1
0
Fork 0
mirror of https://github.com/moby/moby.git synced 2022-11-09 12:21:53 -05:00
moby--moby/integration-cli/docker_api_swarm_test.go
2020-02-11 00:06:42 +01:00

1046 lines
33 KiB
Go

// +build !windows
package main
import (
"context"
"fmt"
"io/ioutil"
"net"
"net/http"
"path/filepath"
"runtime"
"strings"
"sync"
"testing"
"time"
"github.com/cloudflare/cfssl/csr"
"github.com/cloudflare/cfssl/helpers"
"github.com/cloudflare/cfssl/initca"
"github.com/docker/docker/api/types"
"github.com/docker/docker/api/types/container"
"github.com/docker/docker/api/types/swarm"
"github.com/docker/docker/client"
"github.com/docker/docker/integration-cli/checker"
"github.com/docker/docker/integration-cli/daemon"
testdaemon "github.com/docker/docker/testutil/daemon"
"github.com/docker/docker/testutil/request"
"github.com/docker/swarmkit/ca"
"github.com/pkg/errors"
"gotest.tools/v3/assert"
is "gotest.tools/v3/assert/cmp"
"gotest.tools/v3/poll"
)
var defaultReconciliationTimeout = 30 * time.Second
func (s *DockerSwarmSuite) TestAPISwarmInit(c *testing.T) {
// todo: should find a better way to verify that components are running than /info
d1 := s.AddDaemon(c, true, true)
info := d1.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, true)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
assert.Equal(c, info.Cluster.RootRotationInProgress, false)
d2 := s.AddDaemon(c, true, false)
info = d2.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
// Leaving cluster
assert.NilError(c, d2.SwarmLeave(c, false))
info = d2.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
d2.SwarmJoin(c, swarm.JoinRequest{
ListenAddr: d1.SwarmListenAddr(),
JoinToken: d1.JoinTokens(c).Worker,
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
info = d2.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
// Current state restoring after restarts
d1.Stop(c)
d2.Stop(c)
d1.StartNode(c)
d2.StartNode(c)
info = d1.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, true)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
info = d2.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
}
func (s *DockerSwarmSuite) TestAPISwarmJoinToken(c *testing.T) {
d1 := s.AddDaemon(c, false, false)
d1.SwarmInit(c, swarm.InitRequest{})
// todo: error message differs depending if some components of token are valid
d2 := s.AddDaemon(c, false, false)
c2 := d2.NewClientT(c)
err := c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
assert.ErrorContains(c, err, "join token is necessary")
info := d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
err = c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
JoinToken: "foobaz",
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
assert.ErrorContains(c, err, "invalid join token")
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
workerToken := d1.JoinTokens(c).Worker
d2.SwarmJoin(c, swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
JoinToken: workerToken,
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
assert.NilError(c, d2.SwarmLeave(c, false))
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
// change tokens
d1.RotateTokens(c)
err = c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
JoinToken: workerToken,
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
assert.ErrorContains(c, err, "join token is necessary")
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
workerToken = d1.JoinTokens(c).Worker
d2.SwarmJoin(c, swarm.JoinRequest{JoinToken: workerToken, RemoteAddrs: []string{d1.SwarmListenAddr()}})
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
assert.NilError(c, d2.SwarmLeave(c, false))
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
// change spec, don't change tokens
d1.UpdateSwarm(c, func(s *swarm.Spec) {})
err = c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
assert.ErrorContains(c, err, "join token is necessary")
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
