANG models are generated from the MIBs. For JunOS devices the JunOS NED generates a YANG from the JunOS XML Schema. For Schema-less devices like CLI devices, the NED developer writes YANG models corresponding to the CLI structure. The result of this is the device manager and NSO CDB has YANG data models for all devices independent of the underlying protocol.
Throughout this section, we will use the examples.ncs/service-provider/mpls-vpn example. The example network consists of Cisco ASR 9k and Juniper core routers (P and PE) and Cisco IOS-based CE routers.
The central part of the NSO YANG model, in the file tailf-ncs-devices.yang, has the following structure:
Each managed device is uniquely identified by its name, which is a free-form text string. This is typically the DNS name of the managed device but could equally well be the string format of the IP address of the managed device or anything else. Furthermore, each managed device has a mandatory address/port pair that together with the authgroup leaf provides information to NSO on how to connect and authenticate over SSH/NETCONF to the device. Each device also has a mandatory parameter device-type that specifies which southbound protocol to use for communication with the device.
The following device types are available:
NETCONF
CLI: A corresponding CLI NED is needed to communicate with the device. This requires YANG models with the appropriate annotations for the device CLI.
SNMP: The device speaks SNMP, preferably in read-write mode.
Generic NED: A corresponding Generic NED is needed to communicate with the device. This requires YANG models and Java code.
The NSO CLI command below lists the NED types for the devices in the example network.
The empty container /ncs:devices/device/config is used as a mount point for the YANG models from the different managed devices.
As previously mentioned, NSO needs the following information to manage a device:
The IP/Port of the device and authentication information.
Some or all of the YANG data models for the device.
In the example setup, the address and authentication information are provided in the NSO database (CDB) initialization file. There are many different ways to add new managed devices. All of the NSO northbound interfaces can be used to manipulate the set of managed devices. This will be further described later.
Once NSO has started you can inspect the meta information for the managed devices through the NSO CLI. This is an example session:
Alternatively, this information could be retrieved from the NSO northbound NETCONF interface by running the simple Python-based netconf-console program towards the NSO NETCONF server.
All devices in the above two examples (Show Device Configuration in NSO CLI and Show Device Configuration in NETCONF) have /devices/device/state/admin-state set to unlocked, this will be described later in this section.
To communicate with a managed device, a NED for that device type needs to be loaded by NSO. A NED contains the YANG model for the device and corresponding driver code to talk CLI, REST, SNMP, etc. NEDs are distributed as packages.
The CLI command in the above example (Installed Packages) shows all the loaded packages. NSO loads packages at startup and can reload packages at run-time. By default, the packages reside in the packages directory in the NSO run-time directory.
Once you have access to the network information for a managed device, its IP address and authentication information, as well as the data models of the device, you can actually manage the device from NSO.
You start the ncs daemon in a terminal like:
Which is the same as, NSO loads it config from a ncs.conf file
During development, it is sometimes convenient to run ncs in the foreground as:
Once the daemon is running, you can issue the command:
To get more information about options to ncs do:
The ncs --status command produces a lengthy list describing for example which YANG modules are loaded in the system. This is a valuable debug tool.
The same information is also available in the NSO CLI (and thus through all available northbound interfaces, including Maapi for Java programmers)
When the NSO daemon is running and has been initialized with IP/Port and authentication information as well as imported all modules you can start to manage devices through NSO.
NSO provides the ability to synchronize the configuration to or from the device. If you know that the device has the correct configuration you can choose to synchronize from a managed device whereas if you know NSO has the correct device configuration and the device is incorrect, you can choose to synchronize from NSO to the device.
In the normal case, the configuration on the device and the copy of the configuration inside NSO should be identical.
In a cold start situation like in the mpls-vpn example, where NSO is empty and there are network devices to talk to, it makes sense to synchronize from the devices. You can choose to synchronize from one device at a time or from all devices at once. Here is a CLI session to illustrate this.
The command devices sync-from, in example (Synchronize from Devices), is an action that is defined in the NSO data model. It is important to understand the model-driven nature of NSO. All devices are modeled in YANG, network services like MPLS VPN are also modeled in YANG, and the same is true for NSO itself. Anything that can be performed over the NSO CLI or any north-bound is defined in the YANG files. The NSO YANG files are located here:
All packages comes with YANG files as well. For example the directory packages/cisco-ios/src/yang/ contains the YANG definition of an IOS device.
The tailf-ncs.yang is the main part of the NSO YANG data model. The file mode tailf-ncs.yang includes all parts of the model from different files.
The actions sync-from and sync-to are modeled in the file tailf-ncs-devices.yang. The sync action(s) are defined as:
Synchronizing from NSO to the device is common when a device has been configured out-of-band. NSO has no means to enforce that devices are not directly reconfigured behind the scenes of NSO; however, once an out-of-band configuration has been performed, NSO can detect the fact. When this happens it may (or may not, depending on the situation at hand) make sense to synchronize from NSO to the device, i.e. undo the rogue reconfigurations.
The command to do that is:
A dry-run option is available for the action sync-to.
This makes it possible to investigate the changes before they are transmitted to the devices.
sync-fromIt is possible to synchronize a part of the configuration (a certain subtree) from the device using the partial-sync-from action located under /devices. While it is primarily intended to be used by service developers as described in Partial Sync, it is also possible to use directly from the NSO CLI (or any other northbound interface). The example below (Example of Running partial-sync-from Action via CLI) illustrates using this action via CLI, using a router device from examples.ncs/getting-started/developing-with-ncs/0-router-network.
It is now possible to configure several devices through the NSO inside the same network transaction. To illustrate this, start the NSO CLI from a terminal application.
The example above (Configure Devices) illustrates a multi-host transaction. In the same transaction, three hosts were re-configured. Had one of them failed, or been non-operational, the transaction as a whole would have failed.
As seen from the output of the command commit dry-run outformat native, NSO generates the native CLI and NETCONF commands which will be sent to each device when the transaction is committed.
Since the /devices/device/config path contains different models depending on the augmented device model NSO uses the data model prefix in the CLI names; ios, cisco-ios-xr and junos. Different data models might use the same name for elements and the prefix avoids name clashes.
NSO uses different underlying techniques to implement the atomic transactional behavior in case of any error. NETCONF devices are straightforward using confirmed commit. For CLI devices like IOS NSO calculates the reverse diff to restore the configuration to the state before the transaction was applied.
Each managed device needs to be configured with the IP address and the port where the CLI, NETCONF server, etc. of the managed device listens for incoming requests.
