<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Love Arinze on Medium]]></title>
        <description><![CDATA[Stories by Love Arinze on Medium]]></description>
        <link>https://medium.com/@thelovearinze?source=rss-540567ff3019------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*iBSVk82wiF4qrFkg_d6Nwg@2x.jpeg</url>
            <title>Stories by Love Arinze on Medium</title>
            <link>https://medium.com/@thelovearinze?source=rss-540567ff3019------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 05:56:37 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thelovearinze/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AWS PrivateLink Explained Simply]]></title>
            <link>https://towardsaws.com/aws-privatelink-explained-simply-365e1798da5a?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/365e1798da5a</guid>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[aws-networking]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[network]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 11:46:32 GMT</pubDate>
            <atom:updated>2026-10-04T11:46:32.184Z</atom:updated>
            <content:encoded><![CDATA[<p>The Cloud Networking Concept That Changed How I Think About VPCs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PKuR7QeZodv79GiFmEDm9w.png" /></figure><p>What is AWS PrivateLink?</p><p>Not just:</p><p><em>“A service that provides private connectivity between VPCs.”</em></p><p>But really, what does that mean?</p><p>The first time I heard terms like Interface Endpoint, Endpoint Service, Service Provider, Service Consumer, Network Load Balancer, and Private DNS, it felt like AWS had created several names for the same thing.</p><p>At first, it was hard to tell what each one actually did.</p><p>PrivateLink is based on one simple idea:</p><p><em>“Give someone access to your application without giving them access to your entire network.”</em></p><h3>1. Let’s Start With the Problem</h3><p>Imagine two companies.</p><p>Company A has an API running inside its AWS VPC.</p><p>Company B needs to access that API.</p><p>Company B needs private access, but Company A does not want to connect both networks together completely.</p><p>One option could be VPC Peering.</p><p>Another could be a Transit Gateway.</p><p>You could also build VPN connectivity between both environments.</p><p>But all of these approaches are focused on connecting networks.</p><p>What if Company B does not need access to Company A’s network?</p><p>What if they only need access to one application?</p><p>That is where PrivateLink comes in.</p><h3>2. PrivateLink in Simple Words</h3><p>Think about it this way:</p><h3>VPC Peering</h3><p>“Your network can communicate with my network.”</p><h3>PrivateLink</h3><p>“You can communicate with this specific service inside my network.”</p><p>That is the core idea behind PrivateLink.</p><h3>3. The Two Sides You Need to Understand</h3><p>There are two sides to a PrivateLink connection.</p><h3>Service Provider</h3><p>The provider owns the application or service.</p><p>They are basically saying: <em>“I have a service that I want you to access privately.”</em></p><p>On AWS, the provider creates an Endpoint Service.</p><p>The application is typically exposed through a Network Load Balancer.</p><p>So think:</p><p>Provider = owns the service</p><h3>Service Consumer</h3><p>The consumer is the AWS account that needs private access to the service.</p><p>They create an Interface VPC Endpoint inside their VPC.</p><p>So think:</p><p>Consumer = connects to the service</p><p>This is the most important relationship to remember:</p><pre>CONSUMER                     PROVIDER</pre><pre>Interface Endpoint  -----&gt;   Endpoint Service</pre><pre>&quot;I connect&quot;                  &quot;I provide&quot;</pre><p>If you remember that, you already understand a large part of PrivateLink.</p><h3>4. So What Exactly Is an Interface Endpoint?</h3><p>This is where things can get confusing.</p><p>When the consumer creates an Interface VPC Endpoint, AWS creates network interfaces inside the consumer’s selected subnets.</p><p>Those network interfaces receive private IP addresses from the consumer’s VPC.</p><p>For example:</p><pre>Consumer VPC<br>172.16.0.0/16</pre><pre>PrivateLink Endpoint:</pre><pre>172.16.10.25<br>172.16.20.31</pre><p>From the consumer application’s point of view, it is connecting to a private IP inside its own VPC.</p><p>Behind the scenes, AWS carries that traffic through PrivateLink to the provider’s service.</p><p>You do not need normal IP routing into the provider’s entire VPC.</p><p>The consumer only gets access to the service that has been published.</p><h3>5. What Does the Traffic Flow Look Like?</h3><p>Without PrivateLink, you might have something like this:</p><pre>Application<br>     |<br>     |<br>VPC Peering / TGW / VPN<br>     |<br>     |<br>Provider VPC<br>     |<br>Application</pre><p>The networks themselves are connected.</p><p>With PrivateLink, the model looks more like this:</p><pre>Consumer VPC<br>     |<br>Application<br>     |<br>Private IP<br>     |<br>Interface Endpoint<br>     |<br>========================<br>     AWS PrivateLink<br>========================<br>     |<br>Endpoint Service<br>     |<br>Network Load Balancer<br>     |<br>Provider Application</pre><p>The consumer does not need general connectivity into the provider VPC.</p><p>They are simply consuming the published service.</p><h3>6. Where Does DNS Come In?</h3><p>This is another part of PrivateLink that is easy to overlook.</p><p>Let’s say your application normally connects to:</p><pre>https://api.company.com</pre><p>Without PrivateLink, that hostname might resolve to a public endpoint.</p><p>Something like:</p><pre>api.company.com<br>        |<br>        v<br>   Public IP<br>        |<br>        v<br> Public Endpoint</pre><p>When Private DNS is configured for the service and enabled on the endpoint, the same hostname can resolve to the private IP addresses of your Interface Endpoint.</p><p>Now you have:</p><pre>api.company.com<br>        |<br>        v<br>  Private IP<br>        |<br>        v<br>Interface Endpoint<br>        |<br>        v<br>   PrivateLink<br>        |<br>        v<br> Provider Service</pre><p>The interesting part is that your application may not need to change anything.</p><p>It can still call:</p><pre>https://api.company.com</pre><p>DNS simply points it towards the private endpoint instead.</p><p>A simple way to remember this is: <em>PrivateLink provides the private path. DNS tells your application where that path begins.</em></p><h3>7. Provider vs Consumer: Do Not Mix These Up</h3><p>If you remember only one section from this article, remember this:</p><h3>Provider</h3><p>Creates:</p><pre>Endpoint Service</pre><p>The provider owns the application.</p><h3>Consumer</h3><p>Creates:</p><pre>Interface VPC Endpoint</pre><p>The consumer wants to access the application.</p><p>So:</p><pre>Provider = Endpoint Service</pre><pre>Consumer = Interface Endpoint</pre><p>The names sound similar, which is why they are easy to mix up.</p><h3>8. Why Not Just Use VPC Peering?</h3><p>Because they solve different problems.</p><p>With VPC Peering, you are creating network connectivity between two VPCs.</p><p>For example:</p><pre>VPC A<br>172.16.0.0/16</pre><pre>      |</pre><pre>   Peering</pre><pre>      |</pre><pre>VPC B<br>10.20.0.0/16</pre><p>You add routes on both sides to send traffic across the peering connection.</p><p>PrivateLink works differently.</p><p>The consumer does not need a route to the provider’s VPC CIDR.</p><p>Instead, the consumer connects to private IP addresses belonging to the Interface Endpoint inside their own VPC.</p><p>It comes down to what you need to connect: two networks, or one specific service.</p><h3>9. Example</h3><p>Imagine a company called Company A provides a payment API.</p><p>Their API runs inside:</p><pre>VPC A<br>10.10.0.0/16</pre><p>Another company, Company B, needs to call that API from:</p><pre>VPC B<br>172.20.0.0/16</pre><p>Company A does not want to peer both VPCs.</p><p>They only want Company B to access:</p><pre>https://payments.company-a.com</pre><p>Company A becomes the:</p><pre>Service Provider</pre><p>They expose their application through a Network Load Balancer and create an:</p><pre>Endpoint Service</pre><p>Company B becomes the:</p><pre>Service Consumer</pre><p>They create an:</p><pre>Interface VPC Endpoint</pre><p>AWS places private endpoint IP addresses inside Company B’s VPC.</p><p>When Company B’s application connects to the API, the traffic flows:</p><pre>Company B Application<br>        |<br>        v<br>Interface Endpoint<br>        |<br>        v<br>AWS PrivateLink<br>        |<br>        v<br>Company A Endpoint Service<br>        |<br>        v<br>Network Load Balancer<br>        |<br>        v<br>Company A API</pre><p>Company B gets access to the API.</p><p>They do not get access to Company A’s entire VPC.</p><h3>10. What About Security Groups?</h3><p>The Interface Endpoint can have a security group attached to it.</p><p>That means you can control which applications inside the consumer VPC are allowed to connect to the endpoint.</p><p>For example:</p><pre>Application Server<br>172.20.10.50<br>        |<br>        | HTTPS :443<br>        v<br>Interface Endpoint<br>172.20.20.25</pre><p>You might configure the endpoint security group to allow:</p><pre>TCP 443<br>Source: 172.20.10.50/32</pre><p>So even inside the consumer VPC, access to the endpoint can still be restricted.</p><h3>11. Does PrivateLink Work Across Regions?</h3><p>Yes.</p><p>AWS supports Cross-Region PrivateLink.</p><p>That means a service provider can expose a service from one AWS Region and allow consumers to create Interface Endpoints from supported Regions.</p><p>For example:</p><pre>Consumer</pre><pre>us-east-1<br>N. Virginia</pre><pre>      |<br>      |<br>Cross-Region PrivateLink<br>      |<br>      |</pre><pre>Provider</pre><pre>eu-west-1<br>Ireland</pre><p>The roles remain exactly the same.</p><p>The provider still creates the Endpoint Service.</p><p>The consumer still creates the Interface Endpoint.</p><p>The setup is the same, except the provider and consumer are now in different AWS Regions.</p><h3>12. How I Remember PrivateLink</h3><p>Whenever I see PrivateLink, I ask three questions.</p><h3>Who owns the application?</h3><p>That is the:</p><pre>Provider</pre><h3>Who needs to connect to it?</h3><p>That is the:</p><pre>Consumer</pre><h3>What does each side create?</h3><pre>Provider  → Endpoint Service</pre><pre>Consumer  → Interface Endpoint</pre><p>That model removes most of the confusion.</p><h3>13. Let’s Recap</h3><p>PrivateLink gives private access to a specific service across network boundaries.</p><ul><li>The provider publishes the service.</li><li>The consumer creates an Interface Endpoint.</li><li>The endpoint receives private IP addresses inside the consumer VPC.</li><li>The application connects to those private IP addresses.</li><li>AWS carries the traffic privately to the provider.</li><li>With Private DNS, the application can often continue using the same hostname it already uses.</li><li>So the next time you want to define <strong>AWS PrivateLink</strong>, you do not have to start with: <em>“It’s an AWS networking service that uses Interface VPC Endpoints…”</em></li><li>You can simply say: <em>AWS PrivateLink allows a VPC to privately access a specific service through a private endpoint, without requiring direct network connectivity to the service provider’s VPC.</em></li></ul><p><strong>A quick note:</strong> AWS PrivateLink now supports more than just Interface Endpoints and Endpoint Services. AWS also has Resource endpoints and Service Network endpoints for other private connectivity patterns. This article focuses on the Interface Endpoint to Endpoint Service model to keep the explanation simple.</p><h3>What’s Next?</h3><p>Understanding the concept is one thing.</p><p>Building it is where all the small details start to matter.</p><p>In the next article, I will show you <strong>step by step how to set up Cross-Region AWS PrivateLink</strong>, from creating the provider Endpoint Service to creating the consumer Interface Endpoint, configuring DNS, and finally testing the HTTPS connection.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=365e1798da5a" width="1" height="1" alt=""><hr><p><a href="https://towardsaws.com/aws-privatelink-explained-simply-365e1798da5a">AWS PrivateLink Explained Simply</a> was originally published in <a href="https://towardsaws.com/">Towards AWS</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[East-West Inspection in AWS]]></title>
