DOP-C02 Mock Exam 93 — 25 Questions

Question 1

A company uses AWS CodeCommit, CodeBuild, and CodePipeline to manage a multi-module Java application. The pipeline has a single build stage that runs tests and packages the application into a JAR file. Recently, the team split the application into multiple microservices, each in its own CodeCommit repository. They want to create a single pipeline that can build all microservices in parallel and then deploy them together to an Amazon ECS cluster. The pipeline should trigger when any of the repositories receives a push. Currently, the pipeline is configured with a single source stage pointing to one repository. The build stage uses a single build project. The team wants to minimize changes to the existing pipeline structure. What should a DevOps engineer do to achieve this?

Accepted Answer

Add multiple source actions in the source stage, each pointing to a different CodeCommit repository. Configure the build stage to use multiple input artifacts and update the build project to build all modules. Option D is correct because CodePipeline supports multiple source actions in the same stage, and each source can have its own CodeCommit repository. Additionally, CodeBuild can be configured to use multiple input artifacts, allowing the build project to access all repositories. The pipeline trigger can be set to any source change. Option A is wrong because a single source action cannot point to multiple repositories. Option B is wrong because webhooks would require separate pipelines. Option C is wrong because Lambda invocation adds complexity and is not the simplest approach.

Answer

Create a separate pipeline for each microservice and use a webhook to trigger all pipelines on any push

Answer

Create a single source action that uses an S3 bucket as source, and have each repository push to the S3 bucket via a Lambda function

Answer

Create a single source action in CodePipeline that uses a Lambda function to pull from all repositories

Question 2

A company has a CI/CD pipeline using AWS CodePipeline that deploys a critical web application to an Auto Scaling group of EC2 instances. The pipeline includes a deploy stage using AWS CodeDeploy. Recently, the deployment failed because the new application version caused an increase in HTTP 500 errors. The operations team manually rolled back the deployment by redeploying the previous version. However, the team wants to automate this process so that future failed deployments are automatically rolled back. Additionally, they want to ensure that if the rollback itself fails, the system should alert the on-call engineer. Currently, the deployment group is configured with 'rollback when a deployment fails' disabled. The team has also set up a CloudWatch alarm that triggers when the HTTP 500 error rate exceeds a threshold. What should a DevOps engineer do to meet these requirements with minimal operational overhead?

Accepted Answer

Configure the deployment group to automatically roll back when a CloudWatch alarm is triggered, and also enable automatic rollback on deployment failure. Set up a CloudWatch alarm on the deployment group's failure metric to notify the on-call engineer. Option C is correct because enabling automatic rollback on alarm in CodeDeploy will trigger a rollback when the CloudWatch alarm is in ALARM state. Additionally, enabling automatic rollback on deployment failure ensures that if the rollback itself fails, the deployment group status will indicate failure, and a CloudWatch alarm can be set on the deployment group's failure metric to alert the on-call engineer. Option A is wrong because it does not address rollback on alarm. Option B is wrong because a manual approval step would require human intervention. Option D is wrong because SNS notification alone does not perform the rollback.

Answer

Create a CloudWatch Events rule that triggers an AWS Lambda function to roll back the deployment when the alarm state is reached, and configure the Lambda function to send an SNS notification on failure

Answer

Add a manual approval step before the deploy stage in CodePipeline and require the on-call engineer to approve or reject the deployment

Answer

Enable the 'rollback when a deployment fails' option in the deployment group and create a CloudWatch alarm on the deployment group's failure metric to notify the on-call engineer via Amazon SNS

Question 3

A startup is using AWS CodePipeline to deploy a Python web application to AWS Elastic Beanstalk. The pipeline has a source stage (CodeCommit), a build stage (CodeBuild), and a deploy stage (Elastic Beanstalk). The build stage runs unit tests and creates a deployable zip file. The deploy stage uses the Elastic Beanstalk deploy provider. Recently, the deploy stage started failing with the error: 'The API call 'elasticbeanstalk:CreateApplicationVersion' failed with status 403.' The CodePipeline service role has the following permissions: 'elasticbeanstalk:DescribeApplications', 'elasticbeanstalk:DescribeEnvironments', 'elasticbeanstalk:UpdateEnvironment'. What should the DevOps engineer do to resolve the issue?

