feat(pre-deploy): implement pre-deploy hook operations for deployment

pkg/client/deploy/operation/predeploy.go

@@ -0,0 +1,172 @@
+package operation
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/containerd/errdefs"
+	"github.com/docker/compose/v2/pkg/progress"
+	"github.com/docker/docker/api/types/container"
+	cliprogress "github.com/psviderski/uncloud/internal/cli/progress"
+	"github.com/psviderski/uncloud/internal/cli/tui"
+	"github.com/psviderski/uncloud/pkg/api"
+)
+
+// DefaultPreDeployTimeout is the maximum duration to wait for a pre-deploy hook container to complete.
+const DefaultPreDeployTimeout = 5 * time.Minute
+
+// StopPreDeployOperation stops a running pre-deploy hook container from a previous deployment.
+type StopPreDeployOperation struct {
+	MachineID string
+	// MachineName is used for formatting the operation as part of the deployment plan.
+	MachineName string
+	Container   api.ServiceContainer
+}
+
+func (o *StopPreDeployOperation) Execute(ctx context.Context, cli Client) error {
+	if err := cli.StopContainer(ctx, o.Container.ServiceID(), o.Container.ID, container.StopOptions{}); err != nil {
+		if !errdefs.IsNotFound(err) {
+			return fmt.Errorf("stop pre-deploy hook container '%s': %w", o.Container.ID, err)
+		}
+	}
+	return nil
+}
+
+func (o *StopPreDeployOperation) Format() string {
+	displayName := o.Container.ServiceSpec.Name + tui.Faint.Render("/") + o.Container.ShortID()
+	status, _ := o.Container.HumanState()
+
+	return tui.BoldRed.Render("⏹") + "   " +
+		tui.Faint.Render("stop pre-deploy hook") + " " +
+		displayName + " " +
+		tui.Faint.Render("("+status+")") + " " +
+		tui.Faint.Render("on") + " " +
+		o.MachineName
+}
+
+func (o *StopPreDeployOperation) String() string {
+	return fmt.Sprintf("StopPreDeployOperation[machine_id=%s container_id=%s]",
+		o.MachineID, o.Container.ID)
+}
+
+// RunPreDeployOperation runs a one-shot pre-deploy hook container before service deployment.
+type RunPreDeployOperation struct {
+	ServiceID string
+	Spec      api.ServiceSpec
+	MachineID string
+	// MachineName is used for formatting the operation as part of the deployment plan.
+	MachineName string
+	// OldContainerIDs are pre-deploy hook containers from previous deployments to remove.
+	OldContainerIDs []string
+}
+
+func (o *RunPreDeployOperation) Execute(ctx context.Context, cli Client) error {
+	// Remove old pre-deploy containers.
+	for _, id := range o.OldContainerIDs {
+		_ = cli.StopContainer(ctx, o.ServiceID, id, container.StopOptions{})
+		err := cli.RemoveContainer(ctx, o.ServiceID, id, container.RemoveOptions{RemoveVolumes: true})
+		if err != nil && !errdefs.IsNotFound(err) {
+			return fmt.Errorf("remove old pre-deploy hook container '%s': %w", id, err)
+		}
+	}
+
+	resp, err := cli.CreatePreDeployHookContainer(ctx, o.ServiceID, o.Spec, o.MachineID)
+	if err != nil {
+		return fmt.Errorf("create pre-deploy hook container: %w", err)
+	}
+	if err = cli.StartContainer(ctx, o.ServiceID, resp.ID); err != nil {
+		return fmt.Errorf("start pre-deploy hook container: %w", err)
+	}
+
+	timeout := DefaultPreDeployTimeout
+	if o.Spec.PreDeploy.Timeout != nil {
+		timeout = *o.Spec.PreDeploy.Timeout
+	}
+
+	return o.waitForExit(ctx, cli, resp.ID, timeout)
+}
+
+// waitForExit polls the container state until it exits or the timeout is reached.
+func (o *RunPreDeployOperation) waitForExit(
+	ctx context.Context, cli Client, containerID string, timeout time.Duration,
+) error {
+	pw := progress.ContextWriter(ctx)
+	eventID := cliprogress.PreDeployHookEventID(o.Spec.Name, o.MachineName)
+	pw.Event(progress.Waiting(eventID))
+
+	timeoutCtx, cancel := context.WithTimeout(ctx, timeout)
+	defer cancel()
+
+	ticker := time.NewTicker(time.Second)
+	defer ticker.Stop()
+
+	var ctr *api.ServiceContainer
+	for {
+		select {
+		case <-timeoutCtx.Done():
+			// Stop the container on timeout or context cancellation.
+			pw.Event(progress.StoppingEvent(eventID))
+			_ = cli.StopContainer(ctx, o.ServiceID, containerID, container.StopOptions{})
+
+			ctrID := containerID
+			if ctr != nil {
+				ctrID = fmt.Sprintf("%s/%s", o.Spec.Name, ctr.ShortID())
+			}
+			if ctx.Err() != nil {
+				// The parent context has been cancelled before the timeout.
+				pw.Event(progress.NewEvent(eventID, progress.Error, "Cancelled"))
+				return fmt.Errorf("pre-deploy hook container '%s': %w", ctrID, ctx.Err())
+			}
+			pw.Event(progress.NewEvent(eventID, progress.Error, fmt.Sprintf("Timeout (%s)", timeout)))
+			return fmt.Errorf("pre-deploy hook container '%s' timed out after %s. "+
+				"It's stopped and available for inspection. Fetch logs with 'uc logs %s'",
+				ctrID, timeout, o.Spec.Name)
+
+		case <-ticker.C:
+			mc, err := cli.InspectContainer(ctx, o.ServiceID, containerID)
+			if err != nil {
+				return fmt.Errorf("inspect pre-deploy hook container: %w", err)
+			}
+			ctr = &mc.Container
+
+			if ctr.State.Running {
+				if state, err := ctr.Container.HumanState(); err == nil {
+					pw.Event(progress.NewEvent(eventID, progress.Working, fmt.Sprintf("Waiting (%s)", state)))
+				}
+				continue
+			}
+
+			// Hook container has exited successfully.
+			if ctr.State.ExitCode == 0 {
+				pw.Event(progress.Event{
+					ID:     eventID,
+					Status: progress.Done,
+				})
+				return nil
+			}
+
+			pw.Event(progress.ErrorEvent(eventID))
+			ctrID := fmt.Sprintf("%s/%s", o.Spec.Name, ctr.ShortID())
+			return fmt.Errorf("pre-deploy hook container '%s' failed with exit code: %d. "+
+				"It's stopped and available for inspection. Fetch logs with 'uc logs %s'",
+				ctrID, ctr.State.ExitCode, o.Spec.Name)
+		}
+	}
+}
+
+func (o *RunPreDeployOperation) Format() string {
+	cmd := strings.Join(o.Spec.PreDeploy.Command, " ")
+
+	return tui.BoldGreen.Render("▶") + "   " +
+		tui.Faint.Render("run pre-deploy hook") + " " +
+		o.Spec.Name + " (" + cmd + ") " +
+		tui.Faint.Render("on") + " " +
+		o.MachineName
+}
+
+func (o *RunPreDeployOperation) String() string {
+	return fmt.Sprintf("RunPreDeployOperation[machine_id=%s service_id=%s cmd=%v]",
+		o.MachineID, o.ServiceID, o.Spec.PreDeploy.Command)
+}

