feedc0de/my-rook-config
1
0
Fork 0
forked from rook/rook
Code Pull Requests Activity

osd: wait for the pgs to be clean

Browse Source 
If migration of the OSDs is enabled, operator should wait
for the pgs to be clean after the all the OSDs have reconciled.
This will help add the exponential backoff for the operator
restarts when OSD migration is needed but PGs are still rebalancing the
data after last OSD migration.

Signed-off-by: sp98 <sapillai@redhat.com>
This commit is contained in:
sp98
2024-01-19 14:08:07 +05:30
parent f46c0845ee
commit 43d3bf147a
1 changed files with 19 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/operator/ceph/cluster/osd/osd.go
+19 -9
View File
@@ -268,20 +268,30 @@ func (c *Cluster) Start() error {
return errors.Wrapf(err, "failed to update ceph storage status")
}
if c.replaceOSD != nil {
delOpts := &k8sutil.DeleteOptions{MustDelete: true, WaitOptions: k8sutil.WaitOptions{Wait: true}}
err := k8sutil.DeleteConfigMap(c.clusterInfo.Context, c.context.Clientset, OSDReplaceConfigName, namespace, delOpts)
if err != nil {
return errors.Wrapf(err, "failed to delete the %q configmap", OSDReplaceConfigName)
if c.spec.Storage.Store.UpdateStore == OSDStoreUpdateConfirmation {
if c.replaceOSD != nil {
delOpts := &k8sutil.DeleteOptions{MustDelete: true, WaitOptions: k8sutil.WaitOptions{Wait: true}}
err := k8sutil.DeleteConfigMap(c.clusterInfo.Context, c.context.Clientset, OSDReplaceConfigName, namespace, delOpts)
if err != nil {
return errors.Wrapf(err, "failed to delete the %q configmap", OSDReplaceConfigName)
}
}
// Wait for PGs to be healthy before continuing the reconcile
_, err = c.waitForHealthyPGs()
// wait for the pgs to be healthy before attempting to migrate the next OSD
_, err := c.waitForHealthyPGs()
if err != nil {
return errors.Wrapf(err, "failed to wait for pgs to be healhty")
}
return errors.New("reconcile operator to replace OSDs that are pending migration")
// reconcile if migration of one or more OSD is pending.
osdsToReplace, err := c.getOSDWithNonMatchingStore()
if err != nil {
return errors.Wrapf(err, "failed to check if any OSD migration is pending")
}
if len(osdsToReplace) != 0 {
return errors.New("reconcile operator to replace OSDs that are pending migration")
}
}
logger.Infof("finished running OSDs in namespace %q", namespace)
@@ -846,7 +856,7 @@ func (c *Cluster) waitForHealthyPGs() (bool, error) {
if pgClean {
return true, nil
}
logger.Infof("waiting for PGs to be healthy after replacing an OSD, status: %q", pgHealthMsg)
logger.Infof("waiting for PGs to be healthy. PG status: %q", pgHealthMsg)
return false, nil
}