This tutorial uses ExampleService to illustrate how to write and on-board a service in CORD. ExampleService is a multi-tenant service that instantiates a VM instance on behalf of each tenant, and runs an Apache web server in that VM. This web server is then configured to serve a tenant-specified message (a string), where the tenant is able to set this message using CORD's control interface. From a service modeling perspective, ExampleService extends the base Service model with two fields:
service_message
: A string that contains a message to display for the service as a whole (i.e., to all tenants of the service).tenant_message
: A string that is displayed for a specific Tenant.These two fields are a simple illustration of a common pattern. A service model typically includes fields used to configure the service as a whole (service_message
in this example) and fields used to control individual instances of the the service (tenant_message
in this example). It would be common for the operator to set configuration-related fields when the service first starts up, and then set/adjust control-related fields on behalf of individual tenants as the service runs.
Tenant and ServiceInstance are two closely related terms. "Tenant" refers to the user or the consumer of a service. Often we partition a service into logical partitions, each for use by a tenant, thus making it a multi-tenant service. Each one of these tenant-specific partitions is referred to as a ServiceInstance.
The result of preparing ExampleService for on-boarding is the following set of files, all located in the xos
directory of the exampleservice
repository. When checked out, these files live in the CORD_ROOT/orchestration/xos-services/exampleservice
directory on your local development machine.
Component | Source Code (https://github.com/opencord/exampleservice/) |
---|---|
Data Model | xos/synchronizer/models/exampleservice.xproto |
Syncronizer Program | xos/synchronizer/exampleservice-synchronizer.py xos/synchronizer/exampleservice_config.yaml xos/synchronizer/model-deps xos/synchronizer/Dockerfile.synchronizer |
Sync Steps | xos/synchronizer/steps/sync_exampletenant.py xos/synchronizer/steps/exampletenant_playbook.yaml |
Model Policies | xos/synchronizer/model_policies/model_policy_exampleserviceinstance.py |
On-Boarding Spec | xos/exampleservice-onboard.yaml |
In addition to implementing these service-specific files, the final step to on-boarding a service requires you to modify an existing (or write a new) service profile. This tutorial uses the existing R-CORD profile for illustrative purposes. These profile definitions currently live in the https://github.com/opencord/rcord repository. Additional related playbooks reside in the https://github.com/opencord/platform-install/ for historical reasons.
For this tutorial we recommend using a Virtual Pod (CiaB) as your development environment. By default CiaB brings up OpenStack, ONOS, and XOS running the R-CORD collection of services. This tutorial demonstrates how to add a new customer-facing service to R-CORD.
A Virtual Pod includes a build machine, a head node, switches, and a compute node all running as VMs on a single host. Before proceeding you should familiarize yourself with the CiaB environment and the POD Development Loop.
The synchronizer directory holds the model declarations and the synchronizer for the service. Usually this directory is xos/synchronizer
. This tutorial will first walk through creating the models, and then discuss creating the synchronizer itself.
Make a new root directory for your service, and within that directory, create an xos
subdirectory. The xos
subdirectory will hold all xos-related files for your service.
Within the xos
subdirectory, create a synchronizer
subdirectory. The synchronizer
subdirectory holds the subset of files that end up built into the synchronizer
container image.
Your models live in a file named exampleservice.xproto
in your service's xos/synchronizer/models
directory. This file encodes the models in the service in a format called xproto which is a combination of Google Protocol Buffers and some XOS-specific annotations to facilitate the generation of service components, such as the GRPC and REST APIs, security policies, and database models among other things. It consists of two parts:
The XPROTO Header, which contains options that are global to the rest of the file.
The Service model, which manages the service as a whole.
The ServiceInstance model, which manages tenant-specific (per-service-instance) state.
Some options are typically specified at the top of your xproto file:
option name = "exampleservice"; option app_label = "exampleservice";
name
specifies a name for your service. This is used as a default in several places, for example it will be used for app_label
if you don't specifically choose an app_label
. Normally it suffices to set this the name of your service, lower case, with no spaces.
app_label
configures the internal xos database application that is attached to these models. As with name
, it suffices to set this the name of your service, lower case, with no spaces.