pkg/client/deploy/scheduler/state.go

@@ -67,3 +67,11 @@ func (s *ClusterState) Machine(nameOrID string) (*Machine, bool) {
 	}
 	return nil, false
 }
+
+// MachineName returns the machine name by ID from the cluster state. If the id is not found, ("", false) is returned.
+func (s *ClusterState) MachineName(id string) (string, bool) {
+	if m, ok := s.Machine(id); ok {
+		return m.Info.Name, true
+	}
+	return "", false
+}

pkg/client/deploy/strategy.go

@@ -199,6 +199,10 @@ func (s *RollingStrategy) planReplicated(svc *api.Service, spec api.ServiceSpec)
 		}
 	}
 
+	if ops := s.preDeployOperations(svc, plan); len(ops) > 0 {
+		plan.Operations = append(ops, plan.Operations...)
+	}
+
 	return plan, nil
 }
 
@@ -265,6 +269,10 @@ func (s *RollingStrategy) planGlobal(svc *api.Service, spec api.ServiceSpec) (Se
 		}
 	}
 
+	if ops := s.preDeployOperations(svc, plan); len(ops) > 0 {
+		plan.Operations = append(ops, plan.Operations...)
+	}
+
 	return plan, nil
 }
 
@@ -324,6 +332,7 @@ func reconcileGlobalContainer(
 			break
 		}
 		// TODO: handle ContainerNeedsUpdate when update of mutable fields on a container is supported.
+		//  Make sure to update preDeployOperation accordingly.
 	}
 	if upToDate {
 		return ops, nil
@@ -435,6 +444,64 @@ func determineUpdateOrder(oldContainer api.ServiceContainer, spec api.ServiceSpe
 	return api.UpdateOrderStartFirst
 }
 
