group-based-policy/doc/source/devref/service-chain-abstractions.rst

Service Chaining

Group Based Policy provides the intent based application-oriented abstractions for the specification of networking requirements to deploy applications. Network Services are an essential component for the deployment of applications.

One common use-case for using multiple services can be described as a "service-chain" - that is, application of specific services in a specific order for every packet on a specific datapath. This API provides an abstraction to specify that behavior with clear definition of the expected semantics.

The goal of this specification is to provide the user with a tool that captures their high level intent without getting coupled with incidental details of their specific deployment. This then provides a path for migrating such services across technology changes or for hybrid deployments. By design, this specification does not mandate the technology used to providethat service.

Typical scenarios for combining services can usually be specified quite succinctly (as traffic on this interface must first inspected by a firewall and then processed by a load balancer). Unfortunately specifying that can many times get mired in incidental complexity around service insertion and traffic steering issues. If the user is able to describe that intent, we can orchestrate the required service lifecycle events and steer the traffic as required without exposing that complexity (and the resulting coupling) outside the specific implementation details.

This has the added benefits that:

1. As we are only specifying the intent, it is back-end technology agnostic
2. The additional information allows us to provide technology upgrade without breaking service usage from a user perspective
3. It can support hybrid deployments, or migrations across vendors, even when the underlying technology used by those vendors is different.

Also, when specifying abstractions that are required to be implemented across technologies, it is critical that the semantics that are expected by the API (or implied) be clearly defined so that the usage can actually be portable across those technologies.

Database models

1. ServiceChainNode

Attribute Name	Type	Access	Default Value	Validation/ Conversion	Description
id	string (UUID)	RO, all	generated	N/A	identity
name	string	RW, all	''	string	human-readable name
type (flavor?)	string	RW, all	required	foreign-key	service-type
config	string	RW, all	''	string	service configuration (as a HEAT template)
service_params	list of strings	RW, all	''	list of strings	list of required service config param names

2. ServiceChainSpec

Attribute Name	Type	Access	Default Value	Validation/ Conversion	Description
id	string (UUID)	RO, all	generated	N/A	identity
name	string	RW, all	''	string	human-readable name
nodes	string	RW, all	required	list of strings (UUIDs)	list of ServiceChainNode
service_params	list of strings	RO, all	generated	N/A	list of required service config parameter names

service_params is generated by aggregating the service_params of each of the ServiceChainNodes in the ServiceChainSpec. The parameter is not specified in the API to create the ServiceChainSpec resource.

3. ServiceChainInstance

Attribute Name	Type	Access	Default Value	Validation/ Conversion	Description
id	string (UUID)	RO, all	generated	N/A	identity
name	string	RW, all	''	string	human-readable name
service-chain-spec	string	RW, all	required	foreign-key for ServiceChainSpec	service-chain spec for this instance
provider_ptg	string (UUID)	RW, all	required	foreign-key	Destination PolicyTargetGroup
consumer_ptg	string (UUID)	RW, all	required	foreign-key	Source PolicyTargetGroup
classifier	string (UUID)	RW, all	required	foreign-key	Classifier
service_param_values	string	RW, all	required	dictionary	configuration parameter names and values

SEMANTICS:

The expected semantics would be equivalent of:

1. As if the services were created to process traffic from consumer_ptg to provider_ptg that matches the classifier NOTE: This is just specifying that the service chain needs to be applied to all traffic that is traversing between the PolicyTargetGroups. The provider may implement it using any valid insertion strategy.
2. In the order of ServiceChainNodes in the ServiceChainSpec for inbound traffic to the Destination PolicyTargetGroup, and in opposite order for outbound traffic from the Destination PolicyTargetGroup
3. Not all providers will honor arbitrary ordering of services for application of the service. In that case, the provider will raise a "NotImplemented" exception.

USAGE WORKFLOW:

1. Assume a application policy that defines connectivity between a provider PolicyTargetGroup (ptg1) and a consumer PolicyTargetGroup (ptg2)

2. Assume that the semantics that I want to provide are of having all traffic from ptg1 to/from ptg2 needs to be (a) first inspected by a firewall, and then (b) load balanced by a load balancer.

3. Then I would create a ServiceChainSpec with 2 ServiceChainNodes. The first node would be of type FW and the second one LB. The FW node would have config string as the HEAT template for FWaaS configuration and the LB would have the config string as the HEAT template for the LBaaS configuration. CLI for that would look like:

   gbp servicechain-node-create --type flavor_id --config_file fw_heat_template fw_node
   gbp servicechain-node-create --type flavor_id --config_file lb_heat_template lb_node
   
   gbp servicechain-spec-create --nodes "fw_node;lb_node" fwlb_spec

   This creates the ordered-list ["FW", "LB"] as the list of services in the chain.

4. The spec fwlb_spec created in step 3 would be used as the target of a policy-rule in the application policy

5. Finally the GBP provider would create a ServiceChainInstance from this ServiceChainSpec. A equivalent CLI command for that would look like:

   gbp servicechain-instance-create --servicechain_spec_id fwlb_spec --provider_ptg ptg1 --consumer_ptg ptg2 --classifier classifier-all --config_param_values "vip=IP1" service-chain

   This creates a chain that applies services in the order:

   • FW->LB->ptg1 for ingress traffic, and
   • ptg1->LB->FW for egress traffic.

Notifications

1. All updates to service-chain-spec resources are relayed to the configured service-chain-providers

2. Updates to ServiceChainNode or ServiceChainSpec generate notifications to backend to "fixup" the ServiceChainInstances as required.

3. It is assumed that the existing notifications exception handling meets the needs for this API and no new constructs are specified.