AEM Sites Playbook, Chapter 19: Monitoring, maintenance, incidents and recovery
Who detects, investigates and recovers a failing service or content operation?
November 10, 2026
Monitor the service your users need
Use this chapter when establishing operational support, investigating an incident or testing recovery. Define the important customer and author tasks, their acceptable behavior, the alert owner and the response expected. Availability of an endpoint is one useful signal; freshness, working approval, complete media and a functioning integration can require additional observations.
Agree who operates infrastructure, application code, content, connected systems and communication. Adobe's Cloud Service monitoring guidance separates provider monitoring from customer observability. Check current availability before promising a particular integration; its current OpenTelemetry offering is labelled Limited Availability. For 6.5/LTS, document the actual hosting and monitoring arrangement.
Make alerts actionable. Identify the affected service/task, relevant environment, time, evidence location and responsible operator. Test that a representative failure reaches the intended person. Review noisy or ownerless alerts rather than let them become background decoration.
Keep maintenance tied to the deployment
Inventory scheduled jobs, content/version retention, workflow history, index operations and operational configuration. Name the owner who can change each policy and the evidence the team needs to retain. A retention policy should support editorial history and recovery requirements as well as repository operation.
Current Cloud Service maintenance guidance defines configurable versus Adobe-managed responsibilities. The maintenance question has conflicting deployment context and no confirmed successful workflow purge. A health check is not a purge job, and an older accepted response cannot define today's configurable tasks.
Investigate before broad recovery
For an incident, capture the failing task, expected/actual result, affected population, first observation, recent changes and current severity. Preserve relevant sanitized logs and version identities. Correlate the timeline with publication, deployment, dependency or identity changes. Choose containment with the accountable owner, then test a focused hypothesis.
The page-restore discussion illustrates the value of tracing preceding changes. The author eventually identified earlier custom code updating jcr:primaryType; suggestions to remove protected-property safeguards or use elevated privileges were not the repair. Avoid broad permission changes or repeated destructive actions while investigating an unexplained restore failure.
Choose the recovery objective and mechanism
Scroll horizontally if needed.
| Required outcome | Establish before acting |
|---|---|
| Recover one prior page/content state | Suitable retained versions/content copies, references, permissions and publication consequences |
| Recover deleted or damaged content broadly | Available backup scope/time, overwritten newer work, approval and compatibility with current code |
| Reverse a code/configuration release | Supported deployment recovery, prior code identity and content/schema compatibility |
| Rebuild lost media from original files | Metadata, taxonomy, references, renditions, access and delivery state still needing reconstruction |
Cloud Service content restore retains the current custom code and AEM version. Its managed environment restore is distinct from selective content options and code rollback. Review current scope, availability and limits before initiating it. For AEM 6.5, follow the applicable backup/restore guidance and hosting procedure. The asset-export question does not establish that downloading original files creates a complete service backup.
Rehearse recovery in an appropriate non-production setting. State what loss and interruption the organization can tolerate, identify a recovery point, obtain the required approval and verify the restored task. Check references, permissions, workflow/publication state, delivered content, integrations and current-code behavior. A completed restore operation is not the final acceptance.
Identify the failure you need to detect
Community discussions separate exporting files, restoring content and operating maintenance tasks. The examples include an incomplete backup objective, an author-reported custom-code cause and conflicting deployment contexts. Define the operating objective and recovery evidence before selecting a mechanism.
Scroll horizontally if needed.
| Operational need | Useful observation | Named response owner |
|---|---|---|
| Customer journey works | Representative public task, expected revision and dependency result | Channel/application owner |
| Authors can change content | Essential edit/approval task and relevant errors | Content/platform support |
| Integration keeps up | Accepted/applied/failed work and reconciliation | Integration owner |
| Publication is timely | Operation/queue state and delivered freshness | Release/delivery owners |
| History and repository remain operable | Relevant retention/maintenance results | Platform/operations owners |
| Recovery is possible | Available state plus successful representative restoration | Recovery authority and service owner |
An alert should identify the environment, affected task, evidence location and first action. Test the notification and escalation with the actual support arrangement. Provider monitoring does not remove responsibility for application-specific customer and content checks. Use the current Cloud Service monitoring scope or the applicable hosting arrangement.
