func (cfg *InstanceConfig) agentInfo() service.AgentInfo {
return service.NewMachineAgentInfo(
cfg.MachineId,
cfg.DataDir,
cfg.LogDir,
)
}
func (*agentSuite) TestAgentConfMachineLocal(c *gc.C) {
// We use two distinct directories to ensure the paths don't get
// mixed up during the call.
dataDir := c.MkDir()
logDir := c.MkDir()
info := service.NewMachineAgentInfo("0", dataDir, logDir)
renderer, err := shell.NewRenderer("")
c.Assert(err, jc.ErrorIsNil)
conf := service.AgentConf(info, renderer)
jujud := filepath.Join(dataDir, "tools", "machine-0", "jujud"+cmdSuffix)
cmd := strings.Join([]string{
shquote(jujud),
"machine",
"--data-dir", shquote(dataDir),
"--machine-id", "0",
"--debug",
}, " ")
serviceBinary := jujud
serviceArgs := []string{
"machine",
"--data-dir", dataDir,
"--machine-id", "0",
"--debug",
}
c.Check(conf, jc.DeepEquals, common.Conf{
Desc: "juju agent for machine-0",
ExecStart: cmd,
Logfile: filepath.Join(logDir, "machine-0.log"),
Env: osenv.FeatureFlags(),
Limit: map[string]int{
"nofile": 20000,
},
Timeout: 300,
ServiceBinary: serviceBinary,
ServiceArgs: serviceArgs,
})
}
func (*agentSuite) TestAgentConfMachineUbuntu(c *gc.C) {
dataDir := "/var/lib/juju"
logDir := "/var/log/juju"
info := service.NewMachineAgentInfo("0", dataDir, logDir)
renderer, err := shell.NewRenderer("ubuntu")
c.Assert(err, jc.ErrorIsNil)
conf := service.AgentConf(info, renderer)
jujud := dataDir + "/tools/machine-0/jujud"
cmd := strings.Join([]string{
shquote(dataDir + "/tools/machine-0/jujud"),
"machine",
"--data-dir", shquote(dataDir),
"--machine-id", "0",
"--debug",
}, " ")
serviceBinary := jujud
serviceArgs := []string{
"machine",
"--data-dir", dataDir,
"--machine-id", "0",
"--debug",
}
c.Check(conf, jc.DeepEquals, common.Conf{
Desc: "juju agent for machine-0",
ExecStart: cmd,
Logfile: logDir + "/machine-0.log",
Env: osenv.FeatureFlags(),
Limit: map[string]int{
"nofile": 20000,
},
Timeout: 300,
ServiceBinary: serviceBinary,
ServiceArgs: serviceArgs,
})
}
func (*agentSuite) TestAgentConfMachineWindows(c *gc.C) {
dataDir := `C:\Juju\lib\juju`
logDir := `C:\Juju\logs\juju`
info := service.NewMachineAgentInfo("0", dataDir, logDir)
renderer, err := shell.NewRenderer("windows")
c.Assert(err, jc.ErrorIsNil)
conf := service.AgentConf(info, renderer)
jujud := dataDir + `\tools\machine-0\jujud.exe`
cmd := strings.Join([]string{
shquote(jujud),
"machine",
"--data-dir", shquote(dataDir),
"--machine-id", "0",
"--debug",
}, " ")
serviceBinary := jujud
serviceArgs := []string{
"machine",
"--data-dir", dataDir,
"--machine-id", "0",
"--debug",
}
c.Check(conf, jc.DeepEquals, common.Conf{
Desc: "juju agent for machine-0",
ExecStart: cmd,
Logfile: logDir + `\machine-0.log`,
Env: osenv.FeatureFlags(),
Limit: map[string]int{
"nofile": 20000,
},
Timeout: 300,
ServiceBinary: serviceBinary,
ServiceArgs: serviceArgs,
})
}
// Restore handles either returning or creating a controller to a backed up status:
// * extracts the content of the given backup file and:
// * runs mongorestore with the backed up mongo dump
// * updates and writes configuration files
// * updates existing db entries to make sure they hold no references to
// old instances
// * updates config in all agents.
func (b *backups) Restore(backupId string, dbInfo *DBInfo, args RestoreArgs) (names.Tag, error) {
meta, backupReader, err := b.Get(backupId)
if err != nil {
return nil, errors.Annotatef(err, "could not fetch backup %q", backupId)
}
defer backupReader.Close()
workspace, err := NewArchiveWorkspaceReader(backupReader)
if err != nil {
return nil, errors.Annotate(err, "cannot unpack backup file")
}
defer workspace.Close()
// This might actually work, but we don't have a guarantee so we don't allow it.
