<?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 Adam Dunstan on Medium]]></title>
        <description><![CDATA[Stories by Adam Dunstan on Medium]]></description>
        <link>https://medium.com/@oz-adam?source=rss-f8c5a88f31d4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*kMbkGjR2n2uJJ3tnax4qKA.jpeg</url>
            <title>Stories by Adam Dunstan on Medium</title>
            <link>https://medium.com/@oz-adam?source=rss-f8c5a88f31d4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 03:14:24 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@oz-adam/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[Are Cloud & AI companies the next Telecoms]]></title>
            <link>https://oz-adam.medium.com/are-cloud-ai-companies-the-next-telecoms-08585221cccd?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/08585221cccd</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[azure]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Thu, 13 Aug 2026 13:08:46 GMT</pubDate>
            <atom:updated>2026-08-13T13:08:46.031Z</atom:updated>
            <content:encoded><![CDATA[<p>I constantly remind people that Telecom companies are Construction companies not Technology companies. The financial structure and their use of Debt financial supports this assertion. However, this wasn&#39;t always the case. During the introduction of the Internet, Telecom companies participated heavily in Technology development through their involvement in Vendors, Standards Organizations and their own labs. As the technology developed to the deployment stage, construction of the infrastructure took over causing the organizations structure to change.</p><p>During the Cloud adoption wave, infrastructure was partly deb-financed but pre-AI that debt was self-liquidating, asset-matched, and comfortably covered by operating cash flow that was rapidly growing. We also saw rapid introduction of of new Cloud products. Statistics on new product introduction are hard to figure out, but it appears that new production introduction (not feature addition) peaked in 2022. In fact retirement is now a real countervailing force. AWS has moved a couple dozen services to maintenance or sunset since late 2024. For the first decade the portfolio only grew. This could be taken as an early signal that the structure of these businesses is changing. There have been meaningful layoffs in the Cloud business, while growth continues further suggesting that infrastructure instead of tech development is becoming the focus. Some blame it on AI coding and part of that is probably true, but that’s also where the debt starts to accelerate.</p><p>During 2026, the start of the AI boom, the structure of financing change with massive debt issuance. Whats changed is the capex is outrunning cash generation for the first time, debt financing has become a necessity not a financial convenience. In many cases these debt issuance have long duration.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*viem5z5xHbNWDFPzEVk9ww.png" /><figcaption>This is an indication, structure and reporting of participating companies makes it difficult to generate accurate numbers, they are realistically much larger.</figcaption></figure><p>While I don’t see the technology development around AI slowing in the near term. Looking out 10+ years, when most organizations and people have adopted all of the AI they need, just as we have with Broadband, will these companies start to look like Telecom companies managing infrastructure and debt with little new technology development. Simple will the AI and Cloud providers become construction companies?</p><p>I’m not downplaying the contribution that AI and its infrastructure can have to economic growth. I am most certainly not in favor of slowing AI development or Data Center construction. However I do look back to the “build it and they will come” days of Broadband and see similarities beginning in 2026.</p><p>I’m sure there are some who will say its different this time and they may be right…..</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=08585221cccd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The New PureLB]]></title>
            <link>https://medium.com/thermokline/the-new-purelb-56bf82d876ae?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/56bf82d876ae</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[linux]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Mon, 18 May 2026 17:14:30 GMT</pubDate>
            <atom:updated>2026-05-18T17:14:49.287Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/208/1*H9Y-Z2smGRO8s3ocsiUi_Q.png" /></figure><h3>The best k8s Service Load Balancer</h3><p>We have released a major upgrade to our Kubernetes <a href="https://purelb.io/">Service Load Balancer</a>, offering new and improved functionality, and the most stable release yet. Key updates include:</p><ul><li><strong>Subnet-Aware Address Election:</strong> Clusters can use local addresses across multiple subnets.</li><li><strong>BGP:</strong> Features a goBGP sidecar with route import/export and a CR providing access to all of gobgp’s capabilities.</li><li><strong>Krew plugin:</strong> A new CLI with improved visibility and a dashboard command.</li><li><strong>Metrics:</strong> New metrics for both PureLB and goBGP.</li><li><strong>CNI behavior:</strong> Address timers for local addresses enable DHCP-only nodes.</li><li><strong>Stability:</strong> Extensive automated testing scripts and numerous bug fixes, including day-zero race conditions.</li></ul><p>We made significant changes to <a href="https://purelb.io/">PureLB</a>, replacing Hashicorp Memberlist with k8s native leases enabling allocation of local addresses for clusters across multiple subnets. We created a fork of GoBGP (now <em>gobgp-netlink</em>) to add route redistribution which the original maintainers had no interest in implementing, and support for IPv6 BGP neighbors. The <em>k8gobgp</em> controller configures the complete functionality of <em>gobgp-netlink</em> using its native grpc interface. While k8gobgp and gobgp-netlink are bundled with PureLB, they can be used independently and are available in our <a href="https://github.com/purelb">GitHub</a> repo.</p><p><a href="https://purelb.io/">PureLB’</a>s design philosophy is to use the native networking capabilities of Linux (via <em>netlink</em>) to add load balancer addresses, rather than working around them. We originally forked from MetalLB because its maintainers resisted adopting this Linux native solution. Today, little of the original MetalLB code remains as we have completely changed election, address pool management, and how addresses are managed.</p><p>There is much more functionality that we plan to add to the <a href="https://purelb.io/">PureLB</a> core code as well as continue to add plugins in the form of additional sidecars. The on-premise community deserves an actively developed, Linux native, Service LoadBalancer alternative.</p><p>The new release (1.6.x) of PureLB, k8gobgp, and gobgp-netlink are hosted on <a href="https://github.com/purelb">GitHub</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=56bf82d876ae" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thermokline/the-new-purelb-56bf82d876ae">The New PureLB</a> was originally published in <a href="https://medium.com/thermokline">ThermoKline</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI to fork GoBGP for Netlink redistribution]]></title>
            <link>https://oz-adam.medium.com/ai-to-fork-gobgp-for-netlink-redistribution-854fc81c175d?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/854fc81c175d</guid>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[kubernetes]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Wed, 03 Dec 2025 15:03:45 GMT</pubDate>
            <atom:updated>2025-12-03T15:03:45.873Z</atom:updated>
            <content:encoded><![CDATA[<h3>Forking GoBGP with AI: Adding Netlink Redistribution Support</h3><p>This isn’t a tutorial — just a short piece describing my experience using AI to fork GoBGP and add Netlink redistribution support.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ocxq0ZlgBhYXsnAXsCoxMQ.png" /></figure><h3>Background</h3><p>We’re upgrading our Kubernetes Service Load Balancer, PureLB, to add native BGP support. We currently use a simple