The standby server runs, but takeover remains untested

The active Entra Connect server fails. A second server has been in staging mode for months, and its latest imports succeeded. That looks reassuring, but it does not answer the critical questions: Do OU filters and synchronization rules still match? Which objects would the standby server delete? How quickly will new passwords actually reach Entra ID after the switch?

This article covers Microsoft Entra Connect Sync and a controlled takeover test. Microsoft Entra Cloud Sync uses a different agent architecture and is outside this procedure. Hardened sync hosts are a prerequisite. The additional project outcome is an operational acceptance test of the staging server, including export approval, password flows and the decision to switch back.

One active exporter and measurable recovery targets

The target state consists of exactly one active Connect Sync server for the tenant and an up-to-date, operational staging server. Staging mode alone does not provide automatic failover. Assign a decision owner and a deputy; both need a tested administrative access path that does not depend on the failed host.

Define separate acceptance criteria for object synchronization, Password Hash Synchronization and Password Writeback wherever these features are used. A successful export run proves neither current password hash synchronization nor working writeback. Set the maximum acceptable interruption, backlog and observation period in advance. A blanket promise of “available again in five minutes” is not a credible recovery plan without measurement.

Staging suppresses regular exports, not administration

Staging mode allows import and synchronization to continue while suppressing regular exports. Password Hash Synchronization and Password Writeback do not run on that server. It still processes privileged identity data and must be protected as a Tier 0 system.

Staging is not a security boundary: an administrator can trigger a manual export in Synchronization Service Manager. Limit administrative access and explicitly exclude manual export runs from the test plan. A disabled scheduler is also no substitute for staging or effective isolation of the previously active server.

Collect the local state on both hosts with a read-only query:

Import-Module ADSync -ErrorAction Stop

Get-ADSyncScheduler |
  Select-Object SyncCycleEnabled, StagingModeEnabled,
    SyncCycleInProgress, NextSyncCycleStartTimeInUTC

Get-Service -Name ADSync -ErrorAction Stop |
  Select-Object Name, Status, StartType

The output records settings and service status at the time of the query. Also inspect run history for the last successful import, synchronization and export, together with any errors. “Service running” does not mean “data current.”

Resolve configuration drift before the maintenance window

Compare the approved configurations of both servers: supported product version, connected forests, domain and OU filters, enabled features, custom rules including precedence, and the object matching that uses the sourceAnchor. Two successful installations can produce different business outcomes.

A configuration export after a change belongs in protected operational documentation. Automatically generated JSON files do not necessarily capture every change made through PowerShell or synchronization tools. Use the supported export procedure after such changes and verify the result. The JSON file also does not contain a complete backup of connector credentials and replaces neither secure credential reprovisioning nor a rebuild test.

Align the intended configuration without blindly copying server-specific identities or certificates. Verify connector permissions, certificate validity, proxy, DNS, time and required network paths on the standby server itself. A permission or firewall rule for the active host does not automatically apply to the staging host.

Review pending changes as the basis for approval

Let the staging server import and synchronize current data. As a precaution before promotion, plan for a completed initial synchronization cycle; explicitly confirm staging mode beforehand. This is particularly relevant when filters or rules have changed. Then review pending exports in connector space: adds, updates and deletes, with the reason and affected objects for each.

A difference is not automatically an error because the active server may already have exported changes in the meantime. Compare documented points in time and use a short freeze on relevant configuration changes. Unexplained deletions, unexpected new objects or different object matching block approval. Sampling alone is insufficient when unexpectedly large changes appear.

Check the effective accidental deletion protection and the approved threshold. This protection covers neither incorrect attribute updates nor every individual harmful deletion. Investigate a triggered threshold; reflexively disabling it for the takeover test would defeat the acceptance criteria. Export previews contain personal data and belong in an access-controlled project location.

Stop the previous exporter before promoting the standby

For a planned switch, use the following sequence:

  1. Freeze changes, record the current operational state and allow in-progress runs to finish in a controlled manner.
  2. Put the previously active server into staging mode through the supported wizard. Confirm the effective state and that active exports have finished.
  3. Refresh the standby server's data and check its pending exports against the approval again.
  4. Only then disable staging on the standby through the wizard and resume regular synchronization runs under observation.
  5. Monitor run history, errors, object changes and enabled password features, and complete operational acceptance.

If the previously active host cannot be reached, that does not prove it has stopped. Before promotion, verify that it is shut down or isolated so that it cannot resume exports or password processing. Account for AD and cloud paths as well as a possible automatic VM restart. An uncertain state is not a reason to start a second active server on a guess.

Validate password flows with separate test cases

For object synchronization, change an approved, noncritical attribute of a synthetic test account and verify the expected target value. For Password Hash Synchronization, also test a controlled password change and a fresh sign-in through the cloud authentication path actually in use. An existing session or a sign-in through federation or Pass-through Authentication does not prove password hash synchronization works.

After a long staging period, the newly active server may need to process a substantial password change backlog. This can take hours, and new changes may wait behind the backlog. Allow a realistic observation window and monitor progress. Repeatedly restarting the synchronization service does not accelerate catch-up and can impair progress.

If Password Writeback is used, separately test an approved self-service password reset with an unprivileged test account and verify the intended effect in on-premises AD. Record timing and results, never passwords. Explicitly record unused features as “not in use” instead of claiming a successful test.

Switching back does not undo exports

A fallback plan requires two decisions: which host should synchronize going forward, and which already written data needs correction? Returning to the old server does not automatically restore earlier object states. Incorrect exports require a separate, reviewed data correction or recovery procedure.

To switch back, first put the new active exporter into staging or effectively isolate it. Then reassess the previous server's data and pending changes before activating it again. An old VM snapshot is not a substitute. If compromise is suspected, the old host must not be reused without investigation; follow the incident response and recovery process instead.

Benefits of a controlled takeover test

  • Measured recovery time: The team knows the actual duration, including password processing and operational acceptance.
  • Early detection of drift: Differences in filters, rules and permissions emerge before an unplanned failure.
  • Limited change risk: Pending exports receive a documented review and approval.
  • Reliable operational handover: Cover arrangements, evidence and fallback decisions do not depend on one person's knowledge.

Disadvantages and operational limitations

A second server adds patching, monitoring and maintenance work and expands the privileged attack surface. Without continuous configuration alignment, its reliability declines with each change. Shared dependencies such as AD, the tenant, networking or administrative access remain potential causes of failure.

Staging provides no automatic high availability, no complete backup and no protection against a privileged manual export. Production switchover tests need a maintenance window and can temporarily delay password changes or writeback. The test improves demonstrated operational readiness but replaces neither host hardening nor a separate recovery strategy after a security incident.

Project checklist for the next switchover

  • [ ] Document active and staging servers, owners and independent administrative access.
  • [ ] Define separate recovery targets for objects, password hashes and writeback.
  • [ ] Compare versions, filters, rules, object matching and enabled features.
  • [ ] Export configuration securely and verify connector identity reprovisioning.
  • [ ] Review current import and synchronization runs and pending exports against business requirements.
  • [ ] Check effective deletion protection and the approved threshold.
  • [ ] Verify the previous active host is in staging or isolated before promotion.
  • [ ] Test object synchronization and the password flows actually in use separately.
  • [ ] Observe backlog, errors and elapsed time until full acceptance.
  • [ ] Document switching back and data correction as separate procedures.
  • [ ] Clean up test data in a controlled manner and return both servers to their approved operational state.
  • [ ] Repeat the test after relevant changes and at a defined operational interval.