CCNA Data Operations and Support Questions

Options A, B, and C are correct. A: Increasing buffer size and interval allows Firehose to batch more records, reducing the number of PUT requests. B: Increasing the number of shards in the source Kinesis stream provides more write capacity.

C: Requesting a service quota increase for Firehose can raise the default limits. Option D is wrong because S3 bucket versioning does not affect Firehose capacity. Option E is wrong because CloudWatch alarms only alert, they do not resolve capacity issues.

Option D is correct because increasing EBS storage per instance provides more disk space, and using spot instances reduces cost. Option A is wrong because adding more instances may not address the per-instance disk space issue and increases cost. Option B is wrong because increasing core nodes only helps if the shuffle data is distributed; spot instances reduce cost but may cause interruptions.

Option C is wrong because switching to D2 instances is more expensive and may not be needed.

Option B is correct because Glue emits a 'glue.driver.aggregate.numRecords' metric for the number of records processed. Option A is wrong because 'glue.driver.aggregate.elapsedTime' is for time. Option C is wrong because 'glue.driver.aggregate.bytesRead' is for bytes.

Option D is wrong because 'glue.driver.aggregate.recordsRead' is not a standard metric.

Option B and D are correct. Partition pruning is most effective when the partition column is used correctly in the WHERE clause, and using columnar formats like Parquet with predicate pushdown reduces data scanned. Option A is wrong because adding more partitions may not help if the filter is not applied.

Option C is wrong because converting to CSV would increase data scanned. Option E is wrong because using a different file format may not help if partitioning is not leveraged.

Options A and D are correct. A: S3 bucket policies can deny access to specific IP addresses or VPC endpoints; checking them can reveal explicit denies. D: EMR roles may need an IAM policy for S3 access; verifying the policy ensures correct permissions.

Option B is wrong because S3 Block Public Access does not affect IAM-based access. Option C is wrong because S3 endpoints in the VPC are not required if using public internet. Option E is wrong because S3 versioning does not affect access permissions.

A and D are correct: Partitioning reduces scanned data, and converting to Parquet is already done but optimizing file size (e.g., 64 MB) further improves performance. B (compression) is already assumed in Parquet. C (small files) actually hurts performance.

E (no partitioning) is not an optimization.

Option C is correct. The AutomatedSnapshotRetentionPeriod is set to 1 day, which allows only 1 day of point-in-time recovery. To support 35 days, this value must be increased to 35.

Option A is incorrect because the cluster is already encrypted. Option B is incorrect because the number of nodes does not affect snapshot retention. Option D is incorrect because the subnet group does not affect retention.

Options A, B, and E are correct. Job bookmarks can skip data, incorrect mapping can lose columns, and dynamic frame transformations can drop rows if errors occur. Option C is wrong because disabling encryption does not cause data loss.

Option D is wrong because increasing timeout does not cause data loss.

Options B, C, and D are correct. The Lambda function must be created (B), Firehose must be configured to invoke the Lambda function (C), and the IAM role must allow Firehose to invoke Lambda (D). Option A is wrong because Lambda@Edge is for CloudFront.

Option E is wrong because S3 events are not used for Firehose transformation.

Option B is correct because using DMS with change data capture (CDC) captures ongoing changes and replicates them with minimal downtime. Option A is wrong because full load only captures a snapshot. Option C is wrong because AWS Schema Conversion Tool does not handle data replication.

Option D is wrong because exporting and importing does not provide ongoing replication.

Options A, B, and D are correct. Increasing instance memory, tuning memory fractions, and enabling dynamic allocation help manage memory. Option C is wrong because more cores per node increases parallelism, potentially worsening memory pressure.

Option E is wrong because instance store volumes are for temporary storage, not memory.

Option B is correct because MFA Delete must be disabled before versioning can be suspended. Option A is incorrect because MFA Delete is already enabled, not disabled. Option C is incorrect because versioning cannot be suspended directly.

Option D is incorrect because the error is about MFA Delete, not permissions.

Option A is correct because DS_DIST_ALL indicates a cross-node redistribution; using DISTSTYLE KEY on the join columns can reduce data movement. Option C is correct because SORTKEY can speed up joins and aggregations by reducing data scans. Option B is wrong because increasing node count may help but is not a targeted fix.

