Unit 8.3: Capstone Lab: Build the Enterprise Core End-to-End
Introduction
Each module ended with a lean build of one part. This lab assembles those parts into one environment, fc-prod-fra-core, the Frankfurt core of the reference architecture in Unit 8.2. The individual steps are not new; the order and the joins are. Built in the wrong order, a load balancer has no targets, a NAT Gateway has no rules, a security group allows the wrong source, and a database lands on an address that DHCP already handed out.
The lab therefore starts with the dependency plan and the address plan, then walks through the build in the DCD, referencing the unit that teaches each step rather than repeating it, and ends with the gates that prove the core works as one system.
1. Build Order Follows Dependencies
1.1 The Dependencies That Fix the Order
Five platform facts decide the sequence:
- Reserved addresses come first. A public ALB listener needs a reserved public IP, the NAT Gateway needs reserved IP addresses, and the VPN Gateway takes a reserved address from its location (Units 3.3 and 3.6). Reserved IPv4 addresses are bound to their region, so reserve them in Frankfurt.
- Targets before the NLB. The Managed Network Load Balancer needs a private network with targets that are provisioned already (Unit 3.4).
- First provisioning before NAT rules. The NAT Gateway's rules can be configured only after it has been provisioned once (Unit 3.6).
- Clients on the LAN before managed databases. A DBaaS cluster attaches to a private LAN of the VDC, and every client needs a NIC on that LAN (Unit 5.3). Build the data LAN and its servers first.
- An active cluster before node pools. A Kubernetes cluster stays Deploying until it is created; node pools are added once it is Active (Unit 6.2).
1.2 The Build Plan
| Stage | What you build | Depends on | Unit | Decision it realises |
|---|---|---|---|---|
| 0 | Contract, VDC, groups, buckets, certificate | Nothing | 2.1, 2.2, 3.3, 5.2 | Location de/fra fixed; default-deny access |
| 1 | Reserved block fc-prod-fra-edge (3 IPs) plus one IP for the node pool |
Stage 0 | 3.1, 6.2 | Stable public endpoints |
| 2 | LANs fc-app and fc-data, four Dedicated Core Servers with volumes |
Stage 0 | 3.1, 4.1, 4.2 | Private tiers, explicit zones, class per tier |
| 3 | Tier NSGs, default NSG narrowed | Stage 2 | 3.2 | Least privilege by membership |
| 4 | NAT Gateway, then SNAT rules and guest routes | Stages 1, 2 | 3.6 | Egress without public interfaces |
| 5 | ALB fc-alb-edge with target groups |
Stages 1, 2, 3 | 3.3 | Public Layer 7 edge, TLS ends here |
| 6 | NLB fc-nlb-data |
Stage 2 provisioned | 3.4 | Private Layer 4 path, certificates stay on the backend |
| 7 | fc-pg-portal, then fc-imdb-portal |
Stage 2 | 5.3, 5.5 | Zero-loss failover; read scaling by cache and pooler |
| 8 | fc-k8s-portal and its node pools, joined to the ALB |
Stages 1, 2, 5 | 6.2 | ALB in front instead of a LoadBalancer Service |
| 9 | VPN Gateway fc-vpn-onprem and its tunnel |
Stages 1, 2 | 3.6 | Hybrid link for the migration waves |
| 10 | Central Logging, Central Monitoring, egress flow log | Stages 4, 7 | 7.2 | Telemetry planes on |
| 11 | DNS records and the failover monitor | Stage 5 and the Berlin pilot light | 3.7, 7.1 | Cross-site steering after data |
1.3 The Address Plan
Write every managed-service address down before the first click, from the reserved range .2 to .9 of each LAN (Unit 3.1):
| LAN | Address | Member |
|---|---|---|
fc-app |
.2 | NAT Gateway fc-nat-app |
fc-app |
.3 | NLB listener fc-nlb-data |
fc-app |
.4 | VPN Gateway fc-vpn-onprem |
fc-app |
.225 (default) | ALB backend address, the source in fc-nsg-app |
fc-data |
.2 | fc-pg-portal |
fc-data |
.3 | fc-imdb-portal |
fc-data |
.4 to .9 | Kept for fc-mongo-cases and fc-kafka-events (Units 5.4 and 5.6) |
fc-data |
.225 (default) | NLB backend address, the source in fc-nsg-data |
As shown above, fc-data has no free managed-service address left once MongoDB and Kafka join it, which Unit 5.6 already flagged: the next managed service on the data tier needs a new LAN.
