@xen-orchestra/backups 0.23.0 → 0.24.0

package/Backup.js

@@ -6,7 +6,7 @@ const ignoreErrors = require('promise-toolbox/ignoreErrors')
 const { compileTemplate } = require('@xen-orchestra/template')
 const { limitConcurrency } = require('limit-concurrency-decorator')
 
-const { extractIdsFromSimplePattern } = require('./_extractIdsFromSimplePattern.js')
+const { extractIdsFromSimplePattern } = require('./extractIdsFromSimplePattern.js')
 const { PoolMetadataBackup } = require('./_PoolMetadataBackup.js')
 const { Task } = require('./Task.js')
 const { VmBackup } = require('./_VmBackup.js')
@@ -24,6 +24,34 @@ const getAdaptersByRemote = adapters => {
 
 const runTask = (...args) => Task.run(...args).catch(noop) // errors are handled by logs
 
+const DEFAULT_SETTINGS = {
+  reportWhen: 'failure',
+}
+
+const DEFAULT_VM_SETTINGS = {
+  bypassVdiChainsCheck: false,
+  checkpointSnapshot: false,
+  concurrency: 2,
+  copyRetention: 0,
+  deleteFirst: false,
+  exportRetention: 0,
+  fullInterval: 0,
+  healthCheckSr: undefined,
+  healthCheckVmsWithTags: [],
+  maxMergedDeltasPerRun: 2,
+  offlineBackup: false,
+  offlineSnapshot: false,
+  snapshotRetention: 0,
+  timeout: 0,
+  unconditionalSnapshot: false,
+  vmTimeout: 0,
+}
+
+const DEFAULT_METADATA_SETTINGS = {
+  retentionPoolMetadata: 0,
+  retentionXoMetadata: 0,
+}
+
 exports.Backup = class Backup {
   constructor({ config, getAdapter, getConnectedRecord, job, schedule }) {
     this._config = config
@@ -42,17 +70,22 @@ exports.Backup = class Backup {
       '{job.name}': job.name,
       '{vm.name_label}': vm => vm.name_label,
     })
-  }
 
-  run() {
-    const type = this._job.type
+    const { type } = job
+    const baseSettings = { ...DEFAULT_SETTINGS }
     if (type === 'backup') {
-      return this._runVmBackup()
+      Object.assign(baseSettings, DEFAULT_VM_SETTINGS, config.defaultSettings, config.vm?.defaultSettings)
+      this.run = this._runVmBackup
     } else if (type === 'metadataBackup') {
-      return this._runMetadataBackup()
+      Object.assign(baseSettings, DEFAULT_METADATA_SETTINGS, config.defaultSettings, config.metadata?.defaultSettings)
+      this.run = this._runMetadataBackup
     } else {
       throw new Error(`No runner for the backup type ${type}`)
     }
+    Object.assign(baseSettings, job.settings[''])
+
+    this._baseSettings = baseSettings
+    this._settings = { ...baseSettings, ...job.settings[schedule.id] }
   }
 
   async _runMetadataBackup() {
@@ -64,13 +97,6 @@ exports.Backup = class Backup {
     }
 
     const config = this._config
-    const settings = {
-      ...config.defaultSettings,
-      ...config.metadata.defaultSettings,
-      ...job.settings[''],
-      ...job.settings[schedule.id],
-    }
-
     const poolIds = extractIdsFromSimplePattern(job.pools)
     const isEmptyPools = poolIds.length === 0
     const isXoMetadata = job.xoMetadata !== undefined
@@ -78,6 +104,8 @@ exports.Backup = class Backup {
       throw new Error('no metadata mode found')
     }
 
+    const settings = this._settings
+
     const { retentionPoolMetadata, retentionXoMetadata } = settings
 
     if (
@@ -189,14 +217,7 @@ exports.Backup = class Backup {
     const schedule = this._schedule
 
     const config = this._config
-    const { settings } = job
-    const scheduleSettings = {
-      ...config.defaultSettings,
-      ...config.vm.defaultSettings,
-      ...settings[''],
-      ...settings[schedule.id],
-    }
-
+    const settings = this._settings
     await Disposable.use(
       Disposable.all(
         extractIdsFromSimplePattern(job.srs).map(id =>
@@ -224,14 +245,15 @@ exports.Backup = class Backup {
           })
         )
       ),
-      async (srs, remoteAdapters) => {
+      () => settings.healthCheckSr !== undefined ? this._getRecord('SR', settings.healthCheckSr) : undefined,
+      async (srs, remoteAdapters, healthCheckSr) => {
         // remove adapters that failed (already handled)
         remoteAdapters = remoteAdapters.filter(_ => _ !== undefined)
 