Connections are established on demand as they are needed. It is possible to explicitly establish connections, but that functionality is mostly there for troubleshooting connection establishment. We can, for example, do:
We were able to connect to all managed devices. It is also possible to explicitly attempt to test connections to individual managed devices:
Established connections are typically not closed right away when not needed, but rather pooled according to the rules described in Device Session Pooling. This applies to NETCONF sessions as well as sessions established by CLI or generic NEDs via a connection-oriented protocol. In addition to session pooling, underlying SSH connections for NETCONF devices are also reused. Note that a single NETCONF session occupies one SSH channel inside an SSH connection, so multiple NETCONF sessions can co-exist in a single connection. When an SSH connection has been idle (no SSH channels open) for 2 minutes, the SSH connection is closed. If a new connection is needed later, a connection is established on demand.
Three configuration parameters can be used to control the connection establishment: connect-timeout, read-timeout, and write-timeout. In the NSO data model file tailf-ncs-devices.yang, these timeouts are modeled as:
Thus, to change these parameters (globally for all managed devices) you do:
Or, to use a profile:
When NSO connects to a managed device, it requires authentication information for that device. The authgroups are modeled in the NSO data model:
Each managed device must refer to a named authgroup. The purpose of an authentication group is to map local users to remote users together with the relevant SSH authentication information.
Southbound authentication can be done in two ways. One is to configure the stored user and credential components as shown in the example below (Configured authgroup) and the next example (authgroup default-map). The other way is to configure a callback to retrieve user and credentials on demand as shown in the example below (authgroup-callback).
In the example above (Configured authgroup) in the auth group named default, the two local users oper and admin shall use the remote users' name oper and admin respectively with identical passwords.
Inside an authgroup, all local users need to be enumerated. Each local user name must have credentials configured which should be used for the remote host. In centralized AAA environments, this is usually a bad strategy. You may also choose to instantiate a default-map. If you do that it probably only makes sense to specify the same user name/password pair should be used remotely as the pair that was used to log into NSO.
In the example (Configured authgroup), only two users admin and oper were configured. If the default-map in example (authgroup default-map) is configured, all local users not found in the umap list will end up in the default-map. For example, if the user rocky logs in to NSO with the password secret. Since NSO has a built-in SSH server and also a built-in HTTPS server, NSO will be able to pick up the clear text passwords and can then reuse the same password when NSO attempts to establish southbound SSH connections. The user rocky will end up in the default-map and when NSO attempts to propagate rocky's changes towards the managed devices, NSO will use the remote user name rocky with whatever password rocky used to log into NSO.
Authenticating southbound using stored configuration has two main components to define remote user and remote credentials. This is defined by the authgroup. As for the southbound user, there exist two options, the same user logged in to NSO or another user, as specified in the authgroup. As for the credentials, there are three options.
Regular password.
Public key. This means that a private key, either from a file in the user's SSH key directory, or one that is configured in the /ssh/private-key list in the NSO configuration, is used for authentication. Refer to Publickey Authentication for the details on how the private key is selected.
Finally, an interesting option is to use the 'same-pass' option. Since NSO runs its own SSH server and its own SSL server, NSO can pick up the password of a user in clear text. Hence, if the 'same-pass' option is chosen for an authgroup, NSO will reuse the same password when attempting to connect southbound to a managed device.
NSO can connect to a device that is using multi-factor authentication. For this, the authgroup must be configured with an executable for handling the keyboard-interactive part, and optionally some opaque data that is passed to the executable. ie., the /devices/authgroups/group/umap/mfa/executable and /devices/authgroups/group/umap/mfa/opaque (or under default-map for users that are not in umap) must be configured.
The prompts from the SSH server (including the password prompt and any additional challenge prompts) are passed to the stdin of the executable along with some other relevant data. The executable must write a single line to its stdout as the reply to the prompt. This is the reply that NSO sends to the SSH server.
For example, with the above configured for the authgroup, if the user admin is trying to login to the device dev0 with password admin, this is the line that is sent to the stdin of the handle_mfa.py script:
The input to the script is the device, username, password, opaque data, as well as the name, instruction, and prompt from the SSH server. All these fields are base64 encoded, and separated by a semi-colon (';'). So, the above line in effect encodes the following:
A small Python program can be used to implement the keyboard-interactive authentication towards a device, such as:
This script will then be invoked with the above fields for every prompt from the server, and the corresponding output from the script will be sent as the reply to the server.
In the case of authenticating southbound using a callback, remote user and remote credentials are obtained by an action invocation. The action is defined by the callback-node and action-name as in the example below (authgroup-callback) and supported credentials are remote password and optionally a secondary password for the provided local user, authgroup, and device.
With remote passwords, you may encounter issues if you use special characters, such as quotes (") and backslash (\) in your password. See Configure Mode for recommendations on how to avoid running into password issues.
In the example above (authgroup-callback), the configuration for the umap entry of the oper user is changed to use a callback to retrieve southbound authentication credentials. Thus, NSO is going to invoke the action auth-cb defined in the callback-node callback. The callback node is of type instance-identifier and refers to the container called callback defined in the example, (authgroup-callback.yang), which includes an action defined by action-name auth-cb and uses groupings authgroup-callback-input-params and authgroup-callback-output-params for input and output parameters respectively. In the example, (authgroup-callback), authgroup-callback module was loaded in NSO within an example package. Package development and action callbacks are not described here but more can be read in Package Development, the section called DP API and Python API Overview.
Authentication groups and the functionality they bring come with some limitations on where and how it is used.
The callback option that enables authgroup-callback feature is not applicable for members of snmp-group list.
Generic devices that implement their own authentication scheme do not use any mapping or callback functionality provided by Authgroups.
Cluster nodes use their own authgroups and mapping model, thus functionality differs, e.g. callback option is not applicable.
Opening a session towards a managed device is potentially time and resource-consuming. Also, the probability that a recently accessed device is still subject to further requests is reasonably high. These are motives for having a managed devices session pool in NSO.
The NSO device session pool is by default active and normally needs no maintenance. However, under certain circumstances, it might be of interest to modify its behavior. Examples can be when some device type has characteristics that make session pooling undesired, or when connections to a specific device are very costly, and therefore the time that open sessions can stay in the pool should increase.
Changes from the default configuration of the NSO session pool should only be performed when absolutely necessary and when all effects of the change are understood.
NSO presents operational data that represent the current state of the session pool. To visualize this, we use the CLI to connect to NSO and force connection to all known devices:
We can now list all open sessions in the session-pool. But note that this is a live pool. Sessions will only remain open for a certain amount of time, the idle time.
In addition to the idle time for sessions, we can also see the type of device, current number of pooled sessions, and maximum number of pooled sessions.
We can close pooled sessions for specific devices.
And we can close all pooled sessions in the session pool.
The session pool configuration is found in the tailf-ncs-devices.yang submodel. The following part of the YANG device-profile-parameters grouping controls how the session pool is configured:
This grouping can be found in the NSO model under /ncs:devices/global-settings/session-pool, /ncs:devices/profiles/profile/session-pool and /ncs:devices/device/session-pool to be able to control session pooling for all devices, a group of devices, and a specific device respectively.