            <link>https://towardsaws.com/east-west-inspection-in-aws-98ec08ed4315?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/98ec08ed4315</guid>
            <category><![CDATA[terraform]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[palo-alto-networks]]></category>
            <category><![CDATA[network-security]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Tue, 01 Sep 2026 02:11:12 GMT</pubDate>
            <atom:updated>2026-09-01T21:49:39.804Z</atom:updated>
            <content:encoded><![CDATA[<p>How I used AWS Transit Gateway, Gateway Load Balancer, Palo Alto VM-Series, and Terraform to enforce security between VPCs.</p><p>A technical interview recently gave me a question that sounded straightforward until the follow-up questions began:</p><blockquote><em>How would you securely inspect traffic between multiple VPCs connected through AWS Transit Gateway?</em></blockquote><p>I understood the high-level design: connect the application VPCs to a Transit Gateway, place a firewall in a dedicated Inspection VPC, and steer traffic through it.</p><p>But a high-level answer was not enough.</p><p>How does a packet enter the Inspection VPC? Which route table is evaluated at each stage? How does the packet reach the firewall? What exactly does appliance mode solve? How do you prove that the firewall, rather than a security group or a missing route, made the final decision?</p><p>Instead of memorizing a better answer, I built the architecture.</p><p>The result was a Terraform-managed AWS lab with four VPCs, an AWS Transit Gateway, a Gateway Load Balancer, a Gateway Load Balancer Endpoint, and a Palo Alto Networks VM-Series firewall.</p><p>More importantly, I proved three things:</p><ul><li>The service-insertion path was operational, with a healthy firewall target and an available Gateway Load Balancer Endpoint.</li><li>Production-to-Development ICMP traffic crossed the VM-Series firewall and was allowed by an explicit security rule.</li><li>Production-to-Development TCP port 22 traffic was denied by the firewall and recorded with the end reason policy-deny.</li></ul><p>This article explains the design as a packet journey. It covers the routing decisions, the validation evidence, the mistakes I encountered, and the changes required before using this pattern in production.</p><h3>Lab goals and constraints</h3><p>The lab had six design goals:</p><ol><li>Production, Development, Shared Services, and Inspection would use separate VPCs.</li><li>Spoke VPCs would not receive unrestricted direct routes to one another.</li><li>East-west traffic would first enter a dedicated Inspection VPC.</li><li>Gateway Load Balancer would insert Palo Alto VM-Series transparently into the traffic path.</li><li>Forward and response traffic would traverse a symmetric stateful inspection path.</li><li>The result would be validated through AWS state, firewall sessions, explicit policy, and traffic logs.</li></ol><p>The test workloads had no public IP addresses. Terraform created and destroyed the AWS infrastructure.</p><p>This was deliberately a cost-controlled proof of concept. Although the network foundation included multiple Availability Zones, the validated path used one active firewall and one Gateway Load Balancer Endpoint in eu-west-1a.</p><p>It was not a production high-availability deployment.</p><h3>Addressing plan</h3><p>I assigned one private /16 to each functional boundary:</p><ul><li><strong>Production VPC:</strong> 172.20.0.0/16</li><li><strong>Development VPC:</strong> 172.21.0.0/16</li><li><strong>Shared Services VPC:</strong> 172.22.0.0/16</li><li><strong>Inspection VPC:</strong> 172.23.0.0/16</li><li><strong>Reserved simulated on-premises prefix:</strong> 172.24.0.0/16</li></ul><p>The validation traffic flowed between:</p><ul><li><strong>Production test instance:</strong> 172.20.10.10</li><li><strong>Development test instance:</strong> 172.21.10.10</li></ul><h3>Architecture at a glance</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9zdY8Xq2RKczjGWJt7cuEQ.png" /><figcaption><em>Figure 1: The validated single-AZ inspection path from Production through Transit Gateway, GWLBE, GWLB, and Palo Alto VM-Series to Development.</em></figcaption></figure><p>The forward path was:</p><pre>Production VPC<br>→ AWS Transit Gateway<br>→ Inspection VPC attachment<br>→ Gateway Load Balancer Endpoint<br>→ Gateway Load Balancer<br>→ Palo Alto VM-Series<br>→ Gateway Load Balancer<br>→ Gateway Load Balancer Endpoint<br>→ AWS Transit Gateway<br>→ Development VPC</pre><p>The response followed the same inspection chain in reverse.</p><p>Each component had a specific responsibility:</p><ul><li><strong>AWS Transit Gateway</strong> provided hub connectivity and separate routing domains.</li><li><strong>Transit Gateway route tables</strong> selected the next attachment based on where the packet entered.</li><li><strong>Inspection VPC subnet route tables</strong> inserted the Gateway Load Balancer Endpoint into the path.</li><li><strong>Gateway Load Balancer Endpoint</strong>, or GWLBE, provided the routable service-insertion target.</li><li><strong>Gateway Load Balancer</strong>, or GWLB, distributed GENEVE-encapsulated traffic to a healthy firewall target.</li><li><strong>Palo Alto VM-Series</strong> performed the stateful inspection and enforced security policy.</li></ul><p>That distinction is important.</p><p>Transit Gateway did not inspect the packet, and GWLB did not decide whether ICMP or SSH should be permitted. The firewall made the policy decision.</p><h3>Routing was the heart of the design</h3><p>Creating VPCs and Transit Gateway attachments was the easy part.</p><p>The security behavior came from the association, propagation, and subnet-routing decisions.</p><h3>The spoke TGW route table forced inspection</h3><p>The Production, Development, and Shared Services attachments were associated with a <strong>Spoke TGW route table</strong>.</p><p>The decisive route was:</p><pre>0.0.0.0/0 → Inspection VPC attachment</pre><p>The spoke route table did not provide an unrestricted direct route from Production to Development.</p><p>When a packet from 172.20.10.10 entered Transit Gateway through the Production attachment, the spoke route table sent it to the Inspection VPC attachment.</p><h3>Inspection VPC subnet route tables inserted the GWLBE</h3><p>After Transit Gateway delivered the packet to the Inspection VPC, the route table associated with the Transit Gateway attachment subnet directed the traffic to the GWLBE.</p><p>The endpoint inserted the GWLB service into the route path. GWLB encapsulated the original packet using GENEVE over UDP port 6081 and forwarded it to the Palo Alto target.</p><p>After inspection, the packet returned through GWLB and the GWLBE before being sent back toward Transit Gateway.</p><p>This was a separate routing decision from the Transit Gateway route-table decision:</p><ul><li><strong>TGW route tables</strong> chose the next attachment.</li><li><strong>Inspection VPC subnet route tables</strong> steered traffic through the GWLBE.</li></ul><h3>The Inspection TGW route table selected the destination</h3><p>The Inspection VPC attachment was associated with a separate <strong>Inspection TGW route table</strong>.</p><p>That table contained the spoke destination routes:</p><pre>172.20.0.0/16 → Production attachment<br>172.21.0.0/16 → Development attachment<br>172.22.0.0/16 → Shared Services attachment</pre><p>Once the inspected packet re-entered Transit Gateway through the Inspection attachment, this route table selected the Development attachment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pQSTNG3jAGpvqAEkk-u4BQ.png" /><figcaption><em>Figure 2: The spoke TGW route table sends traffic to the Inspection VPC, while the Inspection TGW route table contains propagated routes back to the Production, Development, and Shared Services attachments.</em></figcaption></figure><p>This is how the same Transit Gateway made different forwarding decisions depending on the packet’s ingress attachment.</p><p>I also created a third <strong>Hybrid TGW route table</strong> for future shared-services and on-premises patterns. It was not part of the Production-to-Development validation flow.</p><h3>Why appliance mode mattered</h3><p>A stateful firewall must observe both directions of a session.</p><p>Without the correct path controls, the forward packet and the response can traverse different Availability Zone attachment interfaces. A stateful appliance may then see only half of the conversation and drop the flow.</p><p>I enabled appliance mode only on the Inspection VPC attachment. The workload attachments remained in their default mode.</p><p>Appliance mode and GWLB flow stickiness solve related but different problems:</p><ul><li><strong>Transit Gateway appliance mode</strong> maintains the selected Availability Zone path through the appliance VPC for the lifetime of the flow.</li><li><strong>GWLB flow stickiness</strong> keeps a flow mapped to a specific target appliance.</li><li><strong>Route tables</strong> still determine whether both directions traverse the inspection chain.</li></ul><p>Appliance mode does not replace correct routing, and it does not configure the firewall.</p><h3>Deploying Palo Alto VM-Series</h3><p>I subscribed to the Palo Alto Networks VM-Series PAYG offer through AWS Marketplace and verified that the selected AMI contained the expected Marketplace product code before using it in Terraform.</p><p>The firewall used two network interfaces:</p><ul><li><strong>Dataplane interface:</strong> AWS device index 0, presented as ethernet1/1 after the management-interface swap, with address 172.23.30.10</li><li><strong>Management interface:</strong> AWS device index 1, with address 172.23.40.10</li></ul><p>The bootstrap configuration enabled the interface swap and GWLB inspection support:</p><pre>type=dhcp-client<br>hostname=tgw-inspection-fw-a<br>op-command-modes=mgmt-interface-swap<br>plugin-op-commands=aws-gwlb-inspect:enable</pre><p>For temporary lab administration, the management interface received an Elastic IP. Its security group allowed HTTPS and SSH only from my administrator public /32.</p><p>I configured ethernet1/1 as a Layer 3 DHCP interface, added it to the default virtual router and the inspection zone, and attached an interface-management profile for the GWLB health check.</p><p>The first checkpoint was not application connectivity. It was service readiness.</p><h3>Validation layer 1: Service readiness</h3><p>The GWLB target group reported the firewall target as healthy on GENEVE port 6081.</p><p>The GWLBE reached the available state.</p><pre>Firewall target: healthy<br>GENEVE port:      6081<br>GWLBE state:      available</pre><p>This proved that the service-insertion components were operational.</p><p>It did not yet prove that Production-to-Development traffic traversed the firewall.