         // remove srs that failed (already handled)
         srs = srs.filter(_ => _ !== undefined)
 
-        if (remoteAdapters.length === 0 && srs.length === 0 && scheduleSettings.snapshotRetention === 0) {
+        if (remoteAdapters.length === 0 && srs.length === 0 && settings.snapshotRetention === 0) {
           return
         }
 
@@ -241,23 +263,27 @@ exports.Backup = class Backup {
 
         remoteAdapters = getAdaptersByRemote(remoteAdapters)
 
+        const allSettings = this._job.settings
+        const baseSettings = this._baseSettings
+
         const handleVm = vmUuid =>
           runTask({ name: 'backup VM', data: { type: 'VM', id: vmUuid } }, () =>
             Disposable.use(this._getRecord('VM', vmUuid), vm =>
               new VmBackup({
+                baseSettings,
                 config,
                 getSnapshotNameLabel,
+                healthCheckSr,
                 job,
-                // remotes,
                 remoteAdapters,
                 schedule,
-                settings: { ...scheduleSettings, ...settings[vmUuid] },
+                settings: { ...settings, ...allSettings[vm.uuid] },
                 srs,
                 vm,
               }).run()
             )
           )
-        const { concurrency } = scheduleSettings
+        const { concurrency } = settings
         await asyncMapSettled(vmIds, concurrency === 0 ? handleVm : limitConcurrency(concurrency)(handleVm))
       }
     )

package/HealthCheckVmBackup.js

@@ -0,0 +1,64 @@
+'use strict'
+
+const { Task } = require('./Task')
+
+exports.HealthCheckVmBackup = class HealthCheckVmBackup {
+  #xapi
+  #restoredVm
+
+  constructor({ restoredVm, xapi }) {
+    this.#restoredVm = restoredVm
+    this.#xapi = xapi
+  }
+
+  async run() {
+    return Task.run(
+      {
+        name: 'vmstart',
+      },
+      async () => {
+        let restoredVm = this.#restoredVm
+        const xapi = this.#xapi
+        const restoredId = restoredVm.uuid
+
+        // remove vifs
+        await Promise.all(restoredVm.$VIFs.map(vif => xapi.callAsync('VIF.destroy', vif.$ref)))
+
+        const start = new Date()
+        // start Vm
+
+        await xapi.callAsync(
+          'VM.start',
+          restoredVm.$ref,
+          false, // Start paused?
+          false // Skip pre-boot checks?
+        )
+        const started = new Date()
+        const timeout = 10 * 60 * 1000
+        const startDuration = started - start
+
+        let remainingTimeout = timeout - startDuration
+
+        if (remainingTimeout < 0) {
+          throw new Error(`VM ${restoredId} not started after ${timeout / 1000} second`)
+        }
+
+        // wait for the 'Running' event to be really stored in local xapi object cache
+        restoredVm = await xapi.waitObjectState(restoredVm.$ref, vm => vm.power_state === 'Running', {
+          timeout: remainingTimeout,
+        })
+
+        const running = new Date()
+        remainingTimeout -= running - started
+
+        if (remainingTimeout < 0) {
+          throw new Error(`local xapi  did not get Runnig state for VM ${restoredId} after ${timeout / 1000} second`)
+        }
+        // wait for the guest tool version to be defined
+        await xapi.waitObjectState(restoredVm.guest_metrics, gm => gm?.PV_drivers_version?.major !== undefined, {
+          timeout: remainingTimeout,
+        })
+      }
+    )
+  }
+}

package/RemoteAdapter.js

@@ -1,6 +1,7 @@
 'use strict'
 