Accepted Answer

Add the 'elasticbeanstalk:CreateApplicationVersion' and 'elasticbeanstalk:DeleteApplicationVersion' permissions to the CodePipeline service role. Option A is correct because the error indicates the pipeline role lacks permission to create an application version, which is necessary for deployment. Adding elasticbeanstalk:CreateApplicationVersion and related actions will resolve the issue. Option B is wrong because the Elastic Beanstalk service role is for the environment, not the pipeline. Option C is wrong because the pipeline does not need S3 permissions for Elastic Beanstalk deployment. Option D is wrong because the error is about permissions, not environment configuration.

Answer

Change the deploy provider to use CodeDeploy instead of Elastic Beanstalk

Answer

Add 's3:PutObject' permission to the CodePipeline service role to allow it to upload the zip file to S3

Answer

Update the Elastic Beanstalk environment's service role to allow CodePipeline to deploy

Question 4

A company runs a critical e-commerce application on AWS. They use AWS CodePipeline to manage deployments. The pipeline has a source stage (CodeCommit), a build stage (CodeBuild), and a deploy stage (CodeDeploy to an Auto Scaling group). Recently, a deployment caused a 5-minute outage because the new application version had a bug that caused the health checks to fail. The Auto Scaling group marked instances as unhealthy and replaced them, but during the replacement, traffic was routed to the remaining instances, which also failed health checks, causing a full outage. The company wants to implement a deployment strategy that prevents any traffic from being routed to unhealthy instances and automatically rolls back if the deployment fails. They also want to minimize deployment time and cost. Which solution should the DevOps team implement?

Accepted Answer

Use CodeDeploy blue/green deployment with automatic rollback enabled. The best approach is to use a blue/green deployment with CodeDeploy, which creates a new Auto Scaling group (green) alongside the existing one (blue). Traffic is shifted to the green group only after health checks pass. If health checks fail, the deployment is automatically rolled back by terminating the green group. This avoids any outage. Option B is correct. Option A (in-place with rollback) would still cause downtime during instance replacement. Option C (increase health check grace period) would delay detection but not prevent outage. Option D (manual approval) slows down deployment and doesn't automate rollback.

Answer

Add a manual approval step in CodePipeline before deploy

Answer

Use CodeDeploy in-place deployment with automatic rollback enabled

Answer

Increase the health check grace period in the Auto Scaling group

Question 5

A company uses AWS CloudFormation to manage infrastructure as code. They have a stack that creates an Amazon RDS database instance. The database password is stored as a parameter in AWS Systems Manager Parameter Store. The CloudFormation template references the parameter using the 'resolve:ssm' dynamic reference. Recently, a security audit found that the password was exposed in plaintext in the CloudFormation stack outputs. The team wants to prevent sensitive information from being displayed in stack outputs or logs. Which approach should be taken?

Accepted Answer

Set the 'NoEcho' property to 'true' for the parameter in the template. To prevent sensitive data from being shown in outputs or logs, use the 'NoEcho' property set to 'true' in the CloudFormation template parameter. This will mask the value in outputs. Option B is correct. Option A (encrypt the output) is not possible because outputs are not encrypted. Option C (remove the output) would prevent exposure but may break automation that depends on the output. Option D (use AWS Secrets Manager) is good but does not directly solve the output exposure issue.

Answer

Store the password in AWS Secrets Manager and reference it in the template

Answer

Remove the output from the CloudFormation stack

Answer

Encrypt the output value using AWS KMS

Question 6

A startup is using AWS CodeBuild to build and test their application. The build process takes about 10 minutes. Recently, they noticed that some builds are failing randomly with the error 'Could not download dependencies'. The build environment uses a custom Docker image stored in Amazon ECR. The team suspects that the issue is due to network connectivity problems when pulling the Docker image or dependencies from the internet. They want to ensure reliable and faster builds. Which solution should they implement?

Accepted Answer

Configure CodeBuild to use a VPC with a NAT gateway. To improve reliability and speed, use an Amazon ECR repository in the same region and configure CodeBuild to use a VPC with a NAT gateway for internet access, or use a VPC endpoint for ECR. The best approach is to use a VPC with a NAT gateway to provide consistent internet access. Option A is correct. Option B (increase build timeout) does not fix the root cause. Option C (use a larger instance) may not resolve network issues. Option D (use a public Docker Hub) may still have network issues and is not controlled.