</p><h3>Validation layer 2: The firewall saw the session</h3><p>I deployed two private Amazon Linux 2023 test instances:</p><ul><li>Production: 172.20.10.10</li><li>Development: 172.21.10.10</li></ul><p>Neither instance had a public IP address. I used EC2 Instance Connect Endpoints for private administrative access instead of deploying internet-facing bastion hosts.</p><p>From the Production instance, I tested ICMP:</p><pre>ping -c 4 172.21.10.10</pre><p>The result was:</p><pre>4 packets transmitted, 4 received, 0% packet loss</pre><p>A successful ping proved end-to-end reachability, but reachability alone was insufficient.</p><p>A route or security group could allow the packet without proving that the firewall inspected it.</p><p>I queried the Palo Alto session table using the exact source and destination addresses:</p><pre>show session all filter source 172.20.10.10 destination 172.21.10.10</pre><p>The output showed:</p><pre>Application: ping<br>State:       ACTIVE<br>Source:      172.20.10.10<br>Destination: 172.21.10.10<br>Zone:        inspection</pre><p>The original and translated addresses remained identical.</p><p>That was direct evidence that the packet crossed the firewall without source NAT between the two VPCs.</p><h3>Validation layer 3: Explicit policy allowed ICMP and denied SSH</h3><p>Session visibility proved traversal.</p><p>The next step was proving that Palo Alto policy made the decision.</p><p>I created two ordered security rules:</p><pre>Rule:        allow-prod-to-dev-icmp<br>Source:      172.20.0.0/16<br>Destination: 172.21.0.0/16<br>Application: ping<br>Service:     application-default<br>Action:      allow<br>Log at end:  yes</pre><pre>Rule:        deny-prod-to-dev-all<br>Source:      172.20.0.0/16<br>Destination: 172.21.0.0/16<br>Application: any<br>Service:     any<br>Action:      deny<br>Log at end:  yes</pre><p>After committing the configuration, ICMP continued to succeed with zero packet loss.</p><p>The traffic log associated the flow with:</p><pre>Rule:   allow-prod-to-dev-icmp<br>Action: allow<br>App:    ping</pre><p>I then tested TCP port 22:</p><pre>timeout 5 bash -c &#39;echo &gt; /dev/tcp/172.21.10.10/22&#39;<br>echo $?</pre><p>The command returned:</p><pre>124</pre><p>Exit code 124 showed that the five-second timeout expired without establishing the TCP connection.</p><p>The firewall log provided the decisive evidence:</p><pre>Rule:             deny-prod-to-dev-all<br>Action:           deny<br>Destination port: 22<br>End reason:       policy-deny</pre><p>This distinguished an intentional security-policy decision from a generic network failure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CR5ZK444IhTkPoBEmcFSxA.png" /><figcaption><em>Figure 3: Palo Alto traffic logs showing ICMP allowed by the explicit rule and TCP port 22 denied with the end reason </em><em>policy-deny.</em></figcaption></figure><p>The validation focused on traffic initiated from Production toward Development. Other spoke directions would require equivalent routing and policy validation.</p><h3>Why the validation sequence mattered</h3><p>The strongest part of the lab was not the successful ping. It was the evidence chain:</p><ol><li><strong>Service readiness:</strong> the firewall target was healthy and the GWLBE was available.</li><li><strong>Data-plane traversal:</strong> the VM-Series session table showed the exact Production-to-Development flow.</li><li><strong>Policy enforcement:</strong> the traffic logs tied the allow and deny outcomes to named firewall rules.</li></ol><p>Each layer answered a different question.</p><p>A healthy target does not prove that application traffic uses the path.</p><p>A successful ping does not prove that the firewall allowed it.</p><p>A timeout does not prove that a firewall denied it.</p><p>The session and policy logs closed those gaps.</p><h3>What did not work immediately</h3><h3>The firewall looked unavailable while it was still booting</h3><p>The EC2 system check became healthy before the Palo Alto management plane was ready. My first HTTPS request timed out.</p><p>After VM-Series initialization completed, the same request returned an HTTP 302 redirect to the login page.</p><p>The lesson was simple: EC2 health and application readiness are different signals.</p><h3>A working route did not automatically prove inspection</h3><p>The first successful ping proved reachability, not inspection.</p><p>I needed session output and traffic logs tied to the exact source, destination, rule, and action before I could claim that the security control was working.</p><h3>Stateful inspection depended on two routing layers</h3><p>The flow required both:</p><ul><li>A spoke TGW route table that forced traffic into the Inspection VPC</li><li>Inspection VPC subnet route tables and an Inspection TGW route table that returned the inspected packet to the correct destination</li></ul><p>Treating Transit Gateway as one shared route table would have weakened segmentation and made the packet path harder to reason about.</p><h3><strong>I postponed the Terraform refactor until after validation</strong></h3><p>The AWS provider warned that inline network_interface blocks on aws_instance were deprecated.</p><p>After validating and destroying the live environment, I refactored the configuration to use:</p><ul><li>primary_network_interface for the dataplane interface</li><li>aws_network_interface_attachment for the secondary management interface</li></ul><p>I deliberately avoided changing the EC2 interface model during validation because it could have replaced the working firewall and invalidated the test.</p><h3>Terraform teardown did not cancel the Marketplace agreement</h3><p>terraform destroy removed the AWS infrastructure, but it could not cancel the Palo Alto Marketplace agreement.</p><p>After destroying the managed resources and confirming that Terraform state was empty, I cancelled the Marketplace subscription separately and verified that it moved to inactive subscriptions.</p><h3>What I would change for production</h3><p>This lab validated the inspection pattern. It was not a production firewall platform.</p><h3>High availability and capacity</h3><p>A production deployment should include:</p><ul><li>A complete inspection path in every participating Availability Zone</li><li>At least one healthy firewall target per Availability Zone</li><li>A GWLBE in each inspection path</li><li>Defined scaling and capacity-management procedures</li><li>Tested firewall, endpoint, route, and Availability Zone failure scenarios</li></ul><h3>Firewall configuration and management</h3><p>I would add:</p><ul><li>Automated bootstrap and policy configuration</li><li>Panorama or another controlled centralized management workflow</li><li>Version-controlled policy changes</li><li>Formal change approval and rollback procedures</li></ul><h3>Observability</h3><p>I would centralize:</p><ul><li>Palo Alto traffic, threat, and system logs</li><li>GWLB health and packet-drop metrics</li><li>Transit Gateway and VPC Flow Logs</li><li>Alerts for unhealthy or insufficient firewall capacity</li><li>Dashboards for policy-deny trends and path failures</li></ul><h3>Access, state, and governance</h3><p>I would also use:</p><ul><li>Private administrative access without a long-lived public management interface</li><li>Remote encrypted Terraform state with locking</li><li>CI checks for formatting, validation, security, and policy compliance</li><li>Cost monitoring and Marketplace subscription governance</li></ul><p>The Marketplace trial I used allowed one active firewall instance, and the objective was functional validation rather than high availability.</p><h3>Cost control and teardown</h3><p>This architecture created hourly charges for services including:</p><ul><li>Transit Gateway attachments</li><li>Gateway Load Balancer</li><li>Gateway Load Balancer Endpoint</li><li>EC2</li><li>EBS</li><li>Palo Alto Marketplace software</li></ul><p>I used a saved destroy plan and removed the environment immediately after collecting the evidence:</p><pre>Apply complete! Resources: 0 added, 0 changed, 94 destroyed.</pre><p>I then confirmed that:</p><pre>terraform state list</pre><p>returned no entries.</p><p>Finally, I cancelled the Marketplace agreement separately.</p><p>Infrastructure as code made the AWS teardown repeatable, but responsible cost control still required understanding which commercial resources existed outside Terraform’s lifecycle.</p><h3>Final takeaway</h3><p>The most valuable outcome was learning to explain the design as a sequence of forwarding and policy decisions:</p><ol><li>A packet enters Transit Gateway through the Production attachment.</li><li>The Spoke TGW route table sends it to the Inspection VPC attachment.</li><li>The Inspection VPC subnet route table sends it to the GWLBE.</li><li>GWLB encapsulates the packet with GENEVE and sends it to a healthy VM-Series target.</li><li>Palo Alto evaluates policy using the original source and destination addresses.</li><li>The inspected packet returns through GWLB and the GWLBE.</li><li>The Inspection TGW route table selects the Development attachment.</li><li>The response follows the same inspection chain in reverse.</li><li>Appliance mode and GWLB flow stickiness help preserve the stateful path, while route tables enforce the actual forwarding design.</li></ol><p>I can now answer the original interview question with more than a diagram or a memorized pattern.</p><p>I can explain the attachment associations, the separate route domains, the service-insertion mechanism, the role of appliance mode, the firewall policy, the validation evidence, the production limitations, and the teardown process because I built and tested each part.</p><p>Sometimes the most valuable interview feedback is the question you cannot stop thinking about afterward.</p><h3>Project repository</h3><p>The repository includes the Terraform configuration, Transit Gateway route-table design, Gateway Load Balancer resources, Palo Alto bootstrap configuration, and teardown guidance.</p><p><a href="https://github.com/thelovearinze/tgw-centralized-inspection">View the project on GitHub</a></p><h3>Technical references</h3><ul><li><a href="https://docs.aws.amazon.com/vpc/latest/tgw/how-transit-gateways-work.html">How AWS Transit Gateway works</a></li><li><a href="https://docs.aws.amazon.com/vpc/latest/tgw/tgw-vpc-attachments.html">Amazon VPC attachments in AWS Transit Gateway</a></li><li><a href="https://docs.aws.amazon.com/elasticloadbalancing/latest/gateway/introduction.html">What is a Gateway Load Balancer?</a></li><li><a href="https://docs.aws.amazon.com/whitepapers/latest/building-scalable-secure-multi-vpc-network-infrastructure/using-gwlb-with-tg-for-cns.html">Using Gateway Load Balancer with Transit Gateway for centralized network security</a></li><li><a href="https://docs.paloaltonetworks.com/vm-series/deployment/public-cloud/set-up-the-vm-series-firewall-on-aws/vm-series-integration-with-gateway-load-balancer/enabling-vm-series-integration-with-a-gwlb">Enable VM-Series integration with a Gateway Load Balancer</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=98ec08ed4315" width="1" height="1" alt=""><hr><p><a href="https://towardsaws.com/east-west-inspection-in-aws-98ec08ed4315">East-West Inspection in AWS</a> was originally published in <a href="https://towardsaws.com/">Towards AWS</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building a Self-Healing Egress Architecture for AWS Local Zones]]></title>
            <link>https://medium.com/@thelovearinze/building-a-self-healing-egress-architecture-for-aws-local-zones-235967dadd22?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/235967dadd22</guid>
            <category><![CDATA[aws-networking]]></category>
            <category><![CDATA[network-engineering]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[terraform]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Tue, 30 Jun 2026 19:44:34 GMT</pubDate>
