Now, the backup candidate node is calculate by:
|
// getBackupCandidate returns the hostname of the first not-lagged and |
|
// replicating slave node, else returns the master node. |
|
func (s *jobSyncer) getBackupCandidate() string { |
|
for _, node := range s.cluster.Status.Nodes { |
|
master := s.cluster.GetNodeCondition(node.Name, api.NodeConditionMaster) |
|
replicating := s.cluster.GetNodeCondition(node.Name, api.NodeConditionReplicating) |
|
lagged := s.cluster.GetNodeCondition(node.Name, api.NodeConditionLagged) |
|
|
|
if master == nil || replicating == nil || lagged == nil { |
|
continue |
|
} |
|
|
|
isMaster := master.Status == core.ConditionTrue |
|
isReplicating := replicating.Status == core.ConditionTrue |
|
isLagged := lagged.Status == core.ConditionTrue |
|
|
|
// select a node that is not master is replicating and is not lagged |
|
if !isMaster && isReplicating && !isLagged { |
|
return node.Name |
|
} |
|
} |
|
|
|
log.Info("no healthy slave node found so returns the master node", "master_node", s.cluster.GetMasterHost(), |
|
"key", s.cluster, "backup", s.backup) |
|
return s.cluster.GetMasterHost() |
|
} |
In some cases, we need backup spcify node, for example:
- a node is created only for backup or
- a node has less stress which is more suitable for backup
So, I think we should support manually specify backup candidate node to cover more backup scenario