d2.SwarmJoin(c, swarm.JoinRequest{JoinToken: workerToken, RemoteAddrs: []string{d1.SwarmListenAddr()}})
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
assert.NilError(c, d2.SwarmLeave(c, false))
info = d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
}
func (s *DockerSwarmSuite) TestUpdateSwarmAddExternalCA(c *testing.T) {
d1 := s.AddDaemon(c, false, false)
d1.SwarmInit(c, swarm.InitRequest{})
d1.UpdateSwarm(c, func(s *swarm.Spec) {
s.CAConfig.ExternalCAs = []*swarm.ExternalCA{
{
Protocol: swarm.ExternalCAProtocolCFSSL,
URL: "https://thishasnoca.org",
},
{
Protocol: swarm.ExternalCAProtocolCFSSL,
URL: "https://thishasacacert.org",
CACert: "cacert",
},
}
})
info := d1.SwarmInfo(c)
assert.Equal(c, len(info.Cluster.Spec.CAConfig.ExternalCAs), 2)
assert.Equal(c, info.Cluster.Spec.CAConfig.ExternalCAs[0].CACert, "")
assert.Equal(c, info.Cluster.Spec.CAConfig.ExternalCAs[1].CACert, "cacert")
}
func (s *DockerSwarmSuite) TestAPISwarmCAHash(c *testing.T) {
d1 := s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, false, false)
splitToken := strings.Split(d1.JoinTokens(c).Worker, "-")
splitToken[2] = "1kxftv4ofnc6mt30lmgipg6ngf9luhwqopfk1tz6bdmnkubg0e"
replacementToken := strings.Join(splitToken, "-")
c2 := d2.NewClientT(c)
err := c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
JoinToken: replacementToken,
RemoteAddrs: []string{d1.SwarmListenAddr()},
})
assert.ErrorContains(c, err, "remote CA does not match fingerprint")
}
func (s *DockerSwarmSuite) TestAPISwarmPromoteDemote(c *testing.T) {
d1 := s.AddDaemon(c, false, false)
d1.SwarmInit(c, swarm.InitRequest{})
d2 := s.AddDaemon(c, true, false)
info := d2.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
d1.UpdateNode(c, d2.NodeID(), func(n *swarm.Node) {
n.Spec.Role = swarm.NodeRoleManager
})
poll.WaitOn(c, pollCheck(c, d2.CheckControlAvailable, checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
d1.UpdateNode(c, d2.NodeID(), func(n *swarm.Node) {
n.Spec.Role = swarm.NodeRoleWorker
})
poll.WaitOn(c, pollCheck(c, d2.CheckControlAvailable, checker.False()), poll.WithTimeout(defaultReconciliationTimeout))
// Wait for the role to change to worker in the cert. This is partially
// done because it's something worth testing in its own right, and
// partially because changing the role from manager to worker and then
// back to manager quickly might cause the node to pause for awhile
// while waiting for the role to change to worker, and the test can
// time out during this interval.
poll.WaitOn(c, pollCheck(c, func(c *testing.T) (interface{}, string) {
certBytes, err := ioutil.ReadFile(filepath.Join(d2.Folder, "root", "swarm", "certificates", "swarm-node.crt"))
if err != nil {
return "", fmt.Sprintf("error: %v", err)
}
certs, err := helpers.ParseCertificatesPEM(certBytes)
if err == nil && len(certs) > 0 && len(certs[0].Subject.OrganizationalUnit) > 0 {
return certs[0].Subject.OrganizationalUnit[0], ""
}
return "", "could not get organizational unit from certificate"
}, checker.Equals("swarm-worker")), poll.WithTimeout(defaultReconciliationTimeout))
// Demoting last node should fail
node := d1.GetNode(c, d1.NodeID())
node.Spec.Role = swarm.NodeRoleWorker
url := fmt.Sprintf("/nodes/%s/update?version=%d", node.ID, node.Version.Index)
res, body, err := request.Post(url, request.Host(d1.Sock()), request.JSONBody(node.Spec))
assert.NilError(c, err)
b, err := request.ReadBody(body)
assert.NilError(c, err)
assert.Equal(c, res.StatusCode, http.StatusBadRequest, "output: %q", string(b))
// The warning specific to demoting the last manager is best-effort and
// won't appear until the Role field of the demoted manager has been
// updated.
// Yes, I know this looks silly, but checker.Matches is broken, since
// it anchors the regexp contrary to the documentation, and this makes
// it impossible to match something that includes a line break.
