SOA-C03 domain - 22% of the exam

Monitoring, Logging, Analysis, Remediation, and Performance Optimization

Monitoring, Logging, Analysis, Remediation, and Performance Optimization is 22% of the SOA-C03 exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

82
Practice questions
7
Objectives
12
Easy
70
Medium

Where people go wrong

  • Confusing cluster, partition, and spread placement groups

    The wrong options pick a placement group by its name rather than by what it does to hardware distance. Partition and spread groups separate instances onto different racks or servers, which protects against correlated failure but raises latency between nodes. A cluster group packs instances close together for the lowest latency and highest packets per second. A partition group suits large replicated systems needing rack-level isolation, and a spread group suits a few critical instances, capped at seven per Availability Zone.

    See it in a sample question
  • Resizing an EBS volume to fix the wrong limit

    The wrong options enlarge a volume, raise IOPS or move to a costlier type when a different setting is short. The gp2 baseline scales with size, so growing gp2 reaches an IOPS target only by adding capacity the workload does not need. gp3 decouples IOPS and throughput from capacity, so enlarging a gp3 volume changes neither. IOPS governs random operations and throughput governs sequential MiB/s, so a bandwidth shortfall on gp3 calls for more provisioned throughput above the 125 MiB/s baseline, and io2 Block Express is unnecessary at these levels.

    See it in a sample question
  • Treating Run Command and State Manager as Automation

    The wrong options reach for other Systems Manager tools to run a multi-step remediation. Run Command executes commands on instances in one pass, with no branching and no approval pause. State Manager enforces a desired configuration on a schedule and does not react to alarm state changes. An Automation runbook orders steps, branches on results, can pause at an aws:approve action for a named approver, and can be started by an EventBridge rule that targets it directly.

    See it in a sample question
  • Copying or role switching instead of cross-account observability

    The wrong options move the telemetry or move the engineer. Copying data with PutMetricData or metric streams duplicates it and adds maintenance, and role switching shows one account at a time rather than a combined view. An organisation trail records API activity only, so it carries no metrics, logs or traces. CloudWatch cross-account observability links each source account to a monitoring account, which then reads metrics, logs and traces in place for one dashboard.

    See it in a sample question
  • Expecting detailed monitoring to add new signals

    Detailed monitoring is offered as the fix for gaps it cannot close. It raises the frequency of existing hypervisor metrics to one minute. It adds no in-guest memory or disk values, it is not a setting on custom metrics, and it produces no per-pod metrics or SQL attribution. Memory and disk need the CloudWatch agent, sub-minute custom data needs high-resolution metrics, container metrics need Container Insights, and statement-level load needs Performance Insights.

    See it in a sample question

Objectives in this domain

What to study

The official documents this domain's practice questions cite most.

Sample questions from this domain

Free sampleMonitoring, Logging, Analysis, Remediation, and Performance Optimizationmedium

A distributed big-data cluster of many EC2 instances runs a replicated storage layer where each rack-equivalent group should fail independently of the others, so a single hardware fault must not take down more than one group at once. The team wants AWS to spread the instances across distinct underlying hardware in controllable failure isolation groups. Which EC2 placement group strategy fits this requirement?

  • AA cluster placement group, packing the instances close together in one Availability Zone for the lowest inter-node network latency.
  • BA spread placement group, placing each individual instance on distinct hardware up to the small per-zone instance limit.
  • CA partition placement group, dividing the instances into partitions that each sit on a separate rack of underlying hardware. Correct
  • DNo placement group, relying on launching the instances across several Availability Zones to provide the hardware isolation.
Partition placement groups divide a large fleet into rack-isolated partitions, suiting replicated distributed storage that needs partition-level fault isolation. A partition placement group assigns each partition to a separate set of racks with independent power and network, so a single hardware failure affects only one partition, which aligns a replicated big-data layout with the underlying fault domains.

Why A is wrong: A cluster group optimises latency by packing instances onto nearby hardware, which concentrates rather than isolates failures and works against the requirement.

Why B is wrong: A spread group isolates a handful of critical instances but is capped at seven per Availability Zone, so it cannot host a large distributed cluster.

