terraform: make sure Stop blocks until full completion

terraform/context.go

@@ -93,6 +93,7 @@ type Context struct {
 	parallelSem         Semaphore
 	providerInputConfig map[string]map[string]interface{}
 	runLock             sync.Mutex
+	runCond             *sync.Cond
 	runContext          context.Context
 	runContextCancel    context.CancelFunc
 	shadowErr           error
@@ -648,16 +649,10 @@ func (c *Context) Stop() {
 		c.runContextCancel = nil
 	}
 
-	// Grab the context before we unlock
-	ctx := c.runContext
-
-	// Unlock
+	// Grab the condition var before we exit
+	if cond := c.runCond; cond != nil {
+		cond.Wait()
 		c.l.Unlock()
-
-	// Wait if we have a context
-	if ctx != nil {
-		log.Printf("[WARN] terraform: stop waiting for context completion")
-		<-ctx.Done()
 	}
 
 	log.Printf("[WARN] terraform: stop complete")
@@ -727,23 +722,22 @@ func (c *Context) SetVariable(k string, v interface{}) {
 }
 
 func (c *Context) acquireRun(phase string) func() {
-	// Acquire the runlock first. This is the lock that is held for
-	// the duration of a run to prevent multiple runs.
-	c.runLock.Lock()
-
 	// With the run lock held, grab the context lock to make changes
 	// to the run context.
 	c.l.Lock()
 	defer c.l.Unlock()
 
+	// Wait until we're no longer running
+	for c.runCond != nil {
+		c.runCond.Wait()
+	}
+
+	// Build our lock
+	c.runCond = sync.NewCond(&c.l)
+
 	// Setup debugging
 	dbug.SetPhase(phase)
 
-	// runContext should never be non-nil, check that here
-	if c.runContext != nil {
-		panic("acquireRun called with runContext != nil")
-	}
-
 	// Create a new run context
 	c.runContext, c.runContextCancel = context.WithCancel(context.Background())
 
@@ -772,11 +766,13 @@ func (c *Context) releaseRun() {
 		c.runContextCancel()
 	}
 
+	// Unlock all waiting our condition
+	cond := c.runCond
+	c.runCond = nil
+	cond.Broadcast()
+
 	// Unset the context
 	c.runContext = nil
-
-	// Unlock the run lock
-	c.runLock.Unlock()
 }
 
 func (c *Context) walk(

terraform/context_apply_test.go

@@ -1720,6 +1720,82 @@ func TestContext2Apply_cancel(t *testing.T) {
 	}
 }
 
+func TestContext2Apply_cancelBlock(t *testing.T) {
+	m := testModule(t, "apply-cancel-block")
+	p := testProvider("aws")
+	ctx := testContext2(t, &ContextOpts{
+		Module: m,
+		Providers: map[string]ResourceProviderFactory{
+			"aws": testProviderFuncFixed(p),
+		},
+	})
+
+	applyCh := make(chan struct{})
+	p.DiffFn = testDiffFn
+	p.ApplyFn = func(*InstanceInfo, *InstanceState, *InstanceDiff) (*InstanceState, error) {
+		close(applyCh)
+
+		for !ctx.sh.Stopped() {
+			// Wait for stop to be called
+		}
+
+		// Sleep
+		time.Sleep(100 * time.Millisecond)
+
+		return &InstanceState{
+			ID: "foo",
+		}, nil
+	}
+
+	if _, err := ctx.Plan(); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	// Start the Apply in a goroutine
+	var applyErr error
+	stateCh := make(chan *State)
+	go func() {
+		state, err := ctx.Apply()
+		if err != nil {
+			applyErr = err
+		}
+
+		stateCh <- state
+	}()
+
+	stopDone := make(chan struct{})
+	go func() {
+		defer close(stopDone)
+		<-applyCh
+		ctx.Stop()
+	}()
+
+	// Make sure that stop blocks
+	select {
+	case <-stopDone:
+		t.Fatal("stop should block")
+	case <-time.After(10 * time.Millisecond):
+	}
+
+	// Wait for stop
+	select {
+	case <-stopDone:
+	case <-time.After(500 * time.Millisecond):
+		t.Fatal("stop should be done")
+	}
+
+	// Wait for apply to complete
+	state := <-stateCh
+	if applyErr != nil {
+		t.Fatalf("err: %s", applyErr)
+	}
+
+	checkStateString(t, state, `
+aws_instance.foo:
+  ID = foo
+	`)
+}
+
 func TestContext2Apply_cancelProvisioner(t *testing.T) {
 	m := testModule(t, "apply-cancel-provisioner")
 	p := testProvider("aws")

terraform/test-fixtures/apply-cancel-block/main.tf

@@ -0,0 +1,3 @@
+resource "aws_instance" "foo" {
+    num = "2"
+}