if !strings.Contains(string(b), "last manager of the swarm") {
assert.Assert(c, strings.Contains(string(b), "this would result in a loss of quorum"))
}
info = d1.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
assert.Equal(c, info.ControlAvailable, true)
// Promote already demoted node
d1.UpdateNode(c, d2.NodeID(), func(n *swarm.Node) {
n.Spec.Role = swarm.NodeRoleManager
})
poll.WaitOn(c, pollCheck(c, d2.CheckControlAvailable, checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
}
func (s *DockerSwarmSuite) TestAPISwarmLeaderProxy(c *testing.T) {
// add three managers, one of these is leader
d1 := s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, true, true)
d3 := s.AddDaemon(c, true, true)
// start a service by hitting each of the 3 managers
d1.CreateService(c, simpleTestService, func(s *swarm.Service) {
s.Spec.Name = "test1"
})
d2.CreateService(c, simpleTestService, func(s *swarm.Service) {
s.Spec.Name = "test2"
})
d3.CreateService(c, simpleTestService, func(s *swarm.Service) {
s.Spec.Name = "test3"
})
// 3 services should be started now, because the requests were proxied to leader
// query each node and make sure it returns 3 services
for _, d := range []*daemon.Daemon{d1, d2, d3} {
services := d.ListServices(c)
assert.Equal(c, len(services), 3)
}
}
func (s *DockerSwarmSuite) TestAPISwarmLeaderElection(c *testing.T) {
if runtime.GOARCH == "s390x" {
c.Skip("Disabled on s390x")
}
if runtime.GOARCH == "ppc64le" {
c.Skip("Disabled on ppc64le")
}
// Create 3 nodes
d1 := s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, true, true)
d3 := s.AddDaemon(c, true, true)
// assert that the first node we made is the leader, and the other two are followers
assert.Equal(c, d1.GetNode(c, d1.NodeID()).ManagerStatus.Leader, true)
assert.Equal(c, d1.GetNode(c, d2.NodeID()).ManagerStatus.Leader, false)
assert.Equal(c, d1.GetNode(c, d3.NodeID()).ManagerStatus.Leader, false)
d1.Stop(c)
var (
leader *daemon.Daemon // keep track of leader
followers []*daemon.Daemon // keep track of followers
)
var lastErr error
checkLeader := func(nodes ...*daemon.Daemon) checkF {
return func(c *testing.T) (interface{}, string) {
// clear these out before each run
leader = nil
followers = nil
for _, d := range nodes {
n := d.GetNode(c, d.NodeID(), func(err error) bool {
if strings.Contains(errors.Cause(err).Error(), context.DeadlineExceeded.Error()) || strings.Contains(err.Error(), "swarm does not have a leader") {
lastErr = err
return true
}
return false
})
if n == nil {
return false, fmt.Sprintf("failed to get node: %v", lastErr)
}
if n.ManagerStatus.Leader {
leader = d
} else {
followers = append(followers, d)
}
}
if leader == nil {
return false, "no leader elected"
}
return true, fmt.Sprintf("elected %v", leader.ID())
}
}
// wait for an election to occur
c.Logf("Waiting for election to occur...")
poll.WaitOn(c, pollCheck(c, checkLeader(d2, d3), checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
// assert that we have a new leader
assert.Assert(c, leader != nil)
// Keep track of the current leader, since we want that to be chosen.
stableleader := leader
// add the d1, the initial leader, back
d1.StartNode(c)
// wait for possible election
c.Logf("Waiting for possible election...")
poll.WaitOn(c, pollCheck(c, checkLeader(d1, d2, d3), checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
// pick out the leader and the followers again
// verify that we still only have 1 leader and 2 followers
assert.Assert(c, leader != nil)
assert.Equal(c, len(followers), 2)
// and that after we added d1 back, the leader hasn't changed
assert.Equal(c, leader.NodeID(), stableleader.NodeID())
}
func (s *DockerSwarmSuite) TestAPISwarmRaftQuorum(c *testing.T) {
if runtime.GOARCH == "s390x" {
c.Skip("Disabled on s390x")
}
if runtime.GOARCH == "ppc64le" {
c.Skip("Disabled on ppc64le")
}
d1 := s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, true, true)