            <atom:updated>2026-06-30T20:22:43.680Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Using AWS Transit Gateway Connect, GRE, and FRRouting to eliminate a single point of failure.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aCAcXs73riIbEG05gLwJvA.png" /><figcaption>Sample Workflow</figcaption></figure><p>AWS Local Zones solve a difficult problem: placing compute close to users while keeping workloads inside a specific geographic location.</p><p>For my team, that location was <strong>Lagos, Nigeria</strong>.</p><p>Because of data localization requirements, workloads needed to remain in the Lagos Local Zone while using the South Africa Region as the parent AWS Region.</p><p>The application architecture itself wasn’t the challenge.</p><p>Outbound internet access was.</p><p>Unlike a standard Availability Zone, AWS Local Zones do not currently provide managed NAT Gateways. That meant every private workload depended on a single EC2 instance performing network address translation.</p><p>On paper, the architecture looked simple.</p><pre>Private Workloads<br>        │<br>        ▼<br>  NAT Instance<br>        │<br>        ▼<br>    Internet</pre><p>It worked.</p><p>Until it didn’t.</p><p>If that EC2 instance failed, every workload behind it immediately lost outbound connectivity.</p><p>That raised a simple question.</p><p><strong>Instead of teaching AWS how to recover after a failure, could I teach the network to route around one?</strong></p><p>This article walks through the proof of concept I built to answer that question, including the mistakes, dead ends, and design decisions that eventually led to a self-healing egress architecture using AWS Transit Gateway Connect, GRE tunnels, and containerized FRRouting.</p><h3>The Problem</h3><p>The initial architecture contained:</p><ul><li>Two public subnets</li><li>Two private subnets</li><li>One NAT instance</li><li>One replication server</li></ul><p>Every private workload sent outbound traffic through a single NAT instance.</p><pre>Private Subnets<br>       │<br>       ▼<br>  NAT Instance<br>       │<br>       ▼<br>    Internet</pre><p>This immediately introduced a single point of failure.</p><p>If the NAT instance stopped responding:</p><ul><li>outbound internet access stopped</li><li>software updates failed</li><li>package repositories became unreachable</li><li>recovery depended on replacing the instance or changing route tables</li></ul><p>Neither option was particularly attractive.</p><h3>The Obvious Solution</h3><p>The common cloud-native approach would be something like this.</p><pre>CloudWatch Alarm<br>        │<br>        ▼<br>     Lambda<br>        │<br>        ▼<br> Update Route Table<br>        │<br>        ▼<br> Restore Connectivity</pre><p>This works.</p><p>But it also means failure has already happened before recovery begins.</p><p>Detection takes time.</p><p>Automation takes time.</p><p>Route propagation takes time.</p><p>Every recovery depends on the AWS control plane making infrastructure changes after the outage has already started.</p><p>I wanted something different.</p><h3>A Different Way to Think About the Problem</h3><p>Networks have been solving path failures for decades.</p><p>Routing protocols don’t wait for someone to update a route table.</p><p>They detect that a path no longer exists and select another one automatically.</p><p>Instead of treating the NAT appliance like an EC2 instance that occasionally needs replacing, I started treating it like a network router.</p><p>That completely changed the design.</p><p>Rather than building automation around the NAT instance, I built a routing layer that could make forwarding decisions dynamically.</p><h3>The Architecture</h3><p>The final proof of concept consists of:</p><ul><li>Two NAT appliances running in the Local Zone</li><li>AWS Transit Gateway</li><li>Transit Gateway Connect</li><li>GRE tunnels</li><li>Containerized FRRouting</li><li>Internal BGP</li><li>Equal Cost Multi Path (ECMP)</li></ul><p>When both appliances are healthy, each advertises a default route.</p><p>Transit Gateway installs both routes and distributes traffic across them using ECMP.</p><p>If one appliance disappears, its BGP session drops.</p><p>The route is withdrawn automatically.</p><p>Traffic converges onto the remaining appliance without modifying route tables or invoking Lambda.</p><pre>Internet<br>                        │<br>                Transit Gateway<br>                 /            \<br>          GRE + iBGP      GRE + iBGP<br>             │                │<br>      NAT Appliance 1   NAT Appliance 2<br>       FRRouting          FRRouting<br>             \              /<br>              \            /<br>             Private Workloads</pre><p>The routing protocol handles failover.</p><p>No scripts. No API calls. No manual intervention.</p><h3>The First Problem</h3><p>I originally planned to install FRRouting directly on Amazon Linux 2023.</p><p>That didn’t last long.</p><pre>dnf install frr</pre><p>FRRouting wasn’t available in the default repositories.</p><p>I could have enabled additional repositories and installed it anyway.</p><p>Instead I stopped and asked a different question.</p><p>Did I really want routing software tightly coupled to the operating system?</p><p>Probably not.</p><p>Running FRRouting inside Docker meant:</p><ul><li>consistent deployments</li><li>simpler upgrades</li><li>cleaner rollback</li><li>minimal host dependencies</li></ul><p>The routing software became portable instead of tied to a specific AMI.</p><p>That ended up being the better design.</p><h3>The Second Problem</h3><p>The architecture worked.</p><p>Then BGP stopped working.</p><p>After some packet captures and debugging, I realised the NAT appliance was masquerading the GRE traffic.</p><p>The Transit Gateway wasn’t seeing the packets it expected.</p><p>The fix was surprisingly small.</p><pre>iptables -t nat -I POSTROUTING \<br>-d 169.254.0.0/16 \<br>-j ACCEPT</pre><p>That single rule prevented tunnel traffic from being translated while allowing normal outbound traffic to continue through NAT.</p><p>Sometimes the smallest configuration change fixes the biggest problem.</p><h3>The Third Problem</h3><p>GRE adds overhead.</p><p>Without adjusting TCP MSS, fragmentation became an issue.</p><p>The solution was MSS clamping.</p><pre>iptables -t mangle -A FORWARD \<br>-p tcp --tcp-flags SYN,RST SYN \<br>-j TCPMSS --clamp-mss-to-pmtu</pre><p>Once enabled, traffic flowed normally across the overlay.</p><h3>Testing Failure</h3><p>A routing design isn’t interesting unless it survives failure.</p><p>To validate the proof of concept I started continuous traffic from a private workload.</p><pre>ping 8.8.8.8</pre><p>Then I stopped the FRRouting container on one appliance.</p><pre>docker stop frr</pre><p>Within a few seconds:</p><ul><li>the BGP session disappeared</li><li>Transit Gateway withdrew the route</li><li>traffic converged to the remaining appliance</li><li>outbound connectivity continued</li></ul><p>There were a small number of dropped packets during convergence, which is expected while BGP recalculates the path.</p><p>No route tables were modified.</p><p>No Lambda functions ran.</p><p>No manual recovery was required.</p><h3>What I Learned</h3><p>This project reinforced a few lessons.</p><p>Infrastructure automation is excellent for provisioning infrastructure.</p><p>Routing protocols are better at reacting to network failures.</p><p>Containerising network software can simplify operations more than installing directly on the host.</p><p>AWS Local Zones require different availability patterns than traditional multi-AZ deployments.</p><p>Finally, solving the right problem matters more than choosing the newest service.</p><p>I didn’t need a more complicated automation workflow.</p><p>I needed a better routing design.</p><h3>Where This Could Go Next</h3><p>This proof of concept focuses on validating the architecture.</p><p>The next steps are to:</p><ul><li>benchmark convergence times under different BGP timers</li><li>measure throughput during failover</li><li>modularise the Terraform configuration</li><li>add automated validation tests</li><li>compare the design against newer AWS networking capabilities as they become available</li></ul><p>For environments where Local Zones are required because of latency or data residency, dynamic routing provides a compelling alternative to a single NAT instance and reactive infrastructure automation.</p><h3>Source Code</h3><p>The complete Terraform configuration, FRRouting configuration, and supporting files used in this proof of concept are available on GitHub.</p><p><strong>GitHub</strong></p><p><a href="https://github.com/thelovearinze/localflow-egress">https://github.com/thelovearinze/localflow-egress</a></p><p>If you’re experimenting with AWS Local Zones or exploring alternative approaches to resilient egress, I hope you find it useful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=235967dadd22" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building Atlas: A Lightweight AWS Infrastructure Graphing Tool]]></title>
            <link>https://medium.com/@thelovearinze/building-atlas-a-lightweight-aws-infrastructure-graphing-tool-a16735f71b31?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/a16735f71b31</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[terraform]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Fri, 29 May 2026 17:52:03 GMT</pubDate>
            <atom:updated>2026-05-29T17:52:03.662Z</atom:updated>
            <content:encoded><![CDATA[<p>You start with a few AWS resources. Then the environment grows. VPCs connect to subnets. Instances get attached to security groups. Route tables, internet gateways, and infrastructure changes begin to spread across the account.</p><p>Eventually, understanding how everything connects becomes harder than creating the resources themselves. I wanted to play around with this problem and see if I could build a lightweight way to visualize these environments without the overhead of massive enterprise tools.</p><p>That exploration led to Atlas.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qhapxxE63V81y_fmSA9jkw.jpeg" /><figcaption>Atlas dashboard</figcaption></figure><p>Atlas is a Python-based project I built to map AWS infrastructure into a relationship graph. The goal was simple: discover infrastructure, understand the relationships, and visualize everything in one place just to see the underlying mechanics.</p><p>Instead of clicking through the AWS Console, I wanted a script that builds a graph representation of the environment and renders it inside an interactive dashboard. The project currently pulls:</p><ul><li>VPCs</li><li>Subnets</li><li>Route Tables</li><li>Internet Gateways</li><li>Security Groups</li><li>EC2 Instances</li></ul><p>Once discovered, the tool maps the relationships. For example, a VPC connects to a Subnet, which hosts an EC2 Instance, which uses a Security Group. This turns isolated resources into a connected dependency map.</p><p>For readers interested in the internal architecture and project structure, I added a full architecture diagram in the repository documentation:</p><p><a href="https://github.com/thelovearinze/atlas/tree/main/docs">https://github.com/thelovearinze/atlas/tree/main/docs</a></p><p>I kept the backend straightforward, relying on Python and Boto3. AWS resources are queried directly through the APIs, transformed into nodes, and exported as JSON. For the frontend, I used HTML, CSS, JavaScript, and Vis Network to render the graph visually inside the browser. It is a great way to see the live state of an account instantly without spinning up external databases.</p><p>While tinkering with the relationships, I realized I could add security awareness directly into the graph. Atlas now checks security group rules and flags resources that allow public inbound access from 0.0.0.0/0. When detected, those nodes are highlighted as high risk. This makes wide-open ingress rules immediately visible on the map rather than buried in configuration menus.</p><p>To take the experiment further, I started adding basic Terraform awareness. Atlas parses Terraform files and extracts the defined infrastructure. The idea here is to eventually compare what Terraform thinks is deployed against the live Boto3 data, laying the foundation for visual drift detection.</p><p>The current version is still lightweight, but it successfully provides infrastructure discovery, relationship mapping, security risk detection, and interactive graphing.</p><p>Some areas I want to explore next:</p><ul><li>IAM visualization</li><li>Security Hub integration</li><li>CloudTrail analysis</li><li>Attack path mapping</li><li>Multi-account AWS support</li></ul><p>Building Atlas reminded me that some of the most useful tools start from a simple curiosity:</p><p>“What actually exists inside this environment, and how is it connected?”</p><p>The repository is available here:</p><p><a href="https://github.com/thelovearinze/atlas">https://github.com/thelovearinze/atlas</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a16735f71b31" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A guide to pass AWS Advanced Networking Specialty]]></title>