Option D is wrong because VACUUM reclaims space and sorts, but does not address distribution issues. Option E is wrong because CNAME is a DNS concept, not a Redshift feature.

Option B is correct because VACUUM sorts the data and reclaims space, improving query performance. Option A is wrong because ANALYZE updates statistics but does not sort. Option C is wrong because increasing the number of nodes may help but is not the most direct fix for unsorted data.

Option D is wrong because changing distribution style would require recreating the table.

Partitioning the table and converting to columnar formats like Parquet reduce the amount of data scanned, improving performance. Option C is wrong because using SELECT * scans all columns. Option D is wrong because increasing timeout does not improve performance.

Option E is wrong because it is not a standard optimization.

Option D is correct because after a significant number of rows are inserted, the sort keys may become unsorted, requiring a VACUUM REINDEX to restore the sort order. Option A is incorrect because DISTKEY affects distribution, not sort key efficiency. Option B is incorrect because compression is for storage, not sorting.

Option C is incorrect because WLM configuration affects concurrency, not scan type.

Option B is correct because 'groupFiles' only works when the input data is already small and needs to be coalesced. However, if the input is large and the job writes output, the output file size is determined by the number of Spark partitions, not grouping. The grouping feature only applies to reading input files.

Option A is wrong because the setting is correct. Option C is wrong because grouping is a read-time feature, not write-time. Option D is wrong because grouping does not require repartitioning.

Option B is correct because the log shows the function used only 64 MB of the allocated 128 MB memory, indicating that memory is not the issue. The function timed out after 30 seconds, and the duration is 30001.23 ms, which is just over the default timeout of 30 seconds. Therefore, increasing the timeout will resolve the issue.

Option A is incorrect because memory usage is low. Option C is incorrect because the function is not hitting memory limits. Option D is incorrect because SQS visibility timeout does not affect Lambda execution timeout directly; the Lambda timeout is separate.

Option B is correct because enabling managed scaling allows EMR to automatically add On-Demand instances to replace reclaimed Spot instances, ensuring capacity without manual intervention. Option A is incorrect because using all Spot instances increases risk. Option C is incorrect because using all On-Demand increases cost significantly.

Option D is incorrect because increasing On-Demand count in core node group may help but does not dynamically adjust; managed scaling is more effective.

Option B is correct because object tags are not returned by head-object; they require a separate get-object-tagging call. Option A is wrong because the command works. Option C is wrong because bucket policies affect access, not tag visibility.

Option D is wrong because lifecycle rules do not remove tags.

Option A is correct because EMR's automatic node replacement detects interruptions and launches new instances. Option B is manual. Option C uses on-demand which increases cost.

Option D is for checkpointing but doesn't automatically replace instances.

Option D is correct because the SVL_QUERY_SUMMARY system view provides information about disk spills for each query step, including the number of bytes spilled to disk. Option A is incorrect because STL_SCAN is for table scans, not spills. Option B is incorrect because STV_TBL_PERM shows permanent table storage, not temporary spills.

Option C is incorrect because SYS_QUERY_DETAIL provides general query details but not spill information.

Option A is correct because incrementally processing only new files avoids reprocessing. Option B (full reload) is inefficient. Option C (versioning) handles overwrites but not late-arriving data.

Option D (Kinesis) is for streaming, not batch with late data.

Option B is correct because the error indicates that the executor ran out of memory (10 GB used of 10 GB limit). Increasing the Spark executor memory to 12 GB (since G.1X has 16 GB total, leaving room for overhead) will prevent the container from being killed. Option A is wrong because increasing the number of workers does not increase per-executor memory.

Option C is wrong because reducing the number of workers reduces total memory but does not fix per-executor limits. Option D is wrong because converting to Parquet reduces file size but does not change the memory limit per executor.

Option A is correct because the error is generic; checking CloudWatch Logs provides more details. Option B is wrong because it assumes a JDBC driver issue without evidence. Option C is wrong because testing a small dataset may not reproduce the issue.

