Migrate To Msk

Migrating to MSK Express

Overview

This skill helps customers migrate self-managed Apache Kafka workloads to Amazon MSK Express. It provides three phases — Discovery, Assessment, and an optional Simulation — that can be run end-to-end or individually depending on the customer’s needs.

Scope

This skill covers migrations from self-managed Apache Kafka (on-premises, EC2, Docker, Kubernetes, or other non-MSK deployments) to MSK Express. Migrations from MSK Standard (Provisioned) to MSK Express are out of scope.

Prerequisites

The AWS MCP server is recommended for documentation lookups and informational questions, but is not required. The assessment scripts are pure file processors with no AWS API calls.

Intent Routing

Route the customer’s request based on their intent:

1. Open/exploratory question (“How do I migrate to MSK?”)

Explain what this skill offers:

This skill helps you migrate to MSK Express in three phases:

Phase 1 — Discovery: Inventory your source Kafka cluster — brokers, topics, partition counts, configs, authentication, and workload metrics. I can discover this from IaC files (Terraform, CDK, Docker Compose, Kubernetes manifests), provide commands for you to run on your cluster, or you can provide the information manually. Output: migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json.

Phase 2 — Assessment: Validate your cluster against MSK Express across 5 compatibility pillars (topology, Kafka version, configs, auth, quotas) and produce a target Express specification using the AWS-published MSK Sizing/Pricing workbook. I’ll flag what Express will refuse vs what Express will silently convert. Outputs: compatibility.<cluster_name>.json, the filled MSK_Sizing_Pricing.<cluster_name>.xlsx, and msk-sizing-inputs.<cluster_name>.json.

Phase 3 — Simulation: Spin up an MSK Express cluster with load-testing infrastructure to see how Express performs on your own workload, then run a vended test (End-to-End Latency or Broker Restart Under Load) and review the results on a CloudWatch dashboard.

Data replication: For migrating data to your Express cluster, you can use MSK Replicator. I can provide guidance on setup and configuration.

Where would you like to start? I can begin with discovery if you point me to your infrastructure code or describe your cluster, or jump to assessment if you already have a cluster-config.json file, or go straight to simulation if you already know your target Express configuration.

Guardrails for this overview response:

• This response is an overview and a routing question only. Do NOT begin, simulate, or pre-empt any phase.
• Do NOT produce or estimate assessment output here — no verdicts, pillar findings, compatibility conclusions, broker counts, instance recommendations, or cost figures. Those values exist only after you run the Phase 2 scripts against a real cluster-config.json.
• Do NOT open, read, or summarize the internals of compatibility.py, sizing.py, simulation_load_test_config.py, or the reference files to explain how a phase works. Describe the phases at the level shown above; do not walk the customer through the implementation.
• When the customer chooses a phase, run that phase’s scripts or flow to produce real results. Always operate the skill to answer — never answer from having read its source. For the exact commands, see “Running the assessment” in references/assessment-compatibility.md (opens in a new tab) for Phase 2, and references/simulation.md (opens in a new tab) for Phase 3.

2. Discovery intent (DEFAULT when IaC files are provided)

If the customer provides a directory path, IaC files, or says “here’s our infra” — this is discovery intent. Run ONLY Phase 1 (Discovery). Do NOT run assessment, do NOT suggest migration steps, do NOT mention blockers or compatibility. Produce the migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json file and stop.

3. Assessment intent

Customer explicitly asks to assess or has a migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json file already produced. Run Phase 2 (Assessment) only.

4. Simulation intent

Customer wants to test MSK Express with their workload. They can provide cluster sizing directly (instance type, broker count, Kafka version) or reference an earlier assessment. Proceed directly to Phase 3 — Simulation. An assessment is helpful but not required — the simulation asks for sizing inputs directly.

5. Informational questions

Customer asks about Express capabilities, constraints, configuration differences, authentication support, pricing, or compaction behavior without providing cluster-specific data. Use AWS documentation tools (aws___search_documentation, aws___read_documentation) if available to look up the answer from MSK Express documentation. If MCP tools are not available, reference the MSK Express documentation (opens in a new tab) and answer based on knowledge of AWS MSK.

