New to Stash? Please start here.


What is RestoreSession

A RestoreSession is a Kubernetes CustomResourceDefinition(CRD) which specifies a target to restore and the source of data that will be restored in a Kubernetes native way.

You have to create a RestoreSession object whenever you want to restore. When a RestoreSession object is created, Stash injects an init-container into the target workload and restarts it. The init-container restores the desired data. If the target is a database or a stand-alone PVC, Stash launches a job to perform the restore process.

RestoreSession CRD Specification

Like any official Kubernetes resource, a RestoreSession has TypeMeta, ObjectMeta, Spec and Status sections.

A sample RestoreSession object to restore backed up data of a StatefulSet is shown below:

kind: RestoreSession
  name: statefulset-restore
  namespace: demo
  driver: Restic
    name: local-repo
  # task:
  #   name: workload-restore # task field is not required for workload data backup but it is necessary for database backup.
      apiVersion: apps/v1
      kind: StatefulSet
      name: recovered-statefulset
    - mountPath: /source/data
      name: source-data
  - targetHosts: ["host-3","host-4"] # "host-3" and "host-4" will have restored data of backed up host "host-1"
    sourceHost: "host-1" # source host
    - /source/data
  - targetHosts: [] # empty host match all hosts
    sourceHost: "" # no source host indicates that the host is pod itself
    - /source/data
        - /bin/sh
        - -c
        - echo "Sample PreRestore hook demo"
      containerName: my-app-container
        - /bin/sh
        - -c
        - echo "Sample PostRestore hook demo"
      containerName: my-app-container
          memory: 256M
          memory: 256M
        runAsGroup: 2000
        runAsUser: 2000
        class: 2
        classData: 4
        adjustment: 5
      - name: my-private-registry-secret
      serviceAccountName: my-restore-svc
    disableCache: false
    medium: Memory
    sizeLimit: 2Gi
  totalHosts: 5
  phase: Succeeded
  sessionDuration: 4.148288404s
  - duration: 884.431745ms
    hostname: host-1
    phase: Succeeded
  - duration: 769.924342ms
    hostname: host-2
    phase: Succeeded
  - duration: 868.694738ms
    hostname: host-3
    phase: Succeeded
  - duration: 792.097784ms
    hostname: host-4
    phase: Succeeded
  - duration: 833.139795ms
    hostname: host-0
    phase: Succeeded

Here, we are going to describe the various sections of a RestoreSession object.

RestoreSession Spec

A RestoreSession object has the following fields in the spec section.


spec.driver indicates the mechanism used to restore a target. Currently, Stash supports Restic and VolumeSnapshotter as drivers. The default value of this field is Restic.

ResticUsed to restore workload data, persistent volumes data and databases. It uses restic to restore the target.
VolumeSnapshotterUsed to initialize PersistentVolumeClaim from VolumeSnapshot. It leverages Kubernetes VolumeSnapshot crd and CSI driver to initialize the target PVCs from respective snapshots. Currently, it can restore only in new PVC. field indicates the target where data will be restored. This section consists of the following fields:

  • : refers to the restore target. You have to specify apiVersion, kind and name of the target. Stash will use this information to inject an init-container or to create a restore job.

  • : specifies a list of volumes and their mountPath where the data will be restored. Stash will mount these volumes inside the init-container or restore job.

Note: Stash stores absolute path of the backed up files. Hence, your restored volume must be mounted on the same mountPath as the original volume. Otherwise, the backed up files will not be restored into your desired volume.

  • : You can specify a list of PVC template using field. Stash will create those PVCs then it will restore the desired data into them. Then, you can use those PVCs to deploy your desired workload.

  • : If you want to restore the volumes of a StatefulSet through field, you can specify the number of replicas of the StatefulSet using In this case, you have to use ${POD_ORDINAL} variable suffix in the claim name. Stash will replace that variable with respective ordinal and it will create the volumes for each replica. For more details, please visit here.

spec.repository indicates the Repository crd name that holds necessary backend information from where data will be restored.


spec.task specifies the name and parameters of the Task crd to use to restore the target data.

  • indicates the name of the Task template to use for this restore process.
  • spec.task.params: spec.task.params is an array of custom parameters to use to configure the task.

spec.task section is not necessary for restoring workload data (i.e. Deployment, DaemonSet, StatefulSet etc.). However, it is necessary for restoring database and stand-alone PVC.


spec.rules is an array of restore rules that specify how Stash should restore data for each host. For example, Stash runs restore process in all pod’s of a StatefulSet. You can configure this spec.rules section to control what data will be restored into which pod.

