Kafka Exam Experience Report: Test-Day Flow, Tools, and Pitfalls (CCDAK / CCAAK)

Both CCDAK and CCAAK are online-proctored, and your test-day execution moves the score. This article is based on my own exam experience and carefully separates the stable Kafka fundamentals from the test-operations details that tend to change.

Note that operational details — exam policies, question counts, allotted time — can change, so always check the official Confluent Certification site for the latest. Kafka's behavioral spec (producer idempotence, transactions, consumer groups, log retention, and so on) is stable as long as you follow the official docs.

Test-Day Flow: From Check-In to Finish

The exam is online-proctored. Before launching, you do a system check, identity verification, room scan, and screen-share/mic/camera checks. As a rule, leaving mid-exam is not allowed and external materials are prohibited.

The exact check-in flow and ID verification depend on the proctoring vendor. The flow below is generic — run a rehearsal (system check) in your testing portal before exam day and review the latest guidelines.

• Log in 30–45 minutes early; have your ID and test environment ready
• Quiet, solo room with no extra wall items; clear the desk
• Pre-disable browser extensions and notifications; turning off VPNs and corporate proxies is safer
• During the exam, an online whiteboard is usually the only allowed scratchpad (no physical notes)
• After finishing, scores and reports may take some time to appear

Typical online-exam flow

Quick pre-check shell (network and processes)

# Assumes macOS/Linux. On a corporate PC, mind your management policies.
set -euo pipefail

# Connectivity check (to stable DNS)
for h in 1.1.1.1 8.8.8.8; do
  ping -c 3 "$h" >/dev/null && echo "OK: ping $h" || echo "WARN: ping $h failed"
done

# Check processes currently using camera/microphone (example)
if command -v lsof >/dev/null; then
  lsof -nP | grep -E "(camera|microphone|CoreAudio|avfoundation)" || true
fi

echo "Pause notifications/auto-updates and disconnect dual monitors / external devices"

Test Environment and Tool Setup

The baseline rules are: single monitor, no physical notes, online whiteboard allowed. IME switching is generally fine, but custom key bindings and text snippets risk being flagged, so it is safer to disable them.

Prioritize network stability. Disconnect VPNs and corporate proxies; on Wi-Fi, use the 5 GHz band near the router. Temporarily disable OS notifications, updates, and auto-sleep.

• Single monitor; unplug external displays
• Disable auto-sleep and screensaver; put notifications into focus mode
• Practice drawing and typing on the online whiteboard ahead of time
• Disable input-language auto-conversion and external-dictionary popups
• Turn off browser extensions and prepare a clean profile

Minimal physical setup

Quick network-quality check (Linux/macOS)

# Check sustained connectivity and rough jitter
HOST=8.8.8.8
ping -c 20 "$HOST" | tail -n +2

# HTTP reachability
curl -I https://kafka.apache.org/ || echo "HTTP reachability problem"

echo "Disable VPN/proxy and retest"

Time Management and Using the Online Whiteboard

On pass one, lock in the easy questions quickly and flag the ones you are unsure of; deep-dive on pass two. Speed up decisions with elimination and keyword matching (idempotence, transactions, compaction, ISR, and so on).

The online whiteboard works best with shorthand and diagrams. For example, drawing the relationship between Acks=all, idempotent, tx.commit, and read_committed with arrows helps you avoid trick questions.

• On pass one, decide by instinct; just flag hard ones and do not chase them
• Use elimination — strike clearly-wrong options through on the board
• Shorthand the terms (e.g. idemp, EOS, compaction, ISR)
• Sketch dependencies in diagrams to eliminate ungrounded guessing

Time-budget mental model

Simple local time-box for practice (bash)

# A mini-timer to drill per-question time limits
q=1
while [ $q -le 10 ]; do
  echo "Q$q: decide in 90 seconds (flag if unsure)";
  sleep 90;
  echo "next";
  q=$((q+1))
done

Question Trends and Avoiding Traps (CCDAK/CCAAK)

Most questions come from Kafka's stable spec. Frequent topics include producer idempotence and transactions, consumer isolation levels, partitioning, log retention and compaction, replication and ISR, and assignment strategies (Range/Sticky/Cooperative).