A Service model extends (inherits from) the XOS base Service model. At its head is a set of option declarations such as verbose_name
, which specifies a human-readable name for the service model. Then follows a set of field definitions.
message ExampleService (Service){ option verbose_name = "Example Service"; required string service_message = 1 [help_text = "Service Message to Display", max_length = 254, null = False, db_index = False, blank = False]; }
Your ServiceInstance model will extend the core TenantWithContainer
class, which is a Tenant that creates a VM instance:
message ExampleServiceInstance (TenantWithContainer){ option verbose_name = "Example Service Instance"; required string tenant_message = 1 [help_text = "Tenant Message to Display", max_length = 254, null = False, db_index = False, blank = False]; }
The following field specifies the message that will be displayed on a per-Tenant basis:
tenant_message = models.CharField(max_length=254, help_text="Tenant Message to Display")
Think of this as a tenant-specific (per service instance) parameter.
The second step is to define a synchronizer for the service. Synchronizers are processes that run continuously, checking for changes to service's model(s). When a synchronizer detects a change, it applies that change to the underlying system. For ExampleService, the ServiceInstance
model is the model we will want to synchronize, and the underlying system is a compute instance. In this case, we’re using TenantWithContainer
to create this instance for us.
XOS Synchronizers are typically located in the xos/synchronizer
directory of your service.
Note: Earlier versions included a tool to track model dependencies, but today it is sufficient to create a file named
model-deps
with the contents:{}
.
The Synchronizer has three parts: The synchronizer python program, model policies which enact changes on the data model, and a playbook (typically Ansible) that configures the underlying system. The following describes how to construct these.
First, create a file named exampleservice-synchronizer.py
:
#!/usr/bin/env python # Runs the standard XOS synchronizer import importlib import os import sys from xosconfig import Config config_file = os.path.abspath(os.path.dirname(os.path.realpath(__file__)) + '/exampleservice_config.yaml') Config.init(config_file, 'synchronizer-config-schema.yaml') synchronizer_path = os.path.join(os.path.dirname( os.path.realpath(__file__)), "../../synchronizers/new_base") sys.path.append(synchronizer_path) mod = importlib.import_module("xos-synchronizer") mod.main()
The above is boilerplate. It loads and runs the default xos-synchronizer
module in it’s own Docker container. To configure this module, create a file named exampleservice_config.yaml
, which specifies various configuration and logging options:
name: exampleservice accessor: username: xosadmin@opencord.org password: "@/opt/xos/services/exampleservice/credentials/xosadmin@opencord.org" required_models: - ExampleService - ExampleServiceInstance - ServiceDependency - ServiceMonitoringAgentInfo dependency_graph: "/opt/xos/synchronizers/exampleservice/model-deps" steps_dir: "/opt/xos/synchronizers/exampleservice/steps" sys_dir: "/opt/xos/synchronizers/exampleservice/sys" model_policies_dir: "/opt/xos/synchronizers/exampleservice/model_policies" models_dir: "/opt/xos/synchronizers/exampleservice/models"
Make sure the name
in your synchronizer config file is that same as the app_label in your xproto
file. Otherwise the models won't be dynamically loaded correctly.
NOTE: Historically, synchronizers were named “observers”, so
s/observer/synchronizer/
when you come upon this term in the XOS code and documentation.
Second, create a directory within your synchronizer directory named steps
. In steps, create a file named sync_exampleserviceinstance.py
:
import os import sys from synchronizers.new_base.SyncInstanceUsingAnsible import SyncInstanceUsingAnsible from synchronizers.new_base.modelaccessor import * from xos.logger import Logger, logging parentdir = os.path.join(os.path.dirname(__file__), "..") sys.path.insert(0, parentdir) logger = Logger(level=logging.INFO)
Bring in some basic prerequisites. Also include the models created earlier, and SyncInstanceUsingAnsible
which will run the Ansible playbook in the Instance VM.
class SyncExampleServiceInstance(SyncInstanceUsingAnsible): provides = [ExampleServiceInstance] observes = ExampleServiceInstance requested_interval = 0 template_name = "exampleserviceinstance_playbook.yaml" service_key_name = "/opt/xos/synchronizers/exampleservice/exampleservice_private_key" def __init__(self, *args, **kwargs): super(SyncExampleServiceInstance, self).__init__(*args, **kwargs) def get_exampleservice(self, o): if not o.owner: return None exampleservice = ExampleService.objects.filter(id=o.owner.id) if not exampleservice: return None return exampleservice[0] # Gets the attributes that are used by the Ansible template but are not # part of the set of default attributes. def get_extra_attributes(self, o): fields = {} fields['tenant_message'] = o.tenant_message exampleservice = self.get_exampleservice(o) fields['service_message'] = exampleservice.service_message return fields def delete_record(self, port): # Nothing needs to be done to delete an exampleservice; it goes away # when the instance holding the exampleservice is deleted. pass
Third, create a run-from-api.sh
file for your synchronizer.
