SOA-C02 Exam Domain Deep Dive: Monitoring, Logging & Remediation

TL;DR
- Domain 1 (Monitoring, Logging, and Remediation) is 20% of SOA-C02 — the single largest block — and it's also where the hands-on labs tend to land, so study it in weeks 1–2, not the night before.
- Plan around fixed numbers: 65 questions, 180 minutes, USD 150, scaled score 720/1000 to pass, and labs that will burn 15 minutes each if you work in the wrong region.
- This domain rewards muscle memory over memorization. If
aws logs test-metric-filter,--treat-missing-data notBreaching, and EventBridge→SSM runbooks aren't reflexes yet, drill them until they are.
The domain map: 20% is the largest slice, and everything else borrows from it
SOA-C02 splits into six weighted domains. Here are the published weights alongside what actually trips people up:
| Domain | Weight | Where candidates lose points |
|---|---|---|
| 1. Monitoring, Logging, and Remediation | 20% | alarm state semantics, metric filter syntax, EventBridge targets |
| 2. Reliability and Business Continuity | 16% | RTO/RPO wording, Multi-AZ vs Read Replica vs backup/restore |
| 3. Deployment, Provisioning, and Automation | 18% | CloudFormation drift, StackSets, CodeDeploy lifecycle hooks |
| 4. Security and Compliance | 16% | KMS key policies, IAM permission boundaries, Config rules |
| 5. Networking and Content Delivery | 18% | security groups vs NACLs, Transit Gateway, Route 53 failover |
| 6. Cost and Performance Optimization | 12% | Savings Plans, gp2→gp3, right-sizing with Compute Optimizer |
The reason Domain 1 deserves disproportionate attention isn't just the 20%. It's that questions in Domains 2 through 6 are frequently framed as monitoring questions: "which metric tells you the failover happened," "which alarm action triggers the runbook," "how do you prove the patch applied." You can't dodge CloudWatch by skipping Domain 1.
If you're deciding between certifications, the split is roughly: SAA-C03 asks you to design it, SOA-C02 asks you to keep it alive at 2am. The second one is closer to daily SMB ops work.
What Domain 1 actually tests: detect, then fix
The public exam guide gives Domain 1 two task areas — implement and manage monitoring and logging solutions, and implement and manage remediation and maintenance. In practice that means the CloudWatch agent and its config file, custom namespaces, metric filters against structured logs, composite alarms for noise reduction, CloudTrail for API-level audit trails, and automation via SSM runbooks, EventBridge rules, and Config auto-remediation.
The trap is syntax. Metric filter patterns are not regex, and they are not interchangeable between log formats. A pattern like [level, request_id, status_code = 5*] is a space-delimited pattern that works on plain-text logs; JSON logs need { $.statusCode >= 500 }. Mixing them produces zero matches and no error message at all, which is exactly how alarms end up silent.
Here's the pair of commands I keep in a scratch file for new services:
aws logs put-metric-filter
--log-group-name /aws/lambda/checkout
--filter-name Checkout5xx
--filter-pattern '{ $.statusCode >= 500 }'
--metric-transformations metricName=Checkout5xx,metricNamespace=AppErrors,metricValue=1,defaultValue=0
--region us-east-1
aws cloudwatch put-metric-alarm
--alarm-name checkout-5xx
--namespace AppErrors --metric-name Checkout5xx
--statistic Sum --period 60 --evaluation-periods 3
--threshold 5 --comparison-operator GreaterThanOrEqualToThreshold
--treat-missing-data notBreaching
--alarm-actions arn:aws:sns:us-east-1:111122223333:ops-pager
--region us-east-1
Notice `defaultValue=0` and `--treat-missing-data notBreaching`. Those two flags are the difference between a reliable alarm and one that sits in alarm-limbo forever. Exam questions love this.
