fix: new Compose deployment with volumes and --recreate flag (fixes #176)

pkg/client/deploy/deploy.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 
 	"github.com/psviderski/uncloud/pkg/api"
+	"github.com/psviderski/uncloud/pkg/client/deploy/scheduler"
 )
 
 type Client interface {
@@ -25,6 +26,8 @@ type Deployment struct {
 	Strategy Strategy
 	cli      Client
 	plan     *Plan
+	// state is an optional current and planned cluster state used for scheduling decisions.
+	state *scheduler.ClusterState
 }
 
 type Plan struct {
@@ -47,6 +50,16 @@ func NewDeployment(cli Client, spec api.ServiceSpec, strategy Strategy) *Deploym
 	}
 }
 
+// NewDeploymentWithClusterState creates a new deployment like NewDeployment but also with a provided current cluster
+// state used for scheduling decisions.
+func NewDeploymentWithClusterState(
+	cli Client, spec api.ServiceSpec, strategy Strategy, state *scheduler.ClusterState,
+) *Deployment {
+	d := NewDeployment(cli, spec, strategy)
+	d.state = state
+	return d
+}
+
 // Plan returns a plan of operations to reconcile the service to the desired state.
 // If a plan has already been created, the same plan will be returned.
 func (d *Deployment) Plan(ctx context.Context) (Plan, error) {
@@ -73,7 +86,14 @@ func (d *Deployment) Plan(ctx context.Context) (Plan, error) {
 		return Plan{}, fmt.Errorf("resolve service spec: %w", err)
 	}
 
-	plan, err := d.Strategy.Plan(ctx, d.cli, d.Service, resolvedSpec)
+	if d.state == nil {
+		d.state, err = scheduler.InspectClusterState(ctx, d.cli)
+		if err != nil {
+			return Plan{}, fmt.Errorf("inspect cluster state: %w", err)
+		}
+	}
+
+	plan, err := d.Strategy.Plan(d.state, d.Service, resolvedSpec)
 	if err != nil {
 		return Plan{}, fmt.Errorf("create plan using %s strategy: %w", d.Strategy.Type(), err)
 	}

pkg/client/deploy/scheduler/volume.go

@@ -17,7 +17,7 @@ import (
 //   - If a volume already exists on a machine, it must be used instead of creating a new one.
 //   - A missing volume must only be created on one machine.
 type VolumeScheduler struct {
-	// state is the current state of machines and their resources in the cluster.
+	// state is the current and planned state of machines and their resources in the cluster.
 	state *ClusterState
 	// serviceSpecs is a list of service specifications included in the deployment.
 	serviceSpecs []api.ServiceSpec
@@ -106,6 +106,7 @@ func NewVolumeScheduler(state *ClusterState, specs []api.ServiceSpec) (*VolumeSc
 // Schedule determines what missing volumes should be created and where for services in the multi-service deployment.
 // It returns a map of machine IDs to a list of api.VolumeSpec that should be created on that machine,
 // or an error if services can't be scheduled due to scheduling constraints.
+// It also updates the state of the machines in the cluster state to reflect the scheduled volumes.
 func (s *VolumeScheduler) Schedule() (map[string][]api.VolumeSpec, error) {
 	if len(s.serviceSpecs) == 0 {
 		// No services with volume mounts, nothing to schedule.

pkg/client/deploy/strategy.go

@@ -1,7 +1,6 @@
 package deploy
 
 import (
-	"context"
 	"fmt"
 	"math/rand/v2"
 	"slices"
@@ -18,31 +17,31 @@ type Strategy interface {
 	// Type returns the type of the deployment strategy, e.g. "rolling", "blue-green".
 	Type() string
 	// Plan returns the operation to reconcile the service to the desired state.
-	// If the service does not exist (new deployment), svc will be nil.
-	Plan(ctx context.Context, cli scheduler.Client, svc *api.Service, spec api.ServiceSpec) (Plan, error)
+	// If the service does not exist (new deployment), svc will be nil. state provides the current and planned state
+	// of the cluster for scheduling decisions.
+	Plan(state *scheduler.ClusterState, svc *api.Service, spec api.ServiceSpec) (Plan, error)
 }
 
 // RollingStrategy implements a rolling update deployment pattern where containers are updated one at a time
 // to minimize service disruption.
 type RollingStrategy struct {
-	State         *scheduler.ClusterState
+	// ForceRecreate indicates whether all containers should be recreated during the deployment,
+	// regardless of whether their specifications have changed.
 	ForceRecreate bool
+
+	// state is the current and planned state of the cluster used for scheduling decisions.
+	state *scheduler.ClusterState
 }
 
 func (s *RollingStrategy) Type() string {
 	return "rolling"
 }
 
-func (s *RollingStrategy) Plan(
-	ctx context.Context, cli scheduler.Client, svc *api.Service, spec api.ServiceSpec,
-) (Plan, error) {
-	if s.State == nil {
-		state, err := scheduler.InspectClusterState(ctx, cli)
-		if err != nil {
-			return Plan{}, fmt.Errorf("inspect cluster state: %w", err)
-		}
-		s.State = state
+func (s *RollingStrategy) Plan(state *scheduler.ClusterState, svc *api.Service, spec api.ServiceSpec) (Plan, error) {
+	if state == nil {
+		return Plan{}, fmt.Errorf("cluster state must be provided")
 	}
+	s.state = state
 
 	// We can assume that the spec is valid at this point because it has been validated by the deployment.
 	switch spec.Mode {
@@ -65,7 +64,7 @@ func (s *RollingStrategy) planReplicated(svc *api.Service, spec api.ServiceSpec)
 		return plan, err
 	}
 
-	sched := scheduler.NewServiceScheduler(s.State, spec)
+	sched := scheduler.NewServiceScheduler(s.state, spec)
 	// TODO: return a detailed report on required constraints and which ones are satisfied?
 	availableMachines, err := sched.EligibleMachines()
 	if err != nil {
@@ -221,7 +220,7 @@ func (s *RollingStrategy) planGlobal(svc *api.Service, spec api.ServiceSpec) (Pl
 		}
 	}
 
-	sched := scheduler.NewServiceScheduler(s.State, spec)
+	sched := scheduler.NewServiceScheduler(s.state, spec)
 	availableMachines, err := sched.EligibleMachines()
 	if err != nil {
 		return plan, err