AWS Certified Database Specialty DBS-C01 (DBS-C01) — Questions 1576–1650

Options B and C are correct because using DynamoDB Auto Scaling for writes will adjust capacity based on demand, and reducing provisioned reads to actual usage will save costs. Option A is wrong because switching to on-demand may increase costs if the workload is predictable. Option D is wrong because reducing writes would cause throttling.

Option E is wrong because using DAX does not reduce write costs.

Multi-AZ deployment with synchronous standby provides high availability and data durability, but does not improve read performance. Read replicas (Option B) improve read scalability but may have eventual consistency. Option C (single-AZ) lacks HA.

Option D (cross-region) is for DR.

Option B is correct because using a consistent hashing algorithm ensures that keys are evenly distributed across nodes, reducing evictions and improving hit ratio. Option A is wrong because increasing TTL may cause stale data but does not directly improve hit ratio. Option C is wrong because Cluster Mode Off would limit scalability but not necessarily improve hit ratio.

Option D is wrong because random TTLs can lead to premature evictions and reduce hit ratio.

Option B is correct because point-in-time recovery is performed through the RDS console or API by specifying the time. Option A is wrong because it restores the latest backup. Option C is wrong because it restores a manual snapshot.

Option D is wrong because automated backups are used for point-in-time recovery.

Option D is correct because Amazon CloudWatch Contributor Insights can analyze DynamoDB logs and identify the most accessed partition keys, helping to pinpoint hot partitions. Option A is incorrect because CloudWatch Logs Insights requires enabling VPC Flow Logs, which is not specific to DynamoDB. Option B is incorrect because AWS X-Ray traces requests but does not provide partition-level metrics.

Option C is incorrect because DynamoDB Streams capture changes, not access patterns.

Option B is correct because using AWS DMS with Aurora MySQL as the target allows a full load plus ongoing replication to minimize downtime. Option A is wrong because mysqldump would require application downtime during the dump and restore. Option C is wrong because an Aurora read replica can only be created from an existing Aurora instance, not from on-premises.

Option D is wrong because S3 export/import is for data warehousing, not for minimal downtime migration.

Option C is correct because dc2 nodes have limited local SSD storage; if the cluster runs out of disk space, writes time out. Option A is wrong because Redshift does not have a max_connections parameter like RDS. Option B is wrong because network ACLs would cause connection failures, not timeouts after connection.

Option D is wrong because WLM queue timeouts cause different errors.

Option C is correct because the error indicates connection pool exhaustion. Increasing max_connections allows more concurrent connections. Option A is wrong because the error is about connection slots, not idle sessions.

Option B is wrong because reserving more slots for superusers does not address the root cause. Option D is wrong because the error is not due to replication slots.

AWS DMS supports MongoDB as a source and DocumentDB as a target, with ongoing replication to minimize downtime. Option A (mongorestore) causes high impact. Option C (export/import) causes impact.

Option D (AWS SCT) is not for migration itself.

Option B is correct. The Performance Insights dashboard in Aurora provides SQL query metrics with low overhead and can publish logs to CloudWatch Logs. Option A is wrong because database audit logs are for security compliance, not performance analysis.

Option C is wrong because the general log is high overhead. Option D is wrong because slow query log only captures slow queries, not all queries.

Option C is correct because Amazon RDS for MySQL supports the 'wait_timeout' parameter, which automatically closes idle connections after the specified number of seconds. Reducing this value from the default 28800 seconds will terminate sleeping connections sooner. Option A is incorrect because manually killing connections is not automated.

Option B is incorrect because RDS Proxy does not terminate idle connections; it manages connection pooling. Option D is incorrect because modifying 'max_connections' does not affect idle connections.

Option B is correct because CloudTrail logs all KMS API calls, including encrypt and decrypt. Option A is wrong because CloudWatch Logs can store logs but does not capture KMS API calls by itself. Option C is wrong because AWS Config tracks resource configuration changes, not API calls.

Option D is wrong because S3 access logs are for S3.

Option A is correct because DynamoDB Auto Scaling can adjust capacity based on demand, and setting a target utilization of 70% allows headroom for spikes. Option B is wrong because on-demand capacity is simpler but can be more expensive for predictable workloads. Option C is wrong because increasing WCU to 8000 permanently would be costly.

Option D is wrong because SQS adds latency and does not solve the throttling issue in real-time.

Option C is correct because adding a random suffix to the partition key distributes writes across partitions, avoiding hot partitions. Option A is wrong because increasing WCU does not solve the partition hot spot, as the limit per partition is fixed. Option B is wrong because switching to on-demand still has per-partition limits.