python exampleservice-synchronizer.py
Finally, create a Dockerfile for your synchronizer, name it Dockerfile.synchronizer
and place it in the synchronizer
directory with the other synchronizer files:
FROM xosproject/xos-synchronizer-base:candidate COPY . /opt/xos/synchronizers/exampleservice ENTRYPOINT [] WORKDIR "/opt/xos/synchronizers/exampleservice" # Label image ARG org_label_schema_schema_version=1.0 ARG org_label_schema_name=exampleservice-synchronizer ARG org_label_schema_version=unknown ARG org_label_schema_vcs_url=unknown ARG org_label_schema_vcs_ref=unknown ARG org_label_schema_build_date=unknown ARG org_opencord_vcs_commit_date=unknown ARG org_opencord_component_chameleon_version=unknown ARG org_opencord_component_chameleon_vcs_url=unknown ARG org_opencord_component_chameleon_vcs_ref=unknown ARG org_opencord_component_xos_version=unknown ARG org_opencord_component_xos_vcs_url=unknown ARG org_opencord_component_xos_vcs_ref=unknown LABEL org.label-schema.schema-version=$org_label_schema_schema_version \ org.label-schema.name=$org_label_schema_name \ org.label-schema.version=$org_label_schema_version \ org.label-schema.vcs-url=$org_label_schema_vcs_url \ org.label-schema.vcs-ref=$org_label_schema_vcs_ref \ org.label-schema.build-date=$org_label_schema_build_date \ org.opencord.vcs-commit-date=$org_opencord_vcs_commit_date \ org.opencord.component.chameleon.version=$org_opencord_component_chameleon_version \ org.opencord.component.chameleon.vcs-url=$org_opencord_component_chameleon_vcs_url \ org.opencord.component.chameleon.vcs-ref=$org_opencord_component_chameleon_vcs_ref \ org.opencord.component.xos.version=$org_opencord_component_xos_version \ org.opencord.component.xos.vcs-url=$org_opencord_component_xos_vcs_url \ org.opencord.component.xos.vcs-ref=$org_opencord_component_xos_vcs_ref CMD bash -c "cd /opt/xos/synchronizers/exampleservice; ./run-from-api.sh"
Model policies are used to implement change within the data model. When an ExampleServiceInstance
object is saved, we want an Instance
to be automatically created that will hold the ExampleServiceInstance's web server. Fortunately, there's a base class that implements this functionality for us, so minimal coding needs to be done at this time. Create the model_policies
subdirectory and within that subdirectory create the file model_policy_exampleserviceinstance.py
:
from synchronizers.new_base.modelaccessor import * from synchronizers.new_base.model_policies.model_policy_tenantwithcontainer import TenantWithContainerPolicy class ExampleServiceInstancePolicy(TenantWithContainerPolicy): model_name = "ExampleServiceInstance"
In the same steps
directory where you created sync_exampleserviceinstance.py
, create an Ansible playbook named exampleserviceinstance_playbook.yml
which is the “master playbook” for this set of plays:
# exampletenant_playbook - hosts: "{{ instance_name }}" connection: ssh user: ubuntu sudo: yes gather_facts: no vars: - tenant_message: "{{ tenant_message }}" - service_message: "{{ service_message }}"
This sets some basic configuration, specifies the host this Instance will run on, and the two variables that we’re passing to the playbook.
roles: - install_apache - create_index
This example uses Ansible’s Playbook Roles to organize steps, provide default variables, organize files and templates, and allow for code reuse. Roles are created by using a set directory structure.
In this case, there are two roles, one that installs Apache, and one that creates the index.html
file from a Jinja2 template.
Create a directory named roles
inside steps
, then create two directories named for your roles: install_apache
and create_index
.
Within install_apache
, create a directory named tasks
, then within that directory, a file named main.yml
. This will contain the set of plays for the install_apache
role. To that file add the following:
- name: Install apache using apt apt: name=apache2 update_cache=yes
This will use the Ansible apt module to install Apache.
