Refactor dependency checks from init() into public API

Instead of checking during init(), which could result in major
locking issues when used with pods, make our dependency checks in
the public API instead. This avoids doing them when we start pods
(where, because of the dependency graph, we can reasonably say
all dependencies are up before we start a container).

Signed-off-by: Matthew Heon <matthew.heon@gmail.com>

Closes: #577
Approved by: rhatdan

3 files changed, 89 insertions, 22 deletions

diff --git a/libpod/container_api.go b/libpod/container_api.go
index 639696268..41e754587 100644
--- a/libpod/container_api.go
+++ b/libpod/container_api.go
@@ -5,6 +5,7 @@ import (
 	"io/ioutil"
 	"os"
 	"strconv"
+	"strings"
 	"time"
 
 	"github.com/docker/docker/daemon/caps"
@@ -33,6 +34,15 @@ func (c *Container) Init() (err error) {
 		return errors.Wrapf(ErrCtrExists, "container %s has already been created in runtime", c.ID())
 	}
 
+	notRunning, err := c.checkDependenciesRunning()
+	if err != nil {
+		return errors.Wrapf(err, "error checking dependencies for container %s")
+	}
+	if len(notRunning) > 0 {
+		depString := strings.Join(notRunning, ",")
+		return errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString)
+	}
+
 	if err := c.prepare(); err != nil {
 		return err
 	}
@@ -70,6 +80,15 @@ func (c *Container) Start() (err error) {
 		return errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Created or Stopped state to be started", c.ID())
 	}
 
+	notRunning, err := c.checkDependenciesRunning()
+	if err != nil {
+		return errors.Wrapf(err, "error checking dependencies for container %s")
+	}
+	if len(notRunning) > 0 {
+		depString := strings.Join(notRunning, ",")
+		return errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString)
+	}
+
 	if err := c.prepare(); err != nil {
 		return err
 	}
@@ -124,6 +143,15 @@ func (c *Container) StartAndAttach(noStdin bool, keys string) (attachResChan <-c
 		return nil, errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Created or Stopped state to be started", c.ID())
 	}
 
+	notRunning, err := c.checkDependenciesRunning()
+	if err != nil {
+		return nil, errors.Wrapf(err, "error checking dependencies for container %s")
+	}
+	if len(notRunning) > 0 {
+		depString := strings.Join(notRunning, ",")
+		return nil, errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString)
+	}
+
 	if err := c.prepare(); err != nil {
 		return nil, err
 	}
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index 523d088d5..f75df8c28 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -350,40 +350,72 @@ func (c *Container) save() error {
 	return nil
 }
 
-// Initialize a container, creating it in the runtime
-func (c *Container) init() error {
+// Check if a container's dependencies are running
+// Returns a []string containing the IDs of dependencies that are not running
+func (c *Container) checkDependenciesRunning() ([]string, error) {
 	deps := c.Dependencies()
+	notRunning := []string{}
+
+	// We were not passed a set of dependency containers
+	// Make it ourselves
 	depCtrs := make(map[string]*Container, len(deps))
-	// Ensure that dependencies are up before we begin
 	for _, dep := range deps {
 		// Get the dependency container
 		depCtr, err := c.runtime.state.Container(dep)
 		if err != nil {
-			return errors.Wrapf(err, "error retrieving dependency %s of container %s from state", dep, c.ID())
+			return nil, errors.Wrapf(err, "error retrieving dependency %s of container %s from state", dep, c.ID())
 		}
 
 		// Check the status
 		state, err := depCtr.State()
 		if err != nil {
-			return errors.Wrapf(err, "error retrieving state of dependency %s of container %s", dep, c.ID())
+			return nil, errors.Wrapf(err, "error retrieving state of dependency %s of container %s", dep, c.ID())
 		}
 		if state != ContainerStateRunning {
-			return errors.Wrapf(ErrCtrStateInvalid, "dependency %s of container %s is not running, cannot initialize container", dep, c.ID())
+			notRunning = append(notRunning, dep)
 		}
 		depCtrs[dep] = depCtr
+	}
+
+	return notRunning, nil
+}
+
+// Check if a container's dependencies are running
+// Returns a []string containing the IDs of dependencies that are not running
+// Assumes depencies are already locked, and will be passed in
+// Accepts a map[string]*Container containing, at a minimum, the locked
+// dependency containers
+// (This must be a map from container ID to container)
+func (c *Container) checkDependenciesRunningLocked(depCtrs map[string]*Container) ([]string, error) {
+	deps := c.Dependencies()
+	notRunning := []string{}
+
+	for _, dep := range deps {
+		depCtr, ok := depCtrs[dep]
+		if !ok {
+			return nil, errors.Wrapf(ErrNoSuchCtr, "container %s depends on container %s but it is not on containers passed to checkDependenciesRunning", c.ID(), dep)
+		}
 