            <link>https://medium.com/@thelovearinze/how-i-scored-1000-1000-in-the-aws-advanced-networking-specialty-3389bd57489b?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/3389bd57489b</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[aws-certification]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[aws-advanced-networking]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Sat, 28 Mar 2026 19:00:04 GMT</pubDate>
            <atom:updated>2026-03-31T12:27:44.227Z</atom:updated>
            <content:encoded><![CDATA[<h3>A guide to pass your AWS Advanced Networking Specialty</h3><p>and What Actually Worked!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tFdlSz3Q1T7WeBVCaayfuQ.png" /><figcaption>;🎉🥳</figcaption></figure><p>Last weekend, I sat for my <strong>first </strong>AWS certification, the AWS Certified Advanced Networking Specialty exam.</p><p>I passed with a perfect 1000/1000 on my <strong>first</strong> attempt.</p><p>That score was not a coincidence. It came from reshaping how I think about cloud networking and forcing myself to understand AWS the way AWS wants you to understand it.</p><p>Coming from an Azure background, my new role required deep AWS networking expertise. I spent the last several months rebuilding my mental models from the ground up. I did not just learn service names or memorize features. I focused on understanding how traffic moves, how routing decisions are made, how AWS isolates, scales, and fails over, and why certain design patterns consistently show up in real architectures.</p><p>This article is not a list of topics. It is the system that helped me stop guessing and start recognizing patterns.</p><h3>The Mistake People Make</h3><p>Most people study services. The exam tests decisions.</p><p>You can know exactly what a NAT Gateway does and still miss a question if you do not recognize when to use it over an alternative. The shift for me happened when I started reducing every long, complex scenario to one simple question:</p><p><strong>What is actually broken?</strong></p><p>Here are the 14 patterns and rules that helped me decode the exam.</p><h3>1. Always Identify the Real Problem</h3><p>Exam questions include a lot of background information, but only a small part of it matters.</p><ul><li>High latency from another region -&gt; network path issue</li><li>Need a single IP for firewall rules -&gt; NAT requirement</li><li>Packet loss from remote users -&gt; internet path reliability</li><li>Dynamic HTTPS application -&gt; protocol and delivery optimization</li></ul><p>If you cannot summarize the root problem in one sentence, you are not ready to look at the answers.</p><h3>2. Know When to Use CloudFront vs Global Accelerator</h3><p>This distinction appears constantly.</p><ul><li><strong>CloudFront:</strong> Use this for HTTP or HTTPS applications. It improves delivery, terminates TLS at the edge, and can cache when needed.</li><li><strong>Global Accelerator:</strong> Use this when the issue is the network path itself. This is for long distance traffic, packet loss, non-HTTP protocols like UDP or MQTT, and real-time workloads.</li></ul><p>Simple rule: Web delivery problem means CloudFront. Network path problem means Global Accelerator.</p><h3>3. Understand Traffic Direction</h3><p>Many answers depend entirely on who initiates the connection.</p><ul><li>Outbound only -&gt; NAT Gateway or Egress-Only IGW.</li><li>Bidirectional -&gt; Proper routing through Transit Gateway Appliance Mode.</li><li>Hybrid DNS -&gt; Inbound and Outbound resolver endpoints.</li></ul><p>If you ignore traffic direction, you will pick the wrong answer even if you understand the services.</p><h3>4. Routing Is How You Enforce Isolation</h3><p>Segmentation is not just about Security Groups and Network ACLs.</p><ul><li>Transit Gateway route tables control isolation between business units.</li><li>Cloud WAN policies use segments to isolate traffic globally.</li></ul><p>If you want absolute control over isolation, you control the routes.</p><h3>5. Hybrid Connectivity Is About Combining Services</h3><p>You will rarely use just one service for a complex hybrid setup.</p><ul><li>Direct Connect provides the private, dedicated path.</li><li>VPN sits on top of Direct Connect for IPsec encryption.</li><li>SiteLink allows data center to data center routing over the AWS backbone.</li></ul><p>The right answer often uses multiple services stacked together.</p><h3>6. Centralize Whenever Possible</h3><p>If the question mentions “least operational overhead,” AWS is pointing you toward a centralized architecture.</p><ul><li>Firewalls belong in a centralized inspection VPC.</li><li>DNS resolution belongs in Route 53 Resolver endpoints, not custom DNS servers on EC2.</li><li>Multi-VPC connectivity requires Transit Gateway, not a spiderweb of peerings.</li><li>Multi-account deployments should use shared subnets via RAM, not duplicate VPCs.</li></ul><p>Avoid building the same thing repeatedly when AWS provides a shared model.</p><h3>7. Always Prefer AWS-Managed Services</h3><p>If you are choosing between a managed service and building it yourself, pick the managed option.</p><ul><li>NAT Gateway over a NAT instance.</li><li>AWS Network Firewall or GuardDuty over third-party IDS on EC2.</li><li>Route 53 Resolver over a DNS proxy.</li></ul><p>The exam consistently rewards solutions that reduce operational work.</p><h3>8. Placement Matters</h3><p>Where you put things is a core part of the design.</p><ul><li>ALB listeners are shared.</li><li>Target groups are per application.</li><li>Firewalls sit in a centralized VPC.</li><li>NAT Gateways must sit in public subnets.</li></ul><p>Mixing shared and application-specific components breaks the architecture.</p><h3>9. Some Things Cannot Be Modified</h3><p>This is a very common trap. Some configurations are fixed at creation.</p><ul><li>You often cannot modify an existing live VPN to add IPv6 without disruption.</li><li>You cannot change an ALB target type freely after the fact.</li></ul><p>When you see a requirement to avoid disrupting current traffic while adding a new protocol, the answer is usually to create a new resource alongside the old one, not modify the existing one.</p><h3>10. Know What Services Do Not Support</h3><p>Some questions are solved simply by knowing hard limitations.</p><ul><li>VPC Peering does not support transitive routing.</li><li>SQS does not use interface endpoints in the same way other services do.</li><li>Route 53 does not support zone transfers natively without specific configurations.</li><li>CloudFront is not for UDP traffic.</li></ul><p>Eliminating the impossible often reveals the correct answer immediately.</p><h3>11. Cost Questions Follow Patterns</h3><p>When you see the phrases “most cost-effective” or “at scale”, look for these triggers:</p><ul><li>Avoid NAT Gateways when VPC Gateway Endpoints exist for S3 or DynamoDB.</li><li>Reuse existing Direct Connect connections with new VIFs instead of buying new physical connections.</li><li>Prefer shared and low-cost primitives.</li></ul><p>Cost questions become predictable once you understand the pricing models of data transfer.</p><h3>12 “Least Operational Overhead” Is a Clue</h3><p>When you see this phrase, rule out the following:</p><ul><li>No EC2-based custom routing solutions.</li><li>No custom third-party appliances if a native AWS tool exists.</li><li>No manual configuration per VPC via Lambda scripts.</li></ul><p>Pick the simplest AWS native design that meets the requirement.</p><h3>13. Think in Layers</h3><p>Solve the problem at the right layer of the infrastructure.</p><ul><li>Edge -&gt; CloudFront, Global Accelerator, WAF, Shield.</li><li>Regional -&gt; ALB, NLB, Transit Gateway.</li><li>Compute -&gt; EC2, EKS.</li></ul><p>If users are far away and experiencing latency, solve it at the edge. Do not try to solve it by adding more compute.</p><h3>14. Design for Scale by Default</h3><p>If the question mentions future growth or adding more VPCs later:</p><ul><li>Use Transit Gateway instead of VPC peering.</li><li>Use Resource Access Manager (RAM) instead of duplicating resources.</li><li>Use Transit Gateway Route Propagation instead of static routes.</li></ul><p>Design for tomorrow, not just today.</p><h3>My Study Approach</h3><p>Here are the resources that helped solidify my understanding.</p><p><strong>1. AWS Sample Exam Questions: </strong>Start here to test your baseline and see how AWS phrases their scenarios. Link: <a href="https://lnkd.in/g-jh24xC">https://lnkd.in/g-jh24xC</a></p><p><strong>2. AWS re:Invent Videos:</strong> Skip the marketing videos and focus on the deep technical dives for Direct Connect, hybrid connectivity, routing, and load balancing.</p><ul><li><a href="https://lnkd.in/gu8pDhQE">https://lnkd.in/gu8pDhQE</a></li><li><a href="https://lnkd.in/gDjc9P_z">https://lnkd.in/gDjc9P_z</a></li><li><a href="https://lnkd.in/gDUrgeMA">https://lnkd.in/gDUrgeMA</a></li><li><a href="https://lnkd.in/gXG9hjdd">https://lnkd.in/gXG9hjdd</a></li></ul><p><strong>3. AWS Whitepapers: </strong>These documents helped me understand the actual architecture decisions behind the services.</p><ul><li>Introduction to direct connect: <a href="https://lnkd.in/gkT5rhQv">https://lnkd.in/gkT5rhQv</a></li><li>Adding MacSec: <a href="https://lnkd.in/g-_GvH8Z">https://lnkd.in/g-_GvH8Z</a></li><li>Troubleshooting NAT: <a href="https://lnkd.in/ghXUnECM">https://lnkd.in/ghXUnECM</a></li><li>Upgrading AWS Direct Connect to 100 Gbps in 5 steps: <a href="https://lnkd.in/gZ_yMwms">https://lnkd.in/gZ_yMwms</a></li><li>Scaling VPN throughput using AWS Transit Gateway: <a href="https://lnkd.in/gBV7mhMS">https://lnkd.in/gBV7mhMS</a></li></ul><p><strong>4. Routing Loop Sessions:</strong> This YouTube channel is incredible. They provide real explanations of packet flow, not just high level slides. I watched every video and joined a few live sessions. Link: <a href="https://www.youtube.com/@theroutingloop8540">https://www.youtube.com/@theroutingloop8540</a></p><p><strong>5. Udemy Course:</strong> I used the course by Chetan Agrawal and Stéphane Maarek. The video is good, but following along with the hands-on labs matters much more than just watching. Link: <a href="https://www.udemy.com/course/aws-certified-advanced-networking-specialty-ans/">https://www.udemy.com/course/aws-certified-advanced-networking-specialty-ans/</a></p><h3>Final Thought</h3><p>This exam is about recognizing patterns and choosing the simplest design that works within the constraints provided. Once that clicked for me, everything changed. Build the labs, break the routing tables, and figure out why traffic drops. That is how you pass.</p><p>It is a <strong>difficult test</strong>, but what you learn completely changes how you build. I highly recommend it to anyone working with AWS networking. Take your time, follow these steps, and you will be ready.</p><p>Thank you for reading, and good luck!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3389bd57489b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Zero-Trust Global Service Mesh on AWS]]></title>