Next, within create_index
, create two directories, tasks
and templates
. In templates
, create a file named index.html.j2
, with the contents:
ExampleService
Service Message: "{{ service_message }}"
Tenant Message: "{{ tenant_message }}"
These Jinja2 Expressions will be replaced with the values of the variables set in the master playbook.
In the tasks
directory, create a file named main.yml
, with the contents:
- name: Write index.html file to apache document root template: src=index.html.j2 dest=/var/www/html/index.html
This uses the Ansible template module to load and process the Jinja2 template then put it in the dest
location. Note that there is no path given for the src parameter: Ansible knows to look in the templates directory for templates used within a role.
As a final step, you can check your playbooks for best practices with ansible-lint
if you have it available.
The next step is to define an on-boarding recipe for the service. By convention, we use <servicename>-onboard.yaml
, and place it in the xos
directory of the service.
The on-boarding recipe is a TOSCA specification that lists all of the resources for your synchronizer. For example, here is the on-boarding recipe for ExampleService:
--- tosca_definitions_version: tosca_simple_yaml_1_0 description: Onboard the exampleservice imports: - custom_types/xos.yaml topology_template: node_templates: exampleservice: type: tosca.nodes.ServiceController properties: base_url: file:///opt/xos-services/exampleservice/xos/ # The following will concatenate with base_url automatically, if # base_url is non-null. private_key: file:///opt/xos/key_import/exampleservice_rsa public_key: file:///opt/xos/key_import/exampleservice_rsa.pub
This is a legacy recipe that (when executed) on-boards ExampleService in the sense that it registers the service with the system, but it does not provision the service or create instances of the service. These latter steps can be done through CORD's GUI or REST API, or by submitting yet other TOSCA workflows to a running CORD POD (all based on end-points that are auto-generated from these on-boarded models). Additional information on how to provision and use the service is given in the last section of this tutorial.
NOTE: This file may soon be removed.
The final step to on-boarding a service is to include it in one or more service profiles that are to be built and installed. Service profiles are currently defined as part of the CORD build system, so this involves editing some build-related configuration files. These files can be found in the CORD_ROOT/orchestration/profiles/rcord
and CORD_ROOT/build/platform-install
directories of the checkout out source code.
Inserting ExampleService in a service profile requires creating or modifying one of the .yml
files in orchestration/profiles/rcord
of your local repo. In the following, we use rcord.yml
as an illustrative example. There are potentially three sections of this file that need attention.
First, modify the xos-services
section to identify exampleservice
as a service to include in the profile. Doing this effectively points the build system at the model and synchronizer specifications you've just defined.
xos-services: ... (lines omitted)... - name: exampleservice path: orchestration/xos-services/exampleservice keypair: exampleservice_rsa synchronizer: true
Second, optionally tell the build system to download and install an image into CORD. In our particular case, ExampleService uses the trusty-server-multi-nic
that is included in R-CORD for other purposes.
xos_images: - name: "trusty-server-multi-nic". url: "http://www.vicci.org/opencloud/trusty-server-cloudimg-amd64-disk1.img.20170201" checksum: "sha256:ebf007ba3ec1043b7cd011fc6668e2a1d1d4c69c41071e8513ab355df7a057cb" ... (lines omitted)...
Third, optionally specify any GUI extensions associated with the service. This is done in the enabled_gui_extensions
section of the profile manifest. ExampleService does not include a GUI extension.
Today, a few other build-related configuration files require editing to produce a deployment that includes ExampleService. (These details will be hidden in future releases.)
Add the service's synchronizer image to build/docker_images.yml
Because the build system is integrated with the git
and repo
tools, if your service is not already checked into gerrit.opencord.org
, you will also need to add the service to the manifest file CORD_ROOT/.repo/manifest.xml
. Then run git init
in the service’s source tree.
Once ExampleService is on-boarded into a running POD, it is available to be provisioned, controlled and used. This can be done via the CORD GUI or REST API, but the most common way is to input a TOSCA workflow into the running POD. Typically, each service contributes a TOSCA recipe to run as soon as the POD comes up (i.e., as the last stage of the build system), so as to verify that the installation was successful.
This recipe is generated from a Jinja2
template, which is customized at build-time with specific details for the target POD (e.g., the site that hosts the POD). This results in a .yaml
TOSCA file that is passed to the deployed POD and executed.
