Discovery Chain for Service Mesh Traffic Management
1.6.0+: This feature is available in Consul versions 1.6.0 and newer.
This topic is part of a low-level API primarily targeted at developers building external Consul service mesh proxy integrations.
The service discovery process can be modeled as a "discovery chain" which passes through three distinct stages: routing, splitting, and resolution. Each of these stages is controlled by a set of configuration entries. By configuring different phases of the discovery chain a user can control how proxy upstreams are ultimately resolved to specific instances for load balancing.
Note: The discovery chain is currently only used to discover service mesh proxy upstreams.
Configuration
The configuration entries used in the discovery chain are designed to be simple to read and modify for narrowly tailored changes, but at discovery-time the various configuration entries interact in more complex ways. For example:
If a
service-resolver
is created with a service redirect defined, then all references made to the original service in any other configuration entry is replaced with the redirect destination.If a
service-resolver
is created with a default subset defined then all references made to the original service in any other configuration entry that did not specify a subset will be replaced with the default.If a
service-splitter
is created with a service split, and the target service has its ownservice-splitter
then the overall effect is flattened and only a single aggregate traffic split is ultimately configured in the proxy.service-resolver
redirect loops must be rejected as invalid.service-router
andservice-splitter
configuration entries require an L7 compatible protocol be set for the service via either aservice-defaults
orproxy-defaults
config entry. Violations must be rejected as invalid.When an upstream configuration
datacenter
parameter is defined, any configuration entry that does not explicitly refer to a desired datacenter uses the value defined in the upstream.
Compilation
To correctly interpret a collection of configuration entries as a valid discovery chain, we first compile them into a form more directly usable by the layers responsible for configuring service mesh sidecar proxies.
You can interact with the compiler directly using the discovery-chain API.
Compilation Parameters
- Service Name - The service being discovered by name.
- Datacenter - The datacenter to use as the basis of compilation.
- Overrides - Discovery-time tweaks to apply when compiling. These should be derived from either the proxy or upstream configurations if either are set.
Compilation Results
The response is a single wrapped CompiledDiscoveryChain
field:
CompiledDiscoveryChain
The chain encodes a digraph of nodes and targets. Nodes are the compiled representation of various discovery chain stages and targets are instructions on how to use the health API to retrieve relevant service instance lists.
You should traverse the nodes starting with StartNode
. The
nodes can be resolved by name using the Nodes
field. Targets can be
resolved by name using the Targets
field.
ServiceName
(string)
- The requested service.Partition
(string)
- The requested partition.Namespace
(string)
- The requested namespace.Datacenter
(string)
- The requested datacenter.CustomizationHash
(string: <optional>)
- A unique hash of any overrides that affected the compilation of the discovery chain.If set, this value should be used to prefix/suffix any generated load balancer data plane objects to avoid sharing customized and non-customized versions.
Default
(bool: <optional>)
- Indicates if this discovery chain is based on noservice-resolver
,service-splitter
, orservice-router
config entries.Protocol
(string)
- The overall protocol shared by everything in the chain.ServiceMeta
(map<string|string>)
- The metadata from the underlyingservice-defaults
config entry for the service namedServiceName
.StartNode
(string)
- The first key into theNodes
map that should be followed when traversing the discovery chain.Nodes
(map<string|DiscoveryGraphNode>)
- All nodes available for traversal in the chain keyed by a unique name. You can walk this by starting withStartNode
.The names should be treated as opaque values and are only guaranteed to be consistent within a single compilation.
Targets
(map<string|DiscoveryTarget>)
- A list of all targets used in this chain.The names should be treated as opaque values and are only guaranteed to be consistent within a single compilation.
DiscoveryGraphNode
A single node in the compiled discovery chain.
Type
(string)
- The type of the node. Valid values are:router
,splitter
, andresolver
.Name
(string)
- The unique name of the node.Routes
(array<DiscoveryRoute>)
- Only set forType:router
. List of routes to render.Definition
(ServiceRoute)
- Relevant portion of underlyingservice-router
route.NextNode
(string)
- The name of the next node in the chain inNodes
.
Splits
(array<DiscoverySplit>)
- Only set forType:splitter
. List of traffic splits.Resolver
(DiscoveryResolver: <optional>)
- Only set forType:resolver
. How to resolve the service instances.Default
(bool)
- Set to true if noservice-resolver
config entry is defined for this node and the default was synthesized.ConnectTimeout
(duration)
- Copy of the underlyingservice-resolver
ConnectTimeout
field. If one is not defined the default of5s
is returned.Target
(string)
- The name of the target to use found inTargets
.Failover
(DiscoveryFailover: <optional>)
- Compiled form of the underlyingservice-resolver
Failover
definition to use for this request.LoadBalancer
(LoadBalancer: <optional>
) - Copy of the underlyingservice-resolver
LoadBalancer
field.If a
service-splitter
splits between services with differingLoadBalancer
configuration the first hash-based load balancing policy is copied.
DiscoveryTarget
ID
(string)
- The unique name of this target.Service
(string)
- The service to query when resolving a list of service instances.ServiceSubset
(string: <optional>)
- The subset of the service to resolve.Partition
(string)
- The partition to use when resolving a list of service instances.Namespace
(string)
- The namespace to use when resolving a list of service instances.Datacenter
(string)
- The datacenter to use when resolving a list of service instances.Subset
(ServiceResolverSubset)
- Copy of the underlyingservice-resolver
Subsets
definition for this target.Filter
(string: "")
- The filter expression to be used for selecting instances of the requested service. If empty all healthy instances are returned.OnlyPassing
(bool: false)
- Specifies the behavior of the resolver's health check interpretation. If this is set to false, instances with checks in the passing as well as the warning states will be considered healthy. If this is set to true, only instances with checks in the passing state will be considered healthy.
MeshGateway
(MeshGatewayConfig)
- The mesh gateway configuration to use when connecting to this target's service instances.Mode
(string: "")
- One ofnone
,local
, orremote
.
External
(bool: false)
- True if this target is outside of this consul cluster.ConnectTimeout
(duration)
- Copy of the underlyingservice-resolver
ConnectTimeout
field. If one is not defined the default of5s
is returned.SNI
(string)
- This value should be used as the SNI value when connecting to this set of endpoints over TLS.Name
(string)
- The unique name for this target for use when generating load balancer objects. This has a structure similar to SNI, but will not be affected by SNI customizations such asExternalSNI
.