Revert "Unify warnings about unmovable pods (#389)" (#430)

This reverts commit 4fa09e0dcb.

Reason: the reverted commit bloats the logs
This commit is contained in:
zerg-junior 2018-12-21 17:39:34 +01:00 committed by GitHub
parent 4fa09e0dcb
commit 26670408c4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 28 additions and 42 deletions

View File

@ -7,10 +7,7 @@ import (
"k8s.io/apimachinery/pkg/runtime" "k8s.io/apimachinery/pkg/runtime"
"k8s.io/apimachinery/pkg/watch" "k8s.io/apimachinery/pkg/watch"
"fmt"
"github.com/zalando-incubator/postgres-operator/pkg/cluster" "github.com/zalando-incubator/postgres-operator/pkg/cluster"
"github.com/zalando-incubator/postgres-operator/pkg/spec"
"github.com/zalando-incubator/postgres-operator/pkg/util" "github.com/zalando-incubator/postgres-operator/pkg/util"
) )
@ -58,16 +55,15 @@ func (c *Controller) nodeUpdate(prev, cur interface{}) {
return return
} }
if !c.nodeIsReady(nodePrev) { if util.MapContains(nodeCur.Labels, map[string]string{"master": "true"}) {
c.logger.Debugf("The decommissioned node %v should have already triggered master pod migration. Previous k8s-reported state of the node: %v", util.NameFromMeta(nodePrev.ObjectMeta), nodePrev)
return return
} }
if c.nodeIsReady(nodeCur) { // do nothing if the node should have already triggered an update or
c.logger.Debugf("The decommissioned node %v become schedulable again. Current k8s-reported state of the node: %v", util.NameFromMeta(nodeCur.ObjectMeta), nodeCur) // if only one of the label and the unschedulability criteria are met.
if !c.nodeIsReady(nodePrev) || c.nodeIsReady(nodeCur) {
return return
} }
c.moveMasterPodsOffNode(nodeCur) c.moveMasterPodsOffNode(nodeCur)
} }
@ -77,9 +73,8 @@ func (c *Controller) nodeIsReady(node *v1.Node) bool {
} }
func (c *Controller) moveMasterPodsOffNode(node *v1.Node) { func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
nodeName := util.NameFromMeta(node.ObjectMeta) nodeName := util.NameFromMeta(node.ObjectMeta)
c.logger.Infof("moving pods: node %q became unschedulable and does not have a ready label %q", c.logger.Infof("moving pods: node %q became unschedulable and does not have a ready label: %q",
nodeName, c.opConfig.NodeReadinessLabel) nodeName, c.opConfig.NodeReadinessLabel)
opts := metav1.ListOptions{ opts := metav1.ListOptions{
@ -87,7 +82,7 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
} }
podList, err := c.KubeClient.Pods(c.opConfig.WatchedNamespace).List(opts) podList, err := c.KubeClient.Pods(c.opConfig.WatchedNamespace).List(opts)
if err != nil { if err != nil {
c.logger.Errorf("could not fetch the list of Spilo pods: %v", err) c.logger.Errorf("could not fetch list of the pods: %v", err)
return return
} }
@ -98,25 +93,17 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
} }
} }
movedMasterPods := 0
movableMasterPods := make(map[*v1.Pod]*cluster.Cluster)
unmovablePods := make(map[spec.NamespacedName]string)
clusters := make(map[*cluster.Cluster]bool) clusters := make(map[*cluster.Cluster]bool)
masterPods := make(map[*v1.Pod]*cluster.Cluster)
movedPods := 0
for _, pod := range nodePods { for _, pod := range nodePods {
podName := util.NameFromMeta(pod.ObjectMeta) podName := util.NameFromMeta(pod.ObjectMeta)
role, ok := pod.Labels[c.opConfig.PodRoleLabel] role, ok := pod.Labels[c.opConfig.PodRoleLabel]
if !ok { if !ok || cluster.PostgresRole(role) != cluster.Master {
// pods with an unknown role cannot be safely moved to another node if !ok {