if meta.Origin.Series != args.NewInstSeries {
return nil, errors.Errorf("cannot restore a backup made in a machine with series %q into a machine with series %q, %#v", meta.Origin.Series, args.NewInstSeries, meta)
}
// TODO(perrito666) Create a compatibility table of sorts.
vers := meta.Origin.Version
if vers.Major != 2 {
return nil, errors.Errorf("Juju version %v cannot restore backups made using Juju version %v", version.Current.Minor, vers)
}
backupMachine := names.NewMachineTag(meta.Origin.Machine)
// The path for the config file might change if the tag changed
// and also the rest of the path, so we assume as little as possible.
oldDatadir, err := paths.DataDir(args.NewInstSeries)
if err != nil {
return nil, errors.Annotate(err, "cannot determine DataDir for the restored machine")
}
var oldAgentConfig agent.ConfigSetterWriter
oldAgentConfigFile := agent.ConfigPath(oldDatadir, args.NewInstTag)
if oldAgentConfig, err = agent.ReadConfig(oldAgentConfigFile); err != nil {
return nil, errors.Annotate(err, "cannot load old agent config from disk")
}
logger.Infof("stopping juju-db")
if err = mongo.StopService(); err != nil {
return nil, errors.Annotate(err, "failed to stop mongo")
}
// delete all the files to be replaced
if err := PrepareMachineForRestore(oldAgentConfig.MongoVersion()); err != nil {
return nil, errors.Annotate(err, "cannot delete existing files")
}
logger.Infof("deleted old files to place new")
if err := workspace.UnpackFilesBundle(filesystemRoot()); err != nil {
return nil, errors.Annotate(err, "cannot obtain system files from backup")
}
logger.Infof("placed new restore files")
var agentConfig agent.ConfigSetterWriter
// The path for the config file might change if the tag changed
// and also the rest of the path, so we assume as little as possible.
datadir, err := paths.DataDir(args.NewInstSeries)
if err != nil {
return nil, errors.Annotate(err, "cannot determine DataDir for the restored machine")
}
agentConfigFile := agent.ConfigPath(datadir, backupMachine)
if agentConfig, err = agent.ReadConfig(agentConfigFile); err != nil {
return nil, errors.Annotate(err, "cannot load agent config from disk")
}
ssi, ok := agentConfig.StateServingInfo()
if !ok {
return nil, errors.Errorf("cannot determine state serving info")
}
APIHostPorts := network.NewHostPorts(ssi.APIPort, args.PrivateAddress, args.PublicAddress)
agentConfig.SetAPIHostPorts([][]network.HostPort{APIHostPorts})
if err := agentConfig.Write(); err != nil {
return nil, errors.Annotate(err, "cannot write new agent configuration")
}
logger.Infof("wrote new agent config for restore")
if backupMachine.Id() != "0" {
logger.Infof("extra work needed backup belongs to %q machine", backupMachine.String())
serviceName := "jujud-" + agentConfig.Tag().String()
aInfo := service.NewMachineAgentInfo(
agentConfig.Tag().Id(),
dataDir,
paths.MustSucceed(paths.LogDir(args.NewInstSeries)),
)
// TODO(perrito666) renderer should have a RendererForSeries, for the moment
// restore only works on linuxes.
renderer, _ := shell.NewRenderer("bash")
serviceAgentConf := service.AgentConf(aInfo, renderer)
svc, err := service.NewService(serviceName, serviceAgentConf, args.NewInstSeries)
if err != nil {
return nil, errors.Annotate(err, "cannot generate service for the restored agent.")
}
if err := svc.Install(); err != nil {
return nil, errors.Annotate(err, "cannot install service for the restored agent.")