Why C is correct: A partition placement group splits instances into partitions on distinct racks, so a hardware fault is contained to one partition, matching the replicated storage isolation need.

Why D is wrong: Spreading across zones helps availability but gives no control over rack-level partition isolation, leaving the replicated layout unaligned with hardware fault domains.

Free sampleMonitoring, Logging, Analysis, Remediation, and Performance Optimizationmedium

A transactional database runs on an EC2 instance backed by a 400 GiB General Purpose SSD (gp3) volume left at its default settings. CloudWatch shows the workload now needs roughly 9,000 IOPS and around 400 MiB/s of sustained bandwidth, but it is currently capped at the gp3 baseline. The team wants to meet both needs on the same volume without enlarging it or moving to a costlier volume type. Which two changes together meet this requirement? (Select TWO.)

  • AProvision the gp3 volume's IOPS above the 3,000 baseline up to the figure the workload needs, since gp3 lets IOPS be set independently of volume size. Correct
  • BProvision the gp3 volume's throughput above the 125 MiB/s baseline up to the bandwidth the workload needs, as gp3 sets throughput separately from IOPS. Correct
  • CIncrease the gp3 volume's capacity to a larger size so that its baseline IOPS and throughput scale up automatically in proportion to the new size.
  • DMigrate the volume to Provisioned IOPS SSD io2 Block Express to obtain the required IOPS and throughput from a higher-tier storage class.
  • EEnable EBS optimisation on the instance and pick a larger instance type so the host's aggregate EBS bandwidth ceiling no longer limits the volume.
On a gp3 volume, provisioned IOPS and throughput are configured independently of capacity, so tune each setting directly rather than resizing the volume. gp3 separates the three dials of capacity, IOPS and throughput, so meeting a combined IOPS and bandwidth target means raising provisioned IOPS and raising provisioned throughput on the existing volume. Resizing applies to gp2's size-linked baseline, not gp3; io2 Block Express is the costlier type excluded by the requirement; and EBS optimisation only lifts the instance's aggregate ceiling, leaving the volume's own settings at baseline.

Why A is correct: gp3 decouples IOPS from capacity, so raising provisioned IOPS to about 9,000 directly satisfies the IOPS target without resizing the volume.

Why B is correct: gp3 throughput is a configurable setting up to 1,000 MiB/s, so dialling it to about 400 MiB/s meets the bandwidth need on the same volume.

Why C is wrong: This is the gp2 model and is tempting, but gp3 baseline performance does not scale with size, so resizing alone changes neither IOPS nor throughput.

Why D is wrong: io2 Block Express could deliver the numbers, but it is the costlier volume type the requirement rules out and is unnecessary at these levels.

Why E is wrong: Instance EBS limits can matter, but this changes host-side ceilings rather than the volume's provisioned IOPS and throughput, which still sit at baseline.

Free sampleMonitoring, Logging, Analysis, Remediation, and Performance Optimizationmedium

A team is automating a remediation that scales down a database tier and then deletes orphaned volumes, but a human reviewer must approve the run before the deletion step proceeds, and the whole sequence should be a reusable, version-controlled definition. They want the pause-for-approval and the ordered AWS API steps handled by the remediation engine itself rather than coded by hand. Which Systems Manager approach fits this requirement best?

  • AUse Systems Manager Run Command to send the remediation script to the instances, relying on its built-in approval prompt before the destructive command runs.
  • BWrite a single Lambda function that performs each step in sequence and emails an approver, waiting for the reply inside the function before deleting.
  • CAuthor an Automation runbook that orders the steps and includes an aws:approve action to pause until a named approver responds before the deletion step runs. Correct
  • DDefine a Systems Manager State Manager association that reapplies the steps on a schedule and stops to ask for approval before each reapplication.
Choose an Automation runbook with an aws:approve action when remediation needs ordered steps plus a human approval gate. Automation runbooks express ordered steps and include the aws:approve action that suspends the execution until an approver decides, so the gated, versioned multi-step remediation is handled by the engine rather than hand-written code or stateless Run Command.

