Linux migration / Engineering resource
Linux post-migration validation: the VM boots—what else changed?
Booting proves that a guest can start; it does not prove that its workload survived the migration. Validate the intended target, operating state, data and real application transactions against an agreed source baseline. Keep three layers separate: local system evidence, scoped automation results, and application-owner acceptance over a representative observation window.
01Start with the agreed change and baseline
Bring the readiness record, intended OS changes, source observations and acceptance thresholds to the rebuilt target. Record when and where each observation was taken. Validate the expected release and architecture before interpreting package or configuration differences. On an OS modernization, equality of package names and version strings across distributions is not the goal: the requirement is a compatible, supported runtime that fulfills the workload’s needs.
The examples below are read-only, human-invoked local Linux observations. They do not repair or restart anything. systemctl and journalctl assume systemd; versions, privileges and available options vary. Use your normal approved access and consult the local manuals. Outputs may reveal internal hostnames, addresses, paths or logs; retain them privately and redact them before sharing.
cat /etc/os-release02Separate enabled services from running health
Review failed units and the unit-file enablement inventory, then compare them with the intended service list. Enabled means configured for an activation path; it does not mean currently running or successfully serving requests. Conversely, a socket-activated or oneshot service may not appear as a continuously running process. Check custom units and expected instance names against the application’s deployment design.
Investigate boot errors in context: timestamps, dependencies and whether the error predates the change all matter. An empty failed-unit list does not prove application health, and a historical error does not by itself prove a migration defect. A parity check against a mounted root cannot observe a live running kernel; retain runtime evidence from the actual booted target as a separate layer.
systemctl --failed --no-pager
systemctl list-unit-files --type=service --no-pager
journalctl -b -p err --no-pager03Verify mounts, ownership and usable data
Compare mount targets, sources, filesystem types and options with the design. Confirm persistent mounts survive the expected boot path. A directory existing does not establish that its data volume mounted: an application can quietly write to the underlying root filesystem. Review permissions, numeric ownership, ACLs and security labels for the application identity, not just for an administrator.
Have the data owner check agreed counts, important records and application-visible state from a database-consistent recovery point. Use the database’s supported validation procedures and transaction semantics. Do not substitute blanket file hashes of an active database for consistency evidence. Record any intentionally changed data layout and how the team established completeness.
findmnt -o TARGET,SOURCE,FSTYPE,OPTIONS04Test the client path, not just the routing table
Inspect local addresses and routes, then have the relevant owners validate DNS resolution, firewall policy, proxy settings and the real path from representative clients. A routing table shows configured routes, not external reachability. It does not establish that a remote firewall permits the request, that DNS resolves correctly for clients, or that a load balancer targets the right service.
Check expected application listeners and complete an approved TLS transaction through the intended hostname. Validate certificate names, chain trust and the application response, not merely that a TCP connection opens. Use existing application health checks or known client requests, not an invasive scan. Avoid testing against production write endpoints unless the owner has approved safe test inputs.
ip -brief address
ip route show05Check schedules, monitoring and recovery operations
Validate timers and cron jobs against their intended schedule, timezone, user and dependencies. Confirm that the old system no longer performs the same production work when the new one takes over. Observe an expected execution or use the owner’s approved test mode; seeing a crontab file is not evidence of a successful job.
Confirm metrics, logs, alerts and dashboards identify the new system correctly. Check monitoring from the service that consumes it, so an agent’s local health is not mistaken for end-to-end visibility. Verify backup enrollment and perform an isolated restore check under the backup owner’s process. Do not assume a copied agent configuration has enrolled the new host or retained its recovery keys.
06Move from symptoms to acceptance evidence
| Symptom | Next evidence | Acceptance decision |
|---|---|---|
| Boots, but users cannot sign in | Identity path, application logs, representative login | Application/access owner verifies expected access |
| Service enabled, request fails | Running instance, dependency errors, known transaction | Service serves the intended request |
| Data directory exists, records missing | Actual mount, consistent restore point, database checks | Data owner reconciles agreed records |
| Route present, client times out | Client-side DNS, firewall and TLS path evidence | Representative client completes a safe transaction |
| Low idle CPU, slow batch job | Comparable load, storage latency, job duration | Performance fits pre-agreed tolerances |
| Parity equivalent, optional tests absent | Unverified lanes and owner test results | No acceptance inferred for untested lanes |
07Keep automation and application testing distinct
A package/service/configuration/account comparison can catch rebuild gaps systematically. Read its actual scope and findings rather than assuming it covers the whole application. Virtual Move’s current verify documentation also includes optional data, probe, response-equivalence, soak and database lanes. Whether they were supplied and what they observed matters; their existence does not make every report a database or application test.
Application owners should validate representative transactions, integrations, authorization boundaries, error handling and data reads/writes with approved test data. Database owners should review consistency and upgrade-specific evidence. Record expected results and observed results separately, including failed and unexecuted tests. A test that could not run is outstanding work, not a passing result.
08Observe representative operation and approve explicitly
Compare performance under a comparable workload and account for cache warmup, changed hardware, maintenance tasks and storage behavior. Agree a useful observation window covering the workload’s important cycles: a nightly job, a traffic peak or another operational event. No universal window proves every workload healthy. Keep time-series evidence and operational alerts alongside the point-in-time checks.
Review rollback triggers, data reconciliation and the last safe recovery point before switching traffic or writers. If acceptance fails, follow the approved change plan rather than improvising a restart or reverse migration. Final approval should identify the owners, remaining exceptions and evidence reviewed. Keep the old baseline and new observations so later drift can be distinguished from the migration itself.
Questions engineers ask
- Do clean failed units mean the application is healthy?
- No. They are local systemd evidence. Application transactions, dependencies, data and representative observation still need their own acceptance tests.
- Can I run these commands automatically to fix the VM?
- These examples are manual read-only observations, not a repair script. They do not install packages, restart services or change configuration.
Sources
Product details checked against the current official documentation. Confirm your installed version, exact conversion and operational prerequisites before use.
Continue reading
Apply this to one workload
Virtual Move helps assess, plan and verify supported Linux workloads. Your team owns the migration and acceptance decision.
