func (task *provisionerTask) startMachine(
machine *apiprovisioner.Machine,
provisioningInfo *params.ProvisioningInfo,
startInstanceParams environs.StartInstanceParams,
) error {
result, err := task.broker.StartInstance(startInstanceParams)
if err != nil {
// If this is a retryable error, we retry once
if instance.IsRetryableCreationError(errors.Cause(err)) {
logger.Infof("retryable error received on start instance - retrying instance creation")
result, err = task.broker.StartInstance(startInstanceParams)
if err != nil {
return task.setErrorStatus("cannot start instance for machine after a retry %q: %v", machine, err)
}
} else {
// Set the state to error, so the machine will be skipped next
// time until the error is resolved, but don't return an
// error; just keep going with the other machines.
return task.setErrorStatus("cannot start instance for machine %q: %v", machine, err)
}
}
inst := result.Instance
hardware := result.Hardware
nonce := startInstanceParams.InstanceConfig.MachineNonce
networks, ifaces, err := task.prepareNetworkAndInterfaces(result.NetworkInfo)
if err != nil {
return task.setErrorStatus("cannot prepare network for machine %q: %v", machine, err)
}
volumes := volumesToApiserver(result.Volumes)
volumeAttachments := volumeAttachmentsToApiserver(result.VolumeAttachments)
// TODO(dimitern) In a newer Provisioner API version, change
// SetInstanceInfo or add a new method that takes and saves in
// state all the information available on a network.InterfaceInfo
// for each interface, so we can later manage interfaces
// dynamically at run-time.
err = machine.SetInstanceInfo(inst.Id(), nonce, hardware, networks, ifaces, volumes, volumeAttachments)
if err != nil && params.IsCodeNotImplemented(err) {
return fmt.Errorf("cannot provision instance %v for machine %q with networks: not implemented", inst.Id(), machine)
} else if err == nil {
logger.Infof(
"started machine %s as instance %s with hardware %q, networks %v, interfaces %v, volumes %v, volume attachments %v, subnets to zones %v",
machine, inst.Id(), hardware,
networks, ifaces,
volumes, volumeAttachments,
startInstanceParams.SubnetsToZones,
)
return nil
}
// We need to stop the instance right away here, set error status and go on.
task.setErrorStatus("cannot register instance for machine %v: %v", machine, err)
if err := task.broker.StopInstances(inst.Id()); err != nil {
// We cannot even stop the instance, log the error and quit.
return errors.Annotatef(err, "cannot stop instance %q for machine %v", inst.Id(), machine)
}
return nil
}
func (s *LxcSuite) TestCreateContainerFailsWithInjectedCreateError(c *gc.C) {
errorChannel := make(chan error, 1)
cleanup := mock.PatchCreateTransientErrorInjectionChannel(errorChannel)
defer cleanup()
errorChannel <- errors.New("lxc-create error")
manager := s.makeManager(c, "test")
_, err := containertesting.CreateContainerTest(c, manager, "1/lxc/0")
c.Assert(err, gc.NotNil)
// this should be a retryable error
isRetryable := instance.IsRetryableCreationError(errors.Cause(err))
c.Assert(isRetryable, jc.IsTrue)
c.Assert(err, gc.ErrorMatches, "lxc container creation failed.*")
}
func (s *LxcSuite) TestCreateContainerFailsWithInjectedStartError(c *gc.C) {
errorChannel := make(chan error, 1)
cleanup := mock.PatchStartTransientErrorInjectionChannel(errorChannel)
defer cleanup()
// One injected error means the container creation will fail
// but the destroy function will clean up the remaining container
// resulting in a RetryableCreationError
errorChannel <- errors.New("start error")
manager := s.makeManager(c, "test")
_, err := containertesting.CreateContainerTest(c, manager, "1/lxc/0")
c.Assert(err, gc.NotNil)
// this should be a retryable error
isRetryable := instance.IsRetryableCreationError(errors.Cause(err))
c.Assert(isRetryable, jc.IsTrue)
}
func (s *LxcSuite) TestCreateContainerWithInjectedErrorDestroyFails(c *gc.C) {
errorChannel := make(chan error, 2)
cleanup := mock.PatchStartTransientErrorInjectionChannel(errorChannel)
defer cleanup()
// Two injected errors mean that the container creation and subsequent
// destroy will fail. This should not result in a RetryableCreationError
// as the container was left in an error state
errorChannel <- errors.New("create error")
errorChannel <- errors.New("destroy error")
manager := s.makeManager(c, "test")
_, err := containertesting.CreateContainerTest(c, manager, "1/lxc/0")
c.Assert(err, gc.NotNil)
// this should not be a retryable error
isRetryable := instance.IsRetryableCreationError(errors.Cause(err))
c.Assert(isRetryable, jc.IsFalse)
}
func (task *provisionerTask) startMachine(
machine *apiprovisioner.Machine,
provisioningInfo *params.ProvisioningInfo,
startInstanceParams environs.StartInstanceParams,
) error {
result, err := task.broker.StartInstance(startInstanceParams)
if err != nil {
if !instance.IsRetryableCreationError(errors.Cause(err)) {
// Set the state to error, so the machine will be skipped next
// time until the error is resolved, but don't return an
// error; just keep going with the other machines.
return task.setErrorStatus("cannot start instance for machine %q: %v", machine, err)
}
logger.Infof("retryable error received on start instance: %v", err)
for count := task.retryStartInstanceStrategy.retryCount; count > 0; count-- {
if task.retryStartInstanceStrategy.retryDelay > 0 {
select {
case <-task.catacomb.Dying():
return task.catacomb.ErrDying()
case <-time.After(task.retryStartInstanceStrategy.retryDelay):
}
}
result, err = task.broker.StartInstance(startInstanceParams)
if err == nil {
break
}
// If this was the last attempt and an error was received, set the error
// status on the machine.
if count == 1 {
return task.setErrorStatus("cannot start instance for machine %q: %v", machine, err)
}
}
}
inst := result.Instance
hardware := result.Hardware
nonce := startInstanceParams.InstanceConfig.MachineNonce
networks, ifaces, err := task.prepareNetworkAndInterfaces(result.NetworkInfo)
if err != nil {
return task.setErrorStatus("cannot prepare network for machine %q: %v", machine, err)
}
volumes := volumesToApiserver(result.Volumes)
volumeAttachments := volumeAttachmentsToApiserver(result.VolumeAttachments)
// TODO(dimitern) In a newer Provisioner API version, change
// SetInstanceInfo or add a new method that takes and saves in
// state all the information available on a network.InterfaceInfo
// for each interface, so we can later manage interfaces
// dynamically at run-time.
err = machine.SetInstanceInfo(inst.Id(), nonce, hardware, networks, ifaces, volumes, volumeAttachments)
if err == nil {
logger.Infof(
"started machine %s as instance %s with hardware %q, networks %v, interfaces %v, volumes %v, volume attachments %v, subnets to zones %v",
machine, inst.Id(), hardware,
networks, ifaces,
volumes, volumeAttachments,
startInstanceParams.SubnetsToZones,
)
return nil
}
// We need to stop the instance right away here, set error status and go on.
task.setErrorStatus("cannot register instance for machine %v: %v", machine, err)
if err := task.broker.StopInstances(inst.Id()); err != nil {
// We cannot even stop the instance, log the error and quit.
return errors.Annotatef(err, "cannot stop instance %q for machine %v", inst.Id(), machine)
}
return nil
}
