lock_openFree sampleNetworking and Content Deliveryhard
An EC2 instance in a private subnet cannot reach a partner API over TCP 443, and the team enables VPC Flow Logs on its network interface. For the outbound connection they see an ACCEPT record on the SYN packet leaving the instance but a matching REJECT record for the returning traffic on the same interface. Which configuration most directly explains an accepted outbound packet whose return traffic is logged as REJECT?
- AThe instance security group is missing an outbound rule for TCP 443, so the egress request itself is being discarded before it can leave the network interface.
- BThe security group is stateful, so once the outbound request is allowed the return traffic is automatically denied unless a separate inbound rule names the partner.
- CA network ACL on the subnet allows the outbound request but lacks an inbound entry for the ephemeral return port range, so it drops the reply because ACLs are stateless.check_circle Correct
- DVPC Flow Logs record REJECT whenever return traffic crosses an internet gateway, so the entry reflects normal logging behaviour rather than any blocked packet.
Read VPC Flow Log ACCEPT and REJECT actions to recognise that a stateless network ACL missing the ephemeral return range drops reply traffic. Security groups are stateful, so allowed outbound connections have their replies admitted automatically. Network ACLs are stateless and evaluate inbound and outbound rules separately, so a subnet ACL that permits the egress request but lacks an inbound rule for the ephemeral port range will drop the return packets. Seeing ACCEPT on the outbound flow and REJECT on the return on the same interface points squarely at a stateless ACL gap rather than a security group problem.
Why A is wrong: A missing egress rule would cause the outbound SYN to be rejected, but the flow log shows the outbound packet was ACCEPTed, so the egress security group rule is clearly present.
Why B is wrong: Security group statefulness means allowed outbound traffic has its return automatically permitted, so a stateful group would never be the cause of a rejected reply here.
Why C is correct: Network ACLs are stateless and evaluate each direction independently, so an ACL that permits egress but omits the ephemeral inbound range drops return traffic, matching an ACCEPT out and REJECT back.
Why D is wrong: Flow logs do not mark return traffic as REJECT simply for traversing a gateway; a REJECT action always denotes a packet actually denied by a security group or network ACL.
lock_openFree sampleNetworking and Content Deliverymedium
A public website runs in a primary Region behind an Application Load Balancer, with an identical standby stack in a second Region. The team wants Route 53 to send all visitors to the primary endpoint while it is healthy and to switch traffic to the standby only when the primary stops passing health checks. Which Route 53 routing policy meets this requirement?
- AConfigure a failover routing policy with the primary record as primary and the standby as secondary, each tied to a Route 53 health check.check_circle Correct
- BConfigure a weighted routing policy with a high weight on the primary record and a low weight on the standby so most traffic favours the primary endpoint.
- CConfigure a latency-based routing policy so visitors are sent to whichever Regional endpoint gives them the lowest network latency at query time.
- DConfigure a multivalue answer routing policy returning both endpoints so resolvers can retry the standby whenever the primary endpoint stops responding.
Use a Route 53 failover routing policy with health checks to keep traffic on a primary endpoint and switch to a standby only when the primary is unhealthy. Failover routing implements an active-passive pattern: Route 53 answers with the primary record while its associated health check is healthy and answers with the secondary record only after the primary fails its health check. Weighted, latency, and multivalue policies all distribute traffic across records while both are healthy, so none of them keeps the standby idle until the primary becomes unavailable.
Why A is correct: Failover routing serves the primary record while its health check passes and switches to the secondary only when the primary is unhealthy, which is exactly the active-passive behaviour requested.
Why B is wrong: Weighted routing always sends a proportion of traffic to every record by its weight, so the standby would receive requests even while the primary is healthy, which is not the active-passive behaviour required.
Why C is wrong: Latency routing picks the Region that is fastest for each client rather than honouring a fixed primary, so it would split traffic by latency instead of keeping the standby idle until the primary fails.
Why D is wrong: Multivalue answer routing returns several healthy records at once for client-side load spreading, so it actively hands out the standby alongside the primary rather than reserving it for failover.
lock_openFree sampleNetworking and Content Deliverymedium
A CloudFront distribution sits in front of a dynamic application. Each request carries a unique tracking value in a query string that does not change the rendered page, and CloudFront is currently configured to include the full query string in the cache key. The cache hit ratio is very low because almost every request is treated as unique. The team wants to raise the hit ratio while still passing the tracking value through to the origin. Which change to the cache behaviour achieves this?
- ASet the cache policy to include the tracking query string in the cache key but lower the minimum time to live so objects expire and are revalidated more frequently.
- BDisable caching entirely on the behaviour so every request goes to the origin, which removes the low hit ratio problem by making each response freshly generated.
- CAttach an origin request policy that forwards the tracking query string to the origin, and use a cache policy that excludes that query string from the cache key.check_circle Correct
- DAdd a CloudFront function on the viewer request event that strips the tracking query string from every request before CloudFront builds the cache key for it.
Separate the CloudFront cache key from forwarded values by excluding a query string from the cache policy while forwarding it with an origin request policy. CloudFront builds a cache key from the values named in the cache policy; any query string in that key creates a distinct cached object. A high-cardinality tracking value therefore wrecks the hit ratio. The fix is to remove that value from the cache policy so equivalent pages map to one object, and to add an origin request policy that still forwards the value to the origin. This decouples what makes a cache entry unique from what the origin receives.
