Secure, optimize, and deploy database solutions
Watch for a real control aimed at the wrong problem
An application holding a valid Microsoft Entra ID token calls an Azure Cosmos DB for NoSQL account and gets 403 Forbidden, and granting its identity a broader role changes nothing, because the IP firewall answers every request from an address outside its allowed list with 403, whatever credential the caller holds. That is this domain's pattern; it carries 30–35% of the exam, second only to Design and develop database solutions at 40–45%; the header's 32.5% is the midpoint of this domain's 30–35% range. Where that domain shapes the data model and the code, this one runs the account in production, and every question lands on one of four concerns: Security (who may reach the data, from where, and which field values a read shows), Availability and recovery (what survives an accidental deletion, a zone failure or a region outage), Performance (why an operation costs the request units it does, RUs being the normalized unit of work every operation is billed in, or takes the time it does, and who is warned when it drifts), and Fleet management (how the many accounts of one multitenant application, one per customer or tenant, share throughput and are monitored as one). Each concern offers several genuine controls, so the classic trap is a control from the wrong concern or aimed at the wrong failure: a read region cannot bring back a deleted container, because recovery points come only from backups, raising throughput cannot speed up a call whose time went to a busy client CPU, and a bigger fleet pool (throughput many tenant accounts share, under Fleet management) cannot push one busy physical partition (one of the internal partitions a container's throughput is divided across) past its per-partition ceiling. Name the failure first, then choose the control.
The domain unfolds in four steps, one per concern
Implement database security controls covers Security: network rules (the IP firewall, virtual network service endpoints and private endpoints) decide where a request may come from; a credential (a Microsoft Entra ID token, account keys or a resource token) and native data-plane RBAC (role-based access control stored inside the account, separate from the Azure RBAC that authorizes managing the account itself, the control plane) decide who may read and write which data; and dynamic data masking (DDM) decides which field values a read shows. Implement availability and recovery solutions covers Availability and recovery: for lost data, periodic backup mode, where the Azure Cosmos DB team restores a backup on a support request, or continuous backup mode, whose point-in-time restore lets you restore the data yourself to a chosen second; for outages, read regions, multi-region writes, zone redundancy and the failover mechanisms, per-partition automatic failover (PPAF) among them. Optimize and remediate database performance covers Performance: the four evidence sources (per-query instrumentation, SDK diagnostics, platform metrics and diagnostic logs) show where a cost or a delay comes from, the indexing policy and the integrated cache are the main levers, and Azure Monitor alerts warn you when a condition occurs. Manage and monitor fleets across accounts and subscriptions covers Fleet management: a fleet groups the accounts of one multitenant application, a fleetspace inside it can share a pool of throughput, and fleet analytics (preview) exports usage, cost and settings for every account, aggregated per hour. Unlike database shared throughput, which the containers of one database share, this pool spans the accounts in one fleetspace. The decision tree keeps this order.
When two answers both work, choose the one that grants or spends less
Within each concern, default to the control that grants the least access or adds the least capacity yet still solves the stated problem. In Security that means a managed identity in Microsoft Entra ID with native data-plane RBAC enforcing least privilege rather than account keys, and IP firewall rules or a private endpoint rather than the Accept connections from within Azure datacenters option, which admits requests from all of Azure, other customers' subscriptions included. In Availability and recovery it means adding only needed regions, because adding a read region increases costs for each region. In Performance it means getting the query metrics first (per-query instrumentation) and fixing the query or indexing policy before raising RU/s (request units per second), and in Fleet management it means pooling only when many tenants peak at different times, because with few tenants, or with tenants that are all idle or all busy together, a pool might not provide more benefit.
The four concerns of a production account
| Concern | What it decides | Controls to choose among | Drill into |
|---|---|---|---|
| Security | Who may reach the data, from where, and which field values a read shows | IP firewall, virtual network service endpoint, private endpoint; Microsoft Entra ID, account keys, resource tokens; native data-plane RBAC; DDM | Implement database security controls |
| Availability and recovery | What survives an accidental deletion, a zone failure or a region outage | Periodic or continuous backup mode; read regions, multi-region writes, zone redundancy; PPAF and the other failover mechanisms | Implement availability and recovery solutions |
| Performance | Why an operation costs the RUs or takes the time it does, and who is warned when it drifts | The four evidence sources; the indexing policy; the integrated cache; Azure Monitor alerts | Optimize and remediate database performance |
| Fleet management | How the many per-tenant accounts share throughput and are monitored as one | Fleet and fleetspace; pool; fleet analytics (preview) | Manage and monitor fleets across accounts and subscriptions |