6. Migration strategy questions

Customer asks about MSK Replicator compatibility, version upgrade paths, MirrorMaker 2, or migration strategies. MSK Replicator is the native AWS-supported solution for data replication and works for both MSK-to-MSK and non-MSK-to-MSK migrations. Use AWS documentation tools (aws___search_documentation, aws___read_documentation) if available to retrieve current requirements and supported configurations. If MCP tools are not available, reference the MSK Replicator documentation (opens in a new tab) and answer based on knowledge of AWS MSK.

---

Phase 1 — Discovery

Purpose: Inventory the source cluster to build a migration profile.

Input: One of:

• A directory path containing IaC files (CDK, CloudFormation, Docker Compose, Kubernetes manifests, Terraform)
• Output from Kafka CLI commands the customer runs on their cluster
• Manual information provided by the customer in conversation

Output: migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json — saved to the working directory.

MANDATORY first step for discovery

Before doing ANYTHING else in discovery, you MUST read the reference file: references/discovery.md (located at the skill path shown above).

Use file_read to read the full content of references/discovery.md. This file contains the REQUIRED response template and JSON schema. You MUST follow the template exactly — your response format, forbidden content, and JSON structure are all defined there. Do NOT respond until you have read this file.

Discovery methods

1. IaC analysis — Read infrastructure files and extract cluster metadata.

2. Kafka CLI commands — Display standard Kafka CLI commands for the customer to run on their cluster (kafka-topics.sh, kafka-configs.sh, kafka-broker-api-versions.sh). Do NOT generate or offer Python scripts.

3. Runtime metrics intake — Ingest metrics provided by the customer.

4. Manual conversation — Ask the customer for cluster details.

Discovery rules

• You MUST read references/discovery.md before responding.
• Follow the response template from that file EXACTLY.
• ALWAYS save migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json in the working directory.
• Do NOT proceed to Phase 2 without explicit customer confirmation.

---

Phase 2 — Assessment

Purpose: Assess the cluster against MSK Express requirements and produce a target Express specification (instance type, broker count, monthly cost projection).

Input: migrate-to-msk-skill-artifacts/<cluster_name>/cluster-config.json from Phase 1.

Outputs:

• migrate-to-msk-skill-artifacts/<cluster_name>/compatibility.<cluster_name>.json — five-pillar verdict.
• migrate-to-msk-skill-artifacts/<cluster_name>/MSK_Sizing_Pricing.<cluster_name>.xlsx — the AWS-published MSK Sizing/Pricing workbook (downloaded by the agent) with the six workload inputs filled into the MSK Provisioned sheet. Open it to read the broker count and cost recommendations.
• migrate-to-msk-skill-artifacts/<cluster_name>/msk-sizing-inputs.<cluster_name>.json — a record of the six input values and the cell each maps to.

Assessment is implemented as two file processors (no live AWS API calls):

• scripts/compatibility.py — five-pillar compatibility assessment.
• scripts/sizing.py — computes the six workbook inputs from the discovery contract and fills them into the AWS-published workbook the agent downloads.

Both run via uv run with PEP 723 inline dependencies. For the exact invocation commands, see “Running the assessment” in references/assessment-compatibility.md (opens in a new tab).

Compatibility pillars

compatibility.py validates the source against MSK Express across five pillars:

1. Topology — AZ count, broker count, KRaft vs ZooKeeper, per-cluster broker quota.
2. Kafka version — source version against the Express supported set (3.6, 3.8, 3.9).
3. Configs — broker- and topic-level configs against Express’s editable, read-only, range-restricted, and enforced-value sets (sourced from the Express broker configuration documentation on docs.aws.amazon.com/msk).
4. Auth — checks the source’s authentication mechanism against those MSK Express supports and surfaces any incompatibilities.
5. Quotas — peak workload against absolute Express ceilings (per-broker ingress/ egress, partition count, IAM connection cap, per-partition throughput).

See references/assessment-compatibility.md (opens in a new tab) for the full pseudocode, evidence codes, and verdict mapping.

Verdict vocabulary

Each pillar emits one of three verdicts; the overall is the worst across pillars.

Sizing

sizing.py computes the six workbook inputs from the source workload (peak in/out, total partitions, retention). The agent downloads the AWS-published workbook by reading the Express best-practices page and following its workbook hyperlink, then runs sizing.py --workbook <downloaded.xlsx>, which fills the MSK Provisioned sheet and writes the filled MSK_Sizing_Pricing.<cluster_name>.xlsx (plus a msk-sizing-inputs.<cluster_name>.json record). Open the filled workbook to read the per-instance broker count and monthly cost; its formulas recalculate on open. The workbook is downloaded at assessment time, not packaged with the skill, and the script itself performs no network access (it fills a workbook the agent already downloaded, using the Python standard library). See references/assessment-sizing.md (opens in a new tab) for the cell mapping, the download flow, and caveats.

