A company has developed a service that is deployed on a fleet of Linux-based Amazon EC2 instances that are in an Auto Scaling group. The service occasionally fails unexpectedly because of an error…

Question

Options

• AInstall the Amazon CloudWatch agent on the EC2 instances. Configure the CloudWatch agent to
• BTag the EC2 instances. Create an AWS Lambda function that uses AWS Systems Manager
• CTag the EC2 instances. Use AWS Systems Manager State Manager to create an association that
• DUpdate the EC2 user data that is specified in the Auto Scaling group's launch template to include
• EUpdate the EC2 user data that is specified in the Auto Scaling group's launch template to ensure

• A
  78% (39)
• B
  12% (6)
• C
  8% (4)
• D
  2% (1)

Explanation

Options A and E both address the need for continuous, automated service recovery at the instance level. Option A uses the CloudWatch agent configured to monitor the service process and trigger a restart action (via a CloudWatch alarm + SSM or a custom script) whenever the process dies. Option E uses EC2 user data in the launch template to configure the service as a systemd unit with Restart=always, so the OS itself automatically restarts the crashed process without any external intervention.

Why the distractors fail:

• B (Lambda + SSM): Too much latency and operational overhead - Lambda isn't designed for real-time process-crash detection, so recovery would be delayed and fragile.
• C (SSM State Manager association): State Manager runs on a schedule, not in response to a crash event - it can't provide immediate, event-driven recovery.
• D (User data without persistent restart config): EC2 user data runs only once at instance launch; if the service crashes hours later, the user data script doesn't re-run, so nothing recovers it.

Memory tip: Ask yourself "does this solution react continuously, or only at boot/on a timer?" The two winning options (CloudWatch agent monitoring + systemd auto-restart) are always watching - they react the moment the service crashes. The wrong options are either one-shot (user data alone), scheduled (State Manager), or indirectly triggered (Lambda).

No community discussion yet for this question.