Answer

Switch to using a public Docker image from Docker Hub

Answer

Increase the build timeout in CodeBuild project settings

Answer

Use a larger compute type for the CodeBuild project

Question 7

A company uses AWS CodePipeline to deploy a web application to an Elastic Beanstalk environment. The pipeline has a source stage (S3), a build stage (CodeBuild), and a deploy stage (Elastic Beanstalk). Recently, the deployment started failing with the error 'The Elastic Beanstalk environment is not in a ready state'. The team found that the environment was in an 'Updating' state because a previous deployment was still in progress. They need a solution that prevents concurrent deployments and ensures that the environment is ready before starting a new deployment. Which action should the DevOps engineer take?

Accepted Answer

Configure a 'Wait for Environment to be Ready' action before the deploy stage. To prevent concurrent deployments, configure the pipeline with a concurrency limit or use a stage that checks the environment status before proceeding. The simplest solution is to set the Elastic Beanstalk environment's 'Rolling update type' to 'Immutable' and use a manual approval gate, but the best practice is to use a CodePipeline action that waits for the environment to be ready. Option A is correct. Option B (increase timeout) does not prevent concurrent deployments. Option C (multiple environments) adds complexity. Option D (disable rolling updates) may cause downtime.

Answer

Increase the deployment timeout in the Elastic Beanstalk environment

Answer

Disable rolling updates in the Elastic Beanstalk environment

Answer

Create multiple Elastic Beanstalk environments for blue/green deployment

Question 8

A company uses AWS CloudFormation to deploy infrastructure. They have a template that creates an Amazon EC2 instance and an Elastic IP address. The template uses the AWS::EC2::EIP resource. The team notices that when they delete the stack, the Elastic IP address is not released, leading to charges. They want to ensure that the Elastic IP is automatically released when the stack is deleted. What should they do?

Accepted Answer

Set the DeletionPolicy attribute to 'Delete' on the EIP resource. By default, CloudFormation deletes the EIP when the stack is deleted. However, if the EIP is associated with an instance, the association may prevent deletion. To ensure release, set the 'DeletionPolicy' attribute to 'Delete' (which is default) and ensure that the EIP is not associated with an ENI that is not being deleted. Option A is correct. Option B (DependsOn) does not affect deletion. Option C (Retain policy) would keep the EIP, causing charges. Option D (manual release) is not automated.

Answer

Set the DeletionPolicy attribute to 'Retain' to keep the EIP

Answer

Create a custom resource to release the EIP before stack deletion

Answer

Add a DependsOn clause to ensure proper order of deletion

Question 9

A company runs a containerized microservices application on Amazon ECS with Fargate. The application is deployed using AWS CodePipeline with CodeBuild as the build stage and ECS as the deploy stage. The pipeline uses a deployment controller of type CODE_DEPLOY with a blue/green deployment strategy. Recently, the team noticed that during deployments, the new task set fails health checks and the deployment is rolled back. The application logs indicate that the new containers fail because they cannot connect to the Redis cluster, which is a required dependency. The Redis cluster is running on Amazon ElastiCache and is in the same VPC as the ECS tasks. The team has verified that the security group for the ElastiCache cluster allows inbound traffic from the ECS tasks' security group. The ECS task definition includes the Redis endpoint as an environment variable. What is the MOST likely cause of the connection failure?

Accepted Answer

The security group for the ElastiCache cluster only allows inbound traffic from the security group of the original task set, not the new task set created during deployment.. Option B is correct because in a blue/green deployment with CODE_DEPLOY, the new task set is created in a separate target group, and the security group for the ElastiCache cluster must allow traffic from the new task set's security group. If the security group rule is only attached to the original task set's security group, the new task set cannot connect. Option A is wrong because the ECS service uses the same subnets; the issue is not subnet-related. Option C is wrong because if the environment variable is correctly set, DNS resolution should work. Option D is wrong because the security group for ECS tasks is typically attached to the service, and adding a new security group rule is necessary.

Answer

The new task set is being launched in different private subnets that do not have a route to the ElastiCache cluster.

Answer

The Redis endpoint environment variable is not being passed correctly to the new task set because of a misconfiguration in the task definition.

Answer

The new task set is using a different VPC than the ElastiCache cluster.