Why A is wrong: Run Command executes a document against instances in one pass and has no manual approval pause between steps, so it cannot gate the deletion as required.

Why B is wrong: A Lambda holding a run open for a human reply is custom code that hits execution time limits and is harder to version than a managed runbook, so it fits poorly.

Why C is correct: Automation runbooks order multi-step actions and provide a built-in approval action that pauses the run for a named approver, so the gated sequence is defined declaratively and versioned.

Why D is wrong: State Manager enforces desired state on a recurring schedule and offers no interactive approval gate, so it cannot pause a one-off remediation for sign-off.

Free sampleMonitoring, Logging, Analysis, Remediation, and Performance Optimizationeasy

A central operations team runs workloads in many member accounts of an AWS Organizations structure and wants engineers to view CloudWatch metrics and alarms from all of those accounts inside one monitoring account, without copying any telemetry between accounts. Which CloudWatch feature lets the monitoring account read the source accounts' metrics directly?

  • ACreate a metric stream in each member account that writes metric data points into a shared bucket in the monitoring account for engineers to query later.
  • BGrant each engineer an IAM role in every member account and have them switch role into each account to open that account's own CloudWatch console.
  • CEnable CloudWatch cross-account observability, linking each member account as a source account to the central monitoring account. Correct
  • DConfigure a CloudWatch metric math expression in the monitoring account that references the metric ARNs published in each separate member account directly.
CloudWatch cross-account observability links source accounts to a central monitoring account so telemetry is viewed together without being copied. Cross-account observability creates a one-way link from each source account to a monitoring account, letting that account query metrics, logs and traces where they live, so no telemetry is duplicated and engineers avoid logging into each account.

Why A is wrong: Metric streams export data points to a destination such as a bucket, but that copies the telemetry and is not the native cross-account read feature CloudWatch provides.

Why B is wrong: Role switching does technically reach each account, but it forces engineers to log in and out of every account rather than viewing them together, so it misses the single-pane requirement.

Why C is correct: Cross-account observability links source accounts to a monitoring account so engineers read their metrics, logs and traces in place, which is exactly what the requirement asks for.

Why D is wrong: Metric math combines metrics that the account can already see; it cannot by itself pull metrics across an account boundary, so it does not establish the cross-account access.

Free sampleMonitoring, Logging, Analysis, Remediation, and Performance Optimizationmedium

A trading service emits a custom throughput metric every second, and operators need an alarm that can react to a breach within roughly ten seconds. The metric is currently published with the default resolution. What change to how the metric and alarm are configured will let the alarm evaluate at this speed?

  • AKeep the standard-resolution metric and lower the alarm period to ten seconds, since the alarm period is independent of the metric resolution.
  • BEnable detailed monitoring on the metric so that CloudWatch stores its data points at one-second granularity for faster alarms.
  • CSwitch the alarm to the metric-math RATE function so it recalculates the throughput every second and breaches within ten seconds.
  • DPublish the metric as high resolution with a storage resolution of one second and create a high-resolution alarm using a ten-second period. Correct
Choose high-resolution custom metrics and high-resolution alarms when sub-minute detection of a breach is required. Standard metrics are stored at sixty-second granularity, which caps alarm periods at one minute. Publishing the metric with StorageResolution set to one second makes it a high-resolution metric, and an alarm built on it can use ten-second or thirty-second periods. Only this combination allows CloudWatch to detect and react to a breach within about ten seconds.

Why A is wrong: Standard-resolution metrics are stored at one-minute granularity, so an alarm cannot use a ten-second period regardless of how often the application calls PutMetricData.

Why B is wrong: Detailed monitoring applies to certain AWS service metrics at one-minute frequency and is not a setting on custom metrics, so it cannot deliver one-second storage resolution.

Why C is wrong: Metric math changes how values are computed, not how often data is stored or how short the period can be, so a standard-resolution input still limits the alarm to one minute.

Why D is correct: High-resolution metrics store data points at one-second granularity and support alarm periods of ten or thirty seconds, which is exactly what sub-minute alarm evaluation requires.

Other domains in this exam

See also the SOA-C03 cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Amazon Web Services. Original, blueprint-aligned practice material only.