AZ-305 Cheat Sheet
Quick-reference summary of the key facts for the AZ-305 exam.
Identity, governance and monitoring (25–30%)
- Phishing-resistant auth: FIDO2 security keys, Windows Hello for Business, certificate-based authentication. SMS and voice are the weakest options.
- Hybrid identity: PHS is the default (hashes sync to Entra ID, cloud auth, survives link failure). PTA keeps validation on-premises. Federation (AD FS) only when a specific requirement forces it.
- Conditional Access: always exclude an emergency-access (break-glass) account; test in report-only mode first; target groups, not users.
- Workload identity: managed identities for Azure resources (system-assigned = tied to one resource; user-assigned = shared); workload identity federation for external platforms like GitHub Actions.
- RBAC: control-plane roles (Contributor) do NOT grant data-plane access (need Storage Blob Data Reader etc.). Assign roles to groups at the highest safe scope; permissions inherit downwards.
- PIM: eligible assignments for just-in-time privileged access, with MFA, justification, approval, and time limits.
- Key Vault: Standard = software keys; Premium = HSM (FIPS 140-2 Level 2); Managed HSM = dedicated single-tenant HSM (Level 3). Always enable soft delete + purge protection; use private endpoints.
- Governance: management groups → subscriptions → resource groups; policies inherit downwards. Policy effects: Audit first, then Deny; Modify/Append add tags; DeployIfNotExists remediates. Locks: CanNotDelete vs ReadOnly.
- Monitoring: diagnostic settings route logs to Log Analytics (query/alert), storage (archive), or event hub (SIEM). Azure Monitor Agent + data collection rules (the legacy agent is retired). Metric alerts, log alerts (KQL), activity log alerts, resource health alerts. Sentinel = SIEM + SOAR.
Data storage (20–25%)
- SQL PaaS: Azure SQL Database for new apps; SQL Managed Instance when you need SQL Agent, cross-database queries, or near-100% compatibility; SQL on VMs only when PaaS cannot meet a hard requirement.
- Tiers: General Purpose (most workloads), Business Critical (lowest latency, always-on replicas), Hyperscale (up to 100 TB, near-instant backup), serverless (intermittent workloads), elastic pools (many databases, spiky usage).
- HA/DR for SQL: zone redundancy = datacentre failure; failover groups = automatic regional failover with one listener endpoint; active geo-replication = manual failover, up to 4 readable secondaries; point-in-time restore up to 35 days; long-term retention up to 10 years.
- Blob tiers: Hot (frequent), Cool (30+ days), Cold (90+ days), Archive (180+ days, hours to retrieve). Lifecycle policies automate transitions. Archive + early access = expensive.
- Data Lake Gen2: Blob Storage + hierarchical namespace + ACLs; the analytics lake for Synapse/Databricks/Fabric.
- Cosmos DB: partition key = the critical design choice (high cardinality, matches queries). Consistency: Session is the default; Strong only when reads must never be stale. RUs measure throughput; autoscale or serverless for variable load.
- Redundancy ladder: LRS (1 datacentre) → ZRS (zones) → GRS (paired region) → GZRS (zones + geo); RA- variants add read access to the secondary. Redundancy is NOT backup: soft delete + versioning protect against deletion; immutable storage protects against tampering.
- Integration: Data Factory = batch ETL/ELT (self-hosted IR for on-premises); Event Hubs = high-volume telemetry streams; Stream Analytics = real-time SQL-like processing; Event Grid = discrete event routing; IoT Hub = device management + telemetry. Purview = data governance/catalog.
Business continuity (15–20%)
- RPO = max data loss (time). RTO = max downtime. Read them first; they determine the technology. Over-engineering is as wrong as under-engineering.
- Azure Backup: VMs, SQL, Files, Blobs (operational vs vaulted). Vault storage: LRS/ZRS/GRS. Ransomware protection: soft delete, immutable vaults, multi-user authorisation (resource guard).
- Site Recovery: replicates running workloads for failover (VMs, VMware, Hyper-V). Recovery plans orchestrate ordered failover. Test failovers validate without disrupting production.
- Availability zones protect against datacentre failure; availability sets only within a datacentre (fault + update domains). Two VMs across zones = 99.99% SLA; in a set = 99.95%; single VM = 99.9%.
- Composite SLA = product of component SLAs. Three 99.9% services = 99.7%.
- Active-active needs multi-region-write data (Cosmos DB) or it is pointless; active-passive with failover groups is the simpler default for relational workloads.
Infrastructure (30–35%)
- Compute: VMs = full control; App Service = web/API PaaS with deployment slots; AKS = full Kubernetes; Container Apps = serverless containers without Kubernetes ops; Container Instances = one-off containers; Functions = event-driven (Consumption = scale-to-zero, cold starts; Premium = pre-warmed, VNet); Batch = parallel HPC jobs.
- Messaging: Service Bus = enterprise (ordering, sessions, topics, transactions, dead-lettering); Queue Storage = simple, cheap, high-volume; Event Grid = discrete event routing; Event Hubs = telemetry streams.
- APIs/caching/config: API Management = gateway (auth, rate limits, quotas, portal; Premium = multi-region + VNet). Redis = cache-aside, TTLs, Enterprise for geo-replication. App Configuration = settings + feature flags; secrets stay in Key Vault.
- Migration: landing zone FIRST (CAF). Rehost = fastest; refactor = PaaS with minor changes; rearchitect = rebuild. Azure Migrate = discover/assess/migrate; dependency analysis groups waves. DMS = online database migration. Data Box = offline bulk transfer (calculate network time first).
- Networking: address spaces must not overlap (on-premises or peered VNets). Peering is non-transitive. Hub-and-spoke with firewall + forced tunnelling (0.0.0.0/0 UDR). VPN = cheap, internet-based; ExpressRoute = private, guaranteed bandwidth, higher cost. Virtual WAN = managed multi-branch hub.
- Private access: Private Link/private endpoints = private IP, public access can be disabled. Service endpoints secure the public endpoint but do not remove it. Private DNS zones resolve private endpoint names.
- Load balancing: Front Door = global L7 HTTP(S), anycast, instant failover, WAF, caching. Traffic Manager = DNS-based global, protocol-agnostic, slower failover. Application Gateway = regional L7 (path/host routing, WAF). Load Balancer = regional L4 TCP/UDP.
- Security: Azure Firewall (Premium = TLS inspection + IDPS); WAF in Detection mode first, then Prevention; Bastion = RDP/SSH without public IPs; NAT Gateway = managed outbound, fixes SNAT exhaustion.
Advertisement