2. Integration Gates
A build that provisions is not yet a working core. Each gate tests a join between two modules and must pass before the next stage carries traffic:
| Gate | Test | Proves |
|---|---|---|
| G1 Topology | fc-data shows no internet access element; every server and volume is in Zone 1 or Zone 2, not Auto |
Units 3.1 and 7.1 hold together |
| G2 Filtering | A connection from a portal server to an unopened port on fc-data-01 appears as REJECT in the flow log |
NSGs do what Unit 3.2 designed |
| G3 Egress | A portal server reaches a patch mirror; DNS resolves through the gateway | NAT rules and guest routes (Unit 3.6) |
| G4 Edge | HTTPS to the listener reaches the portal; /api/ reaches fc-tg-api; a stopped portal server drops out of the target group |
ALB rules and health checks (Unit 3.3) |
| G5 Data | The portal reads a profile through fc-imdb-portal, writes through the pooler, and sees its own change at once |
Cache-aside and key deletion (Units 5.3 and 5.5) |
| G6 Internal path | The payment switch answers through fc-nlb-data with mutual TLS intact |
Layer 4 pass-through (Unit 3.4) |
| G7 Cluster | The cluster host on the ALB reaches the ingress controller on the labelled nodes | ALB in front of Kubernetes (Unit 6.2) |
| G8 Hybrid | An on-premises host reaches fc-app over the tunnel and back |
Static routes on both sides (Unit 3.6) |
| G9 Telemetry | PostgreSQL dashboards show data; flow log objects arrive in the bucket | Both forwarding layers on (Unit 7.2) |
DCD Implementation Walkthrough
You will build the enterprise core of FinCorp in fc-prod-fra-core, in the order of Section 1.2, with the addresses of Section 1.3. Each step names the unit whose walkthrough holds the detailed field-by-field procedure; this lab gives the values and the joins. Provision with PROVISION CHANGES: review the Validation tab of the Provision Data Center window, confirm with your password and select Provision Now.
Prerequisites: the production contract and the VDC fc-prod-fra-core in de/fra, created with the default network security group (Unit 2.1); the groups fc-prod-platform-ops and fc-prod-deploy (Unit 2.2); an Administrator account, or the privileges each unit names; the portal certificate imported in Menu > Security > Certificate Manager (Unit 3.3); the buckets fc-flowlogs-fra and fc-k8s-audit-fra (Units 3.2 and 6.2); the Berlin VDC fc-prod-txl-dr with its ALB, and the zone fincorp.example (Units 3.7 and 7.1); the on-premises gateway's public address and subnets.
Build goal: Build the FinCorp enterprise core end to end in the DCD, tying together the module walkthroughs.
Steps (in the Data Center Designer):
- Reserve the addresses (Unit 3.1, step 1; Unit 6.2). In Menu > Network Services > IP Management, reserve the block
fc-prod-fra-edgewith three IPs in Frankfurt for the ALB listener, the NAT Gateway and the VPN Gateway, and one further IPv4 address for the node pool's Reserved IP address field. Write each address into the plan. - Build the private tiers (Unit 3.1, steps 2 to 10). Open
fc-prod-fra-core, createfc-app-01(Zone 1) andfc-app-02(Zone 2) onfc-app, andfc-data-01(Zone 1) andfc-data-02(Zone 2) onfc-data. Do not connect the Internet element to either LAN. Give each portal server a second NIC onfc-data, because every database client needs a NIC on the cluster's LAN (Unit 5.3). - Size the servers and attach the volumes (Units 4.1 and 4.2). Set the Dedicated Core sizing, with AMD EPYC Milan as CPU Architecture for the data tier. Add each boot volume in its server's zone with Balanced, the cloud-init user data and VirtIO; add the Performance data volumes of at least 100 GB to
fc-data-01andfc-data-02. Provision, and check/var/log/cloud-init-output.logbefore any server joins a target group. - Filter the tiers (Unit 3.2, steps 1 to 9). Create
fc-nsg-appwith ingress from the ALB backend address.225onfc-app, andfc-nsg-datawith ingress from the NLB backend address.225onfc-datato the payment switch port. The managed databases are not VMs or NICs, so these groups do not filter them; their protection is the private LAN. Narrow the default NSG's broad rules, attach the tier groups, and add theRejectedflow log onfc-data-01. Provision. - Open the egress path (Unit 3.6, part B). Add
fc-nat-app, connect its interface tofc-app, add a public IP fromfc-prod-fra-edge, and provision. Move its private address to.2in Gateway IPs, then create the TCP SNAT rule for thefc-appsubnet and the UDP rule to the DNS resolvers on port 53. Provision, and set each VM's default route to the gateway's private IP in the guest. - Build the public edge (Unit 3.3, steps 1 to 10). Create
