v0.4.0

This release makes some spec renaming changes.
Upgrade is possible. But it is highly recommended to make a backup before upgrade.
After upgrade, please rename the spec and secret fields according to following notes.

BREAKING CHANGE: Re-naming of TLS spec and TLS secrets' fields.

TLS spec:

• member's clientSecret is changed to serverSecret

TLS secrets:

• member's peerSecret's fields change:
  • peer-crt.pem -> peer.crt
  • peer-key.pem -> peer.key
  • peer-ca-crt.pem -> peer-ca.crt
• member's clientSecret is changed to serverSecret, its fields change:
  • client-crt.pem -> server.crt
  • client-key.pem -> server.key
  • client-ca-crt.pem -> server-ca.crt
• operatorSecret's fields change:
  • etcd-crt.pem -> etcd-client.crt
  • etcd-key.pem -> etcd-client.key
  • etcd-ca-crt.pem -> etcd-client-ca.crt

BREAKING CHANGE: Backup spec: CleanupBackupsOnClusterDelete field is renamed to AutoDelete.

Previous spec like this one

spec:
  backup:
    storageType: "PersistentVolume"
    ...
    cleanupBackupsOnClusterDelete: true

should be changed to

spec:
  backup:
    storageType: "PersistentVolume"
    ...
    autoDelete: true

v0.3.3

[Release 0.3.3]

Added

• Adds ability for users to specify base image for etcd pods in a cluster.
  Default base image is quay.io/coreos/etcd-operator.

Fixed

• [GH-1138] Fixed operator stucks in managing selfhosted cluster when there are not enough nodes to start new etcd member.
• [GH-1196] Fixed etcd operator could not start S3 backup sidecar if given non-root user.

v0.3.2

[Release 0.3.2]

Bug fix release to fix self-hosted etcd issue [GH-1171] .

v0.3.1

[Release 0.3.1]

Notes for self-hosted etcd:
The newly introduced TLS feature for self hosted etcd is a breaking change.
Existing self hosted etcd cluster MUST be recreated for updating to this release.

Added

• Self-hosted etcd: if etcd-hosts.checkpoint file exists under ${datadir}/,
  etcd pod will restore the hosts mapping from it before etcd bootstraps.
• Add static TLS support for self-hosted etcd mode.
• The operator will now post Kubernetes events. To allow this the necessary RBAC rule for the resource events must be added to the clusterrole. See the rbac guide to see how to set up RBAC rules for the operator. If the rbac rule for 'events' is not present then the operator will continue to function normally but will also print out an error message on the failure to post an event.
• Add revision field in backup status.
• Support getting a specific backup with verison and revision from the backup service.

Changed

• Self-hosted etcd: use FQDN for client/peer URL.
• Updated RBAC rules for resources thirdpartyresources and storageclasses to all verbs * . We loose granularity early so that we have more flexibility to use other methods (e.g. Get) later.

Removed

• Update default etcd version to 3.1.8

Fixed

• [GH-1108] selfHosted: fix backup unable to talk to etcd pods

v0.3.0

[Release 0.3.0]

Upgrade Notice

Check upgrade guide.

Added

• Added support for backup policy to be dynamically added, updated
• Added per cluster policy support for S3.

Changed

• Backup sidecar deployment created with Recreate strategy.
• Spec.Backup.MaxBackups meaning change: 0 means unlimited backups; < 0 will be rejected.

Removed

Fixed

• [GH-1068] Backup sidecar deployment stuck at upgrading

v0.2.6

Release 0.2.6

Upgrade Notice

• Once operator is upgraded, all backup-enabled cluster will go through an upgrade process that
  deletes backup sidecar's ReplicaSet and creates new Deployment for sidecar.
  If upgrading failed for any reason, cluster TPR's status.phase will be FAILED.
  Recreate of the cluster TPR is needed on failure case.

Added

• PodPolicy provides EtcdEnv option to add custom env to the etcd process.
• PodPolicy provides Labels option to add custom labels to the etcd pod.
• TLS feature: user can now create TLS-secured cluster via operator.
  See TLS guide.

Changed

• Self-hosted etcd pod's anti-affinity label selector is changed to select {"app": "etcd"}.
  That is, no two etcd pods should sit on the same node, even if they belongs to different clusters.
• Using Deployment to manage backup sidecar instead of ReplicaSet.
• S3 backup path is changed to ${BUCKET_NAME}/v1/${NAMESPACE}/${CLUSTER_NAME}/.

v0.2.5

v0.2.5 Release

Notable Changes

• The previous client lb service named ${clusterName} is changed to headless service, and etcd pods are using hostname + subdomain instead. This requires restart of existing cluster.
  If you have backup enabled you can make a backup, and then restore your new cluster from that backup. Consult this docs to see how to restore from a backup.

Added

• Add "none" PV provisioner option. If operator flag "pv-provisioner" is set to "none",
  operator won’t create any storage class and users couldn’t make use of operator’s PV backup feature.
• Add headless service ${clusterName} which selects etcd pods of given cluster.
• Pod Tolerations.

Changed

• TLSSpec json tag changed as omitempty
• Time related fields in spec, i.e. TransitionTime and CreationTime, is changed to type string.
  This should be backward compatible and no effect on operator upgrade.
• Update default etcd version to 3.1.4
• Self-hosted etcd pod is started with "--metrics extensive" flag.
  This is only available in etcd 3.1+.
• Change client LB service name to ${clusterName}-client.
• Add hostname and subdomain to etcd pods, which makes them have A records formatted in ${memberName}.${clusterName}.${namespace}.svc.cluster.local .
  For more info, see https://kubernetes.io/docs/concepts/services-networking/dns-pod-service/ .
  We also change PeerURL of etcd members to use such hostnames.

Removed

• Individual etcd member's services were removed. Use hostname and subdomain of etcd pod instead.

Fixed

• [GH-910] Operator keeps updating status even if there is no change.

v0.2.4

Bug fix release

[Release 0.2.4]

Fixed

• [GH-900] Fix looping of reconcile skip due to unready members

v0.2.3

Release 0.2.3

This is a bug fix release.

Fixed

• [GH-890] Fix a race that when majority of members went down cluster couldn't recover
• Fix self-hosted cluster reboot case

v0.2.2

Release 0.2.2

Added

• Backup creation time is added in backup status.
• Total size of backups time is added in backup service status.
• Cluster members that are ready and unready to serve requests are tracked via the ClusterStatus fields Members.Ready and Members.Unready

Changed

• PodPolicy resourceRequirements field is renamed to resources
• Default etcd version is changed to 3.1.2
• Self-hosted etcd pod uses hostPath with path /var/etcd/$ns-$member

Fixed

• [GH-851] Fixed a race that caused nil pointer access panic
• [GH-823] Fixed backup service status not shown in TPR status