AZ-305 deep review — September 28, 2026¶
The AZ-305 guide was read in full and mapped to 49 objectives in twelve groups. It now has seven worked examples, ten labs and 48 explained original checks. 38 offline assertions passed: 26 arithmetic/catalog checks and 12 assertions in one published SQLite transaction/replay example. No Azure deployment, SDK integration, migration, failover, restore or paid-content execution occurred. Independent human review remains pending.
Blueprint and certification¶
The official study guide retains its April 17, 2026 baseline; objective and status hashes are unchanged, with no upcoming objective announcement. The credential profile lists ten languages, renewal and Azure Administrator Associate as a prerequisite for earning the Expert certification. This is not a prerequisite merely to sit AZ-305. Exact exam duration was not verified. Practice Assessment and sandbox links were reviewed at their public landing pages only.
The credential page displays a course-availability placeholder, but the standalone AZ-305 course is available and lists four instructor days and eight languages. The guide uses that direct course source and does not infer that official training is absent.
Changes that affect design decisions¶
| Area | Primary evidence and applied change |
|---|---|
| Logging | Inspect the Monitor Logs table-plan matrix with cells preserved. Basic supports Simple Log Alerts; Auxiliary/Lake lacks alerts and workspace replication. Add query costs, eligibility and operational requirements to the decision. |
| Identity and keys | Separate new-vault RBAC defaults and API retirement from existing-vault access migration. Add object backup scope, version constraints and same-subscription/geography restore. |
| Data and recovery | Explain SQL failover-group listeners and customer-managed recovery ownership. Connect Cosmos partition choices to transaction boundaries and skew. |
| Failure analysis | Region pairing is not automatic workload HA. Apply failure mode analysis to serving requests and performing recovery separately. |
| Messaging | Compare metadata-only Geo-Disaster Recovery with message/state Geo-Replication. Distinguish mode, lag, promotion, authorization and preview limits. |
| Business effects | Apply the idempotent-consumer and transactional-outbox boundaries. The original code demonstrates a consumer inbox and local atomic update, not a cloud exactly-once guarantee. |
| Compute and APIs | Separate Functions HTTP deadlines from execution timeout. Compare the APIM v2 matrix with classic Premium multi-region, including primary-region management dependencies. |
| Configuration | App Configuration replicas are eventually consistent. Explain client failover and stale/default behavior instead of implying immediate global propagation. |
| Networking | Scope the private-subnet default to new VNets and the relevant API versions. Add explicit egress and Front Door migration readiness, including postmigration DNS work. |
Lifecycle follow-up¶
| Date | Scope and treatment |
|---|---|
| September 30, 2026 | Service Bus older SDKs retire; SBMP becomes unusable after the deadline. These effects differ. Functions v3 on Linux Consumption stops running after this date. |
| February 27, 2027 | Key Vault control-plane APIs earlier than 2026-02-01 retire. API migration does not automatically change an existing vault's access model. |
| March 31, 2027 | Front Door classic and Redis Enterprise/Enterprise Flash retire; remaining Enterprise caches are disabled starting April 1. |
| September 30, 2028 | Redis Basic/Standard/Premium retire, with disablement from October 1; Linux Consumption hosting also retires September 30. Distinguish hosting retirement from the earlier Functions runtime deadline. |
The maintenance event queue now includes September 30 and October 12 checks, plus advance reviews on February 1 and March 1, 2027. These are source-review triggers; they do not execute cloud migrations. Existing weekly monitoring and age-based checks continue to cover later dates.
Blog intake¶
All three rendered pages returned minimal article shells. Their publicly embedded article bodies and author/date metadata were extracted and hashed in the receipt. Reading was bounded as follows:
- Fault types in Azure, Zoran Jovanovic, April 1, 2026: introduction, FMA/taxonomy and partial-regional discussion through the first Switzerland incident heading. Historical incident details were not verified. The author explicitly disclaims Microsoft endorsement; the registry records personal expert commentary. Adopt the fault-matrix exercise, not qualitative likelihood labels as probabilities.
- APIM Premium v2 GA, Sreekanth Thirthala, November 19, 2025: main announcement read. Use it to introduce separate gateway/backend network designs. Current product exclusions control; launch geography and broad marketing language are not current support guarantees.
- Service Bus Geo-Replication GA, Eldert Grootenboer, December 17, 2025: main announcement read. Add an acknowledgment, latency and promotion worksheet. Current documentation still marks partitioned-namespace support preview and excludes large messages.
Learning resources¶
The public Pluralsight path lists five courses totaling 13h 48m and three labs totaling 4h 15m: 18h 03m combined versus its rounded 18-hour heading. Courses date from 2024–2025; the most recently listed lab is September 18, 2026. New lab metadata does not establish that all video coverage is current.
Direct O'Reilly and Udemy retrieval remains blocked. Indexed O'Reilly provider metadata confirms the ACI/Adam Gordon course at May 2024 / 18h 46m, and the Exam Ref at November 2022 / 192 pages / 5h 32m estimated reading time. The browser could read the Christopher Nett Udemy page: January 2026, 13 sections, 164 lectures and 16h 51m. A provider aggregator's different AZ-305 course duration was not substituted for this exact product.
MeasureUp now lists 164 questions and a September 2026 update, replacing the guide's 148-question / April 2024 metadata. Paid questions and explanations were not opened. Whizlabs review was limited to public planning paragraphs. Savill's video returned a shell, so its historical four-hour estimate is qualified; the linked whiteboard image was not visually inspected. None of these public catalog checks establishes complete coverage or paid-content correctness.
Verification and remaining limits¶
The worked examples cover serial availability, skewed partition demand, queue catch-up, cold-cache load, migration convergence, recovery critical paths and transaction rollback/replay. The 12 code assertions run only in a local in-memory SQLite database; they do not exercise concurrent workers, a broker, external payment APIs or regional recovery. All capacity inputs are fictional.
Local repository validation, reviewed-resource consistency, all 174 unit tests,
strict site build, generated-site validation and whitespace checks passed.
The receipt at ADLC_Docs/operations/2026-09-28-az-305-deep-review.json preserves
the earlier review, objective hashes, 49 mappings, individual source boundaries,
blog body hashes and validation evidence. Retained references were fetched;
long product pages and every linked procedure were not exhaustively reread.
The review process detects source changes and queues work. It does not replace semantic review, live labs or independent human review. Completion of this guide does not mark the remaining Microsoft guides complete; the coverage tracker records the remaining work.