Commonly mixed-up points: the relationship between Acks=all and Exactly-Once (Acks=all is not a necessary-and-sufficient condition), combining compaction with retention, what read_committed actually means, and what triggers a rebalance (member leave, timeout, subscription change). Judge based on the official spec.

• Idempotence: enable.idempotence=true prevents duplicates (resends within a single partition are treated as the same sequence)
• Transactions: set transactional.id and consume with read_committed to ensure consistency
• Log retention: cleanup.policy=delete and compact have different goals (retaining data vs. consolidating the latest value)
• Replication: Acks=all means acknowledgment from all ISR members; combine with min.insync.replicas
• Rebalance: Cooperative Sticky minimizes interruption via incremental hand-off

Logical path: Producer → Broker → Consumer

Minimal idempotent + transactional producer configuration (producer.properties)

bootstrap.servers=broker:9092
aacks=all
enable.idempotence=true
transactional.id=orders-tx-001
max.in.flight.requests.per.connection=5
retries=INT_MAX
linger.ms=10
batch.size=32768

Common Issues and On-the-Day Responses

Issues fall into three buckets: network, device, and OS notifications. A safe triage order is: confirm reproducibility → switch routes (tethering/wired) → restart → notify the proctor.

Proctor chat is often in English, so prepping short templates helps. For example: "Network is unstable. I will reconnect without VPN."

• Network: reboot the router, switch Wi-Fi bands, or move to wired
• Device: explicitly grant camera/mic permission in OS settings
• Notifications: enable focus mode and fully quit messaging apps
• Last resort: explain the situation to the proctor and follow instructions

Simplified troubleshooting decision tree for the day

Suppressing notifications (e.g. macOS, Windows)

# macOS: prefer manually enabling Focus mode (automation often hits permission issues)
# Windows: set Focus Assist to "Alarms only"
# Below only checks the state

# macOS Focus-mode state (reference)
/usr/bin/defaults read ~/Library/Preferences/ByHost/com.apple.notificationcenterui.plist 2>/dev/null || true

echo "Enable Focus mode / Focus Assist from the OS UI"

Cheat Sheet: Frequent Commands and Settings

A minimal collection of commands and settings that come up often in both work and the exam. CLI options and property names follow the official stable spec.

Note: the actual exam is multiple-choice — no CLI execution is required — but knowing these helps with conceptual understanding.

• kafka-topics: create, describe, and learn partition count / replication factor
• kafka-consumer-groups: lag, assignment, and rebalance behavior
• Relationship between cleanup.policy and retention.ms/bytes
• ACL least-privilege (Create/Write/Read/Describe/IdempotentWrite)

Offset commit flow (conceptual)

Snippets of Kafka CLI and configuration

# Create a topic (example)
kafka-topics --bootstrap-server broker:9092 \
  --create --topic orders \
  --partitions 6 --replication-factor 3

# Check consumer group lag
kafka-consumer-groups --bootstrap-server broker:9092 \
  --describe --group app-consumer

# Combined retention + compaction (keep latest long-term while pruning old values)
cleanup.policy=compact,delete
retention.ms=604800000   # 7 days
segment.ms=3600000       # 1 hour

Post-Exam Review and Next-Round Prep

After submitting, wait for the report and score to populate. Map weak spots by theme (e.g. transactional consistency, retention strategy, rebalance) to the official docs and write out the reasoning — repeatability goes up.

Pass or fail, fold what you learned into production config templates and operational runbooks — retention sticks faster.

• Extract keywords from missed questions and link them to the official docs sections
• Tweak time-budget thresholds next time (pass-1 cap, review time)
• Run one validation in production (e.g. read_committed combined with EOS)

The learning loop

Review template (YAML example)

topics:
  - name: Exactly-Once Semantics
    symptoms: ["Confusing Acks=all with EOS", "Missing read_committed"]
    doc_refs:
      - kafka.apache.org/documentation/#semantics
      - docs.confluent.io/
    actions:
      - "Producer: re-verify the minimal idempotence + transaction setup"
      - "Consumer: validate the effect of isolation.level=read_committed"

Check Your Understanding

Frequently Asked Questions