 const { asyncMap, asyncMapSettled } = require('@xen-orchestra/async-map')
+const { synchronized } = require('decorator-synchronized')
 const Disposable = require('promise-toolbox/Disposable')
 const fromCallback = require('promise-toolbox/fromCallback')
 const fromEvent = require('promise-toolbox/fromEvent')
@@ -9,7 +10,7 @@ const groupBy = require('lodash/groupBy.js')
 const pickBy = require('lodash/pickBy.js')
 const { dirname, join, normalize, resolve } = require('path')
 const { createLogger } = require('@xen-orchestra/log')
-const { Constants, createVhdDirectoryFromStream, openVhd, VhdAbstract, VhdDirectory, VhdSynthetic } = require('vhd-lib')
+const { createVhdDirectoryFromStream, openVhd, VhdAbstract, VhdDirectory, VhdSynthetic } = require('vhd-lib')
 const { deduped } = require('@vates/disposable/deduped.js')
 const { decorateMethodsWith } = require('@vates/decorate-with')
 const { compose } = require('@vates/compose')
@@ -17,6 +18,7 @@ const { execFile } = require('child_process')
 const { readdir, stat } = require('fs-extra')
 const { v4: uuidv4 } = require('uuid')
 const { ZipFile } = require('yazl')
+const zlib = require('zlib')
 
 const { BACKUP_DIR } = require('./_getVmBackupDir.js')
 const { cleanVm } = require('./_cleanVm.js')
@@ -78,6 +80,7 @@ class RemoteAdapter {
     this._dirMode = dirMode
     this._handler = handler
     this._vhdDirectoryCompression = vhdDirectoryCompression
+    this._readCacheListVmBackups = synchronized.withKey()(this._readCacheListVmBackups)
   }
 
   get handler() {
@@ -261,7 +264,8 @@ class RemoteAdapter {
   }
 
   async deleteVmBackups(files) {
-    const { delta, full, ...others } = groupBy(await asyncMap(files, file => this.readVmBackupMetadata(file)), 'mode')
+    const metadatas = await asyncMap(files, file => this.readVmBackupMetadata(file))
+    const { delta, full, ...others } = groupBy(metadatas, 'mode')
 
     const unsupportedModes = Object.keys(others)
     if (unsupportedModes.length !== 0) {
@@ -278,6 +282,9 @@ class RemoteAdapter {
       // don't merge in main process, unused VHDs will be merged in the next backup run
       await this.cleanVm(dir, { remove: true, onLog: warn })
     }
+
+    const dedupedVmUuid = new Set(metadatas.map(_ => _.vm.uuid))
+    await asyncMap(dedupedVmUuid, vmUuid => this.invalidateVmBackupListCache(vmUuid))
   }
 
   #getCompressionType() {
@@ -448,34 +455,94 @@ class RemoteAdapter {
     return backupsByPool
   }
 
-  async listVmBackups(vmUuid, predicate) {
+  async invalidateVmBackupListCache(vmUuid) {
+    await this.handler.unlink(`${BACKUP_DIR}/${vmUuid}/cache.json.gz`)
+  }
+
+  async #getCachabledDataListVmBackups(dir) {
     const handler = this._handler
-    const backups = []
+    const backups = {}
 
     try {
-      const files = await handler.list(`${BACKUP_DIR}/${vmUuid}`, {
+      const files = await handler.list(dir, {
         filter: isMetadataFile,
         prependDir: true,
       })
       await asyncMap(files, async file => {
         try {
           const metadata = await this.readVmBackupMetadata(file)
-          if (predicate === undefined || predicate(metadata)) {
-            // inject an id usable by importVmBackupNg()
-            metadata.id = metadata._filename
-
-            backups.push(metadata)
-          }
+          // inject an id usable by importVmBackupNg()
+          metadata.id = metadata._filename
+          backups[file] = metadata
         } catch (error) {
-          warn(`listVmBackups ${file}`, { error })
+          warn(`can't read vm backup metadata`, { error, file, dir })
         }
       })
+      return backups
     } catch (error) {
       let code
       if (error == null || ((code = error.code) !== 'ENOENT' && code !== 'ENOTDIR')) {
         throw error
       }
     }
+  }
+
+  // use _ to mark this method as private by convention
+  // since we decorate it with synchronized.withKey in the constructor
+  // and # function are not writeable.
+  //
+  // read the list of backup of a Vm from cache
+  // if cache is missing  or broken  => regenerate it and return
+
+  async _readCacheListVmBackups(vmUuid) {
+    const dir = `${BACKUP_DIR}/${vmUuid}`
+    const path = `${dir}/cache.json.gz`
+
+    try {
+      const gzipped = await this.handler.readFile(path)
+      const text = await fromCallback(zlib.gunzip, gzipped)
+      return JSON.parse(text)
+    } catch (error) {
+      if (error.code !== 'ENOENT') {
+        warn('Cache file was unreadable', { vmUuid, error })
+      }
+    }
+
+    // nothing cached, or cache unreadable => regenerate it
+    const backups = await this.#getCachabledDataListVmBackups(dir)
+    if (backups === undefined) {
+      return
+    }
+
+    // detached async action, will not reject
+    this.#writeVmBackupsCache(path, backups)
+
+    return backups
+  }
+
+  async #writeVmBackupsCache(cacheFile, backups) {
+    try {
+      const text = JSON.stringify(backups)
+      const zipped = await fromCallback(zlib.gzip, text)
+      await this.handler.writeFile(cacheFile, zipped, { flags: 'w' })
+    } catch (error) {
+      warn('writeVmBackupsCache', { cacheFile, error })
+    }
+  }
+
+  async listVmBackups(vmUuid, predicate) {
+    const backups = []
+    const cached = await this._readCacheListVmBackups(vmUuid)
+
+    if (cached === undefined) {
+      return []
+    }
+
+    Object.values(cached).forEach(metadata => {
+      if (predicate === undefined || predicate(metadata)) {
+        backups.push(metadata)
+      }
+    })
 