In addition under /ncs:devices/global-settings/session-pool/default it is possible to control the global max size of the session pool, as defined by the following yang snippet:
Let's illustrate the possibilities with an example configuration of the session pool:
In the above configuration, the default idle time is set to 100 seconds for all devices. A device profile called small is defined which contains a max-session value of 3 sessions, this profile is set on all ce* devices. The devices pe0 has a max-sessions 0 which implies that this device cannot be pooled. Let's connect all devices and see what happens in the session pool:
Now, we set an upper limit to the maximum number of sessions in the pool. Setting the value to 4 is too small for a real situation but serves the purpose of illustration:
The number of open sessions in the pool will be adjusted accordingly:
Some devices only allow a small number of concurrent sessions, in the extreme case it only allows one (for example through a terminal server). For this reason, NSO can limit the number of concurrent sessions to a device and make operations wait if the maximum number of sessions has been reached.
In other situations, we need to limit the number of concurrent connect attempts made by NSO. For example, the devices managed by NSO talk to the same server for authentication which can only handle a limited number of connections at a time.
The configuration for session limits is found in the tailf-ncs-devices.yang submodel. The following part of the YANG device-profile-parameters grouping controls how the session limits are configured:
This grouping can be found in the NSO model under /ncs:devices/global-settings/session-limits, /ncs:devices/profiles/profile/session-limits and /ncs:devices/device/session-limits to be able to control session limits for all devices, a group of devices, and a specific device respectively.
In addition, under /ncs:devices/global-settings/session-limits, it is possible to control the number of concurrent connect attempts allowed and the maximum time to wait for a device to be available to connect.
It is possible to turn on and off NED traffic tracing. This is often a good way to troubleshoot problems. To understand the trace output, a basic prerequisite is a good understanding of the native device interface. For NETCONF devices, an understanding of NETCONF RPC is a prerequisite. Similarly for CLI NEDs, a good understanding of the CLI capabilities of the managed devices is required.
To turn on southbound traffic tracing, we need to enable the feature and we must also configure a directory where we want the trace output to be written. It is possible to have the trace output in two different formats, pretty and raw. The format of the data depends on the type of the managed device. For NETCONF devices, the pretty mode indents all the XML data for enhanced readability and the raw mode does not. Sometimes when the XML is broken, raw mode is required to see all the data received. Tracing in raw mode will also signal to the corresponding NED to log more verbose tracing information.
To enable tracing, do:
The trace setting only affects new NED connections, so to ensure that we get any tracing data, we can do:
The above command terminates all existing connections.
At this point, if you execute a transaction towards one or several devices and then view the trace data.
It is possible to clear all existing trace files through the command
Finally, it is worth mentioning the trace functionality does not come for free. It is fairly costly to have the trace turned on. Also, there exists no trace log wrapping functionality.
When managing large networks with NSO a good strategy is to consider the NSO copy of the network configuration to be the main primary copy. All device configuration changes must go through NSO and all other device re-configurations are considered rogue.
NSO does not contain any functionality which disallows rogue re-configurations of managed devices, however, it does contain a mechanism whereby it is a very cheap operation to discover if one or several devices have been configured out-of-band.
The underlying mechanism for cheap check-sync is to compare time stamps, transaction IDs, hash-sums, etc., depending on what the device supports. This is in order not to have to read the full configuration to check if the NSO copy is in sync.
The transaction IDs are stored in CDB and can be viewed as:
Some of the devices do not have a transaction ID, this is the case where the NED has not implemented the cheap check-sync mechanism. Although it is called transaction-id, the underlying value in the device can be anything to detect a config change, like for example a time-stamp.
To check for consistency, we execute:
Alternatively for all (or a subset) managed devices:
The following YANG grouping is used for the return value from the check-sync command:
In the previous section, we described how we can easily check if a managed device is in sync. If the device is not in sync, we are interested to know what the difference is. The CLI sequence below shows how to modify ce0 out-of-band using the ncs-netsim tool. Finally, the sequence shows how to do an explicit configuration comparison.
The diff in the above output should be interpreted as: what needs to be done in NSO to become in sync with the device.
Previously in the example (Synchronize from Devices), NSO was brought in sync with the devices by fetching configuration from the devices. In this case, where the device has a rogue re-configuration, NSO has the correct configuration. In such cases, you want to reset the device configuration to what is stored inside NSO.
When you decide to reset the configuration with the copy kept in NSO use the option dry-run in conjunction with sync-to and inspect what will be sent to the device:
As this is the desired data to send to the device a sync-to can now safely be performed.
The device configuration should now be in sync with the copy in NSO and compare-config ought to yield an empty output:
There exist several ways to initialize new devices. The two common ways are to initialize a device from another existing device or to use device templates.
For example, another CE router has been added to our example network. You want to base the configuration of that host on the configuration of the managed device ce0 which has a valid configuration:
If the configuration is accurate you can create a new managed device based on that configuration as:
In the example above (Instantiate Device from Other) the commands first create the new managed device, ce9 and then populates the configuration of the new device based on the configuration of ce0.
This new configuration might not be entirely correct, you can modify any configuration before committing it.
The above concludes the instantiation of a new managed device. The new device configuration is committed and NSO returned OK without the device existing in the network (netsim). Try to force a sync to the device:
The device is southbound locked, this is a mode that is used where you can reconfigure a device, but any changes done to it are never sent to the managed device. This will be thoroughly described in the next section. Devices are by default created southbound locked. Default values are not shown if not explicitly requested:
Another alternative to instantiating a device from the actual working configuration of another device is to have a number of named device templates that manipulate the configuration.
The template tree looks like this:
The tree for device templates is generated from all device YANG models. All constraints are removed and the data type of all leafs is changed to string
A device template is created by setting the desired data in the configuration. The created device template is stored in NSO CDB.
The device template created in the example above (Create ce-initialize template) can now be used to initialize single devices or device groups, see Device Groups.
In the following CLI session, a new device ce10 is created:
Initialize the newly created device ce10 with the device template ce-initialize:
When initializing devices, NSO does not have any knowledge about the capabilities of the device, no connect has been done. This can be overridden by the option accept-empty-capabilities
Inspect the changes made by the template ce-initialize
This section shows how device templates can be used to create and change device configurations. See Introduction in Templates for other ways of using templates.
Device templates are part of the NSO configuration. Device templates are created and changed in the tree /devices/template/config the same way as any other configuration data and are affected by rollbacks and upgrades. Device templates can only manipulate configuration data in the /devices/device/config tree i.e., only device data.
The $NCS_DIR/examples.ncs/service-provider/mpls-vpn example comes with a pre-populated template for SNMP settings.
The variable $DEVICE is used internally by NSO and can not be used in a template.