-		// There is potential race here
-		// A dep container may stop before this container starts
-		// We can't really control this, but it probably can't cause
-		// anything insidious.
+		if err := c.syncContainer(); err != nil {
+			return nil, err
+		}
+
+		if depCtr.state.State != ContainerStateRunning {
+			notRunning = append(notRunning, dep)
+		}
 	}
 
+	return notRunning, nil
+}
+
+// Initialize a container, creating it in the runtime
+func (c *Container) init() error {
 	if err := c.makeBindMounts(); err != nil {
 		return err
 	}
 
 	// Generate the OCI spec
-	spec, err := c.generateSpec(depCtrs)
+	spec, err := c.generateSpec()
 	if err != nil {
 		return err
 	}
@@ -875,7 +907,7 @@ func (c *Container) generateHosts() (string, error) {
 
 // Generate spec for a container
 // Accepts a map of the container's dependencies
-func (c *Container) generateSpec(deps map[string]*Container) (*spec.Spec, error) {
+func (c *Container) generateSpec() (*spec.Spec, error) {
 	g := generate.NewFromSpec(c.config.Spec)
 
 	// If network namespace was requested, add it now
@@ -921,37 +953,37 @@ func (c *Container) generateSpec(deps map[string]*Container) (*spec.Spec, error)
 
 	// Add shared namespaces from other containers
 	if c.config.IPCNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, IPCNS, deps[c.config.IPCNsCtr], spec.IPCNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, IPCNS, c.config.IPCNsCtr, spec.IPCNamespace); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.MountNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, MountNS, deps[c.config.MountNsCtr], spec.MountNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, MountNS, c.config.MountNsCtr, spec.MountNamespace); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.NetNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, NetNS, deps[c.config.NetNsCtr], spec.NetworkNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, NetNS, c.config.NetNsCtr, spec.NetworkNamespace); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.PIDNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, PIDNS, deps[c.config.PIDNsCtr], string(spec.PIDNamespace)); err != nil {
+		if err := c.addNamespaceContainer(&g, PIDNS, c.config.PIDNsCtr, string(spec.PIDNamespace)); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.UserNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, UserNS, deps[c.config.UserNsCtr], spec.UserNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, UserNS, c.config.UserNsCtr, spec.UserNamespace); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.UTSNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, UTSNS, deps[c.config.UTSNsCtr], spec.UTSNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, UTSNS, c.config.UTSNsCtr, spec.UTSNamespace); err != nil {
 			return nil, err
 		}
 	}
 	if c.config.CgroupNsCtr != "" {
-		if err := c.addNamespaceContainer(&g, CgroupNS, deps[c.config.CgroupNsCtr], spec.CgroupNamespace); err != nil {
+		if err := c.addNamespaceContainer(&g, CgroupNS, c.config.CgroupNsCtr, spec.CgroupNamespace); err != nil {
 			return nil, err
 		}
 	}
@@ -979,11 +1011,13 @@ func (c *Container) generateSpec(deps map[string]*Container) (*spec.Spec, error)
 }
 
 // Add an existing container's namespace to the spec
-func (c *Container) addNamespaceContainer(g *generate.Generator, ns LinuxNS, nsCtr *Container, specNS string) error {
-	if nsCtr == nil {
-		return errors.Wrapf(ErrNoSuchCtr, "nil container passed to addNamespaceContainer")
+func (c *Container) addNamespaceContainer(g *generate.Generator, ns LinuxNS, ctr string, specNS string) error {
+	nsCtr, err := c.runtime.state.Container(ctr)
+	if err != nil {
+		return errors.Wrapf(err, "error retrieving dependency %s of container %s from state", ctr, c.ID())
 	}
 
+	// TODO need unlocked version of this for use in pods
 	nsPath, err := nsCtr.NamespacePath(ns)
 	if err != nil {
 		return err
diff --git a/libpod/pod.go b/libpod/pod.go
index f7059800a..7d9e2c81b 100644
--- a/libpod/pod.go
+++ b/libpod/pod.go
@@ -165,6 +165,11 @@ func startNode(node *containerNode, setError bool, ctrErrors map[string]error, c
 	// Going to start the container, mark us as visited
 	ctrsVisited[node.id] = true
 
+	// TODO: Maybe have a checkDependenciesRunningLocked here?
+	// Graph traversal should ensure our deps have been started, but some
+	// might have stopped since?
+	// Potentially will hurt our perf, though
+
 	// Start the container (only if it is not running)
 	ctrErrored := false
 	if node.container.state.State != ContainerStateRunning {