d3 := s.AddDaemon(c, true, true)
d1.CreateService(c, simpleTestService)
d2.Stop(c)
// make sure there is a leader
poll.WaitOn(c, pollCheck(c, d1.CheckLeader, checker.IsNil()), poll.WithTimeout(defaultReconciliationTimeout))
d1.CreateService(c, simpleTestService, func(s *swarm.Service) {
s.Spec.Name = "top1"
})
d3.Stop(c)
var service swarm.Service
simpleTestService(&service)
service.Spec.Name = "top2"
cli := d1.NewClientT(c)
defer cli.Close()
// d1 will eventually step down from leader because there is no longer an active quorum, wait for that to happen
poll.WaitOn(c, pollCheck(c, func(c *testing.T) (interface{}, string) {
_, err := cli.ServiceCreate(context.Background(), service.Spec, types.ServiceCreateOptions{})
return err.Error(), ""
}, checker.Contains("Make sure more than half of the managers are online.")), poll.WithTimeout(defaultReconciliationTimeout*2))
d2.StartNode(c)
// make sure there is a leader
poll.WaitOn(c, pollCheck(c, d1.CheckLeader, checker.IsNil()), poll.WithTimeout(defaultReconciliationTimeout))
d1.CreateService(c, simpleTestService, func(s *swarm.Service) {
s.Spec.Name = "top3"
})
}
func (s *DockerSwarmSuite) TestAPISwarmLeaveRemovesContainer(c *testing.T) {
d := s.AddDaemon(c, true, true)
instances := 2
d.CreateService(c, simpleTestService, setInstances(instances))
id, err := d.Cmd("run", "-d", "busybox", "top")
assert.NilError(c, err, id)
id = strings.TrimSpace(id)
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances+1)), poll.WithTimeout(defaultReconciliationTimeout))
assert.ErrorContains(c, d.SwarmLeave(c, false), "")
assert.NilError(c, d.SwarmLeave(c, true))
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(1)), poll.WithTimeout(defaultReconciliationTimeout))
id2, err := d.Cmd("ps", "-q")
assert.NilError(c, err, id2)
assert.Assert(c, strings.HasPrefix(id, strings.TrimSpace(id2)))
}
// #23629
func (s *DockerSwarmSuite) TestAPISwarmLeaveOnPendingJoin(c *testing.T) {
testRequires(c, Network)
s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, false, false)
id, err := d2.Cmd("run", "-d", "busybox", "top")
assert.NilError(c, err, id)
id = strings.TrimSpace(id)
c2 := d2.NewClientT(c)
err = c2.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d2.SwarmListenAddr(),
RemoteAddrs: []string{"123.123.123.123:1234"},
})
assert.ErrorContains(c, err, "Timeout was reached")
info := d2.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStatePending)
assert.NilError(c, d2.SwarmLeave(c, true))
poll.WaitOn(c, pollCheck(c, d2.CheckActiveContainerCount, checker.Equals(1)), poll.WithTimeout(defaultReconciliationTimeout))
id2, err := d2.Cmd("ps", "-q")
assert.NilError(c, err, id2)
assert.Assert(c, strings.HasPrefix(id, strings.TrimSpace(id2)))
}
// #23705
func (s *DockerSwarmSuite) TestAPISwarmRestoreOnPendingJoin(c *testing.T) {
testRequires(c, Network)
d := s.AddDaemon(c, false, false)
client := d.NewClientT(c)
err := client.SwarmJoin(context.Background(), swarm.JoinRequest{
ListenAddr: d.SwarmListenAddr(),
RemoteAddrs: []string{"123.123.123.123:1234"},
})
assert.ErrorContains(c, err, "Timeout was reached")
poll.WaitOn(c, pollCheck(c, d.CheckLocalNodeState, checker.Equals(swarm.LocalNodeStatePending)), poll.WithTimeout(defaultReconciliationTimeout))
d.RestartNode(c)
info := d.SwarmInfo(c)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateInactive)
}
func (s *DockerSwarmSuite) TestAPISwarmManagerRestore(c *testing.T) {
d1 := s.AddDaemon(c, true, true)
instances := 2
id := d1.CreateService(c, simpleTestService, setInstances(instances))
d1.GetService(c, id)
d1.RestartNode(c)
d1.GetService(c, id)
d2 := s.AddDaemon(c, true, true)
d2.GetService(c, id)
d2.RestartNode(c)
d2.GetService(c, id)
d3 := s.AddDaemon(c, true, true)
d3.GetService(c, id)
d3.RestartNode(c)
d3.GetService(c, id)
err := d3.Kill()
assert.NilError(c, err)
time.Sleep(1 * time.Second) // time to handle signal