            <link>https://aws.plainenglish.io/zero-trust-global-service-mesh-on-aws-f2f89769bd70?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2f89769bd70</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[aws-networking]]></category>
            <category><![CDATA[aws-ec2]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Sat, 28 Feb 2026 06:44:14 GMT</pubDate>
            <atom:updated>2026-02-28T06:44:14.755Z</atom:updated>
            <content:encoded><![CDATA[<p>Bridging VPC Lattice with Transit Gateway</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FBH-QXOwEanofvw_E34jzQ.jpeg" /><figcaption>Architectural Diagram</figcaption></figure><p>Managing a globally distributed enterprise network is notoriously complex.</p><p>Between handling overlapping IP CIDR blocks, managing complex VPC peering matrices, and securing public-facing NAT Gateways, the traditional network perimeter is a security liability waiting to happen.</p><p>Modern infrastructure requires a different approach: a zero-trust architecture where services can securely discover and communicate with each other across continents, entirely isolated from the public internet.</p><p>In this walkthrough, I will break down how I engineered a globally distributed application mesh across North America (Virginia), Europe (London), and Africa (Cape Town) by bridging the Layer 7 service discovery of <strong>AWS VPC Lattice</strong> with the robust Layer 3 backbone of <strong>AWS Transit Gateway</strong>.</p><h3>The Architecture Overview</h3><p>To build this securely, the architecture was divided into three distinct layers:</p><ol><li>The Compute Layer: Amazon Linux 2023 EC2 instances deployed exclusively in private subnets across the three regions. No Internet Gateways. No NAT Gateways.</li><li>The Global Backbone (Layer 3)<strong>:</strong> Inter-region Transit Gateway peering connections linking the isolated VPCs, allowing cross-continental traffic.</li><li>The Application Mesh (Layer 7): VPC Lattice Service Networks establishing a logical application boundary, handling cross-vpc service directory and proxying.</li></ol><p>Building this with Terraform was straightforward, but the true engineering lessons came from the strict zero-trust constraints I enforced. Here are the hurdles encountered and how I solved them.</p><h3>Hurdle 1: The EC2 Instance Connect Endpoint (EICE) Trap</h3><p>Because the instances were in strictly private subnets, SSH access via a bastion host or public IP was out of the question. I opted for EC2 Instance Connect Endpoints (EICE) for secure, terminal-based administrative access.</p><p>Immediately, the Terraform deployment failed with an InstanceConnectEndpointAccessDenied error.</p><p><strong>The Fix:</strong> EICE requires AWS to build a Service-Linked Role in the background. My deployment IAM user lacked the iam:CreateServiceLinkedRole permission. Once granted, the endpoints provisioned, but connections still failed.</p><p>Why? By default, Terraform sets preserve_client_ip = true on EICE resources. This means the endpoint passes your local home Wi-Fi IP address through to the EC2 instance. Because my strict zero-trust Security Groups were configured to <em>only</em> accept traffic from the internal VPC CIDR ranges, the firewall silently dropped my connection. Setting preserve_client_ip = false forced the endpoint to mask my IP with a valid internal VPC address, instantly restoring access.</p><h3>Hurdle 2: The 77-Millisecond Timeout</h3><p>With the EICE working, I initiated a cross-region test. I connected to the London EC2 instance and attempted to curl the VPC Lattice service domain of the Virginia primary hub. The terminal hung, eventually throwing a timeout error exactly 77 milliseconds later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pBYiRVWvrGMcjPaFCVf_nA.png" /><figcaption><em>77ms timeout</em></figcaption></figure><p>In network engineering, a 77‑millisecond timeout closely mirrors the expected round‑trip latency of transatlantic fiber between London and Virginia, confirming that the Layer 3 Transit Gateway backbone was functioning as intended. So why the timeout?</p><p>Two major architectural constraints collided here:</p><ol><li><strong>Layer 7 vs. Layer 3 Routing:</strong> VPC Lattice acts as a regional proxy. Its domain names resolve to link-local IP addresses (the 169.254.x.x range). By design, link-local addresses cannot traverse a Transit Gateway.</li><li><strong>Security Group Isolation:</strong> To test the Transit Gateway, I had to route directly to the Virginia instance’s private IP (Layer 3). However, the Virginia Security Group was strictly configured to only allow Port 80 traffic from its own 172.16.0.0/16 network. The London traffic (172.17.0.0/16) reached the Virginia region, only to be blocked by the Security Group rules.</li></ol><p><strong>The Fix:</strong> I updated the Security Groups to explicitly trust the CIDR blocks of the global mesh (172.16.0.0/16, 172.17.0.0/16, and 172.18.0.0/16), allowing inter-region communication.</p><h3>Hurdle 3: Bypassing the NAT Gateway Dilemma</h3><p>With the firewalls open, the curl command still failed. The blocker turned out to be the application layer.</p><p>During instance provisioning, my user_data script was configured to run yum install -y httpd to start an Apache web server. However, because the instance was in an entirely private subnet without a NAT Gateway, it had no route to the public internet to reach the Amazon Linux package repositories. The installation failed silently in the background.</p><p>Deploying three NAT Gateways across three continents just to download a web server for a proof-of-concept is an unnecessary cloud expense.</p><p><strong>The Fix:</strong> I bypassed the need for internet access entirely. Amazon Linux 2023 comes with Python 3 pre-installed. I SSH’d into the Virginia instance and spun up Python’s built-in HTTP module:</p><pre>mkdir -p my-website<br>cd my-website<br>echo &quot;&lt;h1&gt;Hello from Primary Hub - us-east-1&lt;/h1&gt;&quot; &gt; index.html<br>sudo python3 -m http.server 80 &amp;</pre><h3>The Final Validation</h3><p>With the Python web server actively listening on Port 80 in Virginia, I switched back to the London terminal and executed the final test against the Virginia private IP:</p><pre>curl http://172.16.1.248</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Vxu2gGQWFEN2O3gKbGSHsA.png" /><figcaption><em>London terminal</em></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s_3nWL05auv4umyFaQicMQ.png" /><figcaption><em>Virginia HTTP GET request</em></figcaption></figure><p>The response was instant. A packet successfully originated in a private subnet in Europe, traversed the London Transit Gateway, crossed the transatlantic peering connection, entered the Virginia Transit Gateway, passed through the zero-trust Security Group, and hit the private Python web service.</p><h3>Conclusion</h3><p>Building a zero-trust network requires a deep understanding of how packets traverse different OSI layers. By bridging VPC Lattice for regional service discovery with Transit Gateway for global Layer 3 routing, enterprises can achieve a highly scalable, secure global footprint without exposing a single resource to the public internet.</p><p>You can view the complete Terraform infrastructure-as-code for this deployment on my GitHub: <a href="https://github.com/thelovearinze/aws-vpc-lattice-mesh">https://github.com/thelovearinze/aws-vpc-lattice-mesh</a></p><p>If you end up pulling the Terraform code from GitHub to deploy this mesh, let me know how it goes in the comments. And if you found this post useful, don’t forget to clap and share.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2f89769bd70" width="1" height="1" alt=""><hr><p><a href="https://aws.plainenglish.io/zero-trust-global-service-mesh-on-aws-f2f89769bd70">Zero-Trust Global Service Mesh on AWS</a> was originally published in <a href="https://aws.plainenglish.io/">AWS in Plain English</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Centralized Traffic Inspection Hub on AWS using Gateway Load Balancer]]></title>
            <link>https://aws.plainenglish.io/a-centralized-traffic-inspection-hub-on-aws-using-gateway-load-balancer-f6ed4ee3b56d?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6ed4ee3b56d</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[aws-networking]]></category>
            <category><![CDATA[network]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[aws]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Thu, 29 Jan 2026 20:03:55 GMT</pubDate>
            <atom:updated>2026-01-30T07:56:12.888Z</atom:updated>
            <content:encoded><![CDATA[<p>A step-by-step guide to implementing transparent “bump-in-the-wire” security with Terraform and GENEVE.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Otk_IECkaqKoqt72g_I5pA.png" /><figcaption>Architectural Diagram</figcaption></figure><h3>Introduction</h3><p>In enterprise cloud environments, balancing security with developer velocity is a constant trade-off. Security teams typically require deep packet inspection for compliance, while application teams demand speed and autonomy. The traditional approach of deploying a firewall in every single VPC is expensive and difficult to manage at scale.</p><p>A superior alternative is the Centralized Inspection Architecture. By using the AWS Gateway Load Balancer (GWLB),<strong> </strong>we can route traffic from multiple spoke VPCs through a central security hub. This creates a “bump-in-the-wire” design where traffic is inspected transparently without requiring changes to the application or complex VPN configurations.</p><p>In this guide, I will walk through how I architected and deployed this solution using Terraform. I will also detail the specific routing challenges encountered during implementation and how I resolved them to ensure a production-ready design.</p><h3>The Architecture</h3><p>The design simulates a real-world hub-and-spoke topology split into two Virtual Private Clouds (VPCs):</p><ol><li>The Security VPC (Hub): This hosts the inspection infrastructure, including the Gateway Load Balancer and the Linux-based firewall appliance.</li><li>The Consumer VPC (Spoke): This hosts the application server. It routes outbound traffic to the hub for inspection before it reaches the internet.</li></ol><h3>Step 1: Deploying the Infrastructure as Code</h3><p>I chose Terraform for deployment to ensure the environment was reproducible and modular. The first step was defining the network boundaries.</p><p>I created two distinct VPCs with non-overlapping CIDR blocks to simulate a routed network.</p><ul><li>Security VPC: 198.18.0.0/16</li><li>Spoke VPC: 198.19.0.0/16</li></ul><p>Using Terraform aws_subnet and aws_vpc resources allowed me to isolate the &quot;untrusted&quot; application zone from the &quot;trusted&quot; security zone. This separation of duties is critical for audit compliance.</p><h3>Step 2: Configuring the Gateway Load Balancer</h3><p>The core of this architecture is the GWLB. Unlike a standard Application Load Balancer that works at Layer 7, the GWLB operates at Layer 3 using the GENEVE protocol.