Option D is wrong because a GSI does not change the base table's write distribution.

Option B is correct because converting to gp3 provides a consistent baseline of 3000 IOPS (or 3000 if using the default 3000 IOPS, though for 100 GB the baseline is 3000) without burst credits, eliminating the burst balance problem. Option A is incorrect because increasing gp2 volume size to 200 GB increases baseline to 600 IOPS and burst credits, but still relies on credits; the workload may still exceed baseline. Option C is incorrect because increasing instance size does not affect storage performance.

Option D is incorrect because enabling Performance Insights does not improve performance.

Option C is correct because a full backup followed by transaction log backups allows restoring to a point in time, minimizing data loss and downtime. Option A is wrong because only full backup requires longer downtime during restore and may lose transactions. Option B is wrong because differential backup alone is not sufficient; you need a full backup chain.

Option D is wrong because transaction log backup alone cannot restore without a full backup.

A and D are correct. Performance Insights provides database performance analysis. Enhanced Monitoring provides OS-level metrics.

B is wrong because CloudTrail logs API calls. C is wrong because VPC Flow Logs show network traffic. E is wrong because AWS Config tracks resource configuration.

Option D is correct because Performance Insights provides a database performance tuning and monitoring feature that helps you quickly assess the load on your database and determine when and where to take action. Option A is wrong because CloudWatch Logs captures database logs, not real-time performance data. Option B is wrong because Enhanced Monitoring provides OS-level metrics, not database query details.

Option C is wrong because AWS Trusted Advisor provides best practice checks but not query-level analysis.

Option C is correct because DynamoDB encrypts all data at rest using a single KMS key per table. The base table and all of its GSIs are encrypted with the same key. It is not possible to use a different KMS key for a GSI.

Option A is wrong because the GSI cannot have a separate key. Option B is wrong because KMS key policies do not allow per-index encryption. Option D is wrong because the table and its GSIs always use the same key.

Option A is correct because AWS KMS supports automatic annual rotation for customer-managed keys when enabled. You can enable automatic rotation using the KMS console or API. Option B is wrong because creating a new key and updating the DynamoDB table requires manual effort and is more complex.

Option C is wrong because AWS KMS does not support importing key material for symmetric keys; you can import key material for asymmetric keys but rotation is not automatic. Option D is wrong because manual rotation is not required and is more error-prone.

Option B is correct because ReadIOPS and WriteIOPS show the actual I/O operations. Option D is correct because DiskQueueDepth indicates pending I/O requests, which suggests contention. Option A is wrong because CPU utilization is compute, not I/O.

Option C is wrong because FreeStorageSpace is capacity, not performance. Option E is wrong because DatabaseConnections is about connections, not I/O.

Option C is correct because DMS needs network connectivity via VPC peering or VPN, and the source security group must allow DMS. Option A is wrong because VPC endpoints are not needed for DMS to source. Option B is wrong because DMS does not use an S3 endpoint for source connectivity.

Option D is wrong because DMS does not use an IAM role for database connectivity; it uses database credentials.

Option C is correct. The output shows 'StorageEncrypted: false' and 'KmsKeyId: null', which means the instance is not encrypted at rest. Option A is incorrect because the instance is available.

Option B is incorrect because the engine is MySQL. Option D is incorrect because the output clearly shows encryption status.

Option B is correct because using DynamoDB transactions ensures atomicity; if the S3 upload fails, the transaction can be rolled back. Option A is wrong because DynamoDB Streams with Lambda introduces eventual consistency and possible duplicates. Option C is wrong because writing to S3 first still leaves inconsistency if DynamoDB fails.

Option D is wrong because idempotency tokens help with duplicates but not atomicity.

Option A (AWS DMS with ongoing replication) minimizes downtime. Option B (Schema Conversion Tool) helps convert Oracle schema to PostgreSQL. Option D (data validation) ensures consistency.

Option C (homogeneous migration) is not needed; DMS works. Option E (Aurora Replicas) is for read scaling, not migration.

Option A is correct because Multi-AZ provides automatic failover within minutes, meeting the RTO. Option D is correct because a cross-region read replica with automatic backups allows replication with low RPO. Option B is wrong because manual snapshots have higher RPO.

Option C is wrong because a single-AZ instance with backups takes longer to restore. Option E is wrong because AWS DMS is for migration, not DR.

Option A (enable TLS) is security, not performance. Option B (create indexes) is critical for $lookup and $group. Option C (increase instance size) helps compute-intensive operations.

