Task 1.4: Apply programming concepts

• ASet the function's provisioned concurrency to 20

  Incorrect

  Provisioned concurrency keeps instances initialized to cut cold starts. It does not stop the function from scaling beyond that number.

• BRaise the function's memory

  Incorrect

  More memory brings more CPU and shorter runs, but a burst of files still starts many instances at once.

• CRaise the function's timeout to 15 minutes so that fewer retries happen

  Incorrect

  The timeout limits how long one invocation runs. It does not limit how many instances run at the same time.

• DSet the function's reserved concurrency to 20

  Correct

  Reserved concurrency is the maximum number of concurrent instances for the function, and it is set aside for that function alone. AWS suggests it to protect downstream resources such as database connections.

To protect a database, cap the writer. Reserved concurrency is both the upper limit and a guaranteed share for one function, while provisioned concurrency is about start-up latency.

• APackage the large reference files in a Lambda layer attached to the function

  Incorrect

  A deployment package with its layers is limited to 250 MB unzipped. The problem is space for data the function creates at run time.

• BRaise the function's timeout to 15 minutes so the unpacking can finish

  Incorrect

  The error is about disk space, not time. A longer timeout leaves /tmp at the same size.

• CSet provisioned concurrency so that the temporary files are kept between invocations

  Incorrect

  Provisioned concurrency keeps environments initialized; it does not enlarge /tmp, and the files need not survive the invocation anyway.

• DIncrease the function's ephemeral storage for /tmp to 10,240 MB

  Correct

  Ephemeral storage in /tmp can be set between 512 MB and 10,240 MB. It is temporary and private to each execution environment, which is all this function needs.

Scratch space that lives only for the invocation belongs in /tmp, which can grow to 10,240 MB. Shared or persistent storage would only be needed if other invocations had to read the files.

• AAWS CodeDeploy for each function

  Incorrect

  CodeDeploy automates deployments of application code to Lambda, EC2 or ECS. It does not define the state machine or the table as infrastructure.

• BAWS SAM, with sam build and sam deploy

  Correct

  AWS SAM templates extend CloudFormation with shorthand serverless resources. The SAM CLI builds the code, invokes functions locally with sam local, and sam deploy deploys through CloudFormation.

• CCloudFormation StackSets, with one stack instance for the account

  Incorrect

  StackSets manage stacks across many accounts and Regions. They do not add shorthand serverless resources, builds or local testing.

• DA shell script that calls aws lambda create-function for each function

  Incorrect

  This creates functions one call at a time. It is not infrastructure as code and does not cover the other resources.

AWS SAM is the serverless layer on top of CloudFormation: a compact template plus a CLI that builds, tests locally and deploys the stack.

• AConfigure provisioned concurrency for the function

  Incorrect

  Provisioned concurrency removes cold-start time. It does not add CPU to a running invocation.

• BIncrease the ephemeral storage that is mounted at /tmp

  Incorrect

  More /tmp space helps when the disk is full. It gives the function no extra processing power.

• CIncrease the memory setting of the function

  Correct

  Lambda allocates CPU power in proportion to memory; at 1,769 MB a function has the equivalent of one vCPU. CPU-bound code runs faster with more memory.

• DSet reserved concurrency to 1,000

  Incorrect

  Reserved concurrency controls how many instances run at once. Each instance keeps the same CPU share.

Memory is the single compute dial in Lambda: raising it raises CPU in proportion, which shortens CPU-bound work.

• AAs one DynamoDB item per account that holds lists of the related devices and cards

  Incorrect

  Each extra step would need more lookups in application code. DynamoDB stores items by key; it does not traverse relationships.

• BAs fact and dimension tables in Amazon Redshift that analysts join nightly

  Incorrect

  A warehouse fits analytical queries over large tables. Repeated self-joins for each hop and nightly loads do not give live, millisecond answers.

• CAs vertices and edges in Amazon Neptune, traversed with Gremlin or openCypher

  Correct

  Neptune is a graph database built to store billions of relationships and to traverse them with millisecond latency, with fraud detection as a typical use case.

• DAs JSON documents in Amazon OpenSearch Service with a keyword field per link

  Incorrect

  OpenSearch Service searches and analyzes documents. It does not follow multi-step connections between them.