Each restore rule has the following fields:

  • targetHosts : targetHosts field contains a list of host names which are subject to this rule. If targetHosts field is empty, this rule applies to all hosts for which there is no specific rule. In the sample RestoreSession given above, the first rule applies to only host-3 and host-4 and the second rule is applicable to all hosts.
  • sourceHost : sourceHost specifies the name of host whose backed up data will be restored by this rule. In the sample RestoreSession, the first rule specify that backed up data of host-0 (i.e. pod-0 of old StatefulSet) will be restored into host-3 and host-4 (i.e. pod-3 and pod-4 of new StatefulSet). If you keep sourceHost field empty as the second rule of the above example, data from a similar restore host will be restored on the respective restore host. That means, backed up data of host-0 will be restored into host-0, backed up data of host-1 will be restored into host-1 and so on.
  • paths : paths specifies a list of file paths that will be restored into the hosts who are subject to this rule.
  • snapshots : snapshots specifies the list of snapshots that will be restored into the hosts who are subject to this rule. If you don’t specify snapshot field, latest snasphot of the file paths specified in paths section will be restored.

Restore rules comply with the following conditions:

  • There could be at most one rule with empty targetHosts field.
  • No two rules with non-emtpy targetHosts can’t be matched for a single host.
  • Stash restore only one file path in a single snapshot. So, if you specify snapshots field in a rule, you can’t specify paths field as it may cause restore failure if a file path wasn’t backed up in the snapshot specified in the snapshots field.
  • If no rule matches for a host, no data will be restored on that host.
  • The order of the rules does not have any effect on the restore process.


spec.hooks allows performing some actions before and after the restore process. You can send HTTP requests to a remote server via httpGet or httpPost hooks. You can check whether a TCP socket is open using tcpSocket hook. You can also execute some commands into your application pod using exec hook.

  • spec.hooks.preRestore: spec.hooks.preRestore hooks are executed before the restore process.
  • spec.hooks.postRestore: spec.hooks.postRestore hooks are executed after the restore process.

For more details on how hooks work in Stash and how to configure different types of hook, please visit here.


spec.runtimeSettings allows to configure runtime environment for restore init-container or job. You can specify runtime settings in both pod level and container level.

  • spec.runtimeSettings.container

    spec.runtimeSettings.container is used to configure restore init-container/job in container level. You can configure the following container level parameters,

resourcesCompute resources required by restore init-container or restore job. To know how to manage resources for containers, please visit here.
livenessProbePeriodic probe of restore init-container/job’s container liveness. Container will be restarted if the probe fails.
readinessProbePeriodic probe of restore init-container/job’s container readiness. Container will be removed from service endpoints if the probe fails.
lifecycleActions that the management system should take in response to container lifecycle events.
securityContextSecurity options that restore init-container/job’s container should run with. For more details, please visit here.
niceSet CPU scheduling priority for the restore process. For more details about nice, please visit here.
ioniceSet I/O scheduling class and priority for the restore process. For more details about ionice, please visit here.
envA list of the environment variables to set in the restore init-container/job’s container.
envFromThis allows to set environment variables to the the restore init-container/job’s container from a Secret or ConfigMap.
  • spec.runtimeSettings.pod

    spec.runtimeSettings.pod is used to configure restore job in pod level. You can configure following pod level parameters,

serviceAccountNameName of the ServiceAccount to use for restore job. Stash init-container will use the same ServiceAccount as the target.
nodeSelectorSelector which must be true for restore job pod to fit on a node.
automountServiceAccountTokenIndicates whether a service account token should be automatically mounted into the restore job’s pod.
nodeNameNodeName is used to request to schedule restore job’s pod onto a specific node.
securityContextSecurity options that restore job’s pod should run with. For more details, please visit here.
imagePullSecretsA list of secret names in the same namespace that will be used to pull image from private docker registry. For more details, please visit here.
affinityAffinity and anti-affinity to schedule restore job’s pod in the desired node. For more details, please visit here.
schedulerNameName of the scheduler that should dispatch the restore job.
tolerationsTaints and Tolerations to ensure that restore job’s pod is not scheduled in inappropriate nodes. For more details about toleration, please visit here.
priorityClassNameIndicates the restore job pod’s priority class. For more details, please visit here.
priorityIndicates the restore job pod’s priority value.
readinessGatesSpecifies additional conditions to be evaluated for Pod readiness. For more details, please visit here.
runtimeClassNameRuntimeClass is used for selecting the container runtime configuration. For more details, please visit here
enableServiceLinksEnableServiceLinks indicates whether information about services should be injected into pod’s environment variables.


Stash mounts an emptyDir for holding temporary files. It is also used for caching for faster restore performance. You can configure the emptyDir using spec.tempDir section. You can also disable caching using this field. The following fields are configurable in spec.tempDir section:

  • spec.tempDir.medium : Specifies the type of storage medium should back this file path.
  • spec.tempDir.sizeLimit : Maximum limit of storage for this volume.
  • spec.tempDir.disableCaching : Disable caching while restoring. This may negatively impact restore performance. This field is set to false by default.