Option D (disable journaling) is not recommended for durability. Option E (use parallel scans) can speed up aggregations.

Option C is correct because ConsumedWriteCapacityUnits shows actual usage; comparing with ProvisionedWriteCapacityUnits helps identify throttling. Option A is wrong because UserErrors is for client-side errors. Option B is wrong because ThrottledRequests is a metric but not a CloudWatch metric name; the actual metric is ConsumedWriteCapacityUnits.

Option D is wrong because SuccessfulRequestLatency is about latency, not throttling.

AWS Database Migration Service (DMS) supports heterogeneous migrations from Oracle to Aurora PostgreSQL and can perform ongoing replication to minimize downtime. Option A is correct. Option B (AWS Snowball) is for large data transfer but not for ongoing replication.

Option C (SCT) helps with schema conversion but not data migration. Option D (Data Pipeline) is not designed for database migrations.

Option C is correct. The 'Memory limit exceeded' error often occurs when DMS cannot keep up with the volume of changes, especially for write-heavy workloads. Increasing the replication instance size provides more memory for buffering.

Option A is wrong because stopping and restarting the task does not address the root cause. Option B is wrong because reducing the batch size may not help; the issue is memory capacity. Option D is wrong because enabling multi-AZ does not increase memory; it provides high availability.

Option B is correct because DynamoDB with a fan-out on write pattern pre-joins posts into per-user feed tables, enabling fast reads with low latency. Option A (RDS with read replicas) would struggle with the high write volume and join complexity. Option C (ElastiCache as primary store) loses durability.

Option D (DynamoDB single table with GSI) would require expensive scans to filter followed users' posts.

Options A, C, and D are correct. DynamoDB global tables provide multi-region replication. DAX provides in-memory caching for low-latency reads.

On-demand capacity handles traffic spikes. Option B is wrong because TTL is for data expiration, not performance. Option E is wrong because Streams are for change data capture, not low-latency access.

Options A and D are correct because FreeableMemory shows available memory, and SwapUsage indicates swapping, which occurs when memory is exhausted. Option B (NetworkThroughput) is not memory-related. Option C (ReadIOPS) is I/O-related.

Option E (DatabaseConnections) is connection-related.

Option C is correct. Amazon RDS Multi-AZ with a Single-AZ deployment failover can be slow; using a Multi-AZ DB Cluster provides faster failover (typically under 1 minute) because it uses synchronous replication to two standby instances. Option A (increase instance size) does not affect failover time.

Option B (use Read Replica for failover) adds complexity and is not automatic. Option D (disable Multi-AZ) would increase failover risk.

Option B is correct because the 'service_class' column indicates the workload management (WLM) queue, which can show if the query is queued. Option A is wrong because 'query' is just the query ID. Option C is wrong because 'starttime' shows when the query started.

Option D is wrong because 'aborted' indicates if the query was canceled, not waiting.

Option B is correct because FreeStorageSpace indicates remaining capacity. Option A is wrong because DatabaseConnections is about connections. Option C is wrong because ReadIOPS measures I/O operations.

Option D is wrong because NetworkThroughput measures network traffic.

Option D is correct because the condition restricts DeleteItem to items that have only those attributes. Option A is wrong because the condition is on the delete action, not on the resource. Option B is wrong because the policy does allow DeleteItem with the condition.

Option C is wrong because the function has GetItem and Query permissions.

Option B is correct because you can failover to a specific reader instance by using the --target-db-instance-identifier parameter in the failover-db-cluster command. Option A is wrong because rebooting the writer will cause a failover but may choose any reader, not necessarily instance-2. Option C is wrong because you cannot directly promote a reader to writer; failover must be initiated.

Option D is wrong because modifying the cluster does not trigger failover.

The output shows port 3306, which is the default port for MySQL and MariaDB. Amazon RDS for MySQL uses port 3306. Aurora MySQL also uses 3306, but the endpoint pattern is RDS, not Aurora-specific.

PostgreSQL uses 5432, Oracle uses 1521, SQL Server uses 1433.

Creating appropriate indexes can speed up queries by avoiding table scans. Increasing the instance size provides more memory and CPU for query execution. Enabling query caching can store results of repeated queries.

Changing isolation level and enabling parallel query may not help in all cases and may have side effects.

Option C is correct because IAM database authentication requires the RDS instance to have the IAM DB authentication enabled; if it is not enabled, the authentication token will be rejected. Option A is wrong because the error 'Access denied' suggests authentication failure, not a network issue. Option B is wrong because SSL/TLS is required for IAM auth, but the error would be different if SSL was not used.

