path: root/docs/sections/guides/development_guides/oom_dev_container_orchestration.rst

.. This work is licensed under a Creative Commons Attribution 4.0
.. International License.
.. http://creativecommons.org/licenses/by/4.0
.. Copyright 2018-2020 Amdocs, Bell Canada, Orange, Samsung
.. Modification copyright (C) 2022 Nordix Foundation

.. Links
.. _Kubernetes: https://Kubernetes.io/
.. _AWS Elastic Block Store: https://aws.amazon.com/ebs/
.. _Azure File: https://docs.microsoft.com/en-us/azure/storage/files/storage-files-introduction
.. _GCE Persistent Disk: https://cloud.google.com/compute/docs/disks/
.. _Gluster FS: https://www.gluster.org/
.. _Kubernetes Storage Class: https://Kubernetes.io/docs/concepts/storage/storage-classes/
.. _Assigning Pods to Nodes: https://Kubernetes.io/docs/concepts/configuration/assign-pod-node/


.. _oom_dev_container_orch:

Kubernetes Container Orchestration
##################################

The ONAP components are managed by the Kubernetes_ container management system
which maintains the desired state of the container system as described by one
or more deployment descriptors - similar in concept to OpenStack HEAT
Orchestration Templates. The following sections describe the fundamental
objects managed by Kubernetes, the network these components use to communicate
with each other and other entities outside of ONAP and the templates that
describe the configuration and desired state of the ONAP components.

**Name Spaces**

Within the namespaces are Kubernetes services that provide external
connectivity to pods that host Docker containers.

ONAP Components to Kubernetes Object Relationships
--------------------------------------------------
Kubernetes deployments consist of multiple objects:

- **nodes** - a worker machine - either physical or virtual - that hosts
  multiple containers managed by Kubernetes.
- **services** - an abstraction of a logical set of pods that provide a
  micro-service.
- **pods** - one or more (but typically one) container(s) that provide specific
  application functionality.
- **persistent volumes** - One or more permanent volumes need to be established
  to hold non-ephemeral configuration and state data.

The relationship between these objects is shown in the following figure:

.. .. uml::
..
..   @startuml
..   node PH {
..      component Service {
..         component Pod0
..         component Pod1
..      }
..   }
..
..   database PV
..   @enduml

.. figure:: ../../resources/images/k8s/kubernetes_objects.png

OOM uses these Kubernetes objects as described in the following sections.

Nodes
~~~~~
OOM works with both physical and virtual worker machines.

* Virtual Machine Deployments - If ONAP is to be deployed onto a set of virtual
  machines, the creation of the VMs is outside of the scope of OOM and could be
  done in many ways, such as

  * manually, for example by a user using the OpenStack Horizon dashboard or
    AWS EC2, or
  * automatically, for example with the use of a OpenStack Heat Orchestration
    Template which builds an ONAP stack, Azure ARM template, AWS CloudFormation
    Template, or
  * orchestrated, for example with Cloudify creating the VMs from a TOSCA
    template and controlling their life cycle for the life of the ONAP
    deployment.

* Physical Machine Deployments - If ONAP is to be deployed onto physical
  machines there are several options but the recommendation is to use Rancher
  along with Helm to associate hosts with a Kubernetes cluster.

Pods
~~~~
A group of containers with shared storage and networking can be grouped
together into a Kubernetes pod.  All of the containers within a pod are
co-located and co-scheduled so they operate as a single unit.  Within ONAP
Amsterdam release, pods are mapped one-to-one to docker containers although
this may change in the future.  As explained in the Services section below the
use of Pods within each ONAP component is abstracted from other ONAP
components.

Services
~~~~~~~~
OOM uses the Kubernetes service abstraction to provide a consistent access
point for each of the ONAP components independent of the pod or container
architecture of that component.  For example, the SDNC component may introduce
OpenDaylight clustering as some point and change the number of pods in this
component to three or more but this change will be isolated from the other ONAP
components by the service abstraction.  A service can include a load balancer
on its ingress to distribute traffic between the pods and even react to dynamic
changes in the number of pods if they are part of a replica set.

Persistent Volumes
~~~~~~~~~~~~~~~~~~
To enable ONAP to be deployed into a wide variety of cloud infrastructures a
flexible persistent storage architecture, built on Kubernetes persistent
volumes, provides the ability to define the physical storage in a central
location and have all ONAP components securely store their data.

