fluid-core/internal/runtime/registry.go

// Package runtime implements the FLUID deterministic data plane: the gateway,
// revision resolution, routing, backend connection and telemetry emission.
//
// Nothing here may depend on the evolution control plane to serve a request.
// ArchitectureBlueprint.md section 2 makes this an invariant: the interface
// runtime must continue to function when the Daimon, the model provider, the
// hypothesis store, the experiment controller and the AI budget are all
// unavailable.
package runtime

import (
	"errors"
	"fmt"
	"sync"

	"github.com/tegwick/fluid-core/internal/contract"
)

// Registry is the gateway's cached view of published control-plane state.
//
// It is deliberately a snapshot rather than a client. When the control plane
// dies the registry keeps answering from what it last held, which is what
// ArchitectureBlueprint.md section 34.6 requires: the runtime continues using
// cached published configuration, and no new promotions occur until
// control-plane consistency is restored.
type Registry struct {
	mu        sync.RWMutex
	iface     contract.InterfaceID
	revisions map[contract.RevisionID]contract.Revision
	policy    contract.RoutingPolicy
	hasPolicy bool
}

// NewRegistry returns an empty registry for one interface.
func NewRegistry(iface contract.InterfaceID) *Registry {
	return &Registry{
		iface:     iface,
		revisions: make(map[contract.RevisionID]contract.Revision),
	}
}

var (
	// ErrUnknownRevision is returned for a revision the registry has never seen.
	ErrUnknownRevision = errors.New("unknown revision")
	// ErrRevisionNotRoutable is returned for a revision that exists but must not
	// receive traffic.
	ErrRevisionNotRoutable = errors.New("revision not routable")
	// ErrNoPolicy is returned before any routing policy has been loaded.
	ErrNoPolicy = errors.New("no routing policy loaded")
	// ErrWrongInterface guards against loading another interface's artifacts.
	ErrWrongInterface = errors.New("artifact belongs to a different interface")
	// ErrStalePolicy is returned when an older policy generation is offered.
	ErrStalePolicy = errors.New("routing policy generation is not newer")
)

// PutRevision publishes a revision descriptor into the registry.
//
// The descriptor is expected to have been signature-verified already; this
// method enforces only the structural conditions the router depends on.
func (r *Registry) PutRevision(d contract.Revision) error {
	if d.Interface != r.iface {
		return fmt.Errorf("%w: descriptor is for %q, registry serves %q",
			ErrWrongInterface, d.Interface, r.iface)
	}
	if d.Runtime.Upstream == "" {
		return fmt.Errorf("revision %s: descriptor has no runtime upstream", d.ID)
	}
	if !d.State.Valid() {
		return fmt.Errorf("revision %s: unknown state %q", d.ID, d.State)
	}

	r.mu.Lock()
	defer r.mu.Unlock()
	r.revisions[d.ID] = d
	return nil
}

// Revision returns a published descriptor.
func (r *Registry) Revision(id contract.RevisionID) (contract.Revision, error) {
	r.mu.RLock()
	defer r.mu.RUnlock()
	d, ok := r.revisions[id]
	if !ok {
		return contract.Revision{}, fmt.Errorf("%w: %s", ErrUnknownRevision, id)
	}
	return d, nil
}

// PutPolicy installs a routing policy.
//
// Generations are monotonic: an older policy is refused rather than applied.
// Without this a delayed delivery could silently roll traffic back to a
// superseded allocation, which would corrupt an in-flight experiment's
// measurement window.
func (r *Registry) PutPolicy(p contract.RoutingPolicy) error {
	if p.Interface != r.iface {
		return fmt.Errorf("%w: policy is for %q, registry serves %q",
			ErrWrongInterface, p.Interface, r.iface)
	}

	r.mu.Lock()
	defer r.mu.Unlock()

	if r.hasPolicy && p.Generation <= r.policy.Generation {
		return fmt.Errorf("%w: offered %d, holding %d",
			ErrStalePolicy, p.Generation, r.policy.Generation)
	}
	r.policy = p
	r.hasPolicy = true
	return nil
}

// Policy returns the current routing policy.
func (r *Registry) Policy() (contract.RoutingPolicy, error) {
	r.mu.RLock()
	defer r.mu.RUnlock()
	if !r.hasPolicy {
		return contract.RoutingPolicy{}, ErrNoPolicy
	}
	return r.policy, nil
}

// routableStates lists the descriptor states the router may send traffic to.
//
// ArchitectureBlueprint.md section 5.3 requires rejecting unpublished, failed
// and retired revisions. "created" is unpublished; "verified" has passed tests
// but has not been exposed; "retired" is finished.
var routableStates = map[contract.RevisionState]bool{
	contract.RevisionStateExperiment: true,
	contract.RevisionStateCandidate:  true,
	contract.RevisionStateStable:     true,
	contract.RevisionStateDeprecated: true,
}

// CheckRoutable reports whether a revision may currently receive traffic from
// the given cohort.
func (r *Registry) CheckRoutable(id contract.RevisionID, cohort contract.CohortID) error {
	d, err := r.Revision(id)
	if err != nil {
		return err
	}

	if !routableStates[d.State] {
		return fmt.Errorf("%w: %s is %s", ErrRevisionNotRoutable, id, d.State)
	}
	if d.Policy.SecurityCheck != contract.RevisionPolicySecurityCheckPassed {
		return fmt.Errorf("%w: %s has security_check=%s",
			ErrRevisionNotRoutable, id, d.Policy.SecurityCheck)
	}
	if d.Policy.PolicyCheck != nil && *d.Policy.PolicyCheck == contract.RevisionPolicyPolicyCheckFailed {
		return fmt.Errorf("%w: %s failed its policy check", ErrRevisionNotRoutable, id)
	}

	if d.Routing != nil && len(d.Routing.EligibleCohorts) > 0 {
		if !containsCohort(d.Routing.EligibleCohorts, cohort) {
			return fmt.Errorf("%w: cohort %q is not eligible for %s",
				ErrRevisionNotRoutable, cohort, id)
		}
	}
	return nil
}

func containsCohort(list []contract.CohortID, want contract.CohortID) bool {
	for _, c := range list {
		if c == want {
			return true
		}
	}
	return false
}