explicit failure

Signed-off-by: Victor Vieux <vieux@docker.com>

Author: vieux

api/handlers.go

@@ -438,8 +438,14 @@ func postContainersCreate(c *context, w http.ResponseWriter, r *http.Request) {
 		authConfig = &dockerclient.AuthConfig{}
 		json.Unmarshal(buf, authConfig)
 	}
+	containerConfig := cluster.BuildContainerConfig(config)
 
-	container, err := c.cluster.CreateContainer(cluster.BuildContainerConfig(config), name, authConfig)
+	if err := containerConfig.Validate(); err != nil {
+		httpError(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	container, err := c.cluster.CreateContainer(containerConfig, name, authConfig)
 	if err != nil {
 		if strings.HasPrefix(err.Error(), "Conflict") {
 			httpError(w, err.Error(), http.StatusConflict)

cluster/config.go

@@ -2,6 +2,8 @@ package cluster
 
 import (
 	"encoding/json"
+	"errors"
+	"fmt"
 	"strings"
 
 	"github.com/samalba/dockerclient"
@@ -63,10 +65,10 @@ func consolidateResourceFields(c *dockerclient.ContainerConfig) {
 // BuildContainerConfig creates a cluster.ContainerConfig from a dockerclient.ContainerConfig
 func BuildContainerConfig(c dockerclient.ContainerConfig) *ContainerConfig {
 	var (
-		affinities       []string
-		constraints      []string
-		reschedulePolicy string
-		env              []string
+		affinities         []string
+		constraints        []string
+		reschedulePolicies []string
+		env                []string
 	)
 
 	// only for tests
@@ -84,9 +86,9 @@ func BuildContainerConfig(c dockerclient.ContainerConfig) *ContainerConfig {
 		json.Unmarshal([]byte(labels), &constraints)
 	}
 
-	// parse reschedule policy from labels (ex. docker run --label 'com.docker.swarm.reschedule-policy=on-node-failure')
-	if label, ok := c.Labels[SwarmLabelNamespace+".reschedule-policy"]; ok {
-		reschedulePolicy = label
+	// parse reschedule policy from labels (ex. docker run --label 'com.docker.swarm.reschedule-policies=on-node-failure')
+	if labels, ok := c.Labels[SwarmLabelNamespace+".reschedule-policies"]; ok {
+		json.Unmarshal([]byte(labels), &reschedulePolicies)
 	}
 
 	// parse affinities/constraints/reschedule policies from env (ex. docker run -e affinity:container==redis -e affinity:image==nginx -e constraint:region==us-east -e constraint:storage==ssd -e reschedule:off)
@@ -96,12 +98,7 @@ func BuildContainerConfig(c dockerclient.ContainerConfig) *ContainerConfig {
 		} else if ok && key == "constraint" {
 			constraints = append(constraints, value)
 		} else if ok && key == "reschedule" {
-			if reschedulePolicy == "" {
-				reschedulePolicy = value
-			} else if reschedulePolicy != value {
-				// for off on conflict
-				reschedulePolicy = "off"
-			}
+			reschedulePolicies = append(reschedulePolicies, value)
 		} else {
 			env = append(env, e)
 		}
@@ -124,9 +121,11 @@ func BuildContainerConfig(c dockerclient.ContainerConfig) *ContainerConfig {
 		}
 	}
 
-	// store reschedule policy in labels
-	if reschedulePolicy != "" {
-		c.Labels[SwarmLabelNamespace+".reschedule-policy"] = reschedulePolicy
+	// store reschedule policies in labels
+	if len(reschedulePolicies) > 0 {
+		if labels, err := json.Marshal(reschedulePolicies); err == nil {
+			c.Labels[SwarmLabelNamespace+".reschedule-policies"] = string(labels)
+		}
 	}
 
 	consolidateResourceFields(&c)
@@ -205,11 +204,32 @@ func (c *ContainerConfig) HaveNodeConstraint() bool {
 	return false
 }
 
-// ReschedulePolicy returns the current rescheduling policy
-func (c *ContainerConfig) ReschedulePolicy() string {
-	policy, ok := c.Labels[SwarmLabelNamespace+".reschedule-policy"]
-	if !ok {
-		return "off"
+// HasReschedulePolicy returns true if the specified policy is part of the config
+func (c *ContainerConfig) HasReschedulePolicy(p string) bool {
+	for _, reschedulePolicy := range c.extractExprs("reschedule-policies") {
+		if reschedulePolicy == p {
+			return true
+		}
+	}
+	return false
+}
+
+// Validate returns an error if the config isn't valid
+func (c *ContainerConfig) Validate() error {
+	//TODO: add validation for affinities and constraints
+	reschedulePolicies := c.extractExprs("reschedule-policies")
+	if len(reschedulePolicies) > 1 {
+		return errors.New("too many reschedule policies")
+	} else if len(reschedulePolicies) == 1 {
+		valid := false
+		for _, validReschedulePolicy := range []string{"off", "on-node-failure"} {
+			if reschedulePolicies[0] == validReschedulePolicy {
+				valid = true
+			}
+		}
+		if !valid {
+			return fmt.Errorf("invalid reschedule policy: %s", reschedulePolicies[0])
+		}
 	}
-	return policy
+	return nil
 }