Templates can be created like any configuration data and use the CLI tab completion to navigate. Variables can be used instead of hard-coded values. In the template above the community string is a variable. The template can cover several device types/NEDs, by making use of the namespace information. This will make sure that only devices modeled with this particular namespace will be affected by this part of the template. Hence, it is possible for one template to handle a multitude of devices from various manufacturers.
A template can be applied to a device, a device group, and a range of devices. It can be used as shown in By Template to create the day-zero config for a newly created device.
Applying the snmp1 template, providing a value for the COMMUNITY template variable:
The result of applying the template:
The default operation for templates is to merge the configuration. Tags can be added to templates to have the template merge, replace, delete, create or nocreate configuration. A tag is inherited to its sub-nodes until a new tag is introduced.
merge: Merge with a node if it exists, otherwise create the node. This is the default operation if no operation is explicitly set.
replace: Replace a node if it exists, otherwise create the node.
create: Creates a node. The node can not already exist.
nocreate: Merge with a node if it exists. If it does not exist, it will not be created.
Example of how to set a tag:
Displaying Tags information::
By adding the CLI pipe flag debug template when applying a template, the CLI will output detailed information on what is happening when the template is being applied:
oper-state and admin-stateNSO differentiates between oper-state and admin-state for a managed device. oper-state is the actual state of the device. We have chosen to implement a very simple oper-state model. A managed device oper-state is either enabled or disabled. oper-state can be mapped to an alarm for the device. If the device is disabled, we may have additional error information. For example, the ce9 device created from another device and ce10 created with a device template in the previous section is disabled, and no connection has been established with the device, so its state is completely unknown:
Or, a slightly more interesting CLI usage:
If you manually stop a managed device, for example ce0, NSO doesn't immediately indicate that. NSO may have an active SSH connection to the device, but the device may voluntarily choose to close its end of that (idle) SSH connection. Thus the fact that a socket from the device to NSO is closed by the managed device doesn't indicate anything. The only certain method NSO has to decide a managed device is non-operational - from the point of view of NSO - is NSO cannot SSH connect to it. If you manually stop managed device ce0, you still have:
NSO cannot draw any conclusions from the fact that a managed device closed its end of the SSH connection. It may have done so because it decided to time out an idle SSH connection. Whereas if NSO tried to initiate any operations towards the dead device, the device would be marked as oper-state disabled:
Now, NSO has failed to connect to it, NSO knows that ce0 is dead:
This concludes the oper-state discussion. The next state to be illustrated is the admin-state. The admin-state is what the operator configures, this is the desired state of the managed device.
In tailf-ncs.yang we have the following configuration definition for admin-state:
In the example above (tailf-ncs-devices.yang - admin-state), you can see the four different admin states for a managed device as defined in the YANG model.
locked - This means that all changes to the device are forbidden. Any transaction which attempts to manipulate the configuration of the device will fail. It is still possible to read the configuration of the device.
unlocked -This is the state a device is set into when the device is operational. All changes to the device are attempted to be sent southbound.
southbound-locked - This is the default value. It means that it is possible to manipulate the configuration of the device but changes done to the device configuration are never pushed to the device. This mode is useful during e.g. pre-provisioning, or when we instantiate new devices.
config-locked - This means that any transaction which attempts to manipulate the configuration of the device will fail. It is still possible to read the configuration of the device and send live-status commands or RPCs.
NSO manages a set of devices that are given to NSO through any means like CLI, inventory system integration through XML APIs, or configuration files at startup. The list of devices to manage in an overall integrated network management solution is shared between different tools and therefore it is important to keep an authoritative database of this and share it between different tools including NSO. The purpose of this part is to identify the source of the population of managed devices. The source attribute should indicate the source of the managed device like "inventory", "manual", or "EMS".
These attributes should be automatically set by the integration towards the inventory source, rather than manipulated manually.
added-by-user: Identify the user who loaded the managed device.
context: In what context was the device loaded.
when: When the device was added to NSO.
from-ip: From which IP the load activity was run.
source: Identify the source of the managed device such as the inventory system name or the name of the source file.
The NETCONF protocol mandates that the first thing both the server and the client have to do is to send its list of NETCONF capabilities in the <hello> message. A capability indicates what the peer can do. For example the validate:1.0 indicates that the server can validate a proposed configuration change, whereas the capability http://acme.com/if indicates the device implements the http://acme.com proprietary capability.
The NEDs report the capabilities of the devices at connection time. The NEDs also load the YANG modules for NSO. For a NETCONF/YANG device, all this is straightforward, for non-NETCONF devices the NEDs do the translation.
The capabilities announced by a device also contain the YANG version 1 modules supported. In addition to this, YANG version 1.1 modules are advertised in the YANG library module on the device. NSO checks both the capabilities and the YANG library to find out which YANG modules a device supports.
The capabilities and modules detected by NSO are available in two different lists, /devices/device/capability and devices/device/module. The capability list contains all capabilities announced and all YANG modules in the YANG library. The module list contains all YANG modules announced that are also supported by the NED in NSO.
NSO can be used to handle all or some of the YANG configuration modules for a device. A device may announce several modules through its capability list which NSO ignores. NSO will only handle the YANG modules for a device which are loaded (and compiled through ncsc --ncs-compile-bundle) or ncsc --ncs-compile-module) all other modules for the device are ignored. If you require a situation where NSO is entirely responsible for a device so that complete device backup/configurations are stored in NSO you must ensure NSO indeed has support for all modules for the device. It is not possible to automate this process since a capability URI doesn't necessarily indicate actual configuration.
When a device is added to NSO its NED ID must be set. For a NETCONF device, it is possible to configure the generic NETCONF NED id netconf (defined in the YANG module tailf-ncs-ned). If this NED ID is configured, we can then ask NSO to connect to the device and then check the capability list to see which modules this device implements.
We can also check which modules the loaded NEDs support. Then we can pick the most suitable NED and configure the device with this NED ID.
NSO works best if the managed devices support the NETCONF candidate configuration datastore. However, NSO reads the capabilities of each managed device and executes different sequences of NETCONF commands towards different types of devices.
For implementations of the NETCONF protocol that do not support the candidate datastore, and in particular, devices that do not support NETCONF commit with a timeout, NSO tries to do the best of the situation.
NSO divides devices into the following groups.
start_trans_running: This mode is used for devices that support the Tail-f proprietary transaction extension defined by http://tail-f.com/ns/netconf/transactions/1.0. Read more on this in the Tail-f ConfD user guide. In principle it's a means to - over the NETCONF interface - control transaction processing towards the running data store. This may be more efficient than going through the candidate data store. The downside is that it is Tail-f proprietary non-standardized technology.
lock_candidate: This mode is used for devices that support the candidate data store but disallow direct writes to the running data store.
lock_reset_candidate: This mode is used for devices that support the candidate data and also allow direct writes to the running data store. This is the default mode for Tail-f ConfD NETCONF server. Since the running data store is configurable, we must, before each configuration attempt, copy all of the running to the candidate. (ConfD has optimized this particular usage pattern, so this is a very cheap operation for ConfD)
startup: This mode is used for devices that have writable running, no candidate but do support the startup data store. This is the typical mode for Cisco-like devices.
running-only: This mode is used for devices that only support writable running.