Assessment rules

• Run compatibility.py and sizing.py independently; neither blocks the other.
• Surface any ACTION_REQUIRED evidence to the user for awareness, but do not gate further phases on it. Express may still accept the workload with mitigations.
• Do NOT pivot back into discovery. Assessment operates on the existing cluster-config.json as-is. Partial data is fine — the scripts emit ADVISORY evidence (METRICS_MISSING, AZ_COUNT_UNKNOWN, etc.) for missing fields; surface those findings and stop. Do not propose Kafka CLI commands, IaC walks, scripts, or questionnaires to fill the gaps. Full forbidden-behavior list in references/assessment-compatibility.md (opens in a new tab).
• Your response MUST follow the assessment response template in references/assessment-compatibility.md (opens in a new tab) (section “Response Template”). One template covers both artifacts. Do not freestyle the post-script summary — the template defines required sections, mandatory vocabulary (use the verdict strings verbatim), and forbidden content (no scores, no narrative editorializing, no in-prose cost / instance recommendations — the user reads those from the filled workbook).

---

Phase 3 — Simulation (optional)

Deploy a temporary, isolated MSK Express cluster and client fleet in the customer’s account so they can see how Express performs on their own workload, then run one of two vended tests (End-to-End Latency, Broker Restart Under Load) and hand over a CloudWatch dashboard. Follow the 12-step conversational flow and all deploy, sizing, and guardrail details in references/simulation.md (opens in a new tab); the deterministic artifacts it drives are scripts/simulation_load_test_config.py (opens in a new tab) and the static assets/simulation-stack.yaml (opens in a new tab).

---

Execution model

Scripts run on the customer’s local machine via uv run. They declare their own dependencies (PEP 723) and are pure file processors — no AWS API calls, no network access, and no third-party dependencies (standard library only).

Security Considerations

Apply these controls at every phase. For additional detail, see MSK Security best practices (opens in a new tab) and MSK IAM access control (opens in a new tab).

1. Encryption in transit (mandatory). Enforce TLS for client-broker traffic on the MSK Express target (EncryptionInTransit.ClientBroker = TLS).

2. Encryption at rest (mandatory). Provision the target cluster with a customer-managed KMS key (or AWS-managed if your compliance posture allows).

3. Authentication — prefer IAM over long-lived credentials. Configure the MSK Express target with IAM authentication as the sole client auth method. This gives ephemeral, role-based credentials with full CloudTrail coverage.

4. Credential storage — use AWS Secrets Manager. Store SASL/SCRAM and TLS credentials for source cluster access in Secrets Manager. Never pass passwords as CLI arguments.

5. Network isolation. Deploy MSK clusters in private subnets. Use security groups scoped to specific CIDR ranges or security group references. Do NOT use 0.0.0.0/0 ingress rules.

6. CloudTrail logging and CloudWatch alarms. Ensure CloudTrail is enabled in the target account and covers kafka.amazonaws.com API calls. Configure alarms:

   • ClientAuthenticationFailure — surge indicates credential problems or attack
   • ConnectionCloseCount — abnormal spike may indicate connection-flooding
   • CloudTrail metric filters for denied kafka-cluster:* actions
   • Connection-rate alarms approaching the 100 conn/sec/broker IAM limit

7. Sensitive data handling. Discovery and assessment outputs contain broker addresses, auth hints, and broker config values. Treat these as sensitive — do not paste into public channels or ticketing systems without redaction.

Troubleshooting

Single-broker / single-AZ source. Topology pillar emits BROKER_COUNT_LT_3 / AZ_COUNT_NOT_3 ADVISORY — Express auto-fixes at the target by deploying across 3 AZs with ≥3 brokers regardless of source.

Out-of-range topic configs. max.compaction.lag.ms < 1 day is the only Express-rejected topic-config bound encoded in compatibility.py. Adjust on the source before migration.

Workbook recommendations look blank or stale. The recommendation and cost cells are workbook formulas; they populate once the filled workbook is opened in Excel / LibreOffice / Sheets and its formulas recalculate. sizing.py sets fullCalcOnLoad so this happens automatically on open — if your spreadsheet app has automatic recalculation disabled, trigger a manual recalculation.