When deploying ONAP into a public cloud, available storage services such as
`AWS Elastic Block Store`_, `Azure File`_, or `GCE Persistent Disk`_ are
options.  Alternatively, when deploying into a private cloud the storage
architecture might consist of Fiber Channel, `Gluster FS`_, or iSCSI. Many
other storage options existing, refer to the `Kubernetes Storage Class`_
documentation for a full list of the options. The storage architecture may vary
from deployment to deployment but in all cases a reliable, redundant storage
system must be provided to ONAP with which the state information of all ONAP
components will be securely stored. The Storage Class for a given deployment is
a single parameter listed in the ONAP values.yaml file and therefore is easily
customized. Operation of this storage system is outside the scope of the OOM.

.. code-block:: yaml

  Insert values.yaml code block with storage block here

Once the storage class is selected and the physical storage is provided, the
ONAP deployment step creates a pool of persistent volumes within the given
physical storage that is used by all of the ONAP components. ONAP components
simply make a claim on these persistent volumes (PV), with a persistent volume
claim (PVC), to gain access to their storage.

The following figure illustrates the relationships between the persistent
volume claims, the persistent volumes, the storage class, and the physical
storage.

.. graphviz::

   digraph PV {
      label = "Persistance Volume Claim to Physical Storage Mapping"
      {
         node [shape=cylinder]
         D0 [label="Drive0"]
         D1 [label="Drive1"]
         Dx [label="Drivex"]
      }
      {
         node [shape=Mrecord label="StorageClass:ceph"]
         sc
      }
      {
         node [shape=point]
         p0 p1 p2
         p3 p4 p5
      }
      subgraph clusterSDC {
         label="SDC"
         PVC0
         PVC1
      }
      subgraph clusterSDNC {
         label="SDNC"
         PVC2
      }
      subgraph clusterSO {
         label="SO"
         PVCn
      }
      PV0 -> sc
      PV1 -> sc
      PV2 -> sc
      PVn -> sc

      sc -> {D0 D1 Dx}
      PVC0 -> PV0
      PVC1 -> PV1
      PVC2 -> PV2
      PVCn -> PVn

      # force all of these nodes to the same line in the given order
      subgraph {
         rank = same; PV0;PV1;PV2;PVn;p0;p1;p2
         PV0->PV1->PV2->p0->p1->p2->PVn [style=invis]
      }

      subgraph {
         rank = same; D0;D1;Dx;p3;p4;p5
         D0->D1->p3->p4->p5->Dx [style=invis]
      }

   }

In-order for an ONAP component to use a persistent volume it must make a claim
against a specific persistent volume defined in the ONAP common charts.  Note
that there is a one-to-one relationship between a PVC and PV.  The following is
an excerpt from a component chart that defines a PVC:

.. code-block:: yaml

  Insert PVC example here

OOM Networking with Kubernetes
------------------------------

- DNS
- Ports - Flattening the containers also expose port conflicts between the
  containers which need to be resolved.


Pod Placement Rules
-------------------
OOM will use the rich set of Kubernetes node and pod affinity /
anti-affinity rules to minimize the chance of a single failure resulting in a
loss of ONAP service. Node affinity / anti-affinity is used to guide the
Kubernetes orchestrator in the placement of pods on nodes (physical or virtual
machines).  For example:

- if a container used Intel DPDK technology the pod may state that it as
  affinity to an Intel processor based node, or
- geographical based node labels (such as the Kubernetes standard zone or
  region labels) may be used to ensure placement of a DCAE complex close to the
  VNFs generating high volumes of traffic thus minimizing networking cost.
  Specifically, if nodes were pre-assigned labels East and West, the pod
  deployment spec to distribute pods to these nodes would be:

.. code-block:: yaml

  nodeSelector:
    failure-domain.beta.Kubernetes.io/region: {{ .Values.location }}

- "location: West" is specified in the `values.yaml` file used to deploy
  one DCAE cluster and  "location: East" is specified in a second `values.yaml`
  file (see OOM Configuration Management for more information about
  configuration files like the `values.yaml` file).

Node affinity can also be used to achieve geographic redundancy if pods are
assigned to multiple failure domains. For more information refer to `Assigning
Pods to Nodes`_.