NED: The transaction is controlled by a Network Element Driver. The exact transaction mode depends on the type of the NED.
Which category NSO chooses for a managed device depends on which NETCONF capabilities the device sends to NSO in its NETCONF hello message. You can see in the CLI what NSO has decided for a device as in:
NSO talking to ConfD device running in its standard configuration, thus lock-reset-candidate.
Another important discriminator between managed devices is whether they support the confirmed commit with a timeout capability, i.e., the confirmed-commit:1.0 standard NETCONF capability. If a device supports this capability, NSO utilizes it. This is the case with for example Juniper routers.
If a managed device does not support this capability, NSO attempts to do the best it can.
This is how NSO handles common failure scenarios:
The operator aborts the transaction, or the NSO loses the SSH connection to another managed device which is also participating in the same network transaction. If the device does support the confirmed-commit capability, NSO aborts the outstanding yet-uncommitted transaction simply by closing the SSH connection. When the device does not support the confirmed-commit capability, NSO has the reverse diff and simply sends the precise undo information to the device instead.
The device rejects the transaction in the first place, i.e. the NSO attempts to modify its running data store. This is an easy case since NSO then simply aborts the transaction as a whole in the initial commit confirmed [time] attempt.
NSO loses SSH connectivity to the device during the timeout period. This is a real error case and the configuration is now in an unknown state. NSO will abort the entire transaction, but the configuration of the failing managed device is now probably in error. The correct procedure once network connectivity has been restored to the device is to sync it in the direction from NSO to the device. The NSO copy of the device configuration will be what was configured before the failed transaction.
Thus, even if not all participating devices have first-class NETCONF server implementations, NSO will attempt to fake the confirmed-commit capability.
When the managed device defines top-level NETCONF RPCs or alternatively, define tailf:action points inside the YANG model, these RPCs and actions are also imported into the data model that resides in NSO.
For example, the Juniper NED comes with a set of JunOS RPCs defined in: $NCS_DIR/packages/neds/juniper-junos/src/yang/junos-rpc.yang
Thus, since all RPCs and actions from the devices are accessible through the NSO data model, these actions are also accessible through all NSO northbound APIs, REST, JAVA MAAPI, etc. Hence it is possible to - from user scripts/code - invoke actions and RPCs on all managed devices. The RPCs are augmented below an RPC container:
In the simulated environment of the mpls-vpn example, these RPCs might not have been implemented.
The NSO device manager has a concept of groups of devices. A group is nothing more than a named group of devices. What makes this interesting is that we can invoke several different actions in the group, thus implicitly invoking the action on all members in the group. This is especially interesting for the apply-template action.
The definition of device groups resides at the same layer in the NSO data model as the device list, thus we have:
The MPLS VPN example comes with a couple of pre-defined device-groups:
Device groups are created like below:
Device groups can reference other device groups. There is an operational attribute that flattens all members in the group. The CLI sequence below adds the PE group to my-group. Then it shows the configuration of that group followed by the status of this group. The status for the group contains a members attribute that lists all device members.
Once you have a group, you can sync and check-sync the entire group.
However, what makes device groups really interesting is the ability to apply a template to a group. You can use the pre-populated templates to apply SNMP settings to device groups.
Policies allow you to specify network-wide constraints that always must be true. If someone tries to apply a configuration change over any northbound interface that would be evaluated to false, the configuration change is rejected by NSO. Policies can be of type warning means that it is possible to override them, or error which cannot be overridden.
Assume you would like to enforce all CE routers to have a Gigabit interface 0/1.
As seen in the example above (Policies) , a policy rule has (an optional) for each statement and a mandatory expression and error message. The foreach statement evaluates to a node set, and the expression is then evaluated on each node. So in this example, the expression would be evaluated for every device in NSO which begins with ce. The name variable in the warning message refers to a leaf available from the for-each node set.
Validation is always performed at commit but can also be requested interactively.
Note any configuration can be activated or deactivated. This means that to temporarily turn off a certain policy you can deactivate it. Note also that if the configuration was changed by any other means than NSO by local tools to the device like a CLI, a devices sync-from operation might fail if the device configuration violates the policy.
One of the strengths of NSO is the concept of network-wide transactions. When you commit data to NSO that spans multiple devices in the /ncs:devices/device tree, NSO will - within the NSO transaction - commit the data on all devices or none, keeping the network consistent with CDB. The NSO transaction doesn't return until all participants have acknowledged the proposed configuration change. The downside of this is that the slowest device in each transaction limits the overall transactional throughput in NSO. Such things as out-of-sync checks, network latency, calculation of changes sent southbound, or device deficiencies all affect the throughput.
Typically when automation software north of NSO generates network change requests it may very well be the case more requests arrive than what can be handled. In NSO deployment scenarios where you wish to have higher transactional throughput than what is possible using network-wide transactions, you can use the commit queue instead. The goal of the commit queue is to increase the transactional throughput of NSO while keeping an eventual consistency view of the database. With the commit queue, NSO will compute the configuration change for each participating device, put it in an outbound queue item, and immediately return. The queue is then independently run.
Another use case where you can use the commit queue is when you wish to push a configuration change to a set of devices and don't care about whether all devices accept the change or not. You do not want the default behavior for transactions which is to reject the transaction as a whole if one or more participating devices fail to process its part of the transaction.
An example of the above could be if you wish to set a new NTP server on all managed devices in our entire network, if one or more devices currently are non-operational, you still want to push out the change. You also want the change automatically pushed to the non-operational devices once they go live again.
The big upside of this scheme is that the transactional throughput through NSO is considerably higher. Also, transient devices are handled better. The downsides are:
If a device rejects the proposed change, NSO and the device are now out of sync until any error recovery is performed. Whenever this happens, an NSO alarm (called commit-through-queue-failed) is generated.
While a transaction remains in the queue, i.e., it has been accepted for delivery by NSO but is not yet delivered, the view of the network in NSO is not (yet) correct. Eventually, though, the queued item will be delivered, thus achieving eventual consistency.
To facilitate the two use cases of the commit queue the outbound queue item can be either in an atomic or non-atomic mode.
In atomic mode the outbound queue item will push all configuration changes concurrently once there are no intersecting devices ahead in the queue. If any device rejects the proposed change, all device configuration changes in the queue item will be rejected as a whole, leaving the network in a consistent state. The atomic mode also allows for automatic error recovery to be performed by NSO.
