add command to check deployment state

deployers/commons.go

@@ -29,8 +29,22 @@ const (
 	Swarm DeployerType = "swarm"
 
 	GracefulTimeout = 10 * time.Second
+
+	DefaultStateTimeout = 30 * time.Second
 )
 
+type checkStateOption struct {
+	timeout *time.Duration
+}
+
+type fnStateOption func(c *checkStateOption)
+
+func WithTimeout(duration time.Duration) fnStateOption {
+	return func(c *checkStateOption) {
+		c.timeout = &duration
+	}
+}
+
 // Base struct of the deployers.
 // It handles the main informations to build a deployer.
 //

deployers/swarm.go

@@ -6,6 +6,8 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"sync"
+	"time"
 
 	"gitea.thegux.fr/hmdeploy/connection"
 	"gitea.thegux.fr/hmdeploy/docker"
@@ -14,13 +16,16 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
+const stateTickDuration = 4 * time.Second
+
 var ErrSwarmDeployerNoArchive = errors.New("no archive found to be deployed")
 
 // SwarmDeployer handles the deployment of a Docker service on the swarm instance.
 type SwarmDeployer struct {
 	*deployer
 	conn        connection.IConnection
-	dcli        docker.IClient
+	dloc        docker.IClient
+	drem        *docker.RemoteClient
 	archivePath string
 }
 
@@ -30,7 +35,8 @@ func NewSwarmDeployer(
 	ctx context.Context,
 	project *models.Project,
 	netInfo *models.HMNetInfo,
-	dockerClient docker.IClient,
+	dloc docker.IClient,
+	drem *docker.RemoteClient,
 ) (SwarmDeployer, error) {
 	var sd SwarmDeployer
 
@@ -45,7 +51,8 @@ func NewSwarmDeployer(
 	}
 
 	sd.conn = &conn
-	sd.dcli = dockerClient
+	sd.dloc = dloc
+	sd.drem = drem
 	sd.deployer = newDeployer(ctx, Swarm, project)
 
 	return sd, nil
@@ -100,7 +107,7 @@ func (sd *SwarmDeployer) Build() error {
 
 	filesToArchive := []string{}
 	for idx := range sd.project.ImageNames {
-		tarFile, err := sd.dcli.Save(sd.project.ImageNames[idx], sd.project.Dir)
+		tarFile, err := sd.dloc.Save(sd.project.ImageNames[idx], sd.project.Dir)
 		if err != nil {
 			sd.setDone(err)
 			return err
@@ -189,12 +196,85 @@ func (sd *SwarmDeployer) Deploy() error {
 		return err
 	}
 
+	if err := sd.checkState(docker.Running); err != nil {
+		sd.setDone(err)
+		return err
+	}
+
 	log.Info().Msg("swarm deployment done with success")
 
 	sd.setDone(nil)
 	return nil
 }
 
+// checkState checks the state of the deployment.
+// It loops over all the services deployed for the project (replicas included) and
+// checks if the `target` state match the services states.
+//
+// There's a timeout (default: 30s) that you can set with the options: `WithTimeout`.
+func (sd *SwarmDeployer) checkState(target docker.ServiceStatus, options ...fnStateOption) error {
+	var opts checkStateOption
+	for _, opt := range options {
+		opt(&opts)
+	}
+
+	var checkErr error
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+
+		timeoutDuration := DefaultStateTimeout
+		if opts.timeout != nil {
+			timeoutDuration = *opts.timeout
+		}
+
+		ticker := time.NewTicker(stateTickDuration)
+		ctx, fnCancel := context.WithDeadline(sd.ctx, time.Now().UTC().Add(timeoutDuration))
+		defer fnCancel()
+
+		for {
+			select {
+			case <-ticker.C:
+				log.Info().
+					Str("project", sd.project.Name).
+					Str("state", string(target)).
+					Msg("checking project state...")
+				srvs, err := sd.drem.ExtractServicesDetails(docker.WithName(sd.project.Name))
+				if err != nil {
+					checkErr = err
+					return
+				}
+
+				ready := true
+			mainloop:
+				for idx := range srvs {
+					for idy := range srvs[idx].Replicas {
+						if srvs[idx].Replicas[idy].State != docker.ServiceStatus(target) {
+							log.Info().Dur("retry (ms)", stateTickDuration).Msg("project not in good state yet, retrying...")
+							ready = false
+							break mainloop
+						}
+					}
+				}
+				if ready {
+					return
+				}
+			case <-ctx.Done():
+				msg := "swarm deployment skipped"
+				if errors.Is(ctx.Err(), context.DeadlineExceeded) {
+					msg = "swarm check state timeout"
+				}
+				checkErr = fmt.Errorf("%w, %s", ErrContextDone, msg)
+				return
+			}
+		}
+	}()
+
+	wg.Wait()
+	return checkErr
+}
+
 func (sd *SwarmDeployer) Destroy() error {
 	sd.processing.Store(true)
 	defer sd.processing.Store(false)
@@ -205,6 +285,11 @@ func (sd *SwarmDeployer) Destroy() error {
 		return err
 	}
 