Option D is wrong because the error is not specific to table structure.

Option C is correct because Step Functions has a default execution limit per account; hitting it causes this error. Option A is wrong because Lambda concurrency limits cause throttling errors, not state machine execution limits. Option B is wrong because execution time limits cause 'ExecutionTimedOut' errors.

Option D is wrong because the error is specific to state machine executions, not Lambda.

Option B is correct because disabling job bookmarks forces the Glue job to reprocess all data, which will include the missing records. Option A is wrong because increasing the number of workers does not address the bookmark issue. Option C is wrong because updating the IAM role does not affect bookmarks.

Option D is wrong because job bookmarks do not have a 'reset' API; they are managed via the job configuration.

Options A, C, and E are correct. Option A: Increasing parallelism improves throughput. Option C: Increasing checkpoint interval reduces checkpoint failures.

Option E: Using event time helps with out-of-order data. Option B is wrong because decreasing parallelism reduces throughput. Option D is wrong because disabling checkpointing hurts reliability.

Option C is correct because if the Lambda function's concurrency limit is reached, it cannot process all shards simultaneously, causing lag. Option A is wrong because increasing shards would increase parallelism but is not the root cause. Option B is wrong because Lambda can process from multiple shards.

Option D is wrong because DynamoDB write capacity could be a bottleneck but is less likely than Lambda concurrency.

Options B and D are correct. Option B: PercentageDiskSpace metric with threshold 80. Option D: CPUUtilization metric with threshold 90.

Option A is wrong because WLM is not for disk. Option C is wrong because NetworkReceiveThroughput is not relevant. Option E is wrong because DatabaseConnections is not about CPU.

Option B is correct because concurrency scaling adds additional capacity on demand and is cost-effective for variable workloads. Option A is incorrect because reserved instances require upfront payment and are for steady-state, not variable. Option C is incorrect because reducing node count may impact performance during peak hours.

Option D is incorrect because RA3 nodes are for managed storage, not cost reduction for variable workloads.

Option D is correct because the policy requires the s3:x-amz-server-side-encryption header to be set to AES256. If the request does not include that header, the condition is not met, and the request is denied with a 403 Access Denied error. Option A is wrong because the object is not uploaded.

Option B is wrong because the condition requires AES256, not SSE-C. Option C is wrong because the condition is not met if the header is absent; the request is denied.

Option B is correct. Using Glue job bookmarks enables incremental processing and the ability to resume from the last successful checkpoint. Staging the data in a temporary location and moving it atomically ensures that Athena sees only complete data.

Option A is wrong because increasing worker capacity does not prevent partial writes. Option C is wrong because using Lambda for conversion is less scalable and error-prone. Option D is wrong because partition projection does not solve the atomicity issue.

Option B is correct because using S3 Batch Operations invokes a Lambda function for each object in parallel, handling large volumes efficiently. Option A is wrong because increasing Lambda timeout does not address the root cause of sequential processing. Option C is wrong because Glue jobs have a startup overhead and may cost more.

Option D is wrong because increasing DynamoDB write capacity does not speed up the Lambda processing.

Option D is correct because the 'Rate exceeded' error in AWS Glue when reading from RDS typically indicates that the number of connections or queries per second exceeds the RDS instance's maximum limits. Option A is wrong because AWS Glue does not directly use Amazon Kinesis. Option B is wrong because insufficient IAM permissions would cause an access denied error, not rate exceeded.

Option C is wrong because Amazon S3 does not have a rate exceeded error for writes; it would be a 503 SlowDown error.

Option A is correct because Status: Enabled means versioning is enabled. Option B is wrong because MFADelete is disabled. Option C is wrong because versioning is enabled, not suspended.

Option D is wrong because it does not indicate specific versions.

Option A is correct because Kinesis Data Streams ingests streaming data. Option C is correct because Kinesis Data Firehose can deliver data from the stream to S3. Option B is wrong because SQS is for message queues, not real-time streaming.

Option D is wrong because Glue is for batch ETL, not real-time. Option E is wrong because Athena is a query service, not a delivery service.