Why A is wrong: Keeping the tracking value in the cache key still produces a unique key per request, so the hit ratio stays low regardless of the minimum time to live, which controls only freshness not key uniqueness.
Why B is wrong: Turning caching off means no requests are ever served from the edge, so the hit ratio falls to zero rather than rising and the origin load increases, which is the opposite of the goal.
Why C is correct: Excluding the tracking value from the cache key via the cache policy lets identical pages share one cached object, while the origin request policy still forwards that value to the origin, satisfying both requirements.
Why D is wrong: Stripping the value at the edge would raise the hit ratio but also removes the tracking query string entirely, so the origin never receives it, breaking the requirement to pass it through.
lock_openFree sampleNetworking and Content Deliveryhard
Clients of an application behind an Application Load Balancer report intermittent HTTP 504 responses. Operators need to confirm whether the load balancer is timing out while waiting for the targets to respond, and to see the time spent at each stage of each affected request. Which data source provides the per-request stage timings needed to confirm a target-side delay?
- AApplication Load Balancer access logs, which record the request processing, target processing and response processing times along with the status code for every request.check_circle Correct
- BThe CloudWatch TargetResponseTime metric, which records the exact start and end timestamps of each individual request so operators can replay the slow ones.
- CVPC Flow Logs on the load balancer subnets, which expose the request processing and target processing durations recorded for each connection.
- DAWS CloudTrail data events for the load balancer, which log the latency of every client request and the response the targets returned to the balancer.
Use Application Load Balancer access logs to read per-request processing stage timings and attribute HTTP 504 errors to slow targets. Each ALB access log entry includes request_processing_time, target_processing_time and response_processing_time alongside elb_status_code and target_status_code. A 504 with a target_processing_time set to a large value, or to -1 where the target never responded, confirms the load balancer reached its target timeout while waiting on the backend. CloudWatch metrics aggregate this signal but cannot pinpoint individual requests, and flow logs and CloudTrail operate at the wrong layer.
Why A is correct: ALB access logs capture per-request request, target and response processing times plus the elb_status_code, so a long target processing time alongside a 504 confirms the targets are responding too slowly.
Why B is wrong: TargetResponseTime is an aggregated metric reported as statistics over a period, so it shows trends but cannot reconstruct the stage timings of a single named request.
Why C is wrong: Flow logs record connection metadata such as addresses, ports and byte counts, not HTTP-level processing stage timings, so they cannot attribute a 504 to slow targets.
Why D is wrong: CloudTrail records management and supported data-plane API activity, not per-request HTTP latency for load balancer traffic, so it cannot reveal target response times.
lock_openFree sampleNetworking and Content Deliverymedium
A voice and gaming backend runs on EC2 instances behind Network Load Balancers in eu-west-1 and us-east-1, serving players over both TCP and UDP. Enterprise customers must allowlist a small, fixed set of entry-point IP addresses that never change, and traffic must shift to the healthy Region within seconds if one Region degrades, without waiting for any DNS record to expire. Which TWO actions together meet this requirement? (Select TWO.)
- ACreate an AWS Global Accelerator standard accelerator and assign its two static anycast IP addresses as the published entry points for customer allowlisting.check_circle Correct
- BAdd the Network Load Balancers in both Regions as endpoints in Global Accelerator endpoint groups so edge health checks divert traffic to the healthy Region.check_circle Correct
- CPlace an Amazon CloudFront distribution in front of both Network Load Balancers and rely on origin failover in an origin group to switch Regions.
- DConfigure a Route 53 latency-based routing policy across the two Network Load Balancer endpoints with health checks for failover.
- EEnable cross-zone load balancing on both Network Load Balancers so they distribute connections evenly across all Availability Zones.
Global Accelerator supplies fixed anycast IPs and edge health checks that deliver static entry points and sub-DNS-time-to-live Regional failover for TCP and UDP workloads. The requirement combines two needs: unchanging allowlistable entry points and near-instant cross-Region failover for non-HTTP traffic. A standard Global Accelerator gives two static anycast IPs, and adding the Regional load balancers as endpoint-group members lets edge health checks reroute traffic in seconds without any DNS propagation. CloudFront is HTTP-only with no static IPs, Route 53 failover is bounded by resolver caching of the record, and cross-zone balancing is an in-Region concern, so none of those three meets the stated combination.
Why A is correct: Global Accelerator provides fixed anycast IP addresses at the edge that front TCP and UDP workloads, giving the stable entry points customers can allowlist.
Why B is correct: Endpoint groups with health checks let Global Accelerator stop sending traffic to an unhealthy Region almost immediately at the edge, with no DNS time-to-live to wait out.
Why C is wrong: CloudFront accelerates HTTP and HTTPS only and exposes no fixed allowlistable IPs, so it cannot front a UDP gaming protocol or satisfy the static-IP requirement.
Why D is wrong: Route 53 failover depends on resolvers honouring the record time-to-live, so cached answers keep sending players to a failed Region for longer than seconds, and the IPs are not fixed.
Why E is wrong: Cross-zone balancing spreads load between zones inside one Region but provides no static cross-Region entry points and no automatic Regional failover.