d3.StartNode(c)
d3.GetService(c, id)
}
func (s *DockerSwarmSuite) TestAPISwarmScaleNoRollingUpdate(c *testing.T) {
d := s.AddDaemon(c, true, true)
instances := 2
id := d.CreateService(c, simpleTestService, setInstances(instances))
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
containers := d.ActiveContainers(c)
instances = 4
d.UpdateService(c, d.GetService(c, id), setInstances(instances))
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
containers2 := d.ActiveContainers(c)
loop0:
for _, c1 := range containers {
for _, c2 := range containers2 {
if c1 == c2 {
continue loop0
}
}
c.Errorf("container %v not found in new set %#v", c1, containers2)
}
}
func (s *DockerSwarmSuite) TestAPISwarmInvalidAddress(c *testing.T) {
d := s.AddDaemon(c, false, false)
req := swarm.InitRequest{
ListenAddr: "",
}
res, _, err := request.Post("/swarm/init", request.Host(d.Sock()), request.JSONBody(req))
assert.NilError(c, err)
assert.Equal(c, res.StatusCode, http.StatusBadRequest)
req2 := swarm.JoinRequest{
ListenAddr: "0.0.0.0:2377",
RemoteAddrs: []string{""},
}
res, _, err = request.Post("/swarm/join", request.Host(d.Sock()), request.JSONBody(req2))
assert.NilError(c, err)
assert.Equal(c, res.StatusCode, http.StatusBadRequest)
}
func (s *DockerSwarmSuite) TestAPISwarmForceNewCluster(c *testing.T) {
d1 := s.AddDaemon(c, true, true)
d2 := s.AddDaemon(c, true, true)
instances := 2
id := d1.CreateService(c, simpleTestService, setInstances(instances))
poll.WaitOn(c, pollCheck(c, reducedCheck(sumAsIntegers, d1.CheckActiveContainerCount, d2.CheckActiveContainerCount), checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
// drain d2, all containers should move to d1
d1.UpdateNode(c, d2.NodeID(), func(n *swarm.Node) {
n.Spec.Availability = swarm.NodeAvailabilityDrain
})
poll.WaitOn(c, pollCheck(c, d1.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
poll.WaitOn(c, pollCheck(c, d2.CheckActiveContainerCount, checker.Equals(0)), poll.WithTimeout(defaultReconciliationTimeout))
d2.Stop(c)
d1.SwarmInit(c, swarm.InitRequest{
ForceNewCluster: true,
Spec: swarm.Spec{},
})
poll.WaitOn(c, pollCheck(c, d1.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
d3 := s.AddDaemon(c, true, true)
info := d3.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, true)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
instances = 4
d3.UpdateService(c, d3.GetService(c, id), setInstances(instances))
poll.WaitOn(c, pollCheck(c, reducedCheck(sumAsIntegers, d1.CheckActiveContainerCount, d3.CheckActiveContainerCount), checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
}
func simpleTestService(s *swarm.Service) {
ureplicas := uint64(1)
restartDelay := 100 * time.Millisecond
s.Spec = swarm.ServiceSpec{
TaskTemplate: swarm.TaskSpec{
ContainerSpec: &swarm.ContainerSpec{
Image: "busybox:latest",
Command: []string{"/bin/top"},
},
RestartPolicy: &swarm.RestartPolicy{
Delay: &restartDelay,
},
},
Mode: swarm.ServiceMode{
Replicated: &swarm.ReplicatedService{
Replicas: &ureplicas,
},
},
}
s.Spec.Name = "top"
}
func serviceForUpdate(s *swarm.Service) {
ureplicas := uint64(1)
restartDelay := 100 * time.Millisecond
s.Spec = swarm.ServiceSpec{
TaskTemplate: swarm.TaskSpec{
ContainerSpec: &swarm.ContainerSpec{
Image: "busybox:latest",
Command: []string{"/bin/top"},
},
RestartPolicy: &swarm.RestartPolicy{
Delay: &restartDelay,
},
},
Mode: swarm.ServiceMode{
Replicated: &swarm.ReplicatedService{
Replicas: &ureplicas,
},
},
UpdateConfig: &swarm.UpdateConfig{
Parallelism: 2,
Delay: 4 * time.Second,
FailureAction: swarm.UpdateFailureActionContinue,
},
RollbackConfig: &swarm.UpdateConfig{
Parallelism: 3,
Delay: 4 * time.Second,
FailureAction: swarm.UpdateFailureActionContinue,
},
}
s.Spec.Name = "updatetest"
}
func setInstances(replicas int) testdaemon.ServiceConstructor {
ureplicas := uint64(replicas)
return func(s *swarm.Service) {
s.Spec.Mode = swarm.ServiceMode{
