@xen-orchestra/backups 0.44.7 → 0.45.0

package/_runners/VmsRemote.mjs

@@ -6,11 +6,12 @@ import { extractIdsFromSimplePattern } from '../extractIdsFromSimplePattern.mjs'
 import { Task } from '../Task.mjs'
 import createStreamThrottle from './_createStreamThrottle.mjs'
 import { DEFAULT_SETTINGS, Abstract } from './_Abstract.mjs'
-import { runTask } from './_runTask.mjs'
 import { getAdaptersByRemote } from './_getAdaptersByRemote.mjs'
 import { FullRemote } from './_vmRunners/FullRemote.mjs'
 import { IncrementalRemote } from './_vmRunners/IncrementalRemote.mjs'
 
+const noop = Function.prototype
+
 const DEFAULT_REMOTE_VM_SETTINGS = {
   concurrency: 2,
   copyRetention: 0,
@@ -20,6 +21,7 @@ const DEFAULT_REMOTE_VM_SETTINGS = {
   healthCheckVmsWithTags: [],
   maxExportRate: 0,
   maxMergedDeltasPerRun: Infinity,
+  nRetriesVmBackupFailures: 0,
   timeout: 0,
   validateVhdStreams: false,
   vmTimeout: 0,
@@ -41,6 +43,7 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
     const throttleStream = createStreamThrottle(settings.maxExportRate)
 
     const config = this._config
+
     await Disposable.use(
       () => this._getAdapter(job.sourceRemote),
       () => (settings.healthCheckSr !== undefined ? this._getRecord('SR', settings.healthCheckSr) : undefined),
@@ -62,8 +65,19 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
         const allSettings = this._job.settings
         const baseSettings = this._baseSettings
 
+        const queue = new Set(vmsUuids)
+        const taskByVmId = {}
+        const nTriesByVmId = {}
+
         const handleVm = vmUuid => {
+          if (nTriesByVmId[vmUuid] === undefined) {
+            nTriesByVmId[vmUuid] = 0
+          }
+          nTriesByVmId[vmUuid]++
+
           const taskStart = { name: 'backup VM', data: { type: 'VM', id: vmUuid } }
+          const vmSettings = { ...settings, ...allSettings[vmUuid] }
+          const isLastRun = nTriesByVmId[vmUuid] === vmSettings.nRetriesVmBackupFailures + 1
 
           const opts = {
             baseSettings,
@@ -72,7 +86,7 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
             healthCheckSr,
             remoteAdapters,
             schedule,
-            settings: { ...settings, ...allSettings[vmUuid] },
+            settings: vmSettings,
             sourceRemoteAdapter,
             throttleStream,
             vmUuid,
@@ -90,11 +104,42 @@ export const VmsRemote = class RemoteVmsBackupRunner extends Abstract {
             .listVmBackups(vmUuid, ({ mode }) => mode === job.mode)
             .then(vmBackups => {
               // avoiding to create tasks for empty directories
-              if (vmBackups.length > 0) return runTask(taskStart, () => vmBackup.run())
+              if (vmBackups.length > 0) {
+                if (taskByVmId[vmUuid] === undefined) {
+                  taskByVmId[vmUuid] = new Task(taskStart)
+                }
+                const task = taskByVmId[vmUuid]
+                return task
+                  .run(async () => {
+                    try {
+                      const result = await vmBackup.run()
+                      task.success(result)
+                      return result
+                    } catch (error) {
+                      if (isLastRun) {
+                        throw error
+                      } else {
+                        Task.warning(`Retry the VM mirror backup due to an error`, {
+                          attempt: nTriesByVmId[vmUuid],
+                          error: error.message,
+                        })
+                        queue.add(vmUuid)
+                      }
+                    }
+                  })
+                  .catch(noop)
+              }
             })
         }
         const { concurrency } = settings
-        await asyncMapSettled(vmsUuids, !concurrency ? handleVm : limitConcurrency(concurrency)(handleVm))
+        const _handleVm = !concurrency ? handleVm : limitConcurrency(concurrency)(handleVm)
+
+        while (queue.size > 0) {
+          const vmIds = Array.from(queue)
+          queue.clear()
+
+          await asyncMapSettled(vmIds, _handleVm)
+        }
       }
     )
   }

package/_runners/VmsXapi.mjs

@@ -11,6 +11,8 @@ import { getAdaptersByRemote } from './_getAdaptersByRemote.mjs'
 import { IncrementalXapi } from './_vmRunners/IncrementalXapi.mjs'
 import { FullXapi } from './_vmRunners/FullXapi.mjs'
 
+const noop = Function.prototype
+
 const DEFAULT_XAPI_VM_SETTINGS = {
   bypassVdiChainsCheck: false,
   checkpointSnapshot: false,
@@ -24,6 +26,7 @@ const DEFAULT_XAPI_VM_SETTINGS = {
   healthCheckVmsWithTags: [],
   maxExportRate: 0,
   maxMergedDeltasPerRun: Infinity,
+  nRetriesVmBackupFailures: 0,
   offlineBackup: false,
   offlineSnapshot: false,
   snapshotRetention: 0,
@@ -53,6 +56,7 @@ export const VmsXapi = class VmsXapiBackupRunner extends Abstract {
     const throttleStream = createStreamThrottle(settings.maxExportRate)
 
     const config = this._config
+
     await Disposable.use(
       Disposable.all(
         extractIdsFromSimplePattern(job.srs).map(id =>
@@ -89,48 +93,98 @@ export const VmsXapi = class VmsXapiBackupRunner extends Abstract {
         const allSettings = this._job.settings
         const baseSettings = this._baseSettings
 
