func (s *UpgradeSuite) assertUpgradeSteps(c *gc.C, job state.MachineJob) {
s.agentSuite.PatchValue(&version.Current, s.upgradeToVersion)
err := s.State.SetEnvironAgentVersion(s.upgradeToVersion.Number)
c.Assert(err, gc.IsNil)
oldVersion := s.upgradeToVersion
oldVersion.Major = 1
oldVersion.Minor = 16
var oldConfig agent.Config
s.machine, oldConfig, _ = s.primeAgent(c, oldVersion, job)
a := s.newAgent(c, s.machine)
go func() { c.Check(a.Run(nil), gc.IsNil) }()
defer func() { c.Check(a.Stop(), gc.IsNil) }()
// Wait for upgrade steps to run.
success := false
for attempt := coretesting.LongAttempt.Start(); attempt.Next(); {
conf, err := agent.ReadConfig(agent.ConfigPath(oldConfig.DataDir(), s.machine.Tag()))
c.Assert(err, gc.IsNil)
success = conf.UpgradedToVersion() == s.upgradeToVersion.Number
if success {
break
}
}
// Upgrade worker has completed ok.
c.Assert(success, jc.IsTrue)
}
func ensureMongoService(agentConfig agent.Config) error {
var oplogSize int
if oplogSizeString := agentConfig.Value(agent.MongoOplogSize); oplogSizeString != "" {
var err error
if oplogSize, err = strconv.Atoi(oplogSizeString); err != nil {
return errors.Annotatef(err, "invalid oplog size: %q", oplogSizeString)
}
}
var numaCtlPolicy bool
if numaCtlString := agentConfig.Value(agent.NumaCtlPreference); numaCtlString != "" {
var err error
if numaCtlPolicy, err = strconv.ParseBool(numaCtlString); err != nil {
return errors.Annotatef(err, "invalid numactl preference: %q", numaCtlString)
}
}
si, ok := agentConfig.StateServingInfo()
if !ok {
return errors.Errorf("agent config has no state serving info")
}
err := mongo.EnsureServiceInstalled(agentConfig.DataDir(),
si.StatePort,
oplogSize,
numaCtlPolicy)
return errors.Annotate(err, "cannot ensure that mongo service start/stop scripts are in place")
}
func (a *MachineAgent) uninstallAgent(agentConfig agent.Config) error {
var errors []error
agentServiceName := agentConfig.Value(agent.AgentServiceName)
if agentServiceName == "" {
// For backwards compatibility, handle lack of AgentServiceName.
agentServiceName = os.Getenv("UPSTART_JOB")
}
if agentServiceName != "" {
if err := service.NewService(agentServiceName, common.Conf{}).Remove(); err != nil {
errors = append(errors, fmt.Errorf("cannot remove service %q: %v", agentServiceName, err))
}
}
// Remove the juju-run symlink.
if err := os.Remove(jujuRun); err != nil && !os.IsNotExist(err) {
errors = append(errors, err)
}
namespace := agentConfig.Value(agent.Namespace)
if err := mongo.RemoveService(namespace); err != nil {
errors = append(errors, fmt.Errorf("cannot stop/remove mongo service with namespace %q: %v", namespace, err))
}
if err := os.RemoveAll(agentConfig.DataDir()); err != nil {
errors = append(errors, err)
}
if len(errors) == 0 {
return nil
}
return fmt.Errorf("uninstall failed: %v", errors)
}
// newEnsureServerParams creates an EnsureServerParams from an agent configuration.
func newEnsureServerParams(agentConfig agent.Config) (mongo.EnsureServerParams, error) {
// If oplog size is specified in the agent configuration, use that.
// Otherwise leave the default zero value to indicate to EnsureServer
// that it should calculate the size.
var oplogSize int
if oplogSizeString := agentConfig.Value(agent.MongoOplogSize); oplogSizeString != "" {
var err error
if oplogSize, err = strconv.Atoi(oplogSizeString); err != nil {
return mongo.EnsureServerParams{}, fmt.Errorf("invalid oplog size: %q", oplogSizeString)
}
}
servingInfo, ok := agentConfig.StateServingInfo()
if !ok {
return mongo.EnsureServerParams{}, fmt.Errorf("agent config has no state serving info")
}
params := mongo.EnsureServerParams{
StateServingInfo: servingInfo,
DataDir: agentConfig.DataDir(),
Namespace: agentConfig.Value(agent.Namespace),
OplogSize: oplogSize,
}
return params, nil
}
func waitForUpgradeToFinish(c *gc.C, conf agent.Config) {
success := false
for attempt := coretesting.LongAttempt.Start(); attempt.Next(); {
diskConf := readConfigFromDisk(c, conf.DataDir(), conf.Tag())
success = diskConf.UpgradedToVersion() == version.Current
if success {
break
}
}
c.Assert(success, jc.IsTrue)
}
// NewUpgrader returns a new upgrader worker. It watches changes to the
// current version of the current agent (with the given tag) and tries to
// download the tools for any new version into the given data directory. If
// an upgrade is needed, the worker will exit with an UpgradeReadyError
// holding details of the requested upgrade. The tools will have been
// downloaded and unpacked.
