auricom-home-cluster/.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

x-env: &env
  app: '{{.app}}'
  controller: '{{.controller}}'
  claim: '{{.claim}}'
  puid: '{{.puid}}'
  pgid: '{{.pgid}}'
  ns: '{{.ns}}'
  previous: '{{.previous}}'
  ts: '{{.ts}}'

vars:
  scriptsDir: '{{.ROOT_DIR}}/.taskfiles/volsync/scripts'
  templatesDir: '{{.ROOT_DIR}}/.taskfiles/volsync/templates'
  ts: '{{now | date "150405"}}'

tasks:
  list:
    desc: List snapshots for an application
    summary: |
      Args:
        ns: Namespace the PVC is in (default: default)
        app: Application to list snapshots for (required)
    cmds:
      - envsubst < <(cat {{.templatesDir}}/list.tmpl.yaml) | kubectl apply -f -
      - bash {{.scriptsDir}}/wait.sh list-{{.app}}-{{.ts}} {{.ns}}
      - kubectl -n {{.ns}} wait job/list-{{.app}}-{{.ts}} --for condition=complete --timeout=1m
      - kubectl -n {{.ns}} logs job/list-{{.app}}-{{.ts}} --container list
      - kubectl -n {{.ns}} delete job list-{{.app}}-{{.ts}}
    env: *env
    requires:
      vars: ["app"]
    vars:
      ns: '{{.ns | default "default"}}'
    preconditions:
      - { msg: "Wait script not found", sh: "test -f {{.scriptsDir}}/wait.sh" }
      - { msg: "List template not found", sh: "test -f {{.templatesDir}}/list.tmpl.yaml" }
    silent: true

  unlock:
    desc: Unlock a Restic repository for an application
    summary: |
      Args:
        ns: Namespace the PVC is in (default: default)
        app: Application to unlock (required)
    cmds:
      - envsubst < <(cat {{.templatesDir}}/unlock.tmpl.yaml) | kubectl apply -f -
      - bash {{.scriptsDir}}/wait.sh unlock-{{.app}}-{{.ts}} {{.ns}}
      - kubectl -n {{.ns}} wait job/unlock-{{.app}}-{{.ts}} --for condition=complete --timeout=1m
      - kubectl -n {{.ns}} logs job/{{.app}}-unlock-snapshots-{{.ts}} --container unlock-minio
      - kubectl -n {{.ns}} logs job/{{.app}}-unlock-snapshots-{{.ts}} --container unlock-r2
      - kubectl -n {{.ns}} delete job unlock-{{.app}}-{{.ts}}
    env: *env
    requires:
      vars: ["app"]
    vars:
      ns: '{{.ns | default "default"}}'
    preconditions:
      - { msg: "Wait script not found", sh: "test -f {{.scriptsDir}}/wait.sh" }
      - { msg: "Unlock template not found", sh: "test -f {{.templatesDir}}/unlock.tmpl.yaml" }
    silent: true

