mcp/internal/agent/purge.go

package agent

import (
	"context"
	"fmt"

	mcpv1 "git.wntrmute.dev/kyle/mcp/gen/mcp/v1"
	"git.wntrmute.dev/kyle/mcp/internal/registry"
)

// PurgeComponent removes stale registry entries for components that are both
// gone (observed state is removed/unknown/exited) and unwanted (not in any
// current service definition). It never touches running containers.
func (a *Agent) PurgeComponent(ctx context.Context, req *mcpv1.PurgeRequest) (*mcpv1.PurgeResponse, error) {
	a.Logger.Info("PurgeComponent",
		"service", req.GetService(),
		"component", req.GetComponent(),
		"dry_run", req.GetDryRun(),
	)

	// Build a set of defined service/component pairs for quick lookup.
	defined := make(map[string]bool, len(req.GetDefinedComponents()))
	for _, dc := range req.GetDefinedComponents() {
		defined[dc] = true
	}

	// Determine which services to examine.
	var services []registry.Service
	if req.GetService() != "" {
		svc, err := registry.GetService(a.DB, req.GetService())
		if err != nil {
			return nil, fmt.Errorf("get service %q: %w", req.GetService(), err)
		}
		services = []registry.Service{*svc}
	} else {
		var err error
		services, err = registry.ListServices(a.DB)
		if err != nil {
			return nil, fmt.Errorf("list services: %w", err)
		}
	}

	var results []*mcpv1.PurgeResult

	for _, svc := range services {
		components, err := registry.ListComponents(a.DB, svc.Name)
		if err != nil {
			return nil, fmt.Errorf("list components for %q: %w", svc.Name, err)
		}

		// If a specific component was requested, filter to just that one.
		if req.GetComponent() != "" {
			var filtered []registry.Component
			for _, c := range components {
				if c.Name == req.GetComponent() {
					filtered = append(filtered, c)
				}
			}
			components = filtered
		}

		for _, comp := range components {
			result := a.evaluatePurge(svc.Name, &comp, defined, req.GetDryRun())
			results = append(results, result)
		}

		// If all components of this service were purged (not dry-run),
		// check if the service should be cleaned up too.
		if !req.GetDryRun() {
			remaining, err := registry.ListComponents(a.DB, svc.Name)
			if err != nil {
				a.Logger.Warn("failed to check remaining components", "service", svc.Name, "err", err)
				continue
			}
			if len(remaining) == 0 {
				if err := registry.DeleteService(a.DB, svc.Name); err != nil {
					a.Logger.Warn("failed to delete empty service", "service", svc.Name, "err", err)
				} else {
					a.Logger.Info("purged empty service", "service", svc.Name)
				}
			}
		}
	}

	return &mcpv1.PurgeResponse{Results: results}, nil
}

// purgeableStates are observed states that indicate a component's container
// is gone and the registry entry can be safely removed.
var purgeableStates = map[string]bool{
	"removed": true,
	"unknown": true,
	"exited":  true,
}

// evaluatePurge checks whether a single component is eligible for purge and,
// if not in dry-run mode, deletes it.
func (a *Agent) evaluatePurge(service string, comp *registry.Component, defined map[string]bool, dryRun bool) *mcpv1.PurgeResult {
	key := service + "/" + comp.Name

	// Safety: refuse to purge components with a live container.
	if !purgeableStates[comp.ObservedState] {
		return &mcpv1.PurgeResult{
			Service:   service,
			Component: comp.Name,
			Purged:    false,
			Reason:    fmt.Sprintf("observed=%s, container still exists", comp.ObservedState),
		}
	}

	// Don't purge components that are still in service definitions.
	if defined[key] {
		return &mcpv1.PurgeResult{
			Service:   service,
			Component: comp.Name,
			Purged:    false,
			Reason:    "still in service definitions",
		}
	}

	reason := fmt.Sprintf("observed=%s, not in service definitions", comp.ObservedState)

	if dryRun {
		return &mcpv1.PurgeResult{
			Service:   service,
			Component: comp.Name,
			Purged:    true,
			Reason:    reason,
		}
	}

	// Delete events first (events table has no FK to components).
	if err := registry.DeleteComponentEvents(a.DB, service, comp.Name); err != nil {
		a.Logger.Warn("failed to delete events during purge", "service", service, "component", comp.Name, "err", err)
	}

	// Delete the component (CASCADE handles ports, volumes, cmd).
	if err := registry.DeleteComponent(a.DB, service, comp.Name); err != nil {
		return &mcpv1.PurgeResult{
			Service:   service,
			Component: comp.Name,
			Purged:    false,
			Reason:    fmt.Sprintf("delete failed: %v", err),
		}
	}

	a.Logger.Info("purged component", "service", service, "component", comp.Name, "reason", reason)

	return &mcpv1.PurgeResult{
		Service:   service,
		Component: comp.Name,
		Purged:    true,
		Reason:    reason,
	}
}