When the questions are about connections (who links to whom, in how many steps), the data is a graph. A graph database stores the edges directly and walks them quickly.

• ACall df.coalesce(1) after the join and before each of the aggregations

  Incorrect

  Coalescing to one partition reduces parallelism. Each action would still recompute the join from the source data.

• BAdd a broadcast hint for one of the two large datasets

  Incorrect

  Broadcasting suits a small table that fits in memory. It does not stop later actions from repeating the join.

• CCall df.persist() after the join and unpersist it after the writes

  Correct

  Caching a DataFrame that is used repeatedly keeps the computed result in executor memory and on disk, so later actions reuse it.

• DTurn on job bookmarks for the job and rerun it

  Incorrect

  Job bookmarks track data already processed across runs. They do not reuse results within one run.

Spark evaluates lazily, so every action rebuilds its lineage. Persist a DataFrame that several actions reuse, and release it when done.

• AAs a .zip file split across several Lambda layers

  Incorrect

  The 250 MB unzipped limit covers the function and all of its layers together, so splitting does not help.

• BAs a .zip file uploaded through Amazon S3

  Incorrect

  Uploading through S3 avoids the 50 MB direct upload limit, but the 250 MB unzipped limit still applies.

• CAs a container image pushed to Amazon ECR

  Correct

  Lambda container images can be up to 10 GB uncompressed, including all layers, which fits 1.5 GB of libraries.

• DAs a .zip file, with memory raised to 10,240 MB

  Incorrect

  Memory is a runtime setting. It does not change the limit on deployment package size.

Zip packages and layers stop at 250 MB unzipped. Large dependency sets go in a container image, up to 10 GB.

• ADeletionPolicy: Snapshot

  Incorrect

  Snapshot applies only to resources that support snapshots, such as volumes and database clusters; S3 buckets are not among them.

• BDeletionPolicy: Retain

  Correct

  With Retain, CloudFormation keeps the resource and its contents when the stack is deleted.

• CA DependsOn attribute that names the stack's other resources

  Incorrect

  DependsOn controls the order in which resources are created. It does not keep a resource after deletion.

• DA CreationPolicy attribute with a resource signal count of 1

  Incorrect

  CreationPolicy holds a resource's creation until success signals arrive. It does not affect deletion or replacement.

• EUpdateReplacePolicy: Retain

  Correct

  UpdateReplacePolicy keeps the existing physical resource when a stack update replaces it; by default the old resource is deleted.

Deletion and replacement are separate paths in CloudFormation; protecting stateful resources such as data buckets takes both Retain policies.

• AAdd a filter on the partition column, such as dt = '2026-10-01'

  Correct

  When a query on a partitioned table specifies the partition in the WHERE clause, Athena scans only that partition.

• BRun MSCK REPAIR TABLE on the table before the query runs

  Incorrect

  MSCK REPAIR TABLE adds partitions found in S3 to the catalog metadata. It does not restrict what a query reads.

• CTurn on partition projection for the table and keep the query as it is

  Incorrect

  Projection calculates partition values and locations from table properties. The query still needs a filter on dt to skip partitions.

• DRewrite the query as CREATE TABLE AS SELECT (CTAS)

  Incorrect

  CTAS writes a query's results to a new table in S3. The SELECT inside it would scan the same data.

Partitions help only when the query filters on the partition column itself. A filter on another column leaves Athena reading every partition.

• AValidate the template with the ValidateTemplate operation

  Incorrect

  ValidateTemplate checks that the template is valid JSON or YAML. It reads only the template, not the live resources.

• BTurn on termination protection for the stack

  Incorrect

  Termination protection stops the stack from being deleted. It reports nothing about resource settings.

• CDeploy the template again with CloudFormation StackSets

  Incorrect

  StackSets create, update, or delete stacks across accounts and Regions. Deploying again is a change, not a check.

• DRun drift detection on the stack

  Correct

  Drift detection compares the actual property values of the stack's resources with the values the template expects and returns the drift status of each supported resource. Glue jobs and S3 buckets both support it.

Changes made outside CloudFormation are drift. Drift detection only reads and reports, listing each supported resource as in sync, modified or deleted; correcting the drift is a separate, deliberate step.