  # 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 a PVC for an application
    summary: |
      Args:
        ns: Namespace the PVC is in (default: default)
        app: Application to snapshot (required)
    cmds:
      # # TODO: Only suspend when a arg is passed to the task
      # # Ref: https://github.com/go-task/task/issues/608
      # - task: .suspend
      #   vars: *env
      - kubectl -n {{.ns}} patch replicationsources {{.app}} --type merge -p '{"spec":{"trigger":{"manual":"{{.ts}}"}}}'
      - bash {{.scriptsDir}}/wait.sh volsync-src-{{.app}} {{.ns}}
      - kubectl -n {{.ns}} wait job/volsync-src-{{.app}} --for condition=complete --timeout=120m
    env: *env
    requires:
      vars: ["app"]
    vars:
      ns: '{{.ns | default "default"}}'
      controller:
        sh: true && {{.scriptsDir}}/controller.sh {{.app}} {{.ns}}
    preconditions:
      - { msg: "Controller script not found", sh: "test -f {{.scriptsDir}}/controller.sh" }
      - { msg: "Wait script not found", sh: "test -f {{.scriptsDir}}/wait.sh" }
      - { msg: "RepositorySource not found", sh: "kubectl -n {{.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 a PVC for an application
    summary: |
      Args:
        ns: Namespace the PVC is in (default: default)
        app: Application to restore (required)
        previous: Previous number of snapshots to restore (default: 2)
    cmds:
      - task: .suspend
        vars: *env
      - task: .wipe
        vars: *env
      - task: .restore
        vars: *env
      - task: .resume
        vars: *env
    env: *env
    requires:
      vars: ["app"]
    vars:
      ns: '{{.ns | default "default"}}'
      previous: '{{.previous | default 2}}'
      controller:
        sh: "{{.scriptsDir}}/controller.sh {{.app}} {{.ns}}"
      claim:
        sh: kubectl -n {{.ns}} get replicationsources/{{.app}} -o jsonpath="{.spec.sourcePVC}"
      puid:
        sh: kubectl -n {{.ns}} get replicationsources/{{.app}} -o jsonpath="{.spec.restic.moverSecurityContext.runAsUser}"
      pgid:
        sh: kubectl -n {{.ns}} get replicationsources/{{.app}} -o jsonpath="{.spec.restic.moverSecurityContext.runAsGroup}"
    preconditions:
      - { msg: "Controller script not found", sh: "test -f {{.scriptsDir}}/controller.sh" }
      - { msg: "Wait script not found", sh: "test -f {{.scriptsDir}}/wait.sh" }
      - { msg: "ReplicationDestination script not found", sh: "test -f {{.templatesDir}}/replicationdestination.tmpl.yaml" }
      - { msg: "Wipe template not found", sh: "test -f {{.templatesDir}}/wipe.tmpl.yaml" }

  cleanup:
    desc: Delete volume populator PVCs in all namespaces
    cmds:
      - for: { var: dest }
        cmd: |
          {{- $items := (split "/" .ITEM) }}
          kubectl delete pvc -n {{ $items._0 }} {{ $items._1 }}
      - for: { var: cache }
        cmd: |
          {{- $items := (split "/" .ITEM) }}
          kubectl delete pvc -n {{ $items._0 }} {{ $items._1 }}
    env: *env
    vars:
      dest:
        sh: kubectl get pvc --all-namespaces --no-headers | grep "dst-dest" | awk '{print $1 "/" $2}'
      cache:
        sh: kubectl get pvc --all-namespaces --no-headers | grep "dst-cache" | awk '{print $1 "/" $2}'

  # Suspend the Flux ks and hr
  .suspend:
    internal: true
    cmds:
      - flux -n flux-system suspend kustomization {{.app}}
      - flux -n {{.ns}} suspend helmrelease {{.app}}
      - kubectl -n {{.ns}} scale {{.controller}} --replicas 0
      - kubectl -n {{.ns}} wait pod --for delete --selector="app.kubernetes.io/name={{.app}}" --timeout=2m
    env: *env

  # Wipe the PVC of all data
  .wipe:
    internal: true
    cmds:
      - envsubst < <(cat {{.templatesDir}}/wipe.tmpl.yaml) | kubectl apply -f -
      - bash {{.scriptsDir}}/wait.sh wipe-{{.app}}-{{.ts}} {{.ns}}
      - kubectl -n {{.ns}} wait job/wipe-{{.app}}-{{.ts}} --for condition=complete --timeout=120m
      - kubectl -n {{.ns}} logs job/wipe-{{.app}}-{{.ts}} --container wipe
      - kubectl -n {{.ns}} delete job wipe-{{.app}}-{{.ts}}
    env: *env

  # Create VolSync replicationdestination CR to restore data
  .restore:
    internal: true
    cmds:
      - envsubst < <(cat {{.templatesDir}}/replicationdestination.tmpl.yaml) | kubectl apply -f -
      - bash {{.scriptsDir}}/wait.sh volsync-dst-{{.app}}-{{.ts}} {{.ns}}
      - kubectl -n {{.ns}} wait job/volsync-dst-{{.app}}-{{.ts}} --for condition=complete --timeout=120m
      - kubectl -n {{.ns}} delete replicationdestination {{.app}}-{{.ts}}
    env: *env

  # Resume Flux ks and hr
  .resume:
    internal: true
    cmds:
      - flux -n {{.ns}} resume helmrelease {{.app}}
      - flux -n flux-system resume kustomization {{.app}}
    env: *env