cluster/watchdog.go

@@ -74,8 +74,8 @@ func (w *Watchdog) rescheduleContainers(e *Engine) {
 
 func (w *Watchdog) rescheduleContainer(c *Container) {
 	// Skip containers which don't have an "on-node-failure" reschedule policy.
-	if c.Config.ReschedulePolicy() != "on-node-failure" {
-		log.Debugf("Skipping rescheduling of %s based on rescheduling policy", c.Id)
+	if !c.Config.HasReschedulePolicy("on-node-failure") {
+		log.Debugf("Skipping rescheduling of %s based on rescheduling policies", c.Id)
 		return
 	}
 

docs/scheduler/rescheduling.md

@@ -30,14 +30,14 @@ apply to the container.
 # do not reschedule (default)
 $ docker run -d -e reschedule:off redis
 # or
-$ docker run -d -l com.docker.swarm.reschedule-policy=off redis
+$ docker run -d -l 'com.docker.swarm.reschedule-policy=["off"]' redis
 ```
 
 ```
 # reschedule on node failure
 $ docker run -d -e reschedule:on-node-failure redis
 # or
-$ docker run -d -l com.docker.swarm.reschedule-policy=on-node-failure redis
+$ docker run -d -l 'com.docker.swarm.reschedule-policy=["on-node-failure"]' redis
 ```
 
 - [Docker Swarm overview](../index.md)

test/integration/rescheduling.bats

@@ -12,13 +12,13 @@ function teardown() {
 	swarm_manage --engine-refresh-min-interval=1s --engine-refresh-max-interval=1s --engine-failure-retry=1 ${HOSTS[0]},${HOSTS[1]}
 
 	# c1 on node-0 with reschedule=on-node-failure
-	run docker_swarm run -dit --name c1 -e constraint:node==~node-0 --label com.docker.swarm.reschedule-policy=on-node-failure busybox sh
+	run docker_swarm run -dit --name c1 -e constraint:node==~node-0 --label 'com.docker.swarm.reschedule-policies=["on-node-failure"]' busybox sh
 	[ "$status" -eq 0 ]
 	# c2 on node-0 with reschedule=off
-	run docker_swarm run -dit --name c2 -e constraint:node==~node-0 --label com.docker.swarm.reschedule-policy=off busybox sh
+	run docker_swarm run -dit --name c2 -e constraint:node==~node-0 --label 'com.docker.swarm.reschedule-policies=["off"]' busybox sh
 	[ "$status" -eq 0 ]
 	# c3 on node-1
-	run docker_swarm run -dit --name c3 -e constraint:node==~node-1 --label com.docker.swarm.reschedule-policy=on-node-failure busybox sh
+	run docker_swarm run -dit --name c3 -e constraint:node==~node-1 --label 'com.docker.swarm.reschedule-policies=["on-node-failure"]' busybox sh
 	[ "$status" -eq 0 ]
 
 	run docker_swarm ps -q
@@ -119,11 +119,13 @@ function teardown() {
 	start_docker_with_busybox 2
 	swarm_manage
 
-	docker_swarm run --name c1 -dit -e reschedule:on-node-failure --label com.docker.swarm.reschedule-policy=off busybox sh
-	[[ $(docker_swarm inspect -f '{{ index .Config.Labels "com.docker.swarm.reschedule-policy" }}' c1) == "off" ]]
+	run docker_swarm run --name c1 -dit --label 'com.docker.swarm.reschedule-policies=["false"]' busybox sh
+	[ "$status" -ne 0 ]
+	[[ "${output}" == *'invalid reschedule policy: false'* ]]
 
-	docker_swarm run --name c2 -dit -e reschedule:off --label com.docker.swarm.reschedule-policy=on-node-failure busybox sh
-	[[ $(docker_swarm inspect -f '{{ index .Config.Labels "com.docker.swarm.reschedule-policy" }}' c2) == "off" ]]
+	run docker_swarm run --name c2 -dit -e reschedule:off --label 'com.docker.swarm.reschedule-policies=["on-node-failure"]' -e reschedule:off busybox sh
+	[ "$status" -ne 0 ]
+	[[ "${output}" == *'too many reschedule policies'* ]]
 }
 
 @test "rescheduling pending" {
@@ -185,10 +187,10 @@ function teardown() {
 	swarm_manage --engine-refresh-min-interval=1s --engine-refresh-max-interval=1s --engine-failure-retry=1 ${HOSTS[0]},${HOSTS[1]}
 
 	# c1 on node-0 with reschedule=on-node-failure
-	run docker_swarm run -dit --name c1 -e constraint:node==~node-0 --label com.docker.swarm.reschedule-policy=on-node-failure busybox sh
+	run docker_swarm run -dit --name c1 -e constraint:node==~node-0 --label com.docker.swarm.reschedule-policies=["on-node-failure"] busybox sh
 	[ "$status" -eq 0 ]
 	# c2 on node-1
-	run docker_swarm run -dit --name c2 -e constraint:node==~node-1 --label com.docker.swarm.reschedule-policy=on-node-failure busybox sh
+	run docker_swarm run -dit --name c2 -e constraint:node==~node-1 --label com.docker.swarm.reschedule-policies=["on-node-failure"] busybox sh
 	[ "$status" -eq 0 ]
 
 	run docker_swarm ps -q