Replicated: &swarm.ReplicatedService{
Replicas: &ureplicas,
},
}
}
}
func setUpdateOrder(order string) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
if s.Spec.UpdateConfig == nil {
s.Spec.UpdateConfig = &swarm.UpdateConfig{}
}
s.Spec.UpdateConfig.Order = order
}
}
func setRollbackOrder(order string) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
if s.Spec.RollbackConfig == nil {
s.Spec.RollbackConfig = &swarm.UpdateConfig{}
}
s.Spec.RollbackConfig.Order = order
}
}
func setImage(image string) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
if s.Spec.TaskTemplate.ContainerSpec == nil {
s.Spec.TaskTemplate.ContainerSpec = &swarm.ContainerSpec{}
}
s.Spec.TaskTemplate.ContainerSpec.Image = image
}
}
func setFailureAction(failureAction string) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
s.Spec.UpdateConfig.FailureAction = failureAction
}
}
func setMaxFailureRatio(maxFailureRatio float32) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
s.Spec.UpdateConfig.MaxFailureRatio = maxFailureRatio
}
}
func setParallelism(parallelism uint64) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
s.Spec.UpdateConfig.Parallelism = parallelism
}
}
func setConstraints(constraints []string) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
if s.Spec.TaskTemplate.Placement == nil {
s.Spec.TaskTemplate.Placement = &swarm.Placement{}
}
s.Spec.TaskTemplate.Placement.Constraints = constraints
}
}
func setPlacementPrefs(prefs []swarm.PlacementPreference) testdaemon.ServiceConstructor {
return func(s *swarm.Service) {
if s.Spec.TaskTemplate.Placement == nil {
s.Spec.TaskTemplate.Placement = &swarm.Placement{}
}
s.Spec.TaskTemplate.Placement.Preferences = prefs
}
}
func setGlobalMode(s *swarm.Service) {
s.Spec.Mode = swarm.ServiceMode{
Global: &swarm.GlobalService{},
}
}
func checkClusterHealth(c *testing.T, cl []*daemon.Daemon, managerCount, workerCount int) {
var totalMCount, totalWCount int
for _, d := range cl {
var (
info swarm.Info
)
// check info in a poll.WaitOn(), because if the cluster doesn't have a leader, `info` will return an error
checkInfo := func(c *testing.T) (interface{}, string) {
client := d.NewClientT(c)
daemonInfo, err := client.Info(context.Background())
info = daemonInfo.Swarm
return err, "cluster not ready in time"
}
poll.WaitOn(c, pollCheck(c, checkInfo, checker.IsNil()), poll.WithTimeout(defaultReconciliationTimeout))
if !info.ControlAvailable {
totalWCount++
continue
}
var leaderFound bool
totalMCount++
var mCount, wCount int
for _, n := range d.ListNodes(c) {
waitReady := func(c *testing.T) (interface{}, string) {
if n.Status.State == swarm.NodeStateReady {
return true, ""
}
nn := d.GetNode(c, n.ID)
n = *nn
return n.Status.State == swarm.NodeStateReady, fmt.Sprintf("state of node %s, reported by %s", n.ID, d.NodeID())
}
poll.WaitOn(c, pollCheck(c, waitReady, checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
waitActive := func(c *testing.T) (interface{}, string) {
if n.Spec.Availability == swarm.NodeAvailabilityActive {
return true, ""
}
nn := d.GetNode(c, n.ID)
n = *nn
return n.Spec.Availability == swarm.NodeAvailabilityActive, fmt.Sprintf("availability of node %s, reported by %s", n.ID, d.NodeID())
}
poll.WaitOn(c, pollCheck(c, waitActive, checker.True()), poll.WithTimeout(defaultReconciliationTimeout))
if n.Spec.Role == swarm.NodeRoleManager {
assert.Assert(c, n.ManagerStatus != nil, "manager status of node %s (manager), reported by %s", n.ID, d.NodeID())
if n.ManagerStatus.Leader {
leaderFound = true
}
mCount++
} else {
assert.Assert(c, n.ManagerStatus == nil, "manager status of node %s (worker), reported by %s", n.ID, d.NodeID())
wCount++
}
}
assert.Equal(c, leaderFound, true, "lack of leader reported by node %s", info.NodeID)
assert.Equal(c, mCount, managerCount, "managers count reported by node %s", info.NodeID)
assert.Equal(c, wCount, workerCount, "workers count reported by node %s", info.NodeID)