func NewUpgrader(st *upgrader.State, agentConfig agent.Config) *Upgrader {
u := &Upgrader{
st: st,
dataDir: agentConfig.DataDir(),
tag: agentConfig.Tag(),
}
go func() {
defer u.tomb.Done()
u.tomb.Kill(u.loop())
}()
return u
}
// PreUpgradeSteps runs various checks and prepares for performing an upgrade.
// If any check fails, an error is returned which aborts the upgrade.
func PreUpgradeSteps(st *state.State, agentConf agent.Config, isController, isMaster bool) error {
if err := checkDiskSpace(agentConf.DataDir()); err != nil {
return errors.Trace(err)
}
if isController {
// Update distro info in case the new Juju controller version
// is aware of new supported series. We'll keep going if this
// fails, and the user can manually update it if they need to.
logger.Infof("updating distro-info")
err := updateDistroInfo()
return errors.Annotate(err, "failed to update distro-info")
}
return nil
}
// updateSupportedContainers records in state that a machine can run the specified containers.
// It starts a watcher and when a container of a given type is first added to the machine,
// the watcher is killed, the machine is set up to be able to start containers of the given type,
// and a suitable provisioner is started.
func (a *MachineAgent) updateSupportedContainers(
runner worker.Runner,
st *api.State,
machineTag string,
containers []instance.ContainerType,
agentConfig agent.Config,
) error {
pr := st.Provisioner()
tag, err := names.ParseMachineTag(machineTag)
if err != nil {
return err
}
machine, err := pr.Machine(tag)
if errors.IsNotFound(err) || err == nil && machine.Life() == params.Dead {
return worker.ErrTerminateAgent
}
if err != nil {
return errors.Annotatef(err, "cannot load machine %s from state", tag)
}
if len(containers) == 0 {
if err := machine.SupportsNoContainers(); err != nil {
return errors.Annotatef(err, "clearing supported containers for %s", tag)
}
return nil
}
if err := machine.SetSupportedContainers(containers...); err != nil {
return errors.Annotatef(err, "setting supported containers for %s", tag)
}
initLock, err := hookExecutionLock(agentConfig.DataDir())
if err != nil {
return err
}
// Start the watcher to fire when a container is first requested on the machine.
watcherName := fmt.Sprintf("%s-container-watcher", machine.Id())
handler := provisioner.NewContainerSetupHandler(
runner,
watcherName,
containers,
machine,
pr,
agentConfig,
initLock,
)
a.startWorkerAfterUpgrade(runner, watcherName, func() (worker.Worker, error) {
return worker.NewStringsWorker(handler), nil
})
return nil
}
func (c *BootstrapCommand) startMongo(addrs []network.Address, agentConfig agent.Config) error {
logger.Debugf("starting mongo")
info, ok := agentConfig.StateInfo()
if !ok {
return fmt.Errorf("no state info available")
}
// When bootstrapping, we need to allow enough time for mongo
// to start as there's no retry loop in place.
// 5 minutes should suffice.
bootstrapDialOpts := mongo.DialOpts{Timeout: 5 * time.Minute}
dialInfo, err := mongo.DialInfo(info.Info, bootstrapDialOpts)
if err != nil {
return err
}
servingInfo, ok := agentConfig.StateServingInfo()
if !ok {
return fmt.Errorf("agent config has no state serving info")
}
// Use localhost to dial the mongo server, because it's running in
// auth mode and will refuse to perform any operations unless
// we dial that address.
dialInfo.Addrs = []string{
net.JoinHostPort("127.0.0.1", fmt.Sprint(servingInfo.StatePort)),
}
logger.Debugf("calling ensureMongoServer")
err = ensureMongoServer(
agentConfig.DataDir(),
agentConfig.Value(agent.Namespace),
servingInfo,
)
if err != nil {
return err
}
peerAddr := mongo.SelectPeerAddress(addrs)
if peerAddr == "" {
return fmt.Errorf("no appropriate peer address found in %q", addrs)
}
peerHostPort := net.JoinHostPort(peerAddr, fmt.Sprint(servingInfo.StatePort))
return maybeInitiateMongoServer(peergrouper.InitiateMongoParams{
DialInfo: dialInfo,
MemberHostPort: peerHostPort,
})
}
// NewEnsureServerParams creates an EnsureServerParams from an agent
// configuration.
func NewEnsureServerParams(agentConfig agent.Config) (mongo.EnsureServerParams, error) {
// If oplog size is specified in the agent configuration, use that.