Question 10

A company uses AWS CodeCommit as a Git repository and CodeBuild for continuous integration. The buildspec.yml file includes steps to run unit tests and package the application. The team wants to ensure that only code from the main branch is deployed to production. They have set up a CodePipeline that triggers on changes to any branch. The pipeline includes a build stage that runs CodeBuild, and then a deploy stage that deploys to production. The team noticed that code from feature branches is being deployed to production accidentally. The team wants to modify the pipeline to prevent this. What is the MOST effective solution?

Accepted Answer

In the CodePipeline source stage, configure the branch filter to only allow the main branch to trigger the pipeline.. Option A is correct because adding a filter to the source stage that only triggers on the main branch ensures that feature branch changes do not start the pipeline. Option B is wrong because restricting IAM permissions for developers does not prevent the pipeline from triggering on feature branches if the pipeline is configured to do so. Option C is wrong because a manual approval step prevents unintended deployments but still triggers the pipeline, wasting resources. Option D is wrong because configuring the branch in the build stage does not stop the pipeline from being triggered by feature branches.

Answer

Use IAM policies to restrict developers from pushing to the main branch.

Answer

Add a manual approval step before the deploy stage and require approval from a senior engineer.

Answer

Modify the CodeBuild project to only build the main branch by specifying the branch in the source configuration.

Question 11

A developer is using AWS CloudFormation to deploy a stack that includes an AWS Lambda function. The Lambda function code is stored in an S3 bucket. The CloudFormation template references the S3 bucket and object key. The developer wants to update the Lambda function code by uploading a new zip file to S3 and then updating the stack. The developer updates the S3 object with a new version, but the stack update does not automatically use the new code. What should the developer do to ensure the stack update uses the new code?

Accepted Answer

Upload the new code to a different S3 key or specify a new version ID in the CloudFormation template.. Option C is correct because CloudFormation only detects changes to S3 objects if the S3 key or version changes. By uploading the new code with a different key or setting a new version ID, CloudFormation will detect the change and update the Lambda function. Option A is wrong because the stack policy does not affect resource updates. Option B is wrong because CloudFormation does not automatically poll for S3 changes. Option D is wrong because recreating the entire stack is unnecessary and disruptive.

Answer

Enable S3 event notifications to trigger a CloudFormation stack update when the object is updated.

Answer

Modify the CloudFormation stack policy to allow updates to the Lambda function.

Answer

Delete the stack and recreate it with the new code.

Question 12

A company uses AWS CodePipeline to automate the deployment of a static website hosted on Amazon S3. The pipeline includes a source stage that pulls from a CodeCommit repository and a deploy stage that uses CodeBuild to sync the files to an S3 bucket. The team noticed that the website is not updating after a successful pipeline run. The CodeBuild logs show that the 'aws s3 sync' command completed successfully. However, the website still shows the old content. What is the MOST likely cause?

Accepted Answer

The website is fronted by Amazon CloudFront, which is caching the old content.. Option B is correct because S3 static website hosting serves content from the bucket, but if CloudFront is used as a CDN, it caches the content. The sync command updates the S3 bucket, but CloudFront serves cached content until the TTL expires or the cache is invalidated. Option A is wrong because the bucket policy does not affect serving content if it's already public. Option C is wrong because the sync command does not require public access to update objects. Option D is wrong because website hosting can be enabled on the bucket; the issue is caching.

Answer

The CodeBuild project does not have permission to write to the S3 bucket.

Answer

The S3 bucket is not configured for static website hosting.

Answer

The S3 bucket policy is blocking public access to the updated objects.

Question 13

A company hosts a static website on Amazon S3 with a CloudFront distribution. The website is critical for business operations and must be available even if the primary AWS Region fails. Currently, the S3 bucket is in us-east-1, and CloudFront uses that bucket as the origin. The company has a secondary bucket in us-west-2 with a replica of the data. The company wants to use CloudFront to automatically fail over to the secondary bucket if the primary becomes unavailable. The DevOps engineer needs to implement a solution that requires minimal operational overhead. What should the engineer do?

Accepted Answer

Configure CloudFront Origin Failover by adding both buckets as origins, with the primary in us-east-1 and secondary in us-west-2.. CloudFront Origin Failover allows you to set up a primary and secondary origin. If the primary returns an error (e.g., 503), CloudFront automatically routes requests to the secondary origin.

Answer

Use an Application Load Balancer in front of both S3 buckets and point CloudFront to the ALB.