Transitioning to S3 Standard-IA after 30 days reduces costs for infrequently accessed data while retaining low latency. Option A is wrong because Glacier has retrieval times not suitable for rare but possible access. Option B is wrong because One Zone-IA is less durable.

Option D is wrong because keeping in Standard is more expensive.

Option D is correct because when S3 default encryption uses KMS, the IAM role must have kms:GenerateDataKey and kms:Decrypt permissions on the KMS key. The policy in the exhibit does not include any KMS actions. Option A is wrong because s3:PutObject is already granted.

Option B is wrong because s3:GetObject is allowed for reading. Option C is wrong because s3:ListBucket is allowed for listing. Option E is wrong because Glue GetObject is not a valid action.

Option D is correct because increasing the Lambda timeout gives the function more time to process records. Option A is wrong because increasing shards increases throughput but does not fix timeout. Option B is wrong because it may reduce concurrency but does not extend processing time.

Option C is wrong because decreasing batch size reduces records per invocation but does not extend timeout.

Option A is correct because CloudWatch provides metrics and logs for Kinesis Data Analytics applications. Option B is wrong because CloudTrail records API calls, not performance metrics. Option C is wrong because Athena is an interactive query service.

Option D is wrong because X-Ray traces requests but is not the primary service for monitoring Kinesis Data Analytics performance.

Option A is correct because automated backups with 5-minute transaction logs meet RPO and a Multi-AZ failover meets RTO. Option B is wrong because manual snapshots cannot achieve 5-minute RPO. Option C is wrong because cross-Region replication adds latency.

Option D is wrong because Read Replicas are not for failover.

Option D is correct because enabling S3 default encryption with SSE-S3 provides server-side encryption automatically without any key management overhead. Option A is wrong because client-side encryption requires managing keys. Option B is wrong because SSE-KMS requires managing KMS keys.

Option C is wrong because SSE-C requires managing your own keys.

Options A and D are correct. Option A: Adding a Catch rule to the state machine handles errors. Option D: Setting a timeout prevents stuck executions.

Option B is wrong because exponential backoff is for retry, not catch. Option C is wrong because parallel state is for concurrency, not resilience. Option E is wrong because DLQ is for Lambda, not state machines.

Option B is correct because VACUUM re-sorts rows and reclaims space, improving performance. Option A is wrong because ANALYZE updates statistics but does not physically reorganize data. Option C is wrong because REINDEX rebuilds indexes, but Redshift uses sort keys, not indexes.

Option D is wrong because DROP TABLE deletes data.

Option B is correct because if KeepClusterAliveOnFailure is set to true for a step, the cluster will not terminate even if the step succeeds. Option A is wrong because 'COMPLETED' indicates success. Option C is wrong because the root volume size does not affect termination.

Option D is wrong because cluster is transient, not long-running.

Options B and D are correct. B: Setting the Lambda batch size to 1 and window to 0 ensures one record per invocation, so if the function fails, it will not retry the same record (since at-most-once means no retries). D: Setting the S3 bucket to use Replace or a unique key prevents overwriting.

Option A is wrong because increasing batch size increases chance of partial failures. Option C is wrong because enabling retries violates at-most-once. Option E is wrong because checkpointing is for at-least-once.

Option A is correct because an explicit Deny overrides any Allow. The Deny statement blocks all s3 actions on the confidential prefix, even though the Allow statement grants PutObject. Option B is wrong because the policy allows PutObject on the bucket.

Option C is wrong because the resource is specified correctly. Option D is wrong because the user has permissions on other parts of the bucket.

Option B is correct because automating VACUUM using a scheduled query (e.g., via Amazon EventBridge or Redshift's built-in scheduler) ensures regular maintenance without manual intervention. Option A is wrong because disabling auto vacuum does not automate it. Option C is wrong because dropping and recreating tables is disruptive.

Option D is wrong because manual VACUUM is not automated.

Option B is correct because the Glue job needs permission to write to Redshift. Option A is wrong because the job already reads from S3. Option C is wrong because Redshift needs its own IAM role.

Option D is wrong because KMS is not mentioned.

