public_git/postgres-operator
1
0
Fork 0
mirror of https://github.com/zalando/postgres-operator.git synced 2026-10-21 09:15:48 +02:00
Code Issues Projects Releases Wiki Activity

Migrate only master pods. Migrate single masters. (#199)

Browse Source 
Avoid migrating replica pods, since they will be handled by the
node draining anyway (the PDB specifies that only masters are to
be kept).

Allow migration of the single-pod clusters.
This commit is contained in:
Oleksii Kliukin
2018-01-09 11:55:11 +01:00
committed by GitHub
parent bb5ce6cbbe
commit 23011bdf9a
3 changed files with 38 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/cluster/pod.go
+22 -17
View File
@@ -34,10 +34,6 @@ func (c *Cluster) getRolePods(role PostgresRole) ([]v1.Pod, error) {
return nil, fmt.Errorf("could not get list of pods: %v", err)
}
if len(pods.Items) == 0 {
return nil, fmt.Errorf("no pods")
}
if role == Master && len(pods.Items) > 1 {
return nil, fmt.Errorf("too many masters")
}
@@ -158,6 +154,11 @@ func (c *Cluster) masterCandidate(oldNodeName string) (*v1.Pod, error) {
return nil, fmt.Errorf("could not get replica pods: %v", err)
}
if len(replicas) == 0 {
c.logger.Warningf("single master pod for cluster %q, migration will cause disruption of the service")
return nil, nil
}
for i, pod := range replicas {
// look for replicas running on live nodes. Ignore errors when querying the nodes.
if pod.Spec.NodeName != oldNodeName {
@@ -198,21 +199,25 @@ func (c *Cluster) MigrateMasterPod(podName spec.NamespacedName) error {
return fmt.Errorf("could not get new master candidate: %v", err)
}
pod, err := c.movePodFromEndOfLifeNode(masterCandidatePod)
if err != nil {
return fmt.Errorf("could not move pod: %v", err)
}
// there are two cases for each postgres cluster that has its master pod on the node to migrate from:
// - the cluster has some replicas - migrate one of those if necessary and failover to it
// - there are no replicas - just terminate the master and wait until it respawns
// in both cases the result is the new master up and running on a new node.
if masterCandidatePod != nil {
pod, err := c.movePodFromEndOfLifeNode(masterCandidatePod)
if err != nil {
return fmt.Errorf("could not move pod: %v", err)
}
masterCandidateName := util.NameFromMeta(pod.ObjectMeta)
if err := c.ManualFailover(oldMaster, masterCandidateName); err != nil {
return fmt.Errorf("could not failover to pod %q: %v", masterCandidateName, err)
masterCandidateName := util.NameFromMeta(pod.ObjectMeta)
if err := c.ManualFailover(oldMaster, masterCandidateName); err != nil {
return fmt.Errorf("could not failover to pod %q: %v", masterCandidateName, err)
}
} else {
if _, err = c.movePodFromEndOfLifeNode(oldMaster); err != nil {
return fmt.Errorf("could not move pod: %v", err)
}
}
_, err = c.movePodFromEndOfLifeNode(oldMaster)
if err != nil {
return fmt.Errorf("could not move pod: %v", err)
}
return nil
}
pkg/cluster/resources.go
+3
View File
@@ -95,6 +95,9 @@ func (c *Cluster) preScaleDown(newStatefulSet *v1beta1.StatefulSet) error {
if err != nil {
return fmt.Errorf("could not get master pod: %v", err)
}
if len(masterPod) == 0 {
return fmt.Errorf("no master pod is running in the cluster")
}
podNum, err := getPodIndex(masterPod[0].Name)
if err != nil {