+	if err := sd.checkState(docker.Shutdown); err != nil {
+		sd.setDone(err)
+		return err
+	}
+
 	log.Info().Msg("swarm undeployment done with success")
 
 	sd.setDone(nil)

docker/client.go

@@ -105,8 +105,25 @@ func NewRemoteClient(netInfo *models.HMNetInfo) (RemoteClient, error) {
 	return rc, nil
 }
 
-func (c *RemoteClient) getIDS() ([]string, error) {
-	output, err := c.conn.Execute("docker service ls -q")
+type extractOption struct {
+	filter string
+}
+
+type fnExtractOption func(*extractOption)
+
+func WithName(name string) fnExtractOption {
+	return func(o *extractOption) {
+		o.filter = name
+	}
+}
+
+func (c *RemoteClient) getIDS(name string) ([]string, error) {
+	cmd := "docker service ls -q"
+	if name != "" {
+		cmd += " --filter name=" + name
+	}
+
+	output, err := c.conn.Execute(cmd)
 	if err != nil {
 		return nil, err
 	}
@@ -134,12 +151,7 @@ func (c *RemoteClient) getServiceDetails(id string) (Service, error) {
 	return sc, nil
 }
 
-func (c *RemoteClient) ExtractServicesDetails() (Services, error) {
-	ids, err := c.getIDS()
-	if err != nil {
-		return nil, err
-	}
-
+func (c *RemoteClient) extractServicesDetails(ids ...string) (Services, error) {
 	services := Services{}
 	for _, id := range ids {
 		srv, err := c.getServiceDetails(id)
@@ -151,3 +163,17 @@ func (c *RemoteClient) ExtractServicesDetails() (Services, error) {
 
 	return services, nil
 }
+
+func (c *RemoteClient) ExtractServicesDetails(options ...fnExtractOption) (Services, error) {
+	var opts extractOption
+	for _, opt := range options {
+		opt(&opts)
+	}
+
+	ids, err := c.getIDS(opts.filter)
+	if err != nil {
+		return nil, err
+	}
+
+	return c.extractServicesDetails(ids...)
+}

docker/models.go

@@ -6,6 +6,8 @@ import (
 	"strconv"
 	"strings"
 	"time"
+
+	"github.com/rs/zerolog/log"
 )
 
 const nbImageParts = 2
@@ -170,7 +172,7 @@ func (s *Service) UnmarshalJSON(data []byte) error {
 
 	nbReplicas := ci.Details[0].Spec.Mode.Replicated.Replicas
 	if len(ci.States) < nbReplicas {
-		return fmt.Errorf("must have %d replicas but have %d", nbReplicas, len(ci.States))
+		log.Warn().Msg(fmt.Sprintf("must have %d replicas but have %d", nbReplicas, len(ci.States)))
 	}
 
 	networks := []string{}

main.go

@@ -208,8 +208,13 @@ func initDeployers(
 		return deps, fmt.Errorf("%w, swarm net info does not exist", ErrNetInfoNotFound)
 	}
 
-	dcli := docker.NewLocalClient()
-	sd, err := deployers.NewSwarmDeployer(ctx, project, swarmNet, &dcli)
+	dloc := docker.NewLocalClient()
+	drem, err := docker.NewRemoteClient(swarmNet)
+	if err != nil {
+		return deps, err
+	}
+
+	sd, err := deployers.NewSwarmDeployer(ctx, project, swarmNet, &dloc, &drem)
 	if err != nil {
 		return deps, fmt.Errorf("%w, unable to init swarm deployer, err=%v", ErrDeployerInit, err)
 	}