}
logger.Infof("new machine service")
}
logger.Infof("mongo service will be reinstalled to ensure its presence")
if err := ensureMongoService(agentConfig); err != nil {
return nil, errors.Annotate(err, "failed to reinstall service for juju-db")
}
dialInfo, err := newDialInfo(args.PrivateAddress, agentConfig)
if err != nil {
return nil, errors.Annotate(err, "cannot produce dial information")
}
oldDialInfo, err := newDialInfo(args.PrivateAddress, oldAgentConfig)
if err != nil {
return nil, errors.Annotate(err, "cannot produce dial information for existing mongo")
}
logger.Infof("new mongo will be restored")
mgoVer := agentConfig.MongoVersion()
tagUser, tagUserPassword, err := tagUserCredentials(agentConfig)
if err != nil {
return nil, errors.Trace(err)
}
rArgs := RestorerArgs{
DialInfo: dialInfo,
Version: mgoVer,
TagUser: tagUser,
TagUserPassword: tagUserPassword,
RunCommandFn: runCommand,
StartMongo: mongo.StartService,
StopMongo: mongo.StopService,
NewMongoSession: NewMongoSession,
GetDB: GetDB,
}
// Restore mongodb from backup
restorer, err := NewDBRestorer(rArgs)
if err != nil {
return nil, errors.Annotate(err, "error preparing for restore")
}
if err := restorer.Restore(workspace.DBDumpDir, oldDialInfo); err != nil {
return nil, errors.Annotate(err, "error restoring state from backup")
}
// Re-start replicaset with the new value for server address
logger.Infof("restarting replicaset")
memberHostPort := net.JoinHostPort(args.PrivateAddress, strconv.Itoa(ssi.StatePort))
err = resetReplicaSet(dialInfo, memberHostPort)
if err != nil {
return nil, errors.Annotate(err, "cannot reset replicaSet")
}
err = updateMongoEntries(args.NewInstId, args.NewInstTag.Id(), backupMachine.Id(), dialInfo)
if err != nil {
return nil, errors.Annotate(err, "cannot update mongo entries")
}
// From here we work with the restored controller
mgoInfo, ok := agentConfig.MongoInfo()
if !ok {
return nil, errors.Errorf("cannot retrieve info to connect to mongo")
}
st, err := newStateConnection(agentConfig.Controller(), agentConfig.Model(), mgoInfo)
if err != nil {
return nil, errors.Trace(err)
}
defer st.Close()
machine, err := st.Machine(backupMachine.Id())
if err != nil {
return nil, errors.Trace(err)
}
logger.Infof("updating local machine addresses")
err = updateMachineAddresses(machine, args.PrivateAddress, args.PublicAddress)
if err != nil {
return nil, errors.Annotate(err, "cannot update api server machine addresses")
}
// Update the APIHostPorts as well. Under normal circumstances the API
// Host Ports are only set during bootstrap and by the peergrouper worker.
// Unfortunately right now, the peer grouper is busy restarting and isn't
// guaranteed to set the host ports before the remote machines we are
// about to tell about us. If it doesn't, the remote machine gets its
// agent.conf file updated with this new machine's IP address, it then
// starts, and the "api-address-updater" worker asks for the api host
// ports, and gets told the old IP address of the machine that was backed
// up. It then writes this incorrect file to its agent.conf file, which
// causes it to attempt to reconnect to the api server. Unfortunately it
// now has the wrong address and can never get the correct one.
// So, we set it explicitly here.
if err := st.SetAPIHostPorts([][]network.HostPort{APIHostPorts}); err != nil {
return nil, errors.Annotate(err, "cannot update api server host ports")
}
// update all agents known to the new controller.
// TODO(perrito666): We should never stop process because of this.
// updateAllMachines will not return errors for individual
// agent update failures
models, err := st.AllModels()
if err != nil {
return nil, errors.Trace(err)
}
machines := []machineModel{}
for _, model := range models {
machinesForModel, err := st.AllMachinesFor(model.UUID())
if err != nil {
return nil, errors.Trace(err)
}
for _, machine := range machinesForModel {
machines = append(machines, machineModel{machine: machine, model: model})
}
}
logger.Infof("updating other machine addresses")
if err := updateAllMachines(args.PrivateAddress, args.PublicAddress, machines); err != nil {
return nil, errors.Annotate(err, "cannot update agents")
}
// Mark restoreInfo as Finished so upon restart of the apiserver
// the client can reconnect and determine if we where succesful.
info := st.RestoreInfo()
// In mongo 3.2, even though the backup is made with --oplog, there
// are stale transactions in this collection.
if err := info.PurgeTxn(); err != nil {
return nil, errors.Annotate(err, "cannot purge stale transactions")
}
if err = info.SetStatus(state.RestoreFinished); err != nil {
return nil, errors.Annotate(err, "failed to set status to finished")
}
return backupMachine, nil
}
// Restore handles either returning or creating a controller to a backed up status:
// * extracts the content of the given backup file and:
// * runs mongorestore with the backed up mongo dump
// * updates and writes configuration files
// * updates existing db entries to make sure they hold no references to
// old instances
// * updates config in all agents.
func (b *backups) Restore(backupId string, args RestoreArgs) (names.Tag, error) {
meta, backupReader, err := b.Get(backupId)
if err != nil {
return nil, errors.Annotatef(err, "could not fetch backup %q", backupId)
}
defer backupReader.Close()
workspace, err := NewArchiveWorkspaceReader(backupReader)
if err != nil {
return nil, errors.Annotate(err, "cannot unpack backup file")
}
defer workspace.Close()
// TODO(perrito666) Create a compatibility table of sorts.
