constraintenforcer: Trigger task restarts when appropriate

The constraint enforcer currently sets task desired state to "shutdown"
directly, which means the orchestrator will consider these tasks already
processed, and won't trigger restarts. While this is appropriate for
global services, replicated services should keep the desired number of
replicas running, so each task shut down by the constraint enforcer
should be restarted somewhere else.

This changes the constraint enforcer to trigger a task shutdown by
updating the actual state rather than desired state. This will cause the
orchestrator to restart the task when necessary. It's not a perfect
solution, because it bends rules about field ownership, and may cause
the replacement task to start before the old one stops. However, it's a
good compromise solution for this problem that doesn't require absorbing
the constraint enforcer into each orchestrator (which wouldn't fit the
model well), or adding a third type of state to every task.

Also, update the global orchestrator to only restart a task when the
node still meets the constraints.

Signed-off-by: Aaron Lehmann <aaron.lehmann@docker.com>

Author: Aaron Lehmann

manager/orchestrator/constraintenforcer/constraint_enforcer.go

@@ -1,11 +1,14 @@
 package constraintenforcer
 
 import (
+	"time"
+
 	"github.com/docker/swarmkit/api"
 	"github.com/docker/swarmkit/log"
 	"github.com/docker/swarmkit/manager/constraint"
 	"github.com/docker/swarmkit/manager/state"
 	"github.com/docker/swarmkit/manager/state/store"
+	"github.com/docker/swarmkit/protobuf/ptypes"
 )
 
 // ConstraintEnforcer watches for updates to nodes and shuts down tasks that no
@@ -134,7 +137,16 @@ func (ce *ConstraintEnforcer) shutdownNoncompliantTasks(node *api.Node) {
 						return nil
 					}
 
-					t.DesiredState = api.TaskStateShutdown
+					// We set the observed state to
+					// REJECTED, rather than the desired
+					// state. Desired state is owned by the
+					// orchestrator, and setting it directly
+					// will bypass actions such as
+					// restarting the task on another node
+					// (if applicable).
+					t.Status.State = api.TaskStateRejected
+					t.Status.Message = "assigned node no longer meets constraints"
+					t.Status.Timestamp = ptypes.MustTimestampProto(time.Now())
 					return store.UpdateTask(tx, t)
 				})
 				if err != nil {

manager/orchestrator/constraintenforcer/constraint_enforcer_test.go

@@ -131,9 +131,10 @@ func TestConstraintEnforcer(t *testing.T) {
 
 	go constraintEnforcer.Run()
 
-	// id0 should be killed immediately
-	shutdown1 := testutils.WatchShutdownTask(t, watch)
+	// id0 should be rejected immediately
+	shutdown1 := testutils.WatchTaskUpdate(t, watch)
 	assert.Equal(t, "id0", shutdown1.ID)
+	assert.Equal(t, api.TaskStateRejected, shutdown1.Status.State)
 
 	// Change node id1 to a manager
 	err = s.Update(func(tx store.Tx) error {
@@ -147,8 +148,9 @@ func TestConstraintEnforcer(t *testing.T) {
 	})
 	assert.NoError(t, err)
 
-	shutdown2 := testutils.WatchShutdownTask(t, watch)
+	shutdown2 := testutils.WatchTaskUpdate(t, watch)
 	assert.Equal(t, "id2", shutdown2.ID)
+	assert.Equal(t, api.TaskStateRejected, shutdown2.Status.State)
 
 	// Change resources on node id2
 	err = s.Update(func(tx store.Tx) error {
@@ -162,6 +164,7 @@ func TestConstraintEnforcer(t *testing.T) {
 	})
 	assert.NoError(t, err)
 
-	shutdown3 := testutils.WatchShutdownTask(t, watch)
+	shutdown3 := testutils.WatchTaskUpdate(t, watch)
 	assert.Equal(t, "id4", shutdown3.ID)
+	assert.Equal(t, api.TaskStateRejected, shutdown3.Status.State)
 }

manager/orchestrator/global/global.go

@@ -513,10 +513,22 @@ func (g *Orchestrator) tickTasks(ctx context.Context) {
 				if t == nil || t.DesiredState > api.TaskStateRunning {
 					return nil
 				}
+
 				service := store.GetService(tx, t.ServiceID)
 				if service == nil {
 					return nil
 				}
+
+				node, nodeExists := g.nodes[t.NodeID]
+				serviceEntry, serviceExists := g.globalServices[t.ServiceID]
+				if !nodeExists || !serviceExists {
+					return nil
+				}
+				if !constraint.NodeMatches(serviceEntry.constraints, node) {
+					t.DesiredState = api.TaskStateShutdown
+					return store.UpdateTask(tx, t)
+				}
+
 				return g.restarts.Restart(ctx, tx, g.cluster, service, *t)
 			})
 			if err != nil {