In the non-atomic mode, the outbound queue item will push configuration changes for a device whenever all occurrences of it are completed or it doesn't exist ahead in the queue. The drawback to this mode is that there is no automatic error recovery that can be performed by NSO.
In the following sequences, the simulated device ce0 is stopped to illustrate the commit queue. This can be achieved by the following sequence including returning to the NSO CLI config mode:
By default, the commit queue is turned off. You can configure NSO to run a transaction, device, or device group through the commit queue in a number of different ways, either by providing a flag to the commit command as:
Or, by configuring NSO to always run all transactions through the commit queue as in:
Or, by configuring a number of devices to run through the commit queue as default:
When enabling the commit queue as default on a per device/device group basis, an NSO transaction will compute the configuration change for each participating device, put the devices enabled for the commit queue in the outbound queue, and then proceed with the normal transaction behavior for those devices not commit queue enabled. The transaction will still be successfully committed even if some of the devices added to the outbound queue will fail. If the transaction fails in the validation phase the entire transaction will be aborted, including the configuration change for those devices added to the commit queue. If the transaction fails after the validation phase, the configuration change for the devices in the commit queue will still be delivered.
Do some changes and commit through the commit queue:
In the example above (Commit through Commit Queue), the commit affected three devices, ce0, ce1 and ce2. If you immediately would have launched yet another transaction, as in the second one (see example below), manipulating an interface of ce2, that transaction would have been queued instead of immediately launched. The idea here is to queue entire transactions that touch any device that has anything queued ahead in the queue.
Each transaction committed through the queues becomes a queue item. A queue item has an ID number. A bigger number means that it's scheduled later. Each queue item waits for something to happen. A queue item is in either of three states.
waiting: The queue item is waiting for other queue items to finish. This is because the waiting queue item has participating devices that are part of other queue items, ahead in the queue. It is waiting for a set of devices, to not occur ahead of itself in the queue.
executing: The queue item is currently being processed. Multiple queue items can run currently as long as they don't share any managed devices. Transient errors might be present. These errors occur when NSO fails to communicate with some of the devices. The errors are shown in the leaf-list transient-errors. Retries will take place at intervals specified in /ncs:devices/global-settings/commit-queue/retry-timeout. Examples of transient errors are connection failures and that the changes are rejected due to the device being locked. Transient errors are potentially bad since the queue might grow if new items are added, waiting for the same device.
locked: This queue item is locked and will not be processed until it has been unlocked, see the action /ncs:devices/commit-queue/queue-item/unlock. A locked queue item will block all subsequent queue items that are using any device in the locked queue item.
You can view the queue in the CLI. There are three different view modes, summary, normal, and detailed. Depending on the output, both the summary and the normal look good:
The age field indicated how many seconds a queue item has been in the queue.
You can also view the queue items in detailed mode:
The queue items are stored persistently, thus if NSO is stopped and restarted, the queue remains the same. Similarly, if NSO runs in HA (High Availability) mode, the queue items are replicated, ensuring the queue is processed even in case of failover.
The commit queue is disabled when both HA is enabled, and its HA role is none, i.e., not primary or secondary. See Mode of Operation.
A number of useful actions are available to manipulate the queue:
devices commit-queue add-lock device [ ... ]. This adds a fictive queue item to the commit queue. Any queue item, affecting the same devices, which is entering the commit queue will have to wait for this lock item to be unlocked or deleted. If no devices are specified, all devices in NSO are locked.
devices commit-queue clear. This action clears the entire queue. All devices present in the commit queue will, after this action, have executed be out of sync. The clear action is a rather blunt tool and is not recommended to be used in any normal use case.
devices commit-queue prune device [ ... ] . This action prunes all specified devices from all queue items in the commit queue. The affected devices will, after this action has been executed, be out of sync. Devices that are currently being committed to will not be pruned unless the force option is used. Atomic queue items will not be affected, unless all devices in it are pruned. The force option will brutally kill an ongoing commit. This could leave the device in a bad state. It is not recommended in any normal use case.
devices commit-queue set-atomic-behaviour atomic [ true,false ]. This action sets the atomic behavior of all queue items. If these are set to false, the devices contained in these queue items can start executing if the same devices in other non-atomic queue items ahead of it in the queue are completed. If set to true, the atomic integrity of these queue items is preserved.
devices commit-queue wait-until-empty. This action waits until the commit queue is empty. The default is to wait infinity. A timeout can be specified to wait for a number of seconds. The result is empty if the queue is empty or timeout if there are still items in the queue to be processed.
devices commit-queue queue-item [ id ] lock. This action puts a lock on an existing queue item. A locked queue item will not start executing until it has been unlocked.
devices commit-queue queue-item [ id ] unlock. This action unlocks a locked queue item. Unlocking a queue item that is not locked is silently ignored.
devices commit-queue queue-item [ id ] delete. This action deletes a queue item from the queue. If other queue items are waiting for this (deleted) item, they will all automatically start to run. The devices of the deleted queue item will, after the action has been executed, be out of sync if they haven't started executing. Any error option set for the queue item will also be disregarded. The force option will brutally kill an ongoing commit. This could leave the device in a bad state. It is not recommended in any normal use case.
devices commit-queue queue-item [ id ] prune device [ ... ]. This action prunes the specified devices from the queue item. Devices that are currently being committed to will not be pruned unless the force option is used. Atomic queue items will not be affected, unless all devices in it are pruned. The force option will brutally kill an ongoing commit. This could leave the device in a bad state. It is not recommended in any normal use case.
devices commit-queue queue-item [ id ] set-atomic-behaviour atomic [ true,false ]. This action sets the atomic behavior of this queue item. If this is set to false, the devices contained in this queue item can start executing if the same devices in other non-atomic queue items ahead of it in the queue are completed. If set to true, the atomic integrity of the queue item is preserved.
devices commit-queue queue-item [ id ] wait-until-completed. This action waits until the queue item is completed. The default is to wait infinity. A timeout can be specified to wait for a number of seconds. The result is completed if the queue item is completed or timeout if the timer expired before the queue item was completed.
devices commit-queue queue-item [ id ] retry. This action retries devices with transient errors instead of waiting for the automatic retry attempt. The device option will let you specify the devices to retry.
A typical use scenario is where one or more devices are not operational. In the example above (Viewing Queue Items), there are two queue items, waiting for the device ce0 to come alive. ce0 is listed as a transient error, and this is blocking the entire queue. Whenever a queue item is blocked because another item ahead of it cannot connect to a specific managed device, an alarm is generated:
Block other affecting device ce0 from entering the commit queue:
Now queue item 9577950918 is blocking other items using ce0 from entering the queue.
Prune the usage of the device ce0 from all queue items in the commit queue:
The lock will be in the queue until it has been deleted or unlocked. Queue items affecting other devices are still allowed to enter the queue.