}
assert.Equal(c, totalMCount, managerCount)
assert.Equal(c, totalWCount, workerCount)
}
func (s *DockerSwarmSuite) TestAPISwarmRestartCluster(c *testing.T) {
mCount, wCount := 5, 1
var nodes []*daemon.Daemon
for i := 0; i < mCount; i++ {
manager := s.AddDaemon(c, true, true)
info := manager.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, true)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
nodes = append(nodes, manager)
}
for i := 0; i < wCount; i++ {
worker := s.AddDaemon(c, true, false)
info := worker.SwarmInfo(c)
assert.Equal(c, info.ControlAvailable, false)
assert.Equal(c, info.LocalNodeState, swarm.LocalNodeStateActive)
nodes = append(nodes, worker)
}
// stop whole cluster
{
var wg sync.WaitGroup
wg.Add(len(nodes))
errs := make(chan error, len(nodes))
for _, d := range nodes {
go func(daemon *daemon.Daemon) {
defer wg.Done()
if err := daemon.StopWithError(); err != nil {
errs <- err
}
}(d)
}
wg.Wait()
close(errs)
for err := range errs {
assert.NilError(c, err)
}
}
// start whole cluster
{
var wg sync.WaitGroup
wg.Add(len(nodes))
errs := make(chan error, len(nodes))
for _, d := range nodes {
go func(daemon *daemon.Daemon) {
defer wg.Done()
if err := daemon.StartWithError("--iptables=false"); err != nil {
errs <- err
}
}(d)
}
wg.Wait()
close(errs)
for err := range errs {
assert.NilError(c, err)
}
}
checkClusterHealth(c, nodes, mCount, wCount)
}
func (s *DockerSwarmSuite) TestAPISwarmServicesUpdateWithName(c *testing.T) {
d := s.AddDaemon(c, true, true)
instances := 2
id := d.CreateService(c, simpleTestService, setInstances(instances))
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
service := d.GetService(c, id)
instances = 5
setInstances(instances)(service)
cli := d.NewClientT(c)
defer cli.Close()
_, err := cli.ServiceUpdate(context.Background(), service.Spec.Name, service.Version, service.Spec, types.ServiceUpdateOptions{})
assert.NilError(c, err)
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
}
// Unlocking an unlocked swarm results in an error
func (s *DockerSwarmSuite) TestAPISwarmUnlockNotLocked(c *testing.T) {
d := s.AddDaemon(c, true, true)
err := d.SwarmUnlock(c, swarm.UnlockRequest{UnlockKey: "wrong-key"})
assert.ErrorContains(c, err, "swarm is not locked")
}
// #29885
func (s *DockerSwarmSuite) TestAPISwarmErrorHandling(c *testing.T) {
ln, err := net.Listen("tcp", fmt.Sprintf(":%d", defaultSwarmPort))
assert.NilError(c, err)
defer ln.Close()
d := s.AddDaemon(c, false, false)
client := d.NewClientT(c)
_, err = client.SwarmInit(context.Background(), swarm.InitRequest{
ListenAddr: d.SwarmListenAddr(),
})
assert.ErrorContains(c, err, "address already in use")
}
// Test case for 30242, where duplicate networks, with different drivers `bridge` and `overlay`,
// caused both scopes to be `swarm` for `docker network inspect` and `docker network ls`.
// This test makes sure the fixes correctly output scopes instead.
func (s *DockerSwarmSuite) TestAPIDuplicateNetworks(c *testing.T) {
d := s.AddDaemon(c, true, true)
cli := d.NewClientT(c)
defer cli.Close()
name := "foo"
networkCreate := types.NetworkCreate{
CheckDuplicate: false,
}
networkCreate.Driver = "bridge"
n1, err := cli.NetworkCreate(context.Background(), name, networkCreate)
assert.NilError(c, err)
networkCreate.Driver = "overlay"
n2, err := cli.NetworkCreate(context.Background(), name, networkCreate)
assert.NilError(c, err)
r1, err := cli.NetworkInspect(context.Background(), n1.ID, types.NetworkInspectOptions{})
assert.NilError(c, err)
assert.Equal(c, r1.Scope, "local")
r2, err := cli.NetworkInspect(context.Background(), n2.ID, types.NetworkInspectOptions{})
assert.NilError(c, err)
assert.Equal(c, r2.Scope, "swarm")
}
// Test case for 30178
func (s *DockerSwarmSuite) TestAPISwarmHealthcheckNone(c *testing.T) {
// Issue #36386 can be a independent one, which is worth further investigation.