The ExampleService template is defined by the following file:
build/platform-install/roles/exampleservice-config/templates/test-exampleservice.yaml.j2
It is an historical artifact that this template is in the build/platform-install/roles/exampleservice-config/templates
directory. Templates for new services are instead located in build/platform-install/roles/cord-profile/templates
. For example, see template-service.yaml.j2
in that directory for a template similar to the one used for ExampleService.
The first part of test-exampleservice.yaml.j2
includes some core object reference that ExampleService uses, for example, the trusty-server-multic-nic
image, the small
flavor, and both the management_network
and the public_network
.
--- tosca_definitions_version: tosca_simple_yaml_1_0 imports: - custom_types/slice.yaml - custom_types/site.yaml - custom_types/image.yaml - custom_types/flavor.yaml - custom_types/network.yaml - custom_types/networktemplate.yaml - custom_types/networkslice.yaml - custom_types/exampleservice.yaml - custom_types/exampleserviceinstance.yaml description: configure exampleservice topology_template: node_templates: # site, image, fully created in deployment.yaml {{ site_name }}: type: tosca.nodes.Site properties: must-exist: true name: {{ site_name }} m1.small: type: tosca.nodes.Flavor properties: name: m1.small must-exist: true trusty-server-multi-nic: type: tosca.nodes.Image properties: name: trusty-server-multi-nic must-exist: true # private network template, fully created somewhere else private: type: tosca.nodes.NetworkTemplate properties: must-exist: true name: Private # management networks, fully created in management-net.yaml management_network: type: tosca.nodes.Network properties: must-exist: true name: management # public network, fully created somewhere else public_network: type: tosca.nodes.Network properties: must-exist: true name: public
This is followed by the specification of a private
network used by ExampleService :
exampleservice_network: type: tosca.nodes.Network properties: name: exampleservice_network labels: exampleservice_private_network requirements: - template: node: private relationship: tosca.relationships.BelongsToOne - owner: node: {{ site_name }}_exampleservice relationship: tosca.relationships.BelongsToOne
The next part of the workflow provisions the Slice
(and related instances and networks) in which ExampleService runs. These definitions reference the dependencies established above.
# CORD Slices {{ site_name }}_exampleservice: description: Example Service Slice type: tosca.nodes.Slice properties: name: {{ site_name }}_exampleservice default_isolation: vm network: noauto requirements: - site: node: mysite relationship: tosca.relationships.BelongsToOne - service: node: exampleservice relationship: tosca.relationships.BelongsToOne - default_image: node: trusty-server-multi-nic relationship: tosca.relationships.BelongsToOne - default_flavor: node: m1.small relationship: tosca.relationships.BelongsToOne # CORD NetworkSlices exampleservice_slice_management_network: type: tosca.nodes.NetworkSlice requirements: - network: node: management_network relationship: tosca.relationships.BelongsToOne - slice: node: {{ site_name }}_exampleservice relationship: tosca.relationships.BelongsToOne exampleservice_slice_public_network: type: tosca.nodes.NetworkSlice requirements: - network: node: public_network relationship: tosca.relationships.BelongsToOne - slice: node: {{ site_name }}_exampleservice relationship: tosca.relationships.BelongsToOne exampleservice_slice_exampleservice_network: type: tosca.nodes.NetworkSlice requirements: - network: node: exampleservice_network relationship: tosca.relationships.BelongsToOne - slice: node: {{ site_name }}_exampleservice relationship: tosca.relationships.BelongsToOne
Finally, the recipe instantiates the service object that represents ExampleService (exampleservice
) and spins up a Service Instance on behalf of the first tenant (exampletenant1
).
exampleservice: type: tosca.nodes.ExampleService properties: name: exampleservice public_key: {{ lookup('file', config_cord_profile_dir + '/key_import/exampleservice_rsa.pub') }} private_key_fn: /opt/xos/services/exampleservice/keys/exampleservice_rsa service_message: hello artifacts: pubkey: /opt/cord_profile/key_import/exampleservice_rsa.pub exampletenant1: type: tosca.nodes.ExampleServiceInstance properties: name: exampletenant1 tenant_message: world requirements: - owner: node: exampleservice relationship: tosca.relationships.BelongsToOne
Note that these definitions initialize the service_message
and tenant_message
, respectively. As a consequence, sending an HTTP GET request to ExampleService will result in the response: hello world
. Subsequently, the user can interact with ExampleService via CORD's GUI or REST API to change those values.