For some targets (i.e. some databases), Stash can’t directly pipe the restored data into the target. In this case, it has to store the restored data temporarily before injecting into the target. spec.interimVolumeTemplate specifies a PVC template for holding those data temporarily. Stash will create a PVC according to the template and use it to store the data temporarily. This PVC will be deleted automatically if you delete the RestoreSession.

Note that the usage of this field is different than spec.tempDir which is used for caching purpose. Stash has introduced this field because the emptyDir volume that is used for spec.tempDir does not play nice with large databases( i.e. 100Gi database). Also, it provides debugging capability as Stash keeps it until you delete the RestoreSession.

RestoreSession Status

.status section of RestoreSession shows progress, stats and overall phase of the restore process. The restore init-container or job adds its respective stats in .status section after it completes its task. .status section consists of the following fields:


Not every pods or replica of the target will run restore process. Thus, we refer those entities that runs restore process as a host. status.totalHosts specifies the total number of hosts that will run restore process for this RestoreSession. For more details on how many hosts will run restore process for which types of workload, please visit here.


status.phase indicates the overall phase of the restore process for this RestoreSession. status.phase will be Succeeded only if the phase of all hosts are Succeeded. If any of the hosts fail to complete restore, status.phase will be Failed.


status.sessionDuration indicates the total time taken to complete the restoration of all hosts.


status.stats section is an array of restore statistics of individual hosts. Each host adds their statistics in this array after completing their restore process. This field is only available for the Restic driver but not available for the VolumeSnapshotter driver. The default value of the driver is Restic.

Individual host stats entry consists of the following fields:

  • hostname : hostname indicates the name of the host.
  • phase : phase indicates the restore phase of this host.
  • duration : duration indicates the total time taken to complete restore process for this host.
  • error : error shows the reason for failure if the restore process fails for this host.

Hosts of a restore process

Stash uses two different models for restoring backed up data depending on the target type. It uses init-container model for Kubernetes workloads and job model for rest of the targets. In the init-container model, Stash injects an init-container inside the targeted workload and the init-container is responsible for restoring the desired data on workload restart. In the job model, Stash launches a job to restore the desired data.

Stash uses an identifier called host to identify the entity where the restore process will be run. This host identification process depends on the restore model and the target types. The restore strategy and host identification strategy for different types of target is explained below.

Kubernetes Workloads:

Stash uses init-container model to restore Kubernetes workloads. However, not every init-container will run restore process. How many init-containers will run restore process depends on the type of the workload. We can divide them into the following categories:

  • Deployment, ReplicaSet and ReplicationController: For these types of stateless workloads, all the replicas mount the same volumes. So, restoring into only one replica is enough. In this case, Stash uses leader election to elect the leader pod. Only the init-container inside the leader pod runs the restore process. This leader pod is identified as host-0. The total number of hosts for these types of workloads is 1.
  • StatefulSet: Every replica of a StatefulSet mounts different volumes. So, restoring into each replica is necessary. In this case, init-container inside each replica runs the restore process. Stash identifies pod-0 as host-0, pod-1 as host-1, pod-2 as host-2 and so on. Hence, the total number of hosts for a StatefulSet is the number of replicas.
  • DaemonSet: Daemon replicas on every node may contain different data. So, restoring into each daemon pod is necessary. In this case, init-container inside each daemon pod runs the restore process. Stash considers the individual daemon pod as a separate host and the node name where the daemon pod is running act as their host identifier. The total number of hosts for a DaemonSet is the number of daemon pod running in the cluster.

Stand-alone PVC:

Stash uses job model to restore a stand-alone PVC. Stash launches a job to restore into the targeted PVC. This job is identified as host-0. In this case, the total number of host is 1.


Stash uses job model to restore a database. Stash launches a job to restore into the targeted database. In this case, the number of hosts depends on the database type.

  • Stand-alone database: For stand-alone database, the restore target is identified as host-0 and the total number of host is 1.
  • Replicated cluster: For replicated clustered database such as MongoDB ReplicaSet, all the replicas contain the same data. In this case, Stash restores same data into each replica. Thus, the total number of host is 1 and it is identified as host-0.
  • Sharded cluster: For sharded database cluster, Stash restores the backed up data of individual shard into the respective shard. Hence, the number of hosts for a sharded database is the number of shards and they are identified as host-0, host-1, host-2, etc. However, the number of hosts may increase based on the database type.


Stash uses job model for restoring volume from volume snapshots. Each volume is considered a different host and they are identified by their name. Hence, the number of total hosts is the number of targeted volumes to be restored.

Restore using volumeClaimTemplates:

If volumeClaimTemplates is specified in a RestoreSession, Stash creates the PVCs according to the template then it launches one job for each replica specified by field. In this case, the total number of hosts is number of replicas specified by field. If this field is not provided then the total number of hosts is 1.

Next Steps

  • Learn how restore of workloads data works from here.
  • Learn how restore of databases works from here.
  • Learn how restore stand-alone PVC works from here.