func cleanMetadata(n *cluster.Node) map[string]string {
// iaas-id is ignored because it wasn't created in previous tsuru versions
// and having nodes with and without it would cause unbalanced metadata
// errors.
ignoredMetadata := []string{"iaas-id"}
metadata := n.CleanMetadata()
for _, val := range ignoredMetadata {
delete(metadata, val)
}
return metadata
}
func (h *nodeHealer) healNode(node *cluster.Node) (cluster.Node, error) {
emptyNode := cluster.Node{}
failingAddr := node.Address
nodeMetadata := node.CleanMetadata()
failingHost := urlToHost(failingAddr)
failures := node.FailureCount()
machine, err := iaas.CreateMachineForIaaS(nodeMetadata["iaas"], nodeMetadata)
if err != nil {
node.ResetFailures()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error creating new machine: %s", failures, failingHost, err.Error())
}
err = h.provisioner.getCluster().Unregister(failingAddr)
if err != nil {
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error unregistering old node: %s", failures, failingHost, err.Error())
}
newAddr := machine.FormatNodeAddress()
log.Debugf("New machine created during healing process: %s - Waiting for docker to start...", newAddr)
createdNode, err := h.provisioner.getCluster().WaitAndRegister(newAddr, nodeMetadata, h.waitTimeNewMachine)
if err != nil {
node.ResetFailures()
h.provisioner.getCluster().Register(failingAddr, nodeMetadata)
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error registering new node: %s", failures, failingHost, err.Error())
}
var buf bytes.Buffer
err = h.provisioner.moveContainers(failingHost, "", &buf)
if err != nil {
log.Errorf("Unable to move containers, skipping containers healing %q -> %q: %s: %s", failingHost, machine.Address, err.Error(), buf.String())
}
failingMachine, err := iaas.FindMachineByIdOrAddress(node.Metadata["iaas-id"], failingHost)
if err != nil {
return createdNode, fmt.Errorf("Unable to find failing machine %s in IaaS: %s", failingHost, err.Error())
}
err = failingMachine.Destroy()
if err != nil {
return createdNode, fmt.Errorf("Unable to destroy machine %s from IaaS: %s", failingHost, err.Error())
}
log.Debugf("Done auto-healing node %q, node %q created in its place.", failingHost, machine.Address)
return createdNode, nil
}
func (h *NodeHealer) healNode(node *cluster.Node) (cluster.Node, error) {
emptyNode := cluster.Node{}
failingAddr := node.Address
nodeMetadata := node.CleanMetadata()
failingHost := net.URLToHost(failingAddr)
failures := node.FailureCount()
machine, err := iaas.CreateMachineForIaaS(nodeMetadata["iaas"], nodeMetadata)
if err != nil {
node.ResetFailures()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error creating new machine: %s", failures, failingHost, err.Error())
}
err = h.provisioner.Cluster().Unregister(failingAddr)
if err != nil {
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error unregistering old node: %s", failures, failingHost, err.Error())
}
newAddr := machine.FormatNodeAddress()
log.Debugf("New machine created during healing process: %s - Waiting for docker to start...", newAddr)
createdNode := cluster.Node{
Address: newAddr,
Metadata: nodeMetadata,
CreationStatus: cluster.NodeCreationStatusPending,
}
err = h.provisioner.Cluster().Register(createdNode)
if err != nil {
node.ResetFailures()
h.provisioner.Cluster().Register(cluster.Node{Address: failingAddr, Metadata: nodeMetadata})
machine.Destroy()
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error registering new node: %s", failures, failingHost, err.Error())
}
q, err := queue.Queue()
if err != nil {
return emptyNode, err
}
jobParams := monsterqueue.JobParams{
"endpoint": createdNode.Address,
"machine": machine.Id,
"metadata": createdNode.Metadata,
}
job, err := q.EnqueueWait(bs.QueueTaskName, jobParams, h.waitTimeNewMachine)
if err == nil {
_, err = job.Result()
}
if err != nil {
node.ResetFailures()
h.provisioner.Cluster().Register(cluster.Node{Address: failingAddr, Metadata: nodeMetadata})
return emptyNode, fmt.Errorf("Can't auto-heal after %d failures for node %s: error waiting for the bs task: %s", failures, failingHost, err.Error())
}
var buf bytes.Buffer
err = h.provisioner.MoveContainers(failingHost, "", &buf)
if err != nil {
log.Errorf("Unable to move containers, skipping containers healing %q -> %q: %s: %s", failingHost, machine.Address, err.Error(), buf.String())
}
failingMachine, err := iaas.FindMachineByIdOrAddress(node.Metadata["iaas-id"], failingHost)
if err != nil {
return createdNode, fmt.Errorf("Unable to find failing machine %s in IaaS: %s", failingHost, err.Error())
}
err = failingMachine.Destroy()
if err != nil {
return createdNode, fmt.Errorf("Unable to destroy machine %s from IaaS: %s", failingHost, err.Error())
}
log.Debugf("Done auto-healing node %q, node %q created in its place.", failingHost, machine.Address)
return createdNode, nil
}