// Otherwise leave the default zero value to indicate to EnsureServer
// that it should calculate the size.
var oplogSize int
if oplogSizeString := agentConfig.Value(agent.MongoOplogSize); oplogSizeString != "" {
var err error
if oplogSize, err = strconv.Atoi(oplogSizeString); err != nil {
return mongo.EnsureServerParams{}, fmt.Errorf("invalid oplog size: %q", oplogSizeString)
}
}
// If numa ctl preference is specified in the agent configuration, use that.
// Otherwise leave the default false value to indicate to EnsureServer
// that numactl should not be used.
var numaCtlPolicy bool
if numaCtlString := agentConfig.Value(agent.NumaCtlPreference); numaCtlString != "" {
var err error
if numaCtlPolicy, err = strconv.ParseBool(numaCtlString); err != nil {
return mongo.EnsureServerParams{}, fmt.Errorf("invalid numactl preference: %q", numaCtlString)
}
}
si, ok := agentConfig.StateServingInfo()
if !ok {
return mongo.EnsureServerParams{}, fmt.Errorf("agent config has no state serving info")
}
params := mongo.EnsureServerParams{
APIPort: si.APIPort,
StatePort: si.StatePort,
Cert: si.Cert,
PrivateKey: si.PrivateKey,
CAPrivateKey: si.CAPrivateKey,
SharedSecret: si.SharedSecret,
SystemIdentity: si.SystemIdentity,
DataDir: agentConfig.DataDir(),
OplogSize: oplogSize,
SetNumaControlPolicy: numaCtlPolicy,
Version: agentConfig.MongoVersion(),
}
return params, nil
}
// NewUpgrader returns a new upgrader worker. It watches changes to the
// current version of the current agent (with the given tag) and tries to
// download the tools for any new version into the given data directory. If
// an upgrade is needed, the worker will exit with an UpgradeReadyError
// holding details of the requested upgrade. The tools will have been
// downloaded and unpacked.
func NewUpgrader(
st *upgrader.State,
agentConfig agent.Config,
origAgentVersion version.Number,
isUpgradeRunning func() bool,
) *Upgrader {
u := &Upgrader{
st: st,
dataDir: agentConfig.DataDir(),
tag: agentConfig.Tag(),
origAgentVersion: origAgentVersion,
isUpgradeRunning: isUpgradeRunning,
}
go func() {
defer u.tomb.Done()
u.tomb.Kill(u.loop())
}()
return u
}
// updateSupportedContainers records in state that a machine can run the specified containers.
// It starts a watcher and when a container of a given type is first added to the machine,
// the watcher is killed, the machine is set up to be able to start containers of the given type,
// and a suitable provisioner is started.
func (a *MachineAgent) updateSupportedContainers(
runner worker.Runner,
st *api.State,
tag string,
containers []instance.ContainerType,
agentConfig agent.Config,
) error {
pr := st.Provisioner()
machine, err := pr.Machine(tag)
if err != nil {
return fmt.Errorf("%s is not in state: %v", tag, err)
}
if len(containers) == 0 {
if err := machine.SupportsNoContainers(); err != nil {
return fmt.Errorf("clearing supported containers for %s: %v", tag, err)
}
return nil
}
if err := machine.SetSupportedContainers(containers...); err != nil {
return fmt.Errorf("setting supported containers for %s: %v", tag, err)
}
initLock, err := hookExecutionLock(agentConfig.DataDir())
if err != nil {
return err
}
// Start the watcher to fire when a container is first requested on the machine.
watcherName := fmt.Sprintf("%s-container-watcher", machine.Id())
handler := provisioner.NewContainerSetupHandler(
runner,
watcherName,
containers,
machine,
pr,
agentConfig,
initLock,
)
a.startWorkerAfterUpgrade(runner, watcherName, func() (worker.Worker, error) {
return worker.NewStringsWorker(handler), nil
})
return nil
}
// NewAgentUpgrader returns a new upgrader worker. It watches changes to the
// current version of the current agent (with the given tag) and tries to
// download the tools for any new version into the given data directory. If
// an upgrade is needed, the worker will exit with an UpgradeReadyError
// holding details of the requested upgrade. The tools will have been
// downloaded and unpacked.