Option D is wrong because the RDS resource-based policy cannot grant the IAM user permission to generate the token; that requires an IAM policy attached to the user or role.

Options B, C, and D are correct. IAM roles with least privilege restrict access. AWS CloudTrail logs API calls.

VPC endpoints ensure traffic stays within AWS network and can be controlled via endpoint policies. Option A is for auditing, but CloudTrail is more comprehensive. Option E is for encryption, not access control.

DynamoDB stores items with the same partition key in sorted order by sort key. Queries return items in order by default. Option C is correct because if the application uses eventually consistent reads, the data might be stale.

Option A is wrong because GSIs don't affect base table queries. Option B is wrong because the query returns items in order, but eventually consistent reads may not reflect latest writes. Option D is wrong because pagination is not the cause.

Option D is correct because increasing the node size provides more CPU and memory, reducing evictions and latency. Option A is wrong because a read replica helps with read scaling but the node is a primary. Option B is wrong because Redis Cluster mode requires multiple shards and may add complexity.

Option C is wrong because reducing TTL might increase cache misses.

Option C is correct because CloudWatch provides disk I/O metrics for EC2 instances. Option A is wrong because RDS is for managed databases. Option B is wrong because DynamoDB is NoSQL.

Option D is wrong because S3 is object storage.

Option C is correct because changing the backup retention period takes effect immediately without reboot. Option A is incorrect because it requires a reboot. Option B is incorrect because it requires a reboot.

Option D is incorrect because it requires a reboot.

Option C is correct because Amazon RDS Read Replicas allow offloading read traffic from the primary DB instance without application changes. Option A is wrong because DynamoDB Accelerator (DAX) is for DynamoDB, not RDS. Option B is wrong because Multi-AZ is for high availability, not read scaling.

Option D is wrong because ElastiCache requires application changes to use caching logic.

Option B is correct because AWS DMS with ongoing replication can minimize downtime by capturing changes. Option A is incorrect because native export/import requires downtime. Option C is incorrect because RMAN backup and restore requires downtime for restore.

Option D is incorrect because copying dump files also requires downtime.

Automated backups with a retention period of 35 days provide point-in-time recovery. Cross-region snapshots via AWS Backup provide additional protection. A read replica can be promoted quickly for failover.

Option A (daily manual snapshots) may not meet the 5-minute RPO. Option D (single-AZ) is not as resilient. The correct options are B, C, and E.

Options A, B, and D are correct. AWS SCT converts the schema and code from SQL Server to MySQL (A). AWS DMS performs the full data load and ongoing replication (B).

After migration, the application connection string is updated to point to the Aurora writer endpoint (D). Option C is wrong because creating a read replica from Aurora to on-premises is not supported. Option E is wrong because the reader endpoint is for read traffic, not for the primary application connection.

Option D is correct because the EC2 instance is in the same VPC, and using the security group ID is the most secure and specific. Option A is too broad. Option B is unnecessary since the RDS endpoint is not an IP.

Option C would allow all traffic in the VPC, but using the security group is better.

Options A and D are correct. Increasing the replica instance size gives it more resources to apply changes. Reducing write-heavy DML on the primary reduces the volume of changes to replicate.

Option B is incorrect because enabling Multi-AZ on the primary does not reduce lag. Option C is incorrect because binary logging is required for replication. Option E is incorrect because increasing the replica's allocated storage does not directly reduce lag.

Option B is correct because DMS requires full recovery model to capture all transactions. Option E is correct because DMS uses MS-CDC (change data capture) to capture ongoing changes. Option A is wrong because DMS does not require MS-Replication; it uses its own CDC mechanism.

Option C is wrong because DMS uses a replication instance, not an EC2 instance agent. Option D is wrong because cross-VPC peering is not required; DMS can connect via VPN/Direct Connect.

Option C: Increasing storage to 600 GB increases baseline IOPS from 900 to 1800, reducing reliance on burst credits. This is cost-effective compared to switching to io1. Option A: Increasing to io1 with 3000 IOPS would be more expensive.

Option B: Adding a read replica does not help with write latency. Option D: Enabling Performance Insights only helps diagnose, not resolve.

Option D is correct because the most secure approach is to grant the EC2 IAM role permission to read the secret using an IAM policy attached to the role, and then have the application call the Secrets Manager API to retrieve the secret at runtime. This avoids storing secrets in code or configuration files. Option A is wrong because hardcoding credentials defeats the purpose.

Option B is wrong because storing the secret in a configuration file is not secure. Option C is wrong because you should not attach a resource-based policy to the EC2 instance; IAM roles are the correct mechanism.