</p><p>I defined the aws_lb resource with load_balancer_type = &quot;gateway&quot; and attached it to a Target Group.</p><p>The Implementation Challenge: During the initial deployment, the Target Group health checks failed consistently. The infrastructure existed, but the traffic was not flowing. The root cause was a missing listener configuration. The GWLB requires an explicit listener to forward traffic to the Target Group using the GENEVE protocol on port 6081.</p><p>I resolved this by adding an aws_lb_listener resource to bridge the load balancer and the target group. This experience highlighted the importance of validating Layer 3 connectivity components individually before assuming the system is operational.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7_7zmFnjECzs43AwyD6Brg.png" /><figcaption>TG is healthy and accepting geneve traffic after configuring the listener</figcaption></figure><h3>Step 3: Implementing the “Service Insertion” Routing</h3><p>The most complex part of this project was the routing logic. The goal was to force application traffic through the firewall while keeping the underlying infrastructure accessible for management.</p><p>I encountered an Asymmetric Routing issue during testing. When I routed all traffic (0.0.0.0/0) to the GWLB Endpoint, my SSH connections to the application server timed out. The request packets reached the server, but the response packets were routed into the firewall appliance which dropped them because it lacked the connection context.</p><p>To solve this, I implemented Split Routing in the Spoke VPC Route Table:</p><ol><li>Management Traffic: I routed the default path (0.0.0.0/0) to the Internet Gateway. This ensured that administrative SSH sessions remained direct and stable.</li><li>Inspection Traffic: I created a specific route for test traffic (using 8.8.8.8/32 as the destination) pointing to the Gateway Load Balancer Endpoint (vpce).</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LMwf-uxYe8vbLh-_sXggSg.png" /><figcaption>Split routing to separate management trraffic from inspected traffic</figcaption></figure><h3>Step 4: Validating with GENEVE Decapsulation</h3><p>The final step was proving that inspection was actually happening. Since the GWLB encapsulates traffic, standard logging does not always show the full picture.</p><p>I accessed the inspection appliance via SSH and used tcpdump to listen specifically on UDP port 6081.</p><pre>sudo tcpdump -nni ens5 udp port 6081</pre><p>By generating traffic from the Spoke VPC (ping 8.8.8.8), I captured the packets arriving at the firewall. The logs confirmed that the traffic was wrapped in GENEVE headers, processed by the appliance, and successfully inspected.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EIcvSgVX9MAU8HtBccEeAA.png" /><figcaption>Real-time firewall packet capture confirming receipt of GENEVE traffic from the application server</figcaption></figure><h3>Business Impact and Conclusion</h3><p>Implementing this architecture provides three clear benefits to an organization:</p><ol><li>Cost Reduction: It removes the need to license and run third-party firewalls in every VPC.</li><li>Operational Efficiency: Security rules are managed in one central place rather than fragmented across hundreds of accounts.</li><li>Compliance: It guarantees that all egress traffic passes through a verifiable inspection point.</li></ol><p>This project demonstrated that while the Gateway Load Balancer abstracts much of the complexity of network appliances, a deep understanding of VPC routing and packet flow is still required to build a resilient system.</p><h3>Try It Yourself</h3><p>The best way to understand the GENEVE protocol and VPC routing is to see it in action.</p><p>I have open-sourced the complete, production-ready Terraform configuration for this lab. You can clone the repository, run terraform apply, and inspect the packets yourself in under 10 minutes.</p><p><a href="https://github.com/thelovearinze/aws-gwlb-centralized-inspection"><strong>GitHub Repository: aws-gwlb-centralized-inspection</strong></a></p><p>Feel free to fork the code, experiment with the routing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6ed4ee3b56d" width="1" height="1" alt=""><hr><p><a href="https://aws.plainenglish.io/a-centralized-traffic-inspection-hub-on-aws-using-gateway-load-balancer-f6ed4ee3b56d">A Centralized Traffic Inspection Hub on AWS using Gateway Load Balancer</a> was originally published in <a href="https://aws.plainenglish.io/">AWS in Plain English</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cross Account PrivateLink with Terraform]]></title>
            <link>https://aws.plainenglish.io/cross-account-privatelink-with-terraform-bdab4c95d6d4?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/bdab4c95d6d4</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[cloud]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Mon, 15 Dec 2025 06:29:11 GMT</pubDate>
            <atom:updated>2025-12-15T06:29:11.206Z</atom:updated>
            <content:encoded><![CDATA[<p>A step-by-step guide to building cross account AWS PrivateLink with Terraform.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UiIxYGa0MdEJ3UQnFwlZaw.png" /><figcaption>Architectural diagram</figcaption></figure><p>This project walks through a complete, reproducible, cross-account AWS PrivateLink setup built entirely with Terraform.</p><p>The goal was simple. Run an application privately in one AWS account and consume it from another, without internet gateways, NAT gateways, public subnets, or exposed endpoints.</p><p>Everything stays private.<br>Traffic never leaves the AWS backbone.<br>And both accounts remain fully isolated except for what is intentionally shared.</p><p>This article documents the exact architecture, folder structure, terraform resources, and validation steps used to implement the design.</p><h3>1. What This Deployment Achieves</h3><p>By the end of this setup:</p><h3>Provider Account</h3><ul><li>Hosts a backend EC2 instance in private subnets.</li><li>Exposes that application through an internal Network Load Balancer.</li><li>Publishes an Endpoint Service with acceptance required.</li><li>Approves consumer connections manually.</li></ul><h3>Consumer Account</h3><ul><li>Creates an Interface VPC Endpoint that connects to the provider service.</li><li>Runs all workloads privately, using only VPC-local DNS and routing.</li><li>Uses SSM, SSMMessages, and EC2Messages VPC endpoints for private instance management.</li><li>Deploys a private EC2 instance with no public IP or NAT access.</li></ul><h3>End Result</h3><p>A curl request executed inside the consumer instance flows:</p><pre>client → consumer VPCE → AWS PrivateLink → provider NLB → provider EC2</pre><p>No internet. No NAT. No public routes.<br>A clean, secure application sharing pattern between two AWS accounts.</p><h3>2. Repository Structure</h3><pre>privatelink/<br>├── provider/<br>│   ├── versions.tf<br>│   ├── provider.tf<br>│   ├── vpc.tf<br>│   ├── app.tf<br>│   ├── nlb.tf<br>│   ├── endpoint_service.tf<br>│<br>└── consumer/<br>    ├── versions.tf<br>    ├── provider.tf<br>    ├── vpc.tf<br>    ├── endpoint.tf<br>    ├── iam.tf<br>    ├── ssm_endpoints.tf<br>    ├── client.tf</pre><p>The project is intentionally modular.<br>Both accounts can be deployed independently.</p><h3>3. Provider Account: Build the Application and Service</h3><h3>VPC and Subnets</h3><p>Private subnets only, DNS enabled.</p><h3>Backend Application</h3><p>A small Amazon Linux EC2 instance runs a simple web page served via Python’s built-in HTTP server.</p><h3>Internal Network Load Balancer</h3><p>The EC2 instance registers into a TCP target group behind the internal NLB.</p><h3>Endpoint Service</h3><p>The NLB is published as a PrivateLink Endpoint Service with manual acceptance enabled.</p><p>Terraform output exposes:</p><ul><li>Service ID</li><li>Service Name</li></ul><p>These values are passed to the consumer.</p><h3>4. Consumer Account: Build the Private Endpoint Path</h3><h3>VPC and Private Subnets</h3><p>Two private subnets spread across AZs.</p><h3>Interface VPC Endpoint</h3><p>The consumer connects to the provider using the provider’s service_name.</p><p>A security group allows VPC-local 80 and 443 traffic.</p><h3>Systems Manager Access (Private Only)</h3><p>The consumer VPC gets:</p><ul><li>com.amazonaws.eu-west-1.ssm</li><li>com.amazonaws.eu-west-1.ssmmessages</li><li>com.amazonaws.eu-west-1.ec2messages</li></ul><p>All as Interface Endpoints<br>An IAM instance profile attaches the required</p><p>AmazonSSMManagedInstanceCore policy</p><h3>Private Test EC2 Instance</h3><p>A t3.micro Amazon Linux instance runs without:</p><ul><li>Public IP</li><li>NAT Gateway</li><li>IGW</li></ul><p>All management happens through SSM.</p><h3>5. Connection Approval</h3><p>Once the consumer VPC Endpoint is created, the provider account receives a pending connection.</p><p>After accepting it, the endpoint becomes available on both sides.</p><h3>6. Validation</h3><p>Validation was done using AWS Systems Manager:</p><h3>Check SSM Registration</h3><pre>aws ssm describe-instance-information --query &quot;InstanceInformationList[?InstanceId==&#39;&lt;id&gt;&#39;]</pre><h3>Send a Test Command</h3><pre>aws ssm send-command --document-name &quot;AWS-RunShellScript&quot; --parameters commands=&quot;echo hello</pre><h3>Test PrivateLink Connectivity</h3><p>Inside the instance:</p><ul><li>curl the Endpoint DNS</li><li>confirm the HTML response from provider app</li><li>ensure no internet path is used</li></ul><p>Example output:</p><pre>&lt;html&gt;<br>  &lt;body&gt;<br>    &lt;h1&gt;Demo: Provider EC2 behind NLB&lt;/h1&gt;<br>  &lt;/body&gt;<br>&lt;/html&gt;</pre><p>This confirms the request reached the provider through PrivateLink.</p><h3>7. What This Architecture Proves</h3><p>This deployment demonstrates:</p><ul><li>Cross-account application consumption with zero public exposure</li><li>VPC-to-VPC communication using AWS PrivateLink only</li><li>Private operations using SSM instead of NAT instances or NAT gateways</li><li>Clean separation between provider and consumer accounts</li><li>Full reproducibility using Terraform</li></ul><p>It is a real, production-aligned example of how organizations can share services privately between accounts or even between separate AWS environments.</p><h3>8. Final Notes</h3><p>This pattern works well for:</p><ul><li>SaaS providers</li><li>Internal platform teams</li><li>Regulated or high security environments</li><li>Sensitive backend systems</li><li>Any cross-account integration where exposure must be minimized</li></ul><p>PrivateLink remains one of AWS’s strongest primitives for secure, tightly scoped service sharing.</p><p>Leave a comment if you’d like access to the Github repo, and I’ll publish it once there’s enough demand.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bdab4c95d6d4" width="1" height="1" alt=""><hr><p><a href="https://aws.plainenglish.io/cross-account-privatelink-with-terraform-bdab4c95d6d4">Cross Account PrivateLink with Terraform</a> was originally published in <a href="https://aws.plainenglish.io/">AWS in Plain English</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Set Up a NAT Gateway in AWS (Step-by-Step for Beginners)]]></title>
            <link>https://medium.com/@thelovearinze/how-to-set-up-a-nat-gateway-in-aws-step-by-step-for-beginners-dd0da5e1ea4e?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd0da5e1ea4e</guid>
            <category><![CDATA[network]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[cloud-networking]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[aws]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Thu, 23 Oct 2025 02:02:42 GMT</pubDate>
            <atom:updated>2025-10-23T08:18:59.289Z</atom:updated>