func NewAgentUpgrader(
st *upgrader.State,
agentConfig agent.Config,
origAgentVersion version.Number,
areUpgradeStepsRunning func() bool,
agentUpgradeComplete chan struct{},
) *Upgrader {
u := &Upgrader{
st: st,
dataDir: agentConfig.DataDir(),
tag: agentConfig.Tag(),
origAgentVersion: origAgentVersion,
areUpgradeStepsRunning: areUpgradeStepsRunning,
agentUpgradeComplete: agentUpgradeComplete,
}
go func() {
defer u.tomb.Done()
u.tomb.Kill(u.loop())
}()
return u
}
func (a *MachineAgent) ensureMongoAdminUser(agentConfig agent.Config) (added bool, err error) {
stateInfo, ok1 := agentConfig.MongoInfo()
servingInfo, ok2 := agentConfig.StateServingInfo()
if !ok1 || !ok2 {
return false, fmt.Errorf("no state serving info configuration")
}
dialInfo, err := mongo.DialInfo(stateInfo.Info, mongo.DefaultDialOpts())
if err != nil {
return false, err
}
if len(dialInfo.Addrs) > 1 {
logger.Infof("more than one state server; admin user must exist")
return false, nil
}
return ensureMongoAdminUser(mongo.EnsureAdminUserParams{
DialInfo: dialInfo,
Namespace: agentConfig.Value(agent.Namespace),
DataDir: agentConfig.DataDir(),
Port: servingInfo.StatePort,
User: stateInfo.Tag.String(),
Password: stateInfo.Password,
})
}
// NewAgentUpgrader returns a new upgrader worker. It watches changes to the
// current version of the current agent (with the given tag) and tries to
// download the tools for any new version into the given data directory. If
// an upgrade is needed, the worker will exit with an UpgradeReadyError
// holding details of the requested upgrade. The tools will have been
// downloaded and unpacked.
func NewAgentUpgrader(
st *upgrader.State,
agentConfig agent.Config,
origAgentVersion version.Number,
upgradeStepsWaiter gate.Waiter,
initialUpgradeCheckComplete gate.Unlocker,
) (*Upgrader, error) {
u := &Upgrader{
st: st,
dataDir: agentConfig.DataDir(),
tag: agentConfig.Tag(),
origAgentVersion: origAgentVersion,
upgradeStepsWaiter: upgradeStepsWaiter,
initialUpgradeCheckComplete: initialUpgradeCheckComplete,
}
err := catacomb.Invoke(catacomb.Plan{
Site: &u.catacomb,
Work: u.loop,
})
if err != nil {
return nil, errors.Trace(err)
}
return u, nil
}
func refreshConfig(c *gc.C, config agent.Config) agent.ConfigSetterWriter {
config1, err := agent.ReadConfig(agent.ConfigPath(config.DataDir(), config.Tag()))
c.Assert(err, gc.IsNil)
return config1
}
// updateSupportedContainers records in state that a machine can run the specified containers.
// It starts a watcher and when a container of a given type is first added to the machine,
// the watcher is killed, the machine is set up to be able to start containers of the given type,
// and a suitable provisioner is started.
func (a *MachineAgent) updateSupportedContainers(
runner worker.Runner,
st api.Connection,
containers []instance.ContainerType,
agentConfig agent.Config,
) error {
pr := st.Provisioner()
tag := agentConfig.Tag().(names.MachineTag)
machine, err := pr.Machine(tag)
if errors.IsNotFound(err) || err == nil && machine.Life() == params.Dead {
return worker.ErrTerminateAgent
}
if err != nil {
return errors.Annotatef(err, "cannot load machine %s from state", tag)
}
if len(containers) == 0 {
if err := machine.SupportsNoContainers(); err != nil {
return errors.Annotatef(err, "clearing supported containers for %s", tag)
}
return nil
}
if err := machine.SetSupportedContainers(containers...); err != nil {
return errors.Annotatef(err, "setting supported containers for %s", tag)
}
initLock, err := cmdutil.HookExecutionLock(agentConfig.DataDir())
if err != nil {
return err
}
// Start the watcher to fire when a container is first requested on the machine.
modelUUID, err := st.ModelTag()
if err != nil {
return err
}
watcherName := fmt.Sprintf("%s-container-watcher", machine.Id())
// There may not be a CA certificate private key available, and without
// it we can't ensure that other Juju nodes can connect securely, so only
// use an image URL getter if there's a private key.