unmovablePods[podName] = fmt.Sprintf("could not move pod %q from node %q: pod has no role label %q", podName, nodeName, c.opConfig.PodRoleLabel) c.logger.Warningf("could not move pod %q: pod has no role", podName)
continue }
}
// deployments can transparently re-create replicas so we do not move away such pods
if cluster.PostgresRole(role) == cluster.Replica {
continue continue
} }
@ -126,7 +113,7 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
cl, ok := c.clusters[clusterName] cl, ok := c.clusters[clusterName]
c.clustersMu.RUnlock() c.clustersMu.RUnlock()
if !ok { if !ok {
unmovablePods[podName] = fmt.Sprintf("could not move master pod %q from node %q: pod belongs to an unknown Postgres cluster %q", podName, nodeName, clusterName) c.logger.Warningf("could not move pod %q: pod does not belong to a known cluster", podName)
continue continue
} }
@ -134,20 +121,20 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
clusters[cl] = true clusters[cl] = true
} }
movableMasterPods[pod] = cl masterPods[pod] = cl
} }
for cl := range clusters { for cl := range clusters {
cl.Lock() cl.Lock()
} }
for pod, cl := range movableMasterPods { for pod, cl := range masterPods {
podName := util.NameFromMeta(pod.ObjectMeta) podName := util.NameFromMeta(pod.ObjectMeta)
if err := cl.MigrateMasterPod(podName); err == nil {
movedMasterPods++ if err := cl.MigrateMasterPod(podName); err != nil {
c.logger.Errorf("could not move master pod %q: %v", podName, err)
} else { } else {
unmovablePods[podName] = fmt.Sprintf("could not move master pod %q from node %q: %v", podName, nodeName, err) movedPods++
} }
} }
@ -155,16 +142,15 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
cl.Unlock() cl.Unlock()
} }
if leftPods := len(unmovablePods); leftPods > 0 { totalPods := len(masterPods)
c.logger.Warnf("could not move %d master or unknown role pods from the node %q, you may have to delete them manually",
leftPods, nodeName) c.logger.Infof("%d/%d master pods have been moved out from the %q node",
for _, reason := range unmovablePods { movedPods, totalPods, nodeName)
c.logger.Warning(reason)
} if leftPods := totalPods - movedPods; leftPods > 0 {
c.logger.Warnf("could not move master %d/%d pods from the %q node",
leftPods, totalPods, nodeName)
} }
c.logger.Infof("%d master pods have been moved out from the node %q", movedMasterPods, nodeName)
} }
func (c *Controller) nodeDelete(obj interface{}) { func (c *Controller) nodeDelete(obj interface{}) {

View File

@ -121,7 +121,7 @@ function deploy_self_built_image() {
# update the tag in the postgres operator conf # update the tag in the postgres operator conf
# since the image with this tag already exists on the machine, # since the image with this tag already exists on the machine,
# docker should not attempt to fetch it from the registry due to imagePullPolicy # docker should not attempt to fetch it from the registry due to imagePullPolicy
sed --expression "s/\(image\:.*\:\).*$/\1$TAG/; s/smoke-tested-//" manifests/postgres-operator.yaml > "$PATH_TO_LOCAL_OPERATOR_MANIFEST" sed --expression "s/\(image\:.*\:\).*$/\1$TAG/" manifests/postgres-operator.yaml > "$PATH_TO_LOCAL_OPERATOR_MANIFEST"
retry "kubectl create -f \"$PATH_TO_LOCAL_OPERATOR_MANIFEST\"" "attempt to create $PATH_TO_LOCAL_OPERATOR_MANIFEST resource" retry "kubectl create -f \"$PATH_TO_LOCAL_OPERATOR_MANIFEST\"" "attempt to create $PATH_TO_LOCAL_OPERATOR_MANIFEST resource"
} }