     return backups.sort(compareTimestamp)
   }
@@ -531,46 +598,27 @@ class RemoteAdapter {
     })
   }
 
-  async _createSyntheticStream(handler, paths) {
-    let disposableVhds = []
-
-    // if it's a path : open all hierarchy of parent
-    if (typeof paths === 'string') {
-      let vhd
-      let vhdPath = paths
-      do {
-        const disposable = await openVhd(handler, vhdPath)
-        vhd = disposable.value
-        disposableVhds.push(disposable)
-        vhdPath = resolveRelativeFromFile(vhdPath, vhd.header.parentUnicodeName)
-      } while (vhd.footer.diskType !== Constants.DISK_TYPES.DYNAMIC)
-    } else {
-      // only open the list of path given
-      disposableVhds = paths.map(path => openVhd(handler, path))
-    }
-
+  // open the  hierarchy of ancestors until we find a full one
+  async _createSyntheticStream(handler, path) {
+    const disposableSynthetic = await VhdSynthetic.fromVhdChain(handler, path)
     // I don't want the vhds to be disposed on return
     // but only when the stream is done ( or failed )
-    const disposables = await Disposable.all(disposableVhds)
-    const vhds = disposables.value
 
     let disposed = false
     const disposeOnce = async () => {
       if (!disposed) {
         disposed = true
-
         try {
-          await disposables.dispose()
+          await disposableSynthetic.dispose()
         } catch (error) {
-          warn('_createSyntheticStream: failed to dispose VHDs', { error })
+          warn('openVhd: failed to dispose VHDs', { error })
         }
       }
     }
-
-    const synthetic = new VhdSynthetic(vhds)
-    await synthetic.readHeaderAndFooter()
+    const synthetic = disposableSynthetic.value
     await synthetic.readBlockAllocationTable()
     const stream = await synthetic.stream()
+
     stream.on('end', disposeOnce)
     stream.on('close', disposeOnce)
     stream.on('error', disposeOnce)
@@ -603,7 +651,10 @@ class RemoteAdapter {
   }
 
   async readVmBackupMetadata(path) {
-    return Object.defineProperty(JSON.parse(await this._handler.readFile(path)), '_filename', { value: path })
+    // _filename is a private field used to compute the backup id
+    //
+    // it's enumerable to make it cacheable
+    return { ...JSON.parse(await this._handler.readFile(path)), _filename: path }
   }
 }
 

package/_VmBackup.js

@@ -45,7 +45,18 @@ const forkDeltaExport = deltaExport =>
   })
 
 class VmBackup {
-  constructor({ config, getSnapshotNameLabel, job, remoteAdapters, remotes, schedule, settings, srs, vm }) {
+  constructor({
+    config,
+    getSnapshotNameLabel,
+    healthCheckSr,
+    job,
+    remoteAdapters,
+    remotes,
+    schedule,
+    settings,
+    srs,
+    vm,
+  }) {
     if (vm.other_config['xo:backup:job'] === job.id && 'start' in vm.blocked_operations) {
       // don't match replicated VMs created by this very job otherwise they
       // will be replicated again and again
@@ -55,7 +66,6 @@ class VmBackup {
     this.config = config
     this.job = job
     this.remoteAdapters = remoteAdapters
-    this.remotes = remotes
     this.scheduleId = schedule.id
     this.timestamp = undefined
 