            <content:encoded><![CDATA[<p>A quick, hands-on guide to setting up and testing a NAT Gateway in AWS.</p><p>Have you ever wondered how private EC2 instances in AWS get internet access without having public IPs?</p><p>That’s where NAT Gateways come in and, in this guide, I’ll walk through the <em>entire process</em> of setting one up, testing it, breaking it (on purpose!), and then fixing it again.</p><p>This tutorial is based on a real demo I ran for our intern training session.<br>It’s written in simple, practical language perfect for anyone who wants to understand how private subnets communicate with the internet safely.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jCZ8NMdHw8e5UGgvI53qiQ.png" /><figcaption>Setup diagram</figcaption></figure><h3>The Goal</h3><p>We’ll create:</p><ul><li>A <strong>VPC -</strong>our isolated network environment in AWS.</li><li>An <strong>Internet Gateway (IGW)</strong> -the bridge between our VPC and the internet.</li><li>A <strong>Bastion Host (Public EC2) -</strong>for secure SSH access.</li><li>An <strong>App Instance (Private EC2) -</strong>without a public IP.</li><li>A <strong>NAT Gateway -</strong>to give the private instance outbound internet access.</li></ul><p>Everything will live inside one <strong>VPC</strong> with <strong>public and private subnets</strong>.</p><h3>The VPC Setup</h3><p>Every network starts with one thing: a <strong>VPC</strong>.<br>Think of it as your own private data center inside AWS.</p><ul><li><strong>VPC name:</strong> intern-nat-lab</li><li><strong>CIDR:</strong> 10.10.0.0/16</li><li><strong>Region:</strong> eu-west-1</li></ul><p>This gives us a foundation. Now, we’ll carve it up into subnets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2xfJ7ReVR8zla42cSls9fQ.png" /><figcaption>VPC overview screen</figcaption></figure><h3>2. Splitting the Space: Public and Private Subnets</h3><p>We’ll need:</p><ul><li><strong>Public subnet</strong> (for bastion host): 10.10.1.0/24</li><li><strong>Private subnet</strong> (for internal app): 10.10.2.0/24</li></ul><p>The <strong>public subnet</strong> connects to the internet via an Internet Gateway.<br> The <strong>private subnet</strong> will reach the internet indirectly, through a NAT Gateway.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PrJzWQJKdMXN-dB6ugWw5A.png" /><figcaption>Subnets table</figcaption></figure><h3>3. Internet Gateway and NAT Gateway</h3><h3>Internet Gateway</h3><p>Attached to our VPC.<br>It’s the bridge that lets the public subnet talk to the internet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PLraor5E_MFoGuyVqRczIw.png" /><figcaption>Internet gateway</figcaption></figure><h3>NAT Gateway</h3><p>Lives in the public subnet, but its purpose is to help <strong>private instances</strong> access the internet securely.</p><p>Created in:</p><ul><li><strong>Subnet:</strong> intern-public-subnet</li><li><strong>Elastic IP:</strong> auto-assigned by AWS -pick from the dropdown</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JEggJKv4LaabMbuh-7dTXg.png" /><figcaption>NAT gateway</figcaption></figure><h3>4. Security Groups</h3><p>We created two:</p><p>4.1 Bastion Security Group (intern-bastion-sg)</p><ul><li><strong>Inbound:</strong> SSH (22) from <em>my public IP</em> → 101.xx/32</li><li><strong>Outbound:</strong> All traffic (default)</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rGd-rfdbKz947dhcrh-nSA.png" /><figcaption>Bastion SG inbound rule</figcaption></figure><p>This ensures only my laptop can SSH into the bastion.</p><p>4.2 App Security Group (intern-app-sg)</p><ul><li><strong>Inbound:</strong> SSH (22) <em>only</em> from intern-bastion-sg</li><li><strong>Outbound:</strong> All traffic</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G5ndWfhEfKrrhl55CTyFTA.png" /><figcaption>App sg inbound rule</figcaption></figure><p>This enforces internal-only SSH access no direct public connections.</p><h3>5. Launching the Instances</h3><h3>Bastion Host</h3><ul><li><strong>Subnet:</strong> public</li><li><strong>Security Group:</strong> intern-bastion-sg</li><li><strong>Key Pair:</strong> int-keypair.pem</li><li><strong>Auto-assign public IP:</strong> Enabled</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Uz7Ekp-djn8Xy3jEl2sGVQ.png" /><figcaption>Bastion instance</figcaption></figure><h3>App Instance</h3><ul><li><strong>Subnet:</strong> private</li><li><strong>Security Group:</strong> intern-app-sg</li><li><strong>Auto-assign public IP:</strong> Disabled</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RIg3JJvdL7s-pCCWEsznUQ.png" /><figcaption>Private app instance</figcaption></figure><h3>6. Route Tables — Defining Traffic Flow</h3><p><strong>Public Route Table</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*II7tuXVcrSI4pLb8-GQLvg.png" /></figure><p><strong>Private Route Table</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_8yKdbyV1ON8K21euMwIEg.png" /></figure><h3>7. Testing from the Bastion</h3><p>We first SSH into the bastion using our key:</p><pre>ssh -i &quot;int-keypair.pem.pem&quot; ec2-user@18.202.79.252</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hSr50j0MUA3zd5hctpQj8w.png" /><figcaption>SSH success session</figcaption></figure><p>Then SSH into the private app instance:</p><pre>ssh -i ~/.ssh/int-keypair.pem ec2-user@10.10.2.137</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/950/1*sDL8x-G037JXH8ejiDXAiQ.png" /><figcaption>SSH success session</figcaption></figure><h3>8. Validating NAT Gateway Traffic</h3><p>Inside the private instance:</p><pre>curl -s https://api.ipify.org</pre><p>Output:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LUCb-o67C8qygRyle3pNXg.png" /></figure><p>That’s the NAT Gateway’s public IP, proof the traffic is leaving securely.</p><p>Next, remove the NAT route and re-test:</p><pre>curl -sS --max-time 5 https://api.ipify.org || echo &quot;no internet (expected)&quot;</pre><p>Result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v-HqIBnbbblIvtnKu3N-zw.png" /></figure><p>Re-add NAT route, test again:</p><p>Result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7MsQk8hvOZv4iRoqV_wemg.png" /></figure><h3>9. What We Just Proved</h3><ul><li>Private instances can access the internet <strong>only through NAT</strong></li><li>Bastion host provides controlled SSH entry</li><li>Route tables enforce isolation and routing</li><li>No public IPs on private subnets</li><li>NAT egress path verified end-to-end</li></ul><h3>10. Wrap-Up and Tips</h3><ul><li>Always limit SSH to known IPs.</li><li>Use tags for everything (Project: intern-lab)</li><li>Avoid overlapping CIDRs across labs.</li><li>Delete idle NAT gateways when done — they’re billed hourly.</li></ul><h3>Final Thoughts</h3><p>This simple setup shows how cloud networking works in layers;security, routing, and design principles all tied together.</p><p>It’s a small lab, but it mirrors how real production networks are built.</p><p>Have questions or thoughts? Share them in the comments!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd0da5e1ea4e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When a Quiet Little Switch Took Down a Big Service]]></title>
            <link>https://medium.com/@thelovearinze/when-a-quiet-little-switch-took-down-a-big-service-f86aaf60077e?source=rss-540567ff3019------2</link>
            <guid isPermaLink="false">https://medium.com/p/f86aaf60077e</guid>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[data-center]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[teamwork]]></category>
            <category><![CDATA[network-engineering]]></category>
            <dc:creator><![CDATA[Love Arinze]]></dc:creator>
            <pubDate>Mon, 18 Aug 2025 11:17:08 GMT</pubDate>
            <atom:updated>2025-08-18T11:17:08.230Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/666/1*-LmHoSo8wSBl36qLTwXY8w.png" /><figcaption>An aging switch, a long night</figcaption></figure><p>It was Thursday evening. I was ready to shut my laptop when my phone buzzed: USSD service is down and customers cannot access their accounts. Dinner turned into a Microsoft Teams call that ran from six in the evening to half past ten, with logs, dashboards, and increasingly anxious messages.</p><p><strong>The scene you could not script</strong><br>Engineers joined in from desks and from home, kids in the background. Managers popped in with the inevitable question: Any update? Alerts pinged nonstop, each one saying the same thing in slightly different words. The mood was tense.</p><p>One teammate sighed, not tonight. Another tried to joke and it landed flat. Nobody laughs when banking services are offline.</p><p><strong>The first night</strong><br>We traced the failure to one access switch, old and quiet at the edge of the rack. The logs told it plainly: ERR-DISABLE: LSGroup error detected on port...</p><p>The switch was shutting down ports carrying the banking traffic. Picture a traffic cop who panics and closes the intersection at the first sign of trouble. That is what the switch did.</p><p>We ran every safe command we could think of. Nothing held. By ten thirty the service was still unstable. We wrapped the call tired and frustrated, knowing we would regroup in the morning.</p><p><strong>The Friday morning push</strong><br>At eight sharp the team was back on Teams, coffee in hand and set on a plan. We took a bolder route. We booted the switch into recovery mode, let it start with a bare config, then manually reloaded the saved configuration. One by one the ports came back online.</p><p>A ping succeeded. The USSD service returned. Messages of relief filled the chat. The engineer whose joke had failed the night before let out a quiet laugh.</p><p><strong>The uneasy aftermath</strong><br> A reboot had fixed the immediate outage but not the cause. The logs still showed errors and the switch was more than a decade old, long past its vendor support window.</p><blockquote><em>“A complex system that works is invariably found to have evolved from a simple system that worked.” — John Gall</em></blockquote><p>Our simple system had evolved into a brittle one through years of patches and quick fixes.</p><p><strong>What we learned</strong><br> The likely causes were mundane and stacked: aging optics and fibers, confusing uplink bundles from ESXi hosts, and firmware that had not been updated in years.</p><blockquote><em>“An ounce of prevention is worth a pound of cure.” — Benjamin Franklin</em></blockquote><p>The larger lesson is straightforward. Relying on obsolete hardware for critical services is a bet with downtime as the possible payoff.</p><p><strong>People mattered as much as the commands</strong><br>What stood out was how the team behaved under pressure. The sysadmin who kept calm and pointed to backup paths.</p><p>The manager who did not understand every detail but trusted the engineers. The quiet lift in morale when the service came back. Outages expose weak systems and reveal strong teams.</p><p><strong>Takeaways</strong></p><ol><li>Old hardware is a ticking problem. Still working does not mean reliable.</li><li>Safety features like err disable protect networks but can keep slamming the brakes if the underlying issue is not fixed.</li><li>Clear, calm communication matters as much as the technical response.</li><li>Investing in routine replacement and testing costs less than scrambling for twelve hours while customer trust erodes.</li></ol><p><strong>Reflection</strong><br>From Thursday evening to Friday morning we spent nearly twelve hours on this incident. The service came back, but the lesson stuck: resilience is not only in cloud code. It lives in the dusty boxes in the rack too.</p><p>If you have a story where an old piece of hardware surprised you, share it in the comments. I would love to hear how your team handled it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f86aaf60077e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>