var imageURLGetter container.ImageURLGetter
if agentConfig.Value(agent.AllowsSecureConnection) == "true" {
cfg, err := pr.ModelConfig()
if err != nil {
return errors.Annotate(err, "unable to get environ config")
}
imageURLGetter = container.NewImageURLGetter(
// Explicitly call the non-named constructor so if anyone
// adds additional fields, this fails.
container.ImageURLGetterConfig{
ServerRoot: st.Addr(),
ModelUUID: modelUUID.Id(),
CACert: []byte(agentConfig.CACert()),
CloudimgBaseUrl: cfg.CloudImageBaseURL(),
Stream: cfg.ImageStream(),
ImageDownloadFunc: container.ImageDownloadURL,
})
}
params := provisioner.ContainerSetupParams{
Runner: runner,
WorkerName: watcherName,
SupportedContainers: containers,
ImageURLGetter: imageURLGetter,
Machine: machine,
Provisioner: pr,
Config: agentConfig,
InitLock: initLock,
}
handler := provisioner.NewContainerSetupHandler(params)
a.startWorkerAfterUpgrade(runner, watcherName, func() (worker.Worker, error) {
w, err := watcher.NewStringsWorker(watcher.StringsConfig{
Handler: handler,
})
if err != nil {
return nil, errors.Annotatef(err, "cannot start %s worker", watcherName)
}
return w, nil
})
return nil
}
// ensureMongoServer ensures that mongo is installed and running,
// and ready for opening a state connection.
func (a *MachineAgent) ensureMongoServer(agentConfig agent.Config) error {
a.mongoInitMutex.Lock()
defer a.mongoInitMutex.Unlock()
if a.mongoInitialized {
logger.Debugf("mongo is already initialized")
return nil
}
servingInfo, ok := agentConfig.StateServingInfo()
if !ok {
return fmt.Errorf("state worker was started with no state serving info")
}
namespace := agentConfig.Value(agent.Namespace)
// When upgrading from a pre-HA-capable environment,
// we must add machine-0 to the admin database and
// initiate its replicaset.
//
// TODO(axw) remove this when we no longer need
// to upgrade from pre-HA-capable environments.
var shouldInitiateMongoServer bool
var addrs []network.Address
if isPreHAVersion(agentConfig.UpgradedToVersion()) {
_, err := a.ensureMongoAdminUser(agentConfig)
if err != nil {
return err
}
if servingInfo.SharedSecret == "" {
servingInfo.SharedSecret, err = mongo.GenerateSharedSecret()
if err != nil {
return err
}
if err = a.ChangeConfig(func(config agent.ConfigSetter) {
config.SetStateServingInfo(servingInfo)
}); err != nil {
return err
}
agentConfig = a.CurrentConfig()
}
// Note: we set Direct=true in the mongo options because it's
// possible that we've previously upgraded the mongo server's
// configuration to form a replicaset, but failed to initiate it.
st, m, err := openState(agentConfig, mongo.DialOpts{Direct: true})
if err != nil {
return err
}
if err := st.SetStateServingInfo(servingInfo); err != nil {
st.Close()
return fmt.Errorf("cannot set state serving info: %v", err)
}
st.Close()
addrs = m.Addresses()
shouldInitiateMongoServer = true
}
// ensureMongoServer installs/upgrades the upstart config as necessary.
if err := ensureMongoServer(
agentConfig.DataDir(),
namespace,
servingInfo,
); err != nil {
return err
}
if !shouldInitiateMongoServer {
return nil
}
// Initiate the replicaset for upgraded environments.
//
// TODO(axw) remove this when we no longer need
// to upgrade from pre-HA-capable environments.
stateInfo, ok := agentConfig.StateInfo()
if !ok {
return fmt.Errorf("state worker was started with no state serving info")
}
dialInfo, err := mongo.DialInfo(stateInfo.Info, mongo.DefaultDialOpts())
if err != nil {
return err
}
peerAddr := mongo.SelectPeerAddress(addrs)
if peerAddr == "" {
return fmt.Errorf("no appropriate peer address found in %q", addrs)
}
if err := maybeInitiateMongoServer(peergrouper.InitiateMongoParams{
DialInfo: dialInfo,
MemberHostPort: net.JoinHostPort(peerAddr, fmt.Sprint(servingInfo.StatePort)),
User: stateInfo.Tag,
Password: stateInfo.Password,
}); err != nil {
return err
}
a.mongoInitialized = true
return nil
}