Answer

Create a second CloudFront distribution pointing to the secondary bucket and use Route 53 failover routing between the two distributions.

Answer

Modify the application to switch the CloudFront origin URL using Lambda@Edge when health checks fail.

Question 14

A company uses AWS CloudFormation to deploy infrastructure. During a recent deployment, the stack failed to create an Amazon RDS DB instance because of a parameter validation error. The DevOps engineer fixed the parameter and wants to resume the stack creation without recreating the resources that were already successfully created. The stack template is parameterized and uses nested stacks. What is the MOST efficient way to resume the stack creation?

Accepted Answer

Use the CloudFormation stack update operation with the corrected parameter.. CloudFormation stack updates can be used to fix the issue. By updating the stack with the corrected parameters, CloudFormation will only modify the failed resource and not recreate already created resources.

Answer

Manually create the RDS instance with the corrected parameter and update the stack to import it.

Answer

Delete the entire stack and redeploy with the corrected parameter.

Answer

Use the 'ContinueUpdateRollback' feature to rollback the failed stack and then redeploy.

Question 15

A company runs a microservices application on Amazon ECS with Fargate. The application uses an Application Load Balancer (ALB) to route traffic to services. Each service has a required number of tasks for capacity. The company recently experienced a prolonged outage when a bug caused all tasks of the critical 'payment' service to crash simultaneously. The DevOps team needs to implement a deployment strategy that reduces the risk of a full service outage during updates. The strategy must also allow for quick rollback if a deployment fails. Which deployment strategy should the team implement?

Accepted Answer

Implement blue/green deployment using ECS with target tracking alarms to automate traffic shifting.. Blue/green deployment with target tracking allows you to gradually shift traffic to the new version while monitoring. If issues arise, you can instantly rollback by switching traffic back to the old version.

Answer

Implement a rolling update with a fixed number of tasks to replace at a time.

Answer

Use a canary deployment by creating a new service with a small number of tasks, test, then shift all traffic.

Answer

Deploy changes during maintenance windows with manual approval steps.

Question 16

A financial services company runs a critical application on Amazon ECS with Fargate launch type. The application has strict availability requirements and must survive an Availability Zone failure. The ECS service is configured with a desired count of 4 tasks, spread across two Availability Zones using a spread strategy. The service is fronted by an Application Load Balancer. During a recent AZ outage, one AZ became completely unavailable, but the application continued to serve traffic. However, after the AZ recovered, the ECS service did not automatically place new tasks in the recovered AZ to restore the desired count. The service remains with only 2 tasks in the remaining AZ. What is the most likely cause and solution?

Accepted Answer

The ECS service does not automatically replace tasks in a recovered AZ because the spread strategy is static. Manually update the service (e.g., force new deployment) to trigger rebalancing.. The ECS service uses a spread strategy, which maintains balance across AZs. When an AZ is unhealthy, ECS does not place tasks there. After recovery, the service may not automatically rebalance because the spread strategy is not proactive; it only ensures placement of new tasks. To force rebalancing, one can update the service (e.g., change desired count, then change back).

Answer

The Application Load Balancer is not health-checking the recovered AZ. Configure cross-zone load balancing.

Answer

The ECS service uses a spread strategy that does not automatically rebalance after an AZ recovers. Update the service to use a 'rebalance' strategy.

Answer

The ECS service is configured with a minimum healthy percent of 50%, which prevents replacement. Lower the minimum healthy percent to 0%.

Question 17

A company runs a static website on Amazon S3 with public read access. The website content is stored in an S3 bucket and served through an Amazon CloudFront distribution for better performance and security. Recently, the company noticed that some users are accessing the S3 bucket directly via the S3 endpoint, bypassing CloudFront. This increases costs and exposes the bucket to potential attacks. The company wants to ensure that all access to the website goes through CloudFront only. Which solution should the company implement?

Accepted Answer

Create an origin access identity (OAI) in CloudFront and update the S3 bucket policy to allow only the OAI to read objects.. To restrict access to the S3 bucket only through CloudFront, use an origin access identity (OAI) and a bucket policy that allows only the OAI. This way, direct access via S3 URL is denied.

Answer

Set the S3 bucket policy to deny all requests that do not come from the CloudFront distribution's IP addresses.

Answer

Configure the S3 bucket to use AWS WAF to block requests that do not have a custom header set by CloudFront.

