func openState(agentConfig agent.Config, dialOpts mongo.DialOpts) (_ *state.State, _ *state.Machine, err error) {
info, ok := agentConfig.MongoInfo()
if !ok {
return nil, nil, fmt.Errorf("no state info available")
}
st, err := state.Open(agentConfig.Model(), info, dialOpts, environs.NewStatePolicy())
if err != nil {
return nil, nil, err
}
defer func() {
if err != nil {
st.Close()
}
}()
m0, err := st.FindEntity(agentConfig.Tag())
if err != nil {
if errors.IsNotFound(err) {
err = worker.ErrTerminateAgent
}
return nil, nil, err
}
m := m0.(*state.Machine)
if m.Life() == state.Dead {
return nil, nil, worker.ErrTerminateAgent
}
// Check the machine nonce as provisioned matches the agent.Conf value.
if !m.CheckProvisioned(agentConfig.Nonce()) {
// The agent is running on a different machine to the one it
// should be according to state. It must stop immediately.
logger.Errorf("running machine %v agent on inappropriate instance", m)
return nil, nil, worker.ErrTerminateAgent
}
return st, m, nil
}
func (c *BootstrapCommand) startMongo(addrs []network.Address, agentConfig agent.Config) error {
logger.Debugf("starting mongo")
info, ok := agentConfig.MongoInfo()
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.
mongoDialOpts := mongo.DialOpts{Timeout: 5 * time.Minute}
dialInfo, err := mongo.DialInfo(info.Info, mongoDialOpts)
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")
ensureServerParams, err := cmdutil.NewEnsureServerParams(agentConfig)
if err != nil {
return err
}
err = cmdutil.EnsureMongoServer(ensureServerParams)
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))
if err := initiateMongoServer(peergrouper.InitiateMongoParams{
DialInfo: dialInfo,
MemberHostPort: peerHostPort,
}); err != nil {
return err
}
logger.Infof("started mongo")
return nil
}
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,
})
}
// ensureMongoServer ensures that mongo is installed and running,
// and ready for opening a state connection.
func (a *MachineAgent) ensureMongoServer(agentConfig agent.Config) (err error) {
a.mongoInitMutex.Lock()
defer a.mongoInitMutex.Unlock()
if a.mongoInitialized {
logger.Debugf("mongo is already initialized")
return nil
}
defer func() {
if err == nil {
a.mongoInitialized = true
}
}()
servingInfo, ok := agentConfig.StateServingInfo()
if !ok {
return fmt.Errorf("state worker was started with no state serving info")
}
// 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(a.previousAgentVersion) {
_, 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) error {
config.SetStateServingInfo(servingInfo)
return nil
}); 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
}
ssi := paramsStateServingInfoToStateStateServingInfo(servingInfo)
if err := st.SetStateServingInfo(ssi); 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.
ensureServerParams, err := newEnsureServerParams(agentConfig)
if err != nil {
return err
}
if err := ensureMongoServer(ensureServerParams); 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.MongoInfo()
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)),
// TODO(dfc) InitiateMongoParams should take a Tag
User: stateInfo.Tag.String(),
Password: stateInfo.Password,
}); err != nil {
return err
}
return nil
}
// upgradeWorker runs the required upgrade operations to upgrade to the current Juju version.
func (a *MachineAgent) upgradeWorker(
apiState *api.State,
jobs []params.MachineJob,
agentConfig agent.Config,
) worker.Worker {
return worker.NewSimpleWorker(func(stop <-chan struct{}) error {
select {
case <-a.upgradeComplete:
// Our work is already done (we're probably being restarted
// because the API connection has gone down), so do nothing.
<-stop
return nil
default:
}
// If the machine agent is a state server, flag that state
// needs to be opened before running upgrade steps
needsState := false
for _, job := range jobs {
if job == params.JobManageEnviron {
needsState = true
}
}
// We need a *state.State for upgrades. We open it independently
// of StateWorker, because we have no guarantees about when
// and how often StateWorker might run.
var st *state.State
if needsState {
if err := a.ensureMongoServer(agentConfig); err != nil {
return err
}
var err error
info, ok := agentConfig.MongoInfo()
if !ok {
return fmt.Errorf("no state info available")
}
st, err = state.Open(info, mongo.DialOpts{}, environs.NewStatePolicy())
if err != nil {
return err
}
defer st.Close()
}
err := a.runUpgrades(st, apiState, jobs, agentConfig)
if err == nil {
// Only signal that the upgrade is complete if no error
// was returned.
close(a.upgradeComplete)
} else if !isFatal(err) {
// Only non-fatal errors are returned (this will trigger
// the worker to be restarted).
//
// Fatal upgrade errors are not returned because user
// intervention is required at that point. We don't want
// the upgrade worker or the agent to restart. Status
// output and the logs will report that the upgrade has
// failed.
return err
}
<-stop
return nil
})
}