A green backup job is not a recovered service
A successful Veeam job does not prove that a restored VM will let employees work. Testing must answer three questions: can the backup be read, can the required machines start, and can the expected business workflow succeed? Fictional North Workshop needs to restore a web application and its SQL database. No real customer names, addresses or credentials are used. This guide prepares a focused recovery exercise rather than a complete continuity plan.
1. Define evidence and the restore point
Write acceptance criteria before starting a restore: pilot-user sign-in, reading a reference folder, opening a fictional application record and confirming its date. Select an exact restore point and record its ID, date and time zone, source machine and backup copy. The newest point is not necessarily healthy. Suspected compromised backups require incident-response handling and a dedicated environment; a negative malware scan does not guarantee a clean system. Define RPO as tolerable data loss and measure RTO until service acceptance, not merely Windows startup.
2. Confirm version, platform and licensing
References target Veeam Backup & Replication 13 and primarily VMware vSphere. Record the installed build and use matching documentation for VBR 12, Hyper-V or another hypervisor; do not transpose screens and scripts without validation. SureBackup availability depends on licensing: Veeam guidance distinguishes Enterprise, Enterprise Plus and VUL; Standard can use manual Instant Recovery verification within its capabilities. Confirm subscription entitlements before promising automation. Check supported hypervisor versions, Veeam components, repository access, recoverable decryption keys, datastore/cache space, CPU, RAM and restored application licensing. Current documentation also distinguishes Windows and Linux VBR servers, including some SQL test scripts.
3. Inventory read-only before making changes
In a Veeam PowerShell session already authenticated to the correct server, Veeam.Backup.PowerShell can read backups and restore points. Use a viewing role for inventory and a restore role scoped to the workload for execution; creating a lab may need additional rights. Do not use Domain Admin to read a catalogue. Check installed-version help. Job and VM names are fictional. These reads neither mount backups nor start VMs; they also do not prove that the entire chain is readable.
# Session Veeam déjà connectée / Existing authenticated Veeam session
Import-Module Veeam.Backup.PowerShell
Get-Help Get-VBRRestorePoint -Full
$backup = Get-VBRBackup -Name 'Atelier-VM-Pilote'
Get-VBRRestorePoint -Backup $backup -Name 'APP-PILOTE' |
Sort-Object CreationTime -Descending |
Select-Object Name,CreationTime,Id4. Prepare a genuinely isolated lab
Booting a restored machine is a change: services may send mail, run scheduled tasks or call an API. Validate isolation and its boundaries before starting clones. With SureBackup, review Virtual Lab, network mapping, proxy appliance, masquerading and any static mappings. An appliance with visibility on both sides needs an allowed-path review; the word “lab” alone is not proof of isolation. For a manual restore, first create a separate destination with disconnected NICs or a validated isolated network. Never connect a clone retaining its identity or IP to production. Block outbound mail, payments, backup systems and external integrations, and inspect filtering logs. Stop the exercise if isolation cannot be demonstrated.
5. Start dependencies in the right order
Map dependencies: lab identity and DNS where needed, database, application, then test workstation. In SureBackup, an application group can provide services required by verified machines. Selected points must be mutually compatible; VMs from different times may introduce application inconsistency. Restoring a domain controller requires an AD-specific scenario and strict isolation; do not choose authoritative mode casually. Allow realistic startup timeouts and sufficient memory. A timeout can indicate an undersized lab or incorrect test path rather than an unusable backup.
6. SureBackup: understand what the test actually proves
Full recoverability testing boots machines and executes configured tests; integrity and content-scan-only verification does not provide that application evidence. A VMware Tools heartbeat or ping establishes limited availability. An open SQL or HTTPS port does not prove a business query works. Select relevant roles and scripts for each VM, with expected results and explicit failure handling. Preserve the guest firewall: use an approved narrow test rule if needed rather than selecting blanket disablement. Protect test credentials and never put secrets in visible script arguments.
7. Verify the OS, data and business workflow
In the isolated VM console, inspect expected services, volumes and recent events. Examples below only read local state: adapt service names to the pilot and bound log queries. Running status proves neither SQL health nor application access. The owner must then verify the expected database, an application read of a known fictional record, its consistency and allowed versus denied user permissions. A write transaction needs separate approval and must stay inside the lab. Review the restore point’s VSS/application-aware processing status too: a crash-consistent point may need application recovery.
# Dans la VM isolée uniquement / Inside the isolated VM only
Get-Service -Name 'MSSQLSERVER','W3SVC' -ErrorAction Continue |
Select-Object Name,Status,StartType
Get-Volume | Select-Object DriveLetter,FileSystemLabel,HealthStatus,SizeRemaining
Get-WinEvent -FilterHashtable @{
LogName='Application'; StartTime=(Get-Date).AddHours(-1); Level=2,3
} -MaxEvents 100 | Select-Object TimeCreated,ProviderName,Id,Message8. Measure full recovery and interpret failures
Record repository access, publishing/restoration, dependency startup and business acceptance times. A lab running from backup need not perform like production. Instant Recovery can boot quickly, but moving to permanent storage and validating that state belong in a full RTO exercise. Separate chain failures, capacity limits, test-authentication failures and application errors. Correct one cause at a time on a pilot, retain the original failure and repeat the same criteria. Do not retroactively relax thresholds to turn failure into success.
9. Close the test without losing evidence
After an isolated exercise, retain required reports and screenshots in protected storage, then stop the SureBackup job or Instant Recovery publishing using the Veeam procedure. For this test, do not choose Migrate to production. Verify that temporary VMs, mounts and lab resources have stopped, production is unchanged and regular backup jobs still work. Do not delete restore points or remove immutability to clean up the lab. If testing accidentally reached a real integration, stop and handle the incident; deleting a clone does not undo external effects.
A report that supports decisions
Deliver an application-level result: tested point, repository copy, versions, dependencies, isolation evidence, passed or failed criteria, observed RPO/RTO and untested limits. Obtain business-owner acceptance and assign corrective actions. Repeat after major changes and at a frequency proportionate to criticality and organizational obligations. Periodically include repository or site loss: testing one point does not validate every backup. Storage, backup and infrastructure services can help design these exercises and document usable recovery procedures.