Answer

Change the S3 bucket to be private and use presigned URLs for all requests.

Question 18

Your company has a CI/CD pipeline for a Java web application using AWS CodePipeline, CodeBuild, and CodeDeploy. The pipeline has three stages: Source (CodeCommit), Build (CodeBuild), and Deploy (CodeDeploy to EC2 instances in an Auto Scaling group). Recently, after a change to the buildspec, the Build stage succeeds but the Deploy stage fails with the error 'The deployment failed because the deployment group exceeded the minimum healthy host count.' The CodeDeploy deployment configuration uses a 'OneAtATime' deployment with a minimum healthy host count of 1. The Auto Scaling group has a minimum size of 2 and a maximum size of 4. The application runs on Amazon Linux 2 instances. The CodeDeploy agent is installed and running on all instances. What is the most likely cause of the failure?

Accepted Answer

The new build artifact is larger, causing the instances to run out of disk space during deployment, which prevents the CodeDeploy agent from completing the deployment.. Option B is correct because if the buildspec change introduced a new artifact that requires additional disk space, the instances may run out of space during deployment, causing the agent to fail and the deployment to stop, leading to insufficient healthy hosts. Option A is wrong because if the agent were missing, the error would be different. Option C is wrong because a different AMI would cause launch failures, not deployment failures. Option D is wrong because the CodeDeploy service role is used by CodeDeploy, not the pipeline; if it were missing, the pipeline would fail earlier.

Answer

The CodeDeploy agent on some instances has been stopped or is not responding.

Answer

The Auto Scaling group is using a different Amazon Machine Image (AMI) that lacks the CodeDeploy agent.

Answer

The CodeDeploy service role does not have permission to describe the Auto Scaling group.

Question 19

A startup is using AWS CloudFormation to manage their infrastructure. They have a stack that creates an Amazon S3 bucket and an Amazon DynamoDB table. The stack was created successfully, but when they try to update the stack to add a new S3 bucket, the update fails with the error 'CREATE_FAILED - S3 bucket already exists'. The new bucket name is unique and does not exist. The template uses the same AWS::S3::Bucket resource type. What is the most likely cause?

Accepted Answer

The S3 bucket name was previously used and is still in the process of being deleted (bucket name not yet released).. Option A is correct because the bucket name may have been used in a previous stack and is in a deletion state. Option B is wrong because the account has permissions. Option C is wrong because the resource type is correct. Option D is wrong because the region is correct.

Answer

The IAM user does not have permission to create S3 buckets.

Answer

The stack is in a different region than where the bucket is being created.

Answer

The CloudFormation template uses the wrong resource type for the bucket.

Question 20

You are a DevOps engineer for a company that runs a containerized microservices application on Amazon ECS with Fargate. The CI/CD pipeline uses AWS CodePipeline, with CodeBuild for building Docker images and pushing them to Amazon ECR, and CodeDeploy for deploying to ECS. The pipeline has a manual approval step before production deployment. Recently, the production deployment failed after approval, with the error: 'The service my-service could not be deployed because the task definition arn:aws:ecs:us-east-1:123456789012:task-definition/my-task:5 references an image that does not exist in the repository.' The image was built and pushed successfully in the Build stage. The task definition is updated by CodeDeploy to reference the new image URI. The ECS service is configured with 'deploymentController: CODE_DEPLOY' and uses a blue/green deployment. The CodeDeploy deployment group uses a 'Lambda' compute platform (incorrectly set). The pipeline uses the default CodeDeploy provider for ECS. What is the most likely cause of the failure?

Accepted Answer

The CodeDeploy deployment group's compute platform is set to 'Lambda' instead of 'ECS'.. Option C is correct because CodeDeploy for ECS requires the compute platform to be 'ECS' (or 'Server' for EC2), not 'Lambda'. Setting it to 'Lambda' would cause the deployment to fail as it tries to use a Lambda deployment configuration. Option A is wrong because the image was pushed successfully. Option B is wrong because the pipeline does use CodeDeploy. Option D is wrong because the approval step is manual and would not cause image reference issues.

Answer

The manual approval step changed the task definition ARN to a previous version.

Answer

The ECR repository has a lifecycle policy that expired the image tag before deployment.

Answer

The CodePipeline is not configured to use the CodeDeploy provider for ECS.