theshire/.taskfiles/volsync/Taskfile.yaml

---
# yaml-language-server: $schema=https://taskfile.dev/schema.json
version: '3'

# This taskfile is used to manage certain VolSync tasks for a given application, limitations are described below.
#   1. Fluxtomization, HelmRelease, PVC, ReplicationSource all have the same name (e.g. plex)
#   2. ReplicationSource and ReplicationDestination are a Restic repository
#   3. Applications are deployed as either a Kubernetes Deployment or StatefulSet
#   4. Each application only has one PVC that is being replicated

vars:
  VOLSYNC_RESOURCES_DIR: '{{.ROOT_DIR}}/.taskfiles/volsync/resources'

tasks:

  state-*:
    desc: Suspend or Resume Volsync
    summary: |-
      CLUSTER: Cluster to run command against (default: main)
      STATE: resume or suspend (required)      
    cmds:
      # - until kubectl wait jobs --all --all-namespaces --for=condition=complete --timeout=5m &>/dev/null; do sleep 5; done
      - flux {{.STATE}} kustomization volsync
      - flux --namespace {{.NS}} {{.STATE}} helmrelease volsync
      - kubectl --namespace {{.NS}} scale deployment --all --replicas {{if eq .STATE "suspend"}}0{{else}}1{{end}}
    vars:
      NS: '{{.NS | default "volsync-system"}}'
      STATE: '{{index .MATCH 0}}'
    requires:
      vars: [CLUSTER]

  unlock:
    desc: Unlock all Restic repositories
    summary: |-
      CLUSTER: Cluster to run command against (default: main)      
    cmd: >
      kubectl get replicationsources --all-namespaces --no-headers -A | awk '{print $1, $2}'
      | xargs --max-procs=2 -l bash -c 'kubectl --namespace "$0" patch --field-manager=flux-client-side-apply replicationsources "$1" --type merge --patch "{\"spec\":{\"restic\":{\"unlock\":\"{{now | unixEpoch}}\"}}}"'      
    requires:
      vars: [CLUSTER]

  # To run backup jobs in parallel for all replicationsources:
  #   - kubectl get replicationsources --all-namespaces --no-headers | awk '{print $2, $1}' | xargs --max-procs=4 -l bash -c 'task volsync:snapshot APP=$0 NS=$1'
  snapshot:
    desc: Snapshot an application
    summary: |-
      CLUSTER: Cluster to run command against (default: main)
      NS: Namespace the application is in (default: default)
      APP: Application to snapshot (required)      
    cmds:
      - kubectl --namespace {{.NS}} patch replicationsources {{.APP}} --type merge -p '{"spec":{"trigger":{"manual":"{{now | unixEpoch}}"}}}'
      - until kubectl --namespace {{.NS}} get job/{{.JOB}} &>/dev/null; do sleep 5; done
      - kubectl --namespace {{.NS}} wait job/{{.JOB}} --for=condition=complete --timeout=120m
    vars:
      NS: '{{.NS | default "default"}}'
      JOB: volsync-src-{{.APP}}
    requires:
      vars: [CLUSTER, APP]
    preconditions:
      - kubectl --namespace {{.NS}} get replicationsources {{.APP}}

  # To run restore jobs in parallel for all replicationdestinations:
  #    - kubectl get replicationsources --all-namespaces --no-headers | awk '{print $2, $1}' | xargs --max-procs=4 -l bash -c 'task volsync:restore APP=$0 NS=$1'
  restore:
    desc: Restore an application
    summary: |-
      CLUSTER: Cluster to run command against (default: main)
      NS: Namespace the application is in (default: default)
      APP: Application to restore (required)
      PREVIOUS: Previous number of snapshots to restore (default: 2)      
    cmds:
      - task: .suspend
      - task: .restore
      - task: .resume
    requires:
      vars: [CLUSTER, APP]

  .suspend:
    internal: true
    cmds:
      - flux --namespace flux-system suspend kustomization {{.APP}}
      - flux --namespace {{.NS}} suspend helmrelease {{.APP}}
      - kubectl --namespace {{.NS}} scale {{.CONTROLLER}}/{{.APP}} --replicas 0
      - kubectl --namespace {{.NS}} wait pod --for=delete --selector="app.kubernetes.io/name={{.APP}}" --timeout=5m
    vars:
      NS: '{{.NS | default "default"}}'
      APP: '{{.APP}}'
      CONTROLLER:
        sh: kubectl --namespace {{.NS}} get deployment {{.APP}} &>/dev/null && echo deployment || echo statefulset

  .restore:
    internal: true
    cmds:
      - minijinja-cli --env --trim-blocks --lstrip-blocks --autoescape=none {{.VOLSYNC_RESOURCES_DIR}}/replicationdestination.yaml.j2 | kubectl apply --server-side --filename -
      - until kubectl --namespace {{.NS}} get job/{{.JOB}} &>/dev/null; do sleep 5; done
      - kubectl --namespace {{.NS}} wait job/{{.JOB}} --for=condition=complete --timeout=120m
      - kubectl --namespace {{.NS}} delete replicationdestination {{.JOB}}
    vars:
      NS: '{{.NS | default "default"}}'
      JOB: volsync-dst-{{.APP}}
      PREVIOUS: '{{.PREVIOUS | default 2}}'
      CLAIM:
        sh: kubectl --namespace {{.NS}} get replicationsources/{{.APP}} --output=jsonpath="{.spec.sourcePVC}"
      ACCESS_MODES:
        sh: kubectl --namespace {{.NS}} get replicationsources/{{.APP}} --output=jsonpath="{.spec.restic.accessModes}"
      STORAGE_CLASS_NAME:
        sh: kubectl --namespace {{.NS}} get replicationsources/{{.APP}} --output=jsonpath="{.spec.restic.storageClassName}"
      PUID:
        sh: kubectl --namespace {{.NS}} get replicationsources/{{.APP}} --output=jsonpath="{.spec.restic.moverSecurityContext.runAsUser}"
      PGID:
        sh: kubectl --namespace {{.NS}} get replicationsources/{{.APP}} --output=jsonpath="{.spec.restic.moverSecurityContext.runAsGroup}"
    env:
      NS: '{{.NS}}'
      JOB: '{{.JOB}}'
      APP: '{{.APP}}'
      PREVIOUS: '{{.PREVIOUS}}'
      CLAIM: '{{.CLAIM}}'
      ACCESS_MODES: '{{.ACCESS_MODES}}'
      STORAGE_CLASS_NAME: '{{.STORAGE_CLASS_NAME}}'
      PUID: '{{.PUID}}'
      PGID: '{{.PGID}}'
    preconditions:
      - test -f {{.VOLSYNC_RESOURCES_DIR}}/replicationdestination.yaml.j2

  .resume:
    internal: true
    cmds:
      - flux --namespace {{.NS}} resume helmrelease {{.APP}}
      - flux --namespace flux-system resume kustomization {{.APP}}
      - kubectl --namespace {{.NS}} scale {{.CONTROLLER}}/{{.APP}} --replicas 1
      - kubectl --namespace {{.NS}} wait pod --for=condition=ready --selector="app.kubernetes.io/name={{.APP}}" --timeout=5m
    vars:
      NS: '{{.NS | default "default"}}'
      APP: '{{.APP}}'
      CONTROLLER:
        sh: kubectl --namespace {{.NS}} get deployment {{.APP}} &>/dev/null && echo deployment || echo statefulset