fc-tg-portalandfc-tg-apiwith health checks and the portal and API servers as targets. Drag an application load balancer into the Workspace, connect its northern interface to Internet Access (this LAN becomesfc-edge) and its southern interface tofc-app. Name itfc-alb-edge, select the reserved listener address, add the HTTPS forwarding rule on port 443 with the imported certificate, the Path rule for/api/and the default rule tofc-tg-portal. Confirm the backend address.225in Private IPs, then provision. - Build the internal path (Unit 3.4, steps 1 to 8). With the data servers provisioned, drag the NLB into the Workspace, connect the northern interface to
fc-appand the southern interface tofc-data. Name itfc-nlb-datawith Primary IPv4.3onfc-app, add the forwarding rule with Algorithm Source IP and the payment switch port, addfc-data-01andfc-data-02as targets, set the health check timeouts, and provision. - Create the relational cluster (Unit 5.3, steps 1 to 11). In Menu > Databases > PostgreSQL, create
fc-pg-portalin Frankfurt with3Instances, Synchronization mode Strictly Synchronous, Connection Pooler Transaction, a Backup Location outside Frankfurt,fc-dataas VDC LAN and.2as Private IP, a fixed maintenance period and Enable logs and Enable metrics on. Wait for AVAILABLE and copy the DNS Name. - Create the cache (Unit 5.5, steps 1 to 11). In Menu > Databases > In-Memory DB, create
fc-imdb-portalin Frankfurt with3Instances, AOF persistence,volatile-lrueviction, a snapshot location outside Frankfurt,fc-dataand.3. Point the portal at both DNS names, with TLS on the cache connection. - Add the cluster (Unit 6.2, steps 1 to 12). Create
fc-k8s-portalwith Public node pool access inde/fra, audit logging tofc-k8s-audit-fra, FinCorp's range in Allowed IPs and the change window as maintenance window. When it is Active, addfc-np-portal-z1andfc-np-portal-z2onfc-appwith Dedicated Core templates in Zone 1 and Zone 2. Grantfc-prod-platform-opsEdit in the Security tab, apply the zone-aware storage class, install the ingress controller as a NodePort service, createfc-tg-k8s, and add the cluster's host rule tofc-alb-edge. - Connect the hybrid gateway (Unit 3.6, part A). In Menu > Network Services > VPN Gateway, create
fc-vpn-onpremin Frankfurt with a reserved address, Premium VPN with High Availability, IPSec with IKEv2, and the LAN connection onfc-appat.4. Create the tunnel with a 32-character PSK and the cloud and peer CIDRs, mirror the settings on premises, and add the persistent routes on both sides. - Switch on the telemetry planes (Unit 7.2, steps 1 to 8). Activate Central Logging and Central Monitoring for Frankfurt, create the VM metrics pipeline and the alerts, and add the
Bidirectional,Anyflow log onfc-nat-apptofc-flowlogs-fra. Provision. - Wire the steering layer (Units 3.7 and 7.1). In the zone
fincorp.example, pointportal-fraandportal(TTL60) at the listener address offc-alb-edge, and confirm that the monitor's failover action rewritesportalto the Berlin ALB only after a declared incident. - Run the gates. Work through G1 to G9 of Section 2 in order, record each result, and run the failover drill of Unit 7.1 before the core is signed off.
Common mistakes:
- Creating the ALB, the NAT Gateway or the VPN Gateway before the addresses are reserved. Each of them takes a reserved IP from the Frankfurt block.
- Building the NLB before its targets are provisioned. Provision the data servers first.
- Expecting the NAT rules to exist with the gateway. Provision once, then create the SNAT rules, then set the routes in each guest.
- Forgetting the UDP rule for DNS on the NAT Gateway. Name resolution fails once the default route points at the gateway.
- Writing the NSG sources as "the clients". The targets see the balancers' backend addresses,
.225by default. - Placing a managed service on an address outside .2 to .9. DHCP can hand it to a VM.
- Leaving a server or volume on Auto. Gate G1 fails, and so does the zone argument in the audit.
- Exposing the portal through a Kubernetes LoadBalancer Service. That is a static IP on one node; the ALB is the cluster's entry.
- Letting the monitor move
portalbefore the Berlin database exists. Data first, then steering.
Summary
The capstone turns the module builds into one environment. The order comes from the platform's dependencies: reserved addresses, provisioned targets, a first NAT provisioning and an active cluster each gate the next stage. The address plan reserves every managed-service slot before the first click. Nine integration gates then prove that the joins hold: filtering against the right sources, egress, the edge rules, cache and pooler, the Layer 4 path, the cluster behind the ALB, the hybrid link, telemetry and the cross-site drill.
Key Points:
- Reserve public addresses first; the ALB, NAT Gateway and VPN Gateway all depend on them.
- Provision the data servers before the NLB, and the NAT Gateway before its rules.
- Plan .2 to .9 on each LAN for managed services;
.225is the balancers' default backend address. - Every database client needs a NIC on the cluster's LAN; the portal servers get a second NIC on
fc-data. - Sign the core off only after the integration gates and the Unit 7.1 failover drill pass.