// Updates the channel blocking for recovery
func (e *EtcdBackend) handleRecovery(node *client.Node) {
m := cluster.DeserializeMachine(node.Value)
chn, ok := e.nodeListeners[m.Id]
if ok {
chn <- m
}
}
func (e *EtcdBackend) loadMachines() {
result, err := e.kapi.Get(context.Background(), e.backendConfig.MachinePrefix+"/status/", nil)
if err != nil {
// TODO we probably need to try again, what happens if this is never successful
log.Warning.Printf("Could not load cluster machine list")
return
}
nodes := result.Node.Nodes
for _, node := range nodes {
//log.Info.Printf("Node: %s = %s ", node.Key, node.Value)
m := cluster.DeserializeMachine(node.Value)
if m.Id != e.machine.Id {
e.cluster.AddMachine(m)
}
}
}
func (e *EtcdBackend) handleFailure(node *client.Node) {
e.Status <- "Waiting to recover"
m := cluster.DeserializeMachine(node.Value)
if m.Id == e.machine.Id {
log.Error.Printf("Our key expired but we are still alive! %s", m.Id)
return
}
check := e.cluster.GetMachine(m.Id)
if check == nil {
log.Error.Printf("Received expiration of a machine we weren't aware of %s", m.Id)
return
}
// Create a listener for this machine to be notified
// if the machine returns
// Do we need a buffered channel for this use case?
log.Info.Printf("Handling machine failure of %v", m)
listener := make(chan *cluster.Machine, 8)
e.nodeListeners[m.Id] = listener
select {
case <-listener:
// The machine appears to have recovered, which is great news
// for us because we don't have to do any work
e.Status <- "Not Recovering"
delete(e.nodeListeners, m.Id)
return
case sig := <-e.Signal:
e.Signal <- sig
log.Trace.Printf("Recieved shutdown: aborting recovery")
e.Status <- "Not Recovering"
return
case <-time.After(e.backendConfig.FailoverUnit * time.Duration(e.backendConfig.FailoverTimeout)):
// The machine has expired, so we will start the recovery
// process.
// We'll start by fighting against every machine
// in the cluster to obtain a lock, once we do
// we'll grab all the machines and identify the one
// with the least deployments.
// Then we'll go through all of the downed machine's
// Deployments, and start forwarding them to the
// machine with the least, failing over to the next
// least if the previouys machine already has said
// deployment until all the deployments have been
// reassigned.
e.AttemptRecovery(m)
return
}
}
func (e *EtcdBackend) handleNewNode(node *client.Node) {
m := cluster.DeserializeMachine(node.Value)
e.cluster.AddMachine(m)
}