Worked scenario of recovering an accidentally deleted subtree
The content, timings and outcomes below are hypothetical. A content team discovers that a regional subtree was deleted. New unrelated content has been created since the deletion. The service owner must restore the required material without casually sacrificing that newer work.
The incident record captures affected paths/objects, last known correct state, deletion window, references, publication impact and current code identity. The recovery owner inspects available selective options and the platform's broader restore capabilities. The business approver identifies acceptable interruption and potential loss; those requirements are not assumed from the available backup.
Scroll horizontally if needed.
| Candidate | Question before selection | Verification burden |
|---|---|---|
| Suitable page/content versions | Do the required retained states exist? | Hierarchy, references, permissions and publication |
| Approved content copy/package | Is it trustworthy and correctly scoped? | Avoid overwriting unrelated newer content; verify dependencies |
| Broader managed content restore | Is the scope/time appropriate and approved? | Impact on newer work, current-code compatibility and restored tasks |
| Original asset re-upload | What state is missing beyond the binaries? | Metadata, taxonomy, references, processing and access |
For Cloud Service, content restoration retains current custom code/AEM version and has documented scope/limits. For 6.5, follow its backup procedure and actual hosting responsibilities. The choice is conditional on available evidence and permissions, not a universal preference for one method.
Suppose the team finds a suitable selective content state in the test rehearsal. It restores the subtree into the appropriate controlled setting and checks:
Scroll horizontally if needed.
| Check | Illustrative observation | Decision |
|---|---|---|
| Required content/hierarchy | Present | Continue verification |
| Shared references | One missing asset | Restore or reconcile that dependency |
| Regional permissions | Correct for intended identities | Retain positive/negative evidence |
| Current rendering/editing | Works with the recorded code | Compatibility demonstrated for the fixture |
| Publication/public routes | Need the authorized release operation | Recovery is not yet customer-complete |
| New unrelated content | Remains intact | Selective-scope objective met |
After reconciling the asset and performing the approved publication, the team verifies the customer result and accepts the defined recovery. It records elapsed observations and unresolved limitations without claiming a universal recovery-time guarantee.
Keep incident causality and recovery results separate
Restoring service can succeed before the original cause is confirmed. Record containment, recovery and root-cause hypotheses independently. Preserve a timeline of content, configuration, code and identity changes. The restore case's earlier custom mutation is a reminder to inspect preceding operations, not remove protected-property safeguards.
Assign follow-up actions that can prevent or detect recurrence: a permission correction, deletion approval, regression fixture, content-retention change or alert. Give them owners and review dates. A post-incident document without an operated action adds little protection.
Rehearse maintenance and support handover
Use current maintenance guidance for the deployment and identify which settings are customer-controlled. Test that retention supports review/recovery needs and that failed scheduled work is visible. Keep workload purges, health checks and backups as different operations with different acceptance evidence.
Ask an operator outside the implementation team to follow the runbook, find the relevant evidence and explain when escalation is required. Recovery readiness includes authority, compatible state, dependencies, communication and verification, not merely a backup-exists checkbox.
Close the incident and improve the service
Complete W19 with the timeline, impact, diagnosis, actions, recovery verification and remaining risk. Give each follow-up a named owner and review date. Update the regression set, alert, component guidance or operating procedure that would have detected the problem sooner. Keep factual observations distinct from an unconfirmed cause so the next operator can continue the investigation accurately.
W19 Incident and recovery drill
Use with chapter 19. Distinguish observation, hypothesis and confirmed result.
Scroll horizontally if needed.
| Record | Entry |
|---|---|
| Impact, task, timeline, severity and service/communication owners | [enter] |
| Recent changes, sanitized evidence and cause status | [enter] |
| Containment, diagnosis, action and result | [enter] |
| Required recovery point/interruption and selected mechanism | [enter] |
| Overwritten work, current-code compatibility and authorization | [enter] |
| Restored tasks / references / permissions / delivery and follow-up owners | [enter] |
