100+ Free Tencent Cloud Expert Developer Practice Questions

Explanation: SCF Provisioned Concurrency pre-warms a configurable number of function instances, eliminating cold-start initialization for those instances and removing p99 spikes. Switching triggers changes invocation semantics but does not eliminate cold starts. Increasing timeout does not prevent cold starts—it only prevents timeouts. Moving initialization inside the handler makes cold starts worse, not better.

Explanation: In TCM (which is built on Istio), a DestinationRule defines named subsets (v1, v2) and a VirtualService applies traffic-weight rules (5% to v2, 95% to v1) — the correct two-resource pattern for weighted canary releases. Replica-ratio canary is a poor approximation with no precision. PeerAuthentication controls mTLS, not traffic weights. The nginx annotation applies only to nginx ingress, not to service-mesh sidecar routing.

Explanation: CKafka (managed Kafka) handles millions of messages per second, supports partition-level ordering when producers route by message key, and allows consumer groups to read each partition sequentially. CMQ FIFO does not scale to 200 k/s bursts. RabbitMQ exclusive queues cannot maintain ordering across rebalances at that throughput. SCF+COS triggers are batch-oriented and do not provide per-key ordering semantics.

Explanation: In older versions of the tencentcloud Terraform provider, `cluster_version` was annotated `ForceNew`, meaning any change triggers resource destruction and re-creation. The standard remediation is to pin the version in Terraform state and perform the actual Kubernetes upgrade through the TKE console or tctl, then update the state with `terraform refresh` or a state import. Using `ignore_changes` suppresses the diff but does not perform the upgrade.

Explanation: KEYS * is an O(N) blocking command that holds the Redis event loop for its entire execution, causing latency spikes proportional to database size. Replacing it with SCAN (cursor-based, non-blocking) iterates keys incrementally without blocking other commands. Changing eviction policy affects memory management, not command blocking. Sharding reduces key count per shard but does not prevent KEYS * from blocking within each shard. Enabling slow-query logging diagnoses the problem but does not fix it.

Explanation: CODING DevOps supports native TCR Service Connections that store credentials securely and inject them without exposing secrets in pipeline YAML. The Docker plugin step references the connection by ID, handles login automatically, and avoids any credential leakage. Storing passwords as environment variables risks exposure in logs. Embedding API keys in base64 Kubernetes Secrets is not encrypted by default and still embeds credentials in code. Public TCR namespaces allow unauthenticated pulls, not pushes.

Explanation: PeerAuthentication STRICT enforces mTLS for all pod-to-pod traffic in the mesh. For the IngressGateway (which is outside the mesh peer plane), configuring a Gateway with TLS mode=SIMPLE terminates HTTPS externally and the gateway then communicates internally over mTLS. PASSTHROUGH mode does not terminate TLS at the gateway, so the backend service must handle TLS itself. PERMISSIVE mode allows plaintext, defeating mTLS enforcement. JWT RequestAuthentication is for user identity, not service-to-service mTLS.

Explanation: HPA targeting CPU 70% handles pod-level scale-out; the node pool cluster-autoscaler adds nodes when pods are pending. A preStop lifecycle hook with a sleep (e.g., 30 s) gives the load balancer time to deregister the pod before the container stops, allowing active connections to drain. VPA resizes CPU requests vertically rather than scaling out. KEDA is for custom/event metrics. Scheduled scaling cannot react to dynamic load.

Explanation: COS multipart upload allows resumability: the UploadId returned by InitiateMultipartUpload persists server-side. If the upload fails, you call ListParts with that UploadId to see which parts succeeded, then re-upload only the missing parts before calling CompleteMultipartUpload. Versioning tracks complete object versions, not partial uploads. Transfer Acceleration reduces latency but does not provide resume capability. Presigned URLs handle authentication, not resume.

Explanation: The Saga pattern is the recommended approach for distributed transactions in microservice architectures: each participant executes a local ACID transaction, publishes an event (e.g., to CKafka), and listens for events to trigger compensating actions on failure. 2PC with HTTP endpoints couples services tightly and has poor failure recovery. TencentDB GTM is for distributed MySQL sharding scenarios, not cross-service HTTP transactions. SCF has no built-in database transaction semantics.