The other half of the domain is remediation. Know the difference between detecting and fixing: EventBridge rule on `EC2 Instance State-change Notification` → SSM Automation document to re-attach an EIP is a remediation answer; an SNS email is not. Config rules with automatic remediation are for compliance drift, not for runtime incidents.
| Alarm knob | What it changes | Exam consequence |
|---|---|---|
| `Statistic: Average` vs `Maximum` | averaging hides 30-second spikes | `Maximum` with `period 60` catches spikes you'd otherwise miss |
| `Period × EvaluationPeriods` | total time before firing | 60 × 3 = 3 minutes of sustained breach |
| `treatMissingData` | behavior with no datapoints | `notBreaching` stops false `INSUFFICIENT_DATA` |
| `ComparisonOperator` + threshold | sensitivity | `GreaterThanThreshold` vs `GreaterThanOrEqualToThreshold` changes edge cases |
## Hard-won lesson: 19 minutes of 502s and an alarm that never made a sound
At 02:12 UTC on 2026-11-14, three `m5.xlarge` instances behind an ALB in us-east-1 started returning 502s. The pager stayed quiet. A customer-facing check caught it 19 minutes later. My first guess was SNS — the topic existed, but I assumed the email subscription was still pending confirmation. It wasn't; every subscription showed `Confirmed`.
So I looked at the alarm itself:
```bash
aws cloudwatch describe-alarms --alarm-names checkout-5xx
--query 'MetricAlarms[0].{State:StateValue,Reason:StateReason}' --output json
```json
{"State": "INSUFFICIENT_DATA", "Reason": "Insufficient Data: 1 datapoint was unknown."}
The alarm had never evaluated anything. The metric filter was the culprit: the service logs JSON (`{"level":"error","statusCode":502,...}`), but the filter used a space-delimited pattern copied from an older nginx log group. No tokens matched, no metric was emitted, no datapoints existed, and because we hadn't set `defaultValue` the alarm just waited forever. CloudWatch doesn't warn you about a pattern that matches nothing.
Diagnosis took one command against a captured event:
```bash
echo '["{"level":"error","statusCode":502,"path":"/checkout"}"]' > sample.json
aws logs test-metric-filter
--filter-pattern '[level, request_id, status_code = 5*]'
--log-event-messages file://sample.json
Empty match list. I swapped in `{ $.statusCode >= 500 }` and got a match immediately. After updating the metric filter, I pushed a synthetic event with `aws logs put-log-events` and watched `get-metric-statistics` return `Sum: 1.0` within the next minute — that's how I verified the fix rather than assuming it.
Two follow-ups we now apply everywhere: `defaultValue=0` on every metric transformation, and a composite alarm that pages when *either* the 5xx rate spikes *or* the underlying metric goes missing. A silent alarm is worse than a noisy one.
## Registration, cost, and the 180 minutes
Register through the AWS Certification portal and schedule with Pearson VUE; online proctored delivery is available if your desk and ID pass the OnVUE checks. The exam fee is USD 150, it's 180 minutes, and scoring is scaled 100–1000 with 720 to pass. Results appear on screen at the end of the session; the digital badge follows separately.
| Item | Detail |
|---|---|
| Exam fee | USD 150 (regional pricing varies) |
| Official practice exam | USD 20 at the time of writing, worth it for the lab-style questions |
| Retake wait | 14 days after a failed attempt, full fee again |
| Validity | 3 years from the pass date |
| Study time | 40–60 hours if you already run AWS day to day, 80+ if you don't |
Budget the clock before the content. Sixty-five questions in 180 minutes is 2.7 minutes each — until labs take 15–20 minutes apiece, which leaves you closer to 2 minutes for the rest. Practice at that speed, not at leisure.
## SAA-C03 or SOA-C02 first?
| | SAA-C03 | SOA-C02 |
|---|---|---|
| Time | 130 minutes | 180 minutes |
| Questions | 65, multiple choice/response | 65 including hands-on labs |
| Focus | architecture and design decisions | operations, monitoring, remediation |
|
Leave a Reply