c.Skip("Root cause of Issue #36386 is needed")
d := s.AddDaemon(c, true, true)
out, err := d.Cmd("network", "create", "-d", "overlay", "lb")
assert.NilError(c, err, out)
instances := 1
d.CreateService(c, simpleTestService, setInstances(instances), func(s *swarm.Service) {
if s.Spec.TaskTemplate.ContainerSpec == nil {
s.Spec.TaskTemplate.ContainerSpec = &swarm.ContainerSpec{}
}
s.Spec.TaskTemplate.ContainerSpec.Healthcheck = &container.HealthConfig{}
s.Spec.TaskTemplate.Networks = []swarm.NetworkAttachmentConfig{
{Target: "lb"},
}
})
poll.WaitOn(c, pollCheck(c, d.CheckActiveContainerCount, checker.Equals(instances)), poll.WithTimeout(defaultReconciliationTimeout))
containers := d.ActiveContainers(c)
out, err = d.Cmd("exec", containers[0], "ping", "-c1", "-W3", "top")
assert.NilError(c, err, out)
}
func (s *DockerSwarmSuite) TestSwarmRepeatedRootRotation(c *testing.T) {
m := s.AddDaemon(c, true, true)
w := s.AddDaemon(c, true, false)
info := m.SwarmInfo(c)
currentTrustRoot := info.Cluster.TLSInfo.TrustRoot
// rotate multiple times
for i := 0; i < 4; i++ {
var err error
var cert, key []byte
if i%2 != 0 {
cert, _, key, err = initca.New(&csr.CertificateRequest{
CN: "newRoot",
KeyRequest: csr.NewBasicKeyRequest(),
CA: &csr.CAConfig{Expiry: ca.RootCAExpiration},
})
assert.NilError(c, err)
}
expectedCert := string(cert)
m.UpdateSwarm(c, func(s *swarm.Spec) {
s.CAConfig.SigningCACert = expectedCert
s.CAConfig.SigningCAKey = string(key)
s.CAConfig.ForceRotate++
})
// poll to make sure update succeeds
var clusterTLSInfo swarm.TLSInfo
for j := 0; j < 18; j++ {
info := m.SwarmInfo(c)
// the desired CA cert and key is always redacted
assert.Equal(c, info.Cluster.Spec.CAConfig.SigningCAKey, "")
assert.Equal(c, info.Cluster.Spec.CAConfig.SigningCACert, "")
clusterTLSInfo = info.Cluster.TLSInfo
// if root rotation is done and the trust root has changed, we don't have to poll anymore
if !info.Cluster.RootRotationInProgress && clusterTLSInfo.TrustRoot != currentTrustRoot {
break
}
// root rotation not done
time.Sleep(250 * time.Millisecond)
}
if cert != nil {
assert.Equal(c, clusterTLSInfo.TrustRoot, expectedCert)
}
// could take another second or two for the nodes to trust the new roots after they've all gotten
// new TLS certificates
for j := 0; j < 18; j++ {
mInfo := m.GetNode(c, m.NodeID()).Description.TLSInfo
wInfo := m.GetNode(c, w.NodeID()).Description.TLSInfo
if mInfo.TrustRoot == clusterTLSInfo.TrustRoot && wInfo.TrustRoot == clusterTLSInfo.TrustRoot {
break
}
// nodes don't trust root certs yet
time.Sleep(250 * time.Millisecond)
}
assert.DeepEqual(c, m.GetNode(c, m.NodeID()).Description.TLSInfo, clusterTLSInfo)
assert.DeepEqual(c, m.GetNode(c, w.NodeID()).Description.TLSInfo, clusterTLSInfo)
currentTrustRoot = clusterTLSInfo.TrustRoot
}
}
func (s *DockerSwarmSuite) TestAPINetworkInspectWithScope(c *testing.T) {
d := s.AddDaemon(c, true, true)
name := "test-scoped-network"
ctx := context.Background()
apiclient := d.NewClientT(c)
resp, err := apiclient.NetworkCreate(ctx, name, types.NetworkCreate{Driver: "overlay"})
assert.NilError(c, err)
network, err := apiclient.NetworkInspect(ctx, name, types.NetworkInspectOptions{})
assert.NilError(c, err)
assert.Check(c, is.Equal("swarm", network.Scope))
assert.Check(c, is.Equal(resp.ID, network.ID))
_, err = apiclient.NetworkInspect(ctx, name, types.NetworkInspectOptions{Scope: "local"})
assert.Check(c, client.IsErrNotFound(err))
}