Fix the problem with the device ce0, remove the lock item and sync from the device:
In an LSA cluster, each remote NSO has its own commit queue. When committing through the commit queue on the upper node NSO will automatically create queue items on the lower nodes where the devices in the transaction reside. The progress of the lower node queue items is monitored through a queue item on the upper node. The remote NSO is treated as a device in the queue item and the remote queue items and devices are opaque to the user of the upper node.
Generally, it is not recommended to interfere with the queue items of the lower nodes that have been created by an upper NSO. This can cause the upper queue item to not synchronize with the lower ones correctly.
To be able to track the commit queue on the lower cluster nodes, NSO uses the built-in stream ncs-events that generates northbound notifications for internal events. This stream is required if running the commit queue in a clustered scenario. It is enabled in ncs.conf:
In addition, the commit queue needs to be enabled in the cluster configuration.
For more detailed information on how to set up clustering, see LSA Overview.
The goal of the commit queue is to increase the transactional throughput of NSO while keeping an eventual consistency view of the database. This means no matter if changes committed through the commit queue originate as pure device changes or as the effect of service manipulations the effects on the network should eventually be the same as if performed without a commit queue no matter if they succeed or not. This should apply to a single NSO node as well as NSO nodes in an LSA cluster.
Depending on the selected error-option NSO will store the reverse of the original transaction to be able to undo the transaction changes and get back to the previous state. This data is stored in the /ncs:devices/commit-queue/completed tree from where it can be viewed and invoked with the rollback action. When invoked the data will be removed.
The error option can be configured under /ncs:devices/global-settings/commit-queue/error-option. Possible values are: continue-on-error, rollback-on-error, and stop-on-error. The continue-on-error value means that the commit queue will continue on errors. No rollback data will be created. The rollback-on-error value means that the commit queue item will roll back on errors. The commit queue will place a lock on the failed queue item, thus blocking other queue items with overlapping devices from being executed. The rollback action will then automatically be invoked when the queue item has finished its execution. The lock will be removed as part of the rollback. The stop-on-error means that the commit queue will place a lock on the failed queue item, thus blocking other queue items with overlapping devices from being executed. The lock must then either manually be released when the error is fixed or the rollback action under /devices/commit-queue/completed be invoked. The rollback action is as:
The error option can also be given as a commit parameter.
To guarantee service integrity NSO checks for overlapping service or device modifications against the items in the commit queue and returns an error if such exists. If a service instance does a shared set on the same data as a service instance in the queue actually changed, the reference count will be increased but no actual change is pushed to the device(s). This will give a false positive that the change is actually deployed in the network. The rollback-on-error and stop-on-error error options will automatically create a queue lock on the involved services and devices to prevent such a case.
In a clustered environment, different parts of the resulting configuration change set will end up on different lower nodes. This means on some nodes the queue item could succeed and on others, it could not.
The error option in a cluster environment will originate on the upper node. The reverse of the original transaction will be committed on this node and propagated through the cluster down to the lower nodes. The net effect of this is the state of the network will be the same as before the original change.
As the error option in a cluster environment will originate on the upper node, any configuration on the lower nodes will be meaningless.
When NSO is recovering from a failed commit, the rollback data of the failed queue items in the cluster is applied and committed through the commit queue. In the rollback, the no-networking flag will be set on the commits towards the failed lower nodes or devices to get CDB consistent with the network. Towards the successful nodes or devices, the commit is done as before. This is what the rollback action in /ncs:devices/commit-queue/completed/queue-item does.
TR1; service s1 creates ce0:a and ce1:b. The nodes a and b are created in CDB. In the changes of the queue item, CQ1, a and b are created.
TR2; service s2 creates ce1:c and ce2:d. The nodes c and d are created in CDB. In the changes of the queue item, CQ2, c, and d are created.
The queue item from TR1, CQ1, starts to execute. The node a cannot be created on the device. The node b was created on the device but that change is reverted as a failed to be created.
The reverse of TR1, the rollback of CQ1, TR3, is committed.
TR3; service s1 is applied with the old parameters. Thus the effect of TR1 is reverted. Nothing needs to be pushed towards the network, so no queue item is created.
TR2; as the queue item from TR2, CQ2, is not the same service instance and has no overlapping data on the ce1 device, this queue item executes as normal.
NSO1:TR1; service s1 dispatches the service to NSO2 and NSO3 through the queue item NSO1:CQ1. In the changes of NSO1:CQ1, NSO2:s1 and NSO3:s1 are created.
NSO1:TR2; service s2 dispatches the service to NSO2 through the queue item NSO1:CQ2. In the changes of NSO1:CQ2, NSO2:s2 is created.
The queue item from NSO2:TR1, NSO2:CQ1, starts to execute. The node a cannot be created on the device. The node b was created on the device, but that change is reverted as a failed to be created.
The queue item from NSO3:TR1, NSO3:CQ1, starts to execute. The changes in the queue item are committed successfully to the network.
The reverse of TR1, rollback of CQ1, TR3, is committed on all nodes part of TR1 that failed.
NSO2:TR3; service s1 is applied with the old parameters. Thus the effect of NSO2:TR1 is reverted. Nothing needs to be pushed towards the network, so no queue item is created.
NSO1:TR3; service s1 is applied with the old parameters. Thus the effect of NSO1:TR1 is reverted. A queue item is created to push the transaction changes to the lower nodes that didn't fail.
NSO3:TR3; service s1 is applied with the old parameters. Thus the effect of NSO3:TR1 is reverted. Since the changes in the queue item NSO3:CQ1 was successfully committed to the network a new queue item NSO3:CQ3 is created to revert those changes.
If for some reason the rollback transaction fails there are, depending on the failure, different techniques to reconcile the services involved:
Make sure that the commit queue is blocked to not interfere with the error recovery procedure. Do a sync-from on the non-completed device(s) and then re-deploy the failed service(s) with the reconcile option to reconcile original data, i.e., take control of that data. This option acknowledges other services controlling the same data. The reference count will indicate how many services control the data. Release any queue lock that was created.
Make sure that the commit queue is blocked to not interfere with the error recovery procedure. Use un-deploy with the no-networking option on the service and then do sync-from on the non-completed device(s). Make sure the error is fixed and then re-deploy the failed service(s) with the reconcile option. Release any queue lock that was created.
As the goal of the commit queue is to increase the transactional throughput of NSO it means that we need to calculate the configuration change towards the device(s) outside of the transaction lock. To calculate a configuration change NSO needs a pre-commit running and a running view of the database. The key enabler to support this in the commit queue is to allow different views of the database to live beyond the commit. In NSO, this is implemented by keeping a snapshot database of the configuration tree for devices and storing configuration changes towards this snapshot database on a per-device basis. The snapshot database is updated when a device in the queue has been processed. This snapshot database is stored on disk for persistence (the S.cdb file in the ncs-cdb directory).