@@ -69,6 +79,7 @@ class VmBackup {
     this._fullVdisRequired = undefined
     this._getSnapshotNameLabel = getSnapshotNameLabel
     this._isDelta = job.mode === 'delta'
+    this._healthCheckSr = healthCheckSr
     this._jobId = job.id
     this._jobSnapshots = undefined
     this._xapi = vm.$xapi
@@ -95,7 +106,6 @@ class VmBackup {
         : [FullBackupWriter, FullReplicationWriter]
 
       const allSettings = job.settings
-
       Object.keys(remoteAdapters).forEach(remoteId => {
         const targetSettings = {
           ...settings,
@@ -173,7 +183,10 @@ class VmBackup {
     const settings = this._settings
 
     const doSnapshot =
-      this._isDelta || (!settings.offlineBackup && vm.power_state === 'Running') || settings.snapshotRetention !== 0
+      settings.unconditionalSnapshot ||
+      this._isDelta ||
+      (!settings.offlineBackup && vm.power_state === 'Running') ||
+      settings.snapshotRetention !== 0
     if (doSnapshot) {
       await Task.run({ name: 'snapshot' }, async () => {
         if (!settings.bypassVdiChainsCheck) {
@@ -183,6 +196,7 @@ class VmBackup {
         const snapshotRef = await vm[settings.checkpointSnapshot ? '$checkpoint' : '$snapshot']({
           ignoreNobakVdis: true,
           name_label: this._getSnapshotNameLabel(vm),
+          unplugVusbs: true,
         })
         this.timestamp = Date.now()
 
@@ -304,22 +318,17 @@ class VmBackup {
   }
 
   async _removeUnusedSnapshots() {
-    const jobSettings = this.job.settings
+    const allSettings = this.job.settings
+    const baseSettings = this._baseSettings
     const baseVmRef = this._baseVm?.$ref
-    const { config } = this
-    const baseSettings = {
-      ...config.defaultSettings,
-      ...config.metadata.defaultSettings,
-      ...jobSettings[''],
-    }
 
     const snapshotsPerSchedule = groupBy(this._jobSnapshots, _ => _.other_config['xo:backup:schedule'])
     const xapi = this._xapi
     await asyncMap(Object.entries(snapshotsPerSchedule), ([scheduleId, snapshots]) => {
       const settings = {
         ...baseSettings,
-        ...jobSettings[scheduleId],
-        ...jobSettings[this.vm.uuid],
+        ...allSettings[scheduleId],
+        ...allSettings[this.vm.uuid],
       }
       return asyncMap(getOldEntries(settings.snapshotRetention, snapshots), ({ $ref }) => {
         if ($ref !== baseVmRef) {
@@ -398,6 +407,24 @@ class VmBackup {
     this._fullVdisRequired = fullVdisRequired
   }
 
+  async _healthCheck() {
+    const settings = this._settings
+
+    if (this._healthCheckSr === undefined) {
+      return
+    }
+
+    // check if current VM has tags
+    const { tags } = this.vm
+    const intersect = settings.healthCheckVmsWithTags.some(t => tags.includes(t))
+
+    if (settings.healthCheckVmsWithTags.length !== 0 && !intersect) {
+      return
+    }
+
+    await this._callWriters(writer => writer.healthCheck(this._healthCheckSr), 'writer.healthCheck()')
+  }
+
   async run($defer) {
     const settings = this._settings
     assert(
@@ -407,7 +434,9 @@ class VmBackup {
 
     await this._callWriters(async writer => {
       await writer.beforeBackup()
-      $defer(() => writer.afterBackup())
+      $defer(async () => {
+        await writer.afterBackup()
+      })
     }, 'writer.beforeBackup()')
 
     await this._fetchJobSnapshots()
@@ -443,6 +472,7 @@ class VmBackup {
       await this._fetchJobSnapshots()
       await this._removeUnusedSnapshots()
     }
+    await this._healthCheck()
   }
 }
 exports.VmBackup = VmBackup