mcp/internal/agent/recover.go

package agent

import (
	"context"
	"fmt"

	"git.wntrmute.dev/mc/mcp/internal/registry"
	"git.wntrmute.dev/mc/mcp/internal/runtime"
)

// Recover recreates containers from the agent's registry for all services
// whose desired state is "running" but which don't have a running container
// in podman. This is the recovery path after a podman database loss (e.g.,
// after a UID change or podman reset).
//
// Recover does NOT pull images — it assumes the images are cached locally.
// If an image is missing, that component is skipped with a warning.
func (a *Agent) Recover(ctx context.Context) error {
	services, err := registry.ListServices(a.DB)
	if err != nil {
		return fmt.Errorf("list services: %w", err)
	}

	// Get the list of currently running containers from podman.
	running, err := a.listAllContainers(ctx)
	if err != nil {
		a.Logger.Warn("cannot list containers, assuming none running", "err", err)
		running = nil
	}
	runningSet := make(map[string]bool)
	for _, c := range running {
		runningSet[c.Name] = true
	}

	var recovered, skipped, already int

	for _, svc := range services {
		if !svc.Active {
			continue
		}

		comps, err := registry.ListComponents(a.DB, svc.Name)
		if err != nil {
			a.Logger.Warn("list components", "service", svc.Name, "err", err)
			continue
		}

		for _, comp := range comps {
			if comp.DesiredState != "running" {
				continue
			}

			containerName := svc.Name + "-" + comp.Name
			if comp.Name == svc.Name {
				containerName = svc.Name
			}

			// Skip if container is already running.
			if runningSet[containerName] {
				already++
				continue
			}

			a.Logger.Info("recovering container",
				"service", svc.Name,
				"component", comp.Name,
				"image", comp.Image,
			)

			rt := a.runtimeFor(comp.Runtime)

			// Remove any stale container with the same name.
			_ = rt.Remove(ctx, containerName)

			// Build the container spec from the registry.
			spec := runtime.ContainerSpec{
				Name:     containerName,
				Image:    comp.Image,
				Network:  comp.Network,
				User:     comp.UserSpec,
				Restart:  comp.Restart,
				Volumes:  comp.Volumes,
				Cmd:      comp.Cmd,
				MemoryMB: comp.MemoryMB,
				VCPUs:    comp.VCPUs,
			}

			// Allocate ports from routes if the component has routes.
			if len(comp.Routes) > 0 && a.PortAlloc != nil {
				ports, env, allocErr := a.allocateRoutePorts(svc.Name, comp.Name, comp.Routes)
				if allocErr != nil {
					a.Logger.Warn("allocate route ports", "container", containerName, "err", allocErr)
					spec.Ports = comp.Ports
				} else {
					spec.Ports = append(comp.Ports, ports...)
					spec.Env = append(spec.Env, env...)
				}
			} else {
				spec.Ports = comp.Ports
			}

			if err := rt.Run(ctx, spec); err != nil {
				a.Logger.Error("recover container failed",
					"container", containerName,
					"err", err,
				)
				skipped++
				continue
			}

			// Re-register mc-proxy routes.
			if a.Proxy != nil && len(comp.Routes) > 0 {
				hostPorts, hpErr := registry.GetRouteHostPorts(a.DB, svc.Name, comp.Name)
				if hpErr == nil {
					if proxyErr := a.Proxy.RegisterRoutes(ctx, svc.Name, comp.Routes, hostPorts); proxyErr != nil {
						a.Logger.Warn("re-register routes", "service", svc.Name, "err", proxyErr)
					}
				}
			}

			// Provision TLS certs if needed.
			if a.Certs != nil && hasL7Routes(comp.Routes) {
				hostnames := l7Hostnames(svc.Name, comp.Routes)
				if certErr := a.Certs.EnsureCert(ctx, svc.Name, hostnames); certErr != nil {
					a.Logger.Warn("cert provisioning", "service", svc.Name, "err", certErr)
				}
			}

			recovered++
			a.Logger.Info("container recovered", "container", containerName)
		}
	}

	a.Logger.Info("recovery complete",
		"recovered", recovered,
		"skipped", skipped,
		"already_running", already,
	)

	return nil
}

// hasL7Routes and l7Hostnames are defined in deploy.go.