The snapshot database could be populated in two ways. This is controlled by the /ncs-config/cdb/snapshot/pre-populate setting in the ncs.conf file. The parameter controls whether the snapshot database should be pre-populated during the upgrade or not. Switching this on or off implies different trade-offs.
If set to false, NSO is optimized for the default transaction behavior. The snapshot database is populated in a lazy manner (when a device is committed through the commit queue for the first time after an upgrade). The drawback is that this commit will suffer performance-wise, which is especially true for devices with large configurations. Subsequent commits on the same device will not have the same penalty.
If true, NSO is optimized for systems using the commit queue extensively. This will lead to better performance when committing using the commit queue with no additional penalty for first-time commits. The drawbacks are that the time to do upgrades will increase and also an almost twofold increase in NSO memory consumption.
The NSO device manager has built-in support for the NETCONF Call Home client protocol operations over SSH as defined in RFC 8071.
With NETCONF SSH Call Home, the NETCONF client listens for TCP connection requests from NETCONF servers. The SSH client protocol is started when the connection is accepted. The SSH client validates the server's presented host key with credentials stored in NSO. If no matching host key is found the TCP connection is closed immediately. Otherwise, the SSH connection is established, and NSO is enabled to communicate with the device. The SSH connection is kept open until the device itself terminates the connection, an NSO user disconnects the device, or the idle connection timeout is triggered (configurable in the ncs.conf file).
NSO will generate an asynchronous notification event whenever there is a connection request. An application can subscribe to these events and, for example, add an unknown device to the device tree with the information provided, or invoke actions on the device if it is known.
If an SSH connection is established, any outstanding configuration in the commit queue for the device will be pushed. Any notification stream for the device will also be reconnected.
NETCONF Call Home is enabled and configured under /ncs-config/netconf-call-home in the ncs.conf file. By default NETCONF Call Home is disabled.
A device can be connected through the NETCONF Call Home client only if /devices/device/state/admin-state is set to call-home. This state prevents any southbound communication to the device unless the connection has already been established through the NETCONF Call Home client protocol.
The NSO device manager has built-in support for device notifications. Notifications are a means for the managed devices to send structured data asynchronously to the manager. NSO has native support for NETCONF event notifications (see RFC 5277) but could also receive notifications from other protocols implemented by the Network Element Drivers.
Notifications can be utilized in various use-case scenarios. It can be used to populate alarms in the Alarm manager, collect certain types of errors over time, build a network-wide audit log, react to configuration changes, etc.
The basic mode of operation is the manager subscribes to one or more named notification channels which are announced by the managed device. The manager keeps an open SSH channel towards the managed device, and then, the managed device may asynchronously send structured XML data on the SSH channel.
The notification support in NSO is usable as is without any further programming. However, NSO cannot understand any semantics contained inside the received XML messages, thus for example a notification with a content of "Clear Alarm 456" cannot be processed by NSO without any additional programming.
When you add programs to interpret and act upon notifications, make sure that resulting operations are idempotent. This means that they should be able to be called any number of times while guaranteeing that side effects only occur once. The reason for this is that, for example, replaying notifications can sometimes mean that your program will handle the same notifications multiple times.
In the tailf-ncs.yang data model, you find a YANG data model that can be used to:
Setup subscriptions. A subscription is configuration data from the point of view of NSO, thus if NSO is restarted, all configured subscriptions are automatically resumed.
Inspect which named streams a managed device publishes.
View all received notifications.
Notifications must be defined at the top level of a YANG module. NSO does currently not support defining notifications inside lists or containers as specified in section 7.16 in RFC 7950.
In this section, we will use the examples.ncs/web-server-farm/basic example.
Let's dive into an example session with the NSO CLI. In the NSO example collection, the webserver publishes two NETCONF notification structures, indicating what they intend to send to any interested listeners. They all have the YANG module:
Follow the instructions in the README file if you want to run the example: build the example, start netsim, and start NCS.
The above shows how we can inspect - as status data - which named streams the managed device publishes. Each stream also has some associated data. The data model for that looks like this:
Let's set up a subscription for the stream called interface. The subscriptions are NSO configuration data, thus to create a subscription we need to enter configuration mode:
The above example created subscriptions for the interface stream on all web servers, i.e. managed devices, www0, www1, and www2. Each subscription must have an associated stream to it, this is however not the key for an NSO notification, the key is a free-form text string. This is because we can have multiple subscriptions to the same stream. More on this later when we describe the filter that can be associated with a subscription. Once the notifications start to arrive, they are read by NSO and stored in stable storage as CDB operational data. they are stored under each managed device - and we can view them as:
Each received notification has some associated metadata, such as the time the event was received by NSO, which subscription and which stream is associated with the notification, and also which user created the subscription.
It is fairly instructive to inspect the XML that goes on the wire when we create a subscription and then also receive the first notification. We can do:
Thus, once the subscription has been configured, NSO continuously receives, and stores in CDB oper persistent storage, the notifications sent from the managed device. The notifications are stored in a circular buffer, to set the size of the buffer, we can do:
The default value is 200. Once the size of the circular buffer is exceeded, the older notification is removed.
A running subscription can be in either of three states. The YANG model has:
If a subscription is in the failed state, an optional failure-reason field indicates the reason for the failure. If a subscription fails due to, not being able to connect to the managed device or if the managed device closed its end of the SSH socket, NSO will attempt to automatically reconnect. The re-connect attempt interval is configurable.
SNMP Notifications (v1, v2c, v3) can be received by NSO and acted upon. The SNMP receiver is a stand-alone process and by default, all notifications are ignored. IP addresses must be opted in and a handler must be defined to take actions on certain notifications. This can be used to for example listen to configuration change notifications and trigger a log action or a resync for example
These actions are programmed in Java, see the SNMP Notification Receiver for how to do this.
NSO can configure inactive parameters on the devices that support inactive configuration. Currently, these devices include Juniper devices and devices that announce http://tail-f.com/ns/netconf/inactive/1.0 capability. NSO itself implements http://tail-f.com/ns/netconf/inactive/1.0 capability which is formally defined in tailf-netconf-inactive YANG module.
To recap, a node that is marked as inactive exists in the data store but is not used by the server. The nodes announced as inactive by the device will also be inactive in the device's configuration in NSO, and activating/deactivating a node in NSO will push the corresponding change to the device. This also means that for NSO to be able to manage inactive configuration, both /ncs-config/enable-inactive and /ncs-config/netconf-north-bound/capabilities/inactive need to be enabled in ncs.conf.
If the inactive feature is disabled in ncs.conf, NSO will still be able to manage devices that have inactive configuration in their datastore, but the inactive attribute will be ignored, so the data will appear as active in NSO and it would not be possible for NSO to activate/deactivate such nodes in the device.