version := meta.Origin.Version
backupMachine := names.NewMachineTag(meta.Origin.Machine)
if err := mongo.StopService(); err != nil {
return nil, errors.Annotate(err, "cannot stop mongo to replace files")
}
// delete all the files to be replaced
if err := PrepareMachineForRestore(); err != nil {
return nil, errors.Annotate(err, "cannot delete existing files")
}
logger.Infof("deleted old files to place new")
if err := workspace.UnpackFilesBundle(filesystemRoot()); err != nil {
return nil, errors.Annotate(err, "cannot obtain system files from backup")
}
logger.Infof("placed new files")
var agentConfig agent.ConfigSetterWriter
// The path for the config file might change if the tag changed
// and also the rest of the path, so we assume as little as possible.
datadir, err := paths.DataDir(args.NewInstSeries)
if err != nil {
return nil, errors.Annotate(err, "cannot determine DataDir for the restored machine")
}
agentConfigFile := agent.ConfigPath(datadir, backupMachine)
if agentConfig, err = agent.ReadConfig(agentConfigFile); err != nil {
return nil, errors.Annotate(err, "cannot load agent config from disk")
}
ssi, ok := agentConfig.StateServingInfo()
if !ok {
return nil, errors.Errorf("cannot determine state serving info")
}
APIHostPorts := network.NewHostPorts(ssi.APIPort, args.PrivateAddress)
agentConfig.SetAPIHostPorts([][]network.HostPort{APIHostPorts})
if err := agentConfig.Write(); err != nil {
return nil, errors.Annotate(err, "cannot write new agent configuration")
}
logger.Infof("wrote new agent config")
if backupMachine.Id() != "0" {
logger.Infof("extra work needed backup belongs to %q machine", backupMachine.String())
serviceName := "jujud-" + agentConfig.Tag().String()
aInfo := service.NewMachineAgentInfo(
agentConfig.Tag().Id(),
dataDir,
paths.MustSucceed(paths.LogDir(args.NewInstSeries)),
)
// TODO(perrito666) renderer should have a RendererForSeries, for the moment
// restore only works on linuxes.
renderer, _ := shell.NewRenderer("bash")
serviceAgentConf := service.AgentConf(aInfo, renderer)
svc, err := service.NewService(serviceName, serviceAgentConf, args.NewInstSeries)
if err != nil {
return nil, errors.Annotate(err, "cannot generate service for the restored agent.")
}
if err := svc.Install(); err != nil {
return nil, errors.Annotate(err, "cannot install service for the restored agent.")
}
logger.Infof("new machine service")
}
logger.Infof("mongo service will be reinstalled to ensure its presence")
if err := ensureMongoService(agentConfig); err != nil {
return nil, errors.Annotate(err, "failed to reinstall service for juju-db")
}
logger.Infof("new mongo will be restored")
// Restore mongodb from backup
if err := placeNewMongoService(workspace.DBDumpDir, version); err != nil {
return nil, errors.Annotate(err, "error restoring state from backup")
}
// Re-start replicaset with the new value for server address
dialInfo, err := newDialInfo(args.PrivateAddress, agentConfig)
if err != nil {
return nil, errors.Annotate(err, "cannot produce dial information")
}
logger.Infof("restarting replicaset")
memberHostPort := net.JoinHostPort(args.PrivateAddress, strconv.Itoa(ssi.StatePort))
err = resetReplicaSet(dialInfo, memberHostPort)
if err != nil {
return nil, errors.Annotate(err, "cannot reset replicaSet")
}
err = updateMongoEntries(args.NewInstId, args.NewInstTag.Id(), backupMachine.Id(), dialInfo)
if err != nil {
return nil, errors.Annotate(err, "cannot update mongo entries")
}
// From here we work with the restored controller
mgoInfo, ok := agentConfig.MongoInfo()
if !ok {
return nil, errors.Errorf("cannot retrieve info to connect to mongo")
}
st, err := newStateConnection(agentConfig.Model(), mgoInfo)
if err != nil {
return nil, errors.Trace(err)
}
defer st.Close()
machine, err := st.Machine(backupMachine.Id())
if err != nil {
return nil, errors.Trace(err)
}
err = updateMachineAddresses(machine, args.PrivateAddress, args.PublicAddress)
if err != nil {
return nil, errors.Annotate(err, "cannot update api server machine addresses")
}
// update all agents known to the new controller.
// TODO(perrito666): We should never stop process because of this.
// updateAllMachines will not return errors for individual
// agent update failures
machines, err := st.AllMachines()
if err != nil {
return nil, errors.Trace(err)
}
if err = updateAllMachines(args.PrivateAddress, machines); err != nil {
return nil, errors.Annotate(err, "cannot update agents")
}
info, err := st.RestoreInfoSetter()
if err != nil {
return nil, errors.Trace(err)
}
// Mark restoreInfo as Finished so upon restart of the apiserver
// the client can reconnect and determine if we where succesful.
err = info.SetStatus(state.RestoreFinished)
return backupMachine, errors.Annotate(err, "failed to set status to finished")
}