Option C is correct because Athena relies on the partition metadata stored in the Glue Data Catalog. If the partition folder structure does not match the table's partition definition (e.g., the folder is named 'date=2021-01-01' but the table's partition key is named 'dt'), MSCK REPAIR will not register the partitions. The partition key name must match the folder prefix.

Option A is wrong because Athena supports JSON format. Option B is wrong because MSCK REPAIR does add partitions if the structure matches. Option D is wrong because the data is already in the S3 bucket; no need to copy.

Option C is correct because S3 bucket policies can explicitly deny access even if the IAM role allows it. Option A is wrong because S3 Transfer Acceleration is not related to access denied errors. Option B is wrong because S3 is not region-specific for the error.

Option D is wrong because S3 Server Access Logging does not affect access permissions.

Option A is correct because the error indicates the table does not exist or was deleted. Option B is wrong because IAM role issues would cause Access Denied, not ResourceNotFoundException. Option C is wrong because network issues would cause timeout or connection refused.

Option D is wrong because schema changes could cause type mismatch but not ResourceNotFoundException.

AWS KMS is the managed service for creating and controlling encryption keys used to encrypt data. Option A is wrong because IAM manages access, not keys. Option B is wrong because CloudHSM is a hardware security module but is not the only option.

Option C is wrong because Secrets Manager is for secrets like passwords.

Option C is correct because increasing the buffer size (e.g., to 128 MB) causes Firehose to deliver fewer, larger files. Option A is incorrect because reducing the buffer interval would create more small files. Option B is incorrect because changing the compression algorithm does not affect file size directly; it reduces storage size but not the number of files.

Option D is incorrect because using a Lambda transformation does not inherently change buffering behavior.

Option C is correct because even though the security group allows inbound traffic, the Glue job's VPC may not have a route to the Redshift cluster's VPC if they are in different VPCs. Option A is wrong because IAM permissions are not the issue. Option B is wrong because the Redshift cluster is in a VPC and not publicly accessible.

Option D is wrong because the NAT gateway is for outbound internet, not for connecting to Redshift within the same VPC.

Option C is correct because enabling pushdown predicate filtering in Spark reduces the amount of data read by pruning partitions early. Option A is wrong because Glue jobs already use distributed processing. Option B is wrong because increasing the number of DPUs can help but is not the most direct fix for the described symptom.

Option D is wrong because converting to CSV would increase I/O and processing time.

Option C is correct because cross-account access requires both a bucket policy in the source account and an IAM user/role in the target account with permissions. Option A (disable block public access) is not needed if the bucket policy is not public. Option B (ACLs) are legacy and not recommended.

Option D (lifecycle policy) is unrelated.

Option B is correct because if the S3 bucket uses an SSE-KMS encrypted prefix, the crawler may not have permission to decrypt; the KMS key policy or IAM role must allow kms:Decrypt. Option A is wrong because S3 events are not required for crawling. Option C is wrong because the crawler can handle many objects.

Option D is wrong because the crawler configuration may need to be set to detect partitions, but the most likely cause given encryption is permission.

Options A, C, and E are correct. A: Automated snapshots provide point-in-time recovery. C: Cross-Region snapshot copy protects against regional failures.

E: Multi-node clusters provide data replication within the cluster. Option B is wrong because Multi-AZ is not a feature of Redshift; it uses automatic failover within a single AZ. Option D is wrong because S3 Cross-Region Replication is for S3, not Redshift.

Option A is correct because Kinesis Data Streams provides sequence numbers for records, enabling deduplication. Option C is correct because Kinesis Data Analytics uses the concept of 'idempotent' output to achieve exactly-once processing. Option D is correct because Apache Flink on Amazon EMR provides exactly-once processing through checkpointing.

Option B is incorrect because Kinesis Firehose delivers at-least-once, not exactly-once. Option E is incorrect because DynamoDB Streams is at-least-once.

Option B is correct because increasing the number of DPUs (e.g., to G.2X with 32 GB) provides more memory per worker, directly addressing the heap space error. Option A is wrong because increasing worker count without increasing worker type may not help if each worker still has insufficient memory. Option C is wrong because increasing batch size could increase memory pressure.

Option D is wrong because converting to JSON typically increases file size and memory usage.