+// preDeployOperations returns operations for the pre-deploy hook if the spec has one and the plan updates the service.
+// It prepends StopPreDeployOperations for any running hook containers, followed by a RunPreDeployOperation on the same
+// machine as the first run/replace operation.
+func (s *RollingStrategy) preDeployOperations(svc *api.Service, plan ServicePlan) []operation.Operation {
+	if plan.Spec.PreDeploy == nil {
+		return nil
+	}
+
+	// Find the first run or replace operation to determine the target machine.
+	var machineID, machineName string
+	for _, op := range plan.Operations {
+		switch o := op.(type) {
+		case *operation.RunContainerOperation:
+			machineID = o.MachineID
+			machineName = o.MachineName
+		case *operation.ReplaceContainerOperation:
+			machineID = o.MachineID
+			machineName = o.MachineName
+		default:
+			continue
+		}
+		break
+	}
+
+	// Skip the hook as there are no run or replace operations.
+	if machineID == "" {
+		return nil
+	}
+
+	var ops []operation.Operation
+
+	// Collect old pre-deploy container IDs to clean up and stop any that are still running.
+	var oldContainerIDs []string
+	if svc != nil {
+		for _, c := range svc.HookContainers {
+			oldContainerIDs = append(oldContainerIDs, c.Container.ID)
+			if c.Container.State.Running {
+				hookMachineName, _ := s.state.MachineName(c.MachineID)
+				ops = append(ops, &operation.StopPreDeployOperation{
+					MachineID:   c.MachineID,
+					MachineName: hookMachineName,
+					Container:   c.Container,
+				})
+			}
+		}
+	}
+
+	ops = append(ops, &operation.RunPreDeployOperation{
+		ServiceID:       plan.ServiceID,
+		Spec:            plan.Spec,
+		MachineID:       machineID,
+		MachineName:     machineName,
+		OldContainerIDs: oldContainerIDs,
+	})
+
+	return ops
+}
+
 // newEmptyServicePlan creates a new empty plan for a service deployment with initialised service ID and name.
 func newEmptyServicePlan(svc *api.Service, spec api.ServiceSpec) (ServicePlan, error) {
 	plan := ServicePlan{

pkg/client/deploy/strategy_test.go

@@ -9,6 +9,7 @@ import (
 	"github.com/psviderski/uncloud/internal/machine/api/pb"
 	"github.com/psviderski/uncloud/pkg/api"
 	"github.com/psviderski/uncloud/pkg/client/deploy/operation"
+	"github.com/psviderski/uncloud/pkg/client/deploy/scheduler"
 	"github.com/stretchr/testify/assert"
 )
 
