AZ-204 — Frequently Asked Questions

Multiple revision mode enables traffic splitting and canary testing, but single revision mode is the documented Azure Container Apps zero-downtime deployment mechanism.

No. With ingress enabled, single revision mode keeps the existing revision serving 100% of traffic until the new revision is ready, which is exactly the zero-downtime behavior required.

With ingress enabled, the app receives HTTP traffic and can scale from zero automatically when a request arrives, so it becomes accessible without a custom scale rule.

No, but it doesn't hurt; a custom scale rule can be used with or without ingress, but enabling ingress is the key fix for accessibility.

With ingress disabled and no minReplicas or custom scale rule defined, Container Apps has no HTTP traffic trigger and defaults to zero replicas, leaving no way for the app to start back up.

A running instance alone is not enough — users still need the ingress endpoint to reach the app, so the outage would persist for external callers.

It applies per item (per document), measured as UTF-8 JSON length. Stored-procedure and transactional batch requests are also capped at 2 MB, but that does not reduce the size allowed for a single document.

They treat 2 MB as a shared budget for the whole batch, so File1 plus File2 would exceed it. Because the cap is per document, File2 is valid on its own and option E is too restrictive.

Yes. Cosmos DB for MongoDB supports documents up to 16 MB, which is one reason the NoSQL API's 2 MB per-item limit surprises candidates who have only used the MongoDB API.

Bounded staleness also prevents out-of-order reads, but it adds a staleness window that is stronger than the question requires. Consistent prefix meets the only stated ordering guarantee while maximizing read concurrency.

No. Session consistency guarantees ordering and read-your-writes only within a single client session, so clients in other sessions or regions are not covered by that ordering guarantee.

FeedOptions is used for queries (e.g., CreateDocumentQuery) and also exposes SessionToken, but App1 requires point reads and writes, which use RequestOptions.

No. ConnectionPolicy configures connection mode, retry, and timeout settings; it has no SessionToken property for session consistency.

CDN provides HTTPS and geo-filtering but lacks a Web Application Firewall, which is needed to meet the security requirements and protect against common exploits.

WAF supports geo-match custom rules that allow or block traffic based on the source country/region, satisfying the compliance requirement.

Azure Cache for Redis Standard tier already defaults to volatile-lru, so selecting it changes nothing on the instance. The extra read cache must come from lowering the maxmemory-reserved reservation instead.

You free memory for cached data, but the reservation also covers non-cache work such as replication during failover. Setting it too low risks instability, so lower it gradually and monitor evictions and latency.

Those settings reserve memory for non-cache operations and fragmentation; they do not reduce the connection spike and server load caused when a Redis node fails over during maintenance.

A failover moves all client connections onto the surviving node and server load spikes; recreating the connections redistributes them across both nodes.

Fixed-rate sampling sets an explicit SampleRate so the volume reduction is predictable and statistically consistent, and it synchronizes client and server sampling so related page views and HTTP requests still correlate.

A daily cap simply stops telemetry ingestion once the limit is reached, so data is lost instead of sampled, costs are not truly reduced, and no client-server correlation is preserved.

Event Domains support up to 100,000 topics and provide a single publishing endpoint, making them ideal for managing thousands of customers efficiently.

Yes, you can create event subscriptions on domain topics to filter events by type before processing, meeting the filtering requirement.

A custom topic supports a maximum of 500 event subscriptions, so thousands of customers would exceed that quota.

Event Grid domains allow up to 100,000 topics with per-customer subscriptions and filtering, making them the documented multi-tenant pattern.

Accept only negotiates the response media type, such as application/json. It signals nothing about evolvable enumeration members, so the service will still omit enum values your app was not built for.

If-Match carries an ETag for optimistic concurrency, protecting against lost updates on writes. Prefer expresses an optional behavior such as include-unknown-enum-members on reads.

TrackMetric accepts pre-aggregated metric telemetry with min, max, and standard deviation, while TrackEvent logs each occurrence as a separate custom event, increasing volume and cost.

You combine many measurements into a single telemetry item before sending, so fewer items are ingested and stored while statistical analysis remains valid.

The web test objects returned by Get-AzApplicationInsightsWebTest express the test generation through WebTestKind, not Type, so a Type filter matches nothing and the classic tests are never removed.

"ping" identifies legacy classic URL ping tests, while "standard" identifies the newer standard availability tests that support more validation features and are the migration target.

A workbook only queries and renders Application Insights data. It has no alert-action binding, so the appinsights1 alert still sends email without invoking the function.

Add an action group to the alert rule and include a Function (HTTP trigger) action targeting Function1's endpoint; the alert payload is then POSTed to the function.

An action group only defines who is contacted and how; it needs an alert rule as the trigger. Without a rule watching AI1's failed availability results, nothing ever fires and no email is sent.

The failing App1 test results by themselves notify no one. Azure Monitor delivers the owner email through the alert rule configured over those AI1 availability results.