Lesson 3 of 5
The source database as fallback
9 min
Listen to this lesson
One property makes DMO much less frightening than it sounds. SAP states that during DMO the source database keeps running and its application data is not modified; fallback is available throughout the process through the SUM Reset function. In SAP's DMO business case the original database is kept and can be reactivated as a fallback, which reduces risk because no restore is required.
- UptimeShadow on targetSource fully in use; target holds only the new repository
- Downtime startsSwitchSystem connects to SAP HANA; source no longer receives changes
- DowntimeMigration and conversionSource read, not changed; reset still possible
- ValidationChecks and sign-offFallback still cheap: business has not posted in the target
- Users releasedPoint of no returnNew business data exists only in the target
What fallback means in practice
Because the source was not changed, falling back does not mean restoring a multi-terabyte backup under pressure. It means resetting the procedure and bringing the old system back up on its old database. That is fast and predictable, which is why the cutover plan can set a late point of no return: the last moment before business users post in the target. After that, falling back would lose real postings, so the decision becomes a business decision with very different consequences.
| Situation | Fallback option | Notes |
|---|---|---|
| Run fails in uptime | Reset; source never stopped | Fix and restart; no business impact beyond the planned window |
| Run fails in downtime | Reset and restart the source system | Source unchanged; communicate the cancelled window |
| Validation fails after the run | Fall back before users are released | Needs a clear decision owner and criteria |
| Problems found after users posted | Fix forward | Falling back would lose postings; requires a business decision |
Fallback options by point in time. Agree the decision owner for each row before the weekend.
Test the fallback, too
A fallback that has never been exercised is an assumption. Use one mock run to practice it: after the run, follow the reset procedure, bring the source back, and check that interfaces, jobs and users work on the old system again. Time it, so the point of no return in the runbook is based on a measured fallback duration, not on optimism. With a move variant, also check that the old hosts, host names and connections are still intact, because the source side includes more than the database.
- Has the reset procedure been practiced in a mock?Yes → use its measured duration · No → schedule a practice run before the dress rehearsal
- Are the source hosts and connections preserved?Yes → fallback restores the old landscape · No → fallback is incomplete; fix the plan
- Is the decision owner named?Yes → Luis decides with Grace informed · No → name one before cutover
- Is the point of no return set from measured times?Yes → fallback plan signable · No → recalculate
Keep the source available after a successful go-live, too. Until the process owners have signed the first business days in the new system, the old database is valuable evidence: auditors and finance may want to compare a balance or a document with its pre-conversion state. Decide with Dan and Grace how long the source stays read-only and who may access it, and record the decommissioning as a deliberate step with its own sign-off rather than a cleanup task someone does when disk space runs low.
What would you do?
During Contoso's dress rehearsal, validation finds that open customer items in company code 1710 do not reconcile, at 11:00 Sunday. Users are not yet released. Dan needs at least six hours to investigate. The fallback measured in mock 2 takes nine hours. What should Luis do?
- During DMO the source database keeps running and its application data is not modified.
- SUM's Reset makes fallback possible throughout the procedure, without a database restore.
- The fallback window closes when business users post in the target: the point of no return.
- Practice and time the fallback in a mock, including source hosts and connections.
Check your understanding · not graded
Lesson check
1. Why can a failed DMO run fall back without restoring a backup?
2. When does the cheap fallback window close?
3. Why practice the reset procedure in a mock run?
4. With a move variant, what else must be intact for fallback to work?