.. note::
   One could use Pod to Node assignment to totally constrain Kubernetes when
   doing initial container assignment to replicate the Amsterdam release
   OpenStack Heat based deployment. Should one wish to do this, each VM would
   need a unique node name which would be used to specify a node constaint
   for every component.  These assignment could be specified in an environment
   specific values.yaml file. Constraining Kubernetes in this way is not
   recommended.

Kubernetes has a comprehensive system called Taints and Tolerations that can be
used to force the container orchestrator to repel pods from nodes based on
static events (an administrator assigning a taint to a node) or dynamic events
(such as a node becoming unreachable or running out of disk space). There are
no plans to use taints or tolerations in the ONAP Beijing release.  Pod
affinity / anti-affinity is the concept of creating a spacial relationship
between pods when the Kubernetes orchestrator does assignment (both initially
an in operation) to nodes as explained in Inter-pod affinity and anti-affinity.
For example, one might choose to co-located all of the ONAP SDC containers on a
single node as they are not critical runtime components and co-location
minimizes overhead. On the other hand, one might choose to ensure that all of
the containers in an ODL cluster (SDNC and APPC) are placed on separate nodes
such that a node failure has minimal impact to the operation of the cluster.
An example of how pod affinity / anti-affinity is shown below:

Pod Affinity / Anti-Affinity

.. code-block:: yaml

  apiVersion: v1
  kind: Pod
  metadata:
    name: with-pod-affinity
  spec:
    affinity:
      podAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
        - labelSelector:
            matchExpressions:
        - key: security
          operator: In
          values:
          - S1
          topologyKey: failure-domain.beta.Kubernetes.io/zone
      podAntiAffinity:
        preferredDuringSchedulingIgnoredDuringExecution:
        - weight: 100
          podAffinityTerm:
            labelSelector:
              matchExpressions:
              - key: security
                operator: In
                values:
                - S2
            topologyKey: Kubernetes.io/hostname
       containers:
       - name: with-pod-affinity
         image: gcr.io/google_containers/pause:2.0

This example contains both podAffinity and podAntiAffinity rules, the first
rule is is a must (requiredDuringSchedulingIgnoredDuringExecution) while the
second will be met pending other considerations
(preferredDuringSchedulingIgnoredDuringExecution).  Preemption Another feature
that may assist in achieving a repeatable deployment in the presence of faults
that may have reduced the capacity of the cloud is assigning priority to the
containers such that mission critical components have the ability to evict less
critical components.  Kubernetes provides this capability with Pod Priority and
Preemption.  Prior to having more advanced production grade features available,
the ability to at least be able to re-deploy ONAP (or a subset of) reliably
provides a level of confidence that should an outage occur the system can be
brought back on-line predictably.

Health Checks
-------------

Monitoring of ONAP components is configured in the agents within JSON files and
stored in gerrit under the consul-agent-config, here is an example from the AAI
model loader (aai-model-loader-health.json):

.. code-block:: json

  {
    "service": {
      "name": "A&AI Model Loader",
      "checks": [
        {
          "id": "model-loader-process",
          "name": "Model Loader Presence",
          "script": "/consul/config/scripts/model-loader-script.sh",
          "interval": "15s",
          "timeout": "1s"
        }
      ]
    }
  }

Liveness Probes
---------------

These liveness probes can simply check that a port is available, that a
built-in health check is reporting good health, or that the Consul health check
is positive.  For example, to monitor the SDNC component has following liveness
probe can be found in the SDNC DB deployment specification:

.. code-block:: yaml

  sdnc db liveness probe

  livenessProbe:
    exec:
      command: ["mysqladmin", "ping"]
      initialDelaySeconds: 30 periodSeconds: 10
      timeoutSeconds: 5

The 'initialDelaySeconds' control the period of time between the readiness
probe succeeding and the liveness probe starting. 'periodSeconds' and
'timeoutSeconds' control the actual operation of the probe.  Note that
containers are inherently ephemeral so the healing action destroys failed
containers and any state information within it.  To avoid a loss of state, a
persistent volume should be used to store all data that needs to be persisted
over the re-creation of a container.  Persistent volumes have been created for
the database components of each of the projects and the same technique can be
used for all persistent state information.