+        const queue = new Set(vmIds)
+        const taskByVmId = {}
+        const nTriesByVmId = {}
+
         const handleVm = vmUuid => {
+          const getVmTask = () => {
+            if (taskByVmId[vmUuid] === undefined) {
+              taskByVmId[vmUuid] = new Task(taskStart)
+            }
+            return taskByVmId[vmUuid]
+          }
+          const vmBackupFailed = error => {
+            if (isLastRun) {
+              throw error
+            } else {
+              Task.warning(`Retry the VM backup due to an error`, {
+                attempt: nTriesByVmId[vmUuid],
+                error: error.message,
+              })
+              queue.add(vmUuid)
+            }
+          }
+
+          if (nTriesByVmId[vmUuid] === undefined) {
+            nTriesByVmId[vmUuid] = 0
+          }
+          nTriesByVmId[vmUuid]++
+
+          const vmSettings = { ...settings, ...allSettings[vmUuid] }
           const taskStart = { name: 'backup VM', data: { type: 'VM', id: vmUuid } }
+          const isLastRun = nTriesByVmId[vmUuid] === vmSettings.nRetriesVmBackupFailures + 1
 
           return this._getRecord('VM', vmUuid).then(
             disposableVm =>
-              Disposable.use(disposableVm, vm => {
-                taskStart.data.name_label = vm.name_label
-                return runTask(taskStart, () => {
-                  const opts = {
-                    baseSettings,
-                    config,
-                    getSnapshotNameLabel,
-                    healthCheckSr,
-                    job,
-                    remoteAdapters,
-                    schedule,
-                    settings: { ...settings, ...allSettings[vm.uuid] },
-                    srs,
-                    throttleStream,
-                    vm,
-                  }
-                  let vmBackup
-                  if (job.mode === 'delta') {
-                    vmBackup = new IncrementalXapi(opts)
-                  } else {
-                    if (job.mode === 'full') {
-                      vmBackup = new FullXapi(opts)
+              Disposable.use(disposableVm, async vm => {
+                if (taskStart.data.name_label === undefined) {
+                  taskStart.data.name_label = vm.name_label
+                }
+
+                const task = getVmTask()
+                return task
+                  .run(async () => {
+                    const opts = {
+                      baseSettings,
+                      config,
+                      getSnapshotNameLabel,
+                      healthCheckSr,
+                      job,
+                      remoteAdapters,
+                      schedule,
+                      settings: vmSettings,
+                      srs,
+                      throttleStream,
+                      vm,
+                    }
+
+                    let vmBackup
+                    if (job.mode === 'delta') {
+                      vmBackup = new IncrementalXapi(opts)
                     } else {
-                      throw new Error(`Job mode ${job.mode} not implemented`)
+                      if (job.mode === 'full') {
+                        vmBackup = new FullXapi(opts)
+                      } else {
+                        throw new Error(`Job mode ${job.mode} not implemented`)
+                      }
+                    }
+
+                    try {
+                      const result = await vmBackup.run()
+                      task.success(result)
+                      return result
+                    } catch (error) {
+                      vmBackupFailed(error)
                     }
-                  }
-                  return vmBackup.run()
-                })
+                  })
+                  .catch(noop) // errors are handled by logs
               }),
             error =>
-              runTask(taskStart, () => {
-                throw error
+              getVmTask().run(() => {
+                vmBackupFailed(error)
               })
           )
         }
         const { concurrency } = settings
-        await asyncMapSettled(vmIds, concurrency === 0 ? handleVm : limitConcurrency(concurrency)(handleVm))
+        const _handleVm = concurrency === 0 ? handleVm : limitConcurrency(concurrency)(handleVm)
+
+        while (queue.size > 0) {
+          const vmIds = Array.from(queue)
+          queue.clear()
+
+          await asyncMapSettled(vmIds, _handleVm)
+        }
       }
     )
   }

package/_runners/_vmRunners/_AbstractRemote.mjs

@@ -22,6 +22,7 @@ export const AbstractRemote = class AbstractRemoteVmBackupRunner extends Abstrac
     this.config = config
     this.job = job
     this.remoteAdapters = remoteAdapters
+    this._settings = settings
     this.scheduleId = schedule.id
     this.timestamp = undefined
 

package/_runners/_writers/_MixinRemoteWriter.mjs

@@ -77,11 +77,11 @@ export const MixinRemoteWriter = (BaseClass = Object) =>
     healthCheck() {
       const sr = this._healthCheckSr
       assert.notStrictEqual(sr, undefined, 'SR should be defined before making a health check')
-      assert.notStrictEqual(
-        this._metadataFileName,
-        undefined,
-        'Metadata file name should be defined before making a health check'
-      )
+      if (this._metadataFileName === undefined) {
+        // this can happen when making a mirror backup with nothing to transfer
+        Task.info('no health check, since no backup have been transferred')
+        return
+      }
       return Task.run(
         {
           name: 'health check',

package/package.json

@@ -8,7 +8,7 @@
     "type": "git",
     "url": "https://github.com/vatesfr/xen-orchestra.git"
   },
-  "version": "0.44.7",
+  "version": "0.45.0",
   "engines": {
     "node": ">=14.18"
   },
@@ -56,7 +56,7 @@
     "tmp": "^0.2.1"
   },
   "peerDependencies": {
-    "@xen-orchestra/xapi": "^4.2.1"
+    "@xen-orchestra/xapi": "^4.3.0"
   },
   "license": "AGPL-3.0-or-later",
   "author": {