@@ -423,6 +424,246 @@ func TestReconcileGlobalContainer(t *testing.T) {
 	}
 }
 
+func TestPreDeployOperations(t *testing.T) {
+	hook := &api.PreDeployHook{
+		Command: []string{"db", "migrate"},
+	}
+
+	strategy := &RollingStrategy{
+		state: &scheduler.ClusterState{
+			Machines: []*scheduler.Machine{
+				{Info: &pb.MachineInfo{Id: "m-1", Name: "machine-1"}},
+				{Info: &pb.MachineInfo{Id: "m-2", Name: "machine-2"}},
+				{Info: &pb.MachineInfo{Id: "m-3", Name: "machine-3"}},
+			},
+		},
+	}
+
+	runningHook1 := newServiceContainer("running-hook-1", container.State{Running: true, Status: "running"})
+	runningHook2 := newServiceContainer("running-hook-2", container.State{Running: true, Status: "running"})
+
+	tests := []struct {
+		name     string
+		plan     ServicePlan
+		svc      *api.Service
+		expected []operation.Operation
+	}{
+		{
+			name: "no pre-deploy hook in spec",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.RunContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			expected: nil,
+		},
+		{
+			name: "plan has RunContainerOperation",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.RunContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			expected: []operation.Operation{
+				&operation.RunPreDeployOperation{
+					ServiceID:   "svc-1",
+					MachineID:   "m-1",
+					MachineName: "machine-1",
+				},
+			},
+		},
+		{
+			name: "plan has ReplaceContainerOperation",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.ReplaceContainerOperation{MachineID: "m-2", MachineName: "machine-2"},
+					},
+				},
+			},
+			expected: []operation.Operation{
+				&operation.RunPreDeployOperation{
+					ServiceID:   "svc-1",
+					MachineID:   "m-2",
+					MachineName: "machine-2",
+				},
+			},
+		},
+		{
+			name: "plan has only RemoveContainerOperation",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.RemoveContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			expected: nil,
+		},
+		{
+			name: "plan has only StopContainerOperation",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.StopContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			expected: nil,
+		},
+		{
+			name: "empty plan with no operations",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+			},
+			expected: nil,
+		},
+		{
+			name: "stopped hook containers are collected for cleanup",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.RunContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			svc: &api.Service{
+				HookContainers: []api.MachineServiceContainer{
+					{MachineID: "m-1", Container: newServiceContainer("old-hook-1", container.State{Status: "exited"})},
+				},
+			},
+			expected: []operation.Operation{
+				&operation.RunPreDeployOperation{
+					ServiceID:       "svc-1",
+					MachineID:       "m-1",
+					MachineName:     "machine-1",
+					OldContainerIDs: []string{"old-hook-1"},
+				},
+			},
+		},
+		{
+			name: "running hook containers are stopped before run",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.RunContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+					},
+				},
+			},
+			svc: &api.Service{
+				HookContainers: []api.MachineServiceContainer{
+					{MachineID: "m-2", Container: runningHook1},
+				},
+			},
+			expected: []operation.Operation{
+				&operation.StopPreDeployOperation{
+					MachineID:   "m-2",
+					MachineName: "machine-2",
+					Container:   runningHook1,
+				},
+				&operation.RunPreDeployOperation{
+					ServiceID:       "svc-1",
+					MachineID:       "m-1",
+					MachineName:     "machine-1",
+					OldContainerIDs: []string{"running-hook-1"},
+				},
+			},
+		},
+		{
+			name: "mixed stopped and running hook containers and mixed container operations",
+			plan: ServicePlan{
+				ServiceID:   "svc-1",
+				ServiceName: "app",
+				Spec:        api.ServiceSpec{PreDeploy: hook},
+				SequenceOperation: operation.SequenceOperation{
+					Operations: []operation.Operation{
+						&operation.StopContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+						&operation.RemoveContainerOperation{MachineID: "m-1", MachineName: "machine-1"},
+						&operation.ReplaceContainerOperation{MachineID: "m-2", MachineName: "machine-2"},
+						&operation.RunContainerOperation{MachineID: "m-3", MachineName: "machine-3"},
+						&operation.RemoveContainerOperation{MachineID: "m-3", MachineName: "machine-3"},
+					},
+				},
+			},
+			svc: &api.Service{
+				HookContainers: []api.MachineServiceContainer{
+					{MachineID: "m-1", Container: newServiceContainer("stopped-hook", container.State{Status: "exited"})},
+					{MachineID: "m-1", Container: runningHook1},
+					{MachineID: "m-3", Container: runningHook2},
+				},
+			},
+			expected: []operation.Operation{
+				&operation.StopPreDeployOperation{
+					MachineID:   "m-1",
+					MachineName: "machine-1",
+					Container:   runningHook1,
+				},
+				&operation.StopPreDeployOperation{
+					MachineID:   "m-3",
+					MachineName: "machine-3",
+					Container:   runningHook2,
+				},
+				&operation.RunPreDeployOperation{
+					ServiceID:       "svc-1",
+					MachineID:       "m-2",
+					MachineName:     "machine-2",
+					OldContainerIDs: []string{"stopped-hook", "running-hook-1", "running-hook-2"},
+				},
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := strategy.preDeployOperations(tt.svc, tt.plan)
+			if tt.expected == nil {
+				assert.Nil(t, result)
+				return
+			}
+			assertOperationsEqual(t, tt.expected, result)
+		})
+	}
+}
+
+// newServiceContainer creates an api.ServiceContainer with the given ID and state.
+func newServiceContainer(id string, state container.State) api.ServiceContainer {
+	return api.ServiceContainer{Container: api.Container{
+		InspectResponse: container.InspectResponse{
+			ContainerJSONBase: &container.ContainerJSONBase{
+				ID:    id,
+				State: &state,
+			},
+		},
+	}}
+}
+
 // assertOperationsEqual compares expected and actual operations, ignoring the Spec field
 // which is passed separately to the function and not the focus of these tests.
 func assertOperationsEqual(t *testing.T, expected, actual []operation.Operation) {
@@ -430,6 +671,7 @@ func assertOperationsEqual(t *testing.T, expected, actual []operation.Operation)
 	opts := cmp.Options{
 		cmpopts.IgnoreFields(operation.RunContainerOperation{}, "Spec"),
 		cmpopts.IgnoreFields(operation.ReplaceContainerOperation{}, "Spec"),
+		cmpopts.IgnoreFields(operation.RunPreDeployOperation{}, "Spec"),
 		cmpopts.IgnoreUnexported(api.Container{}),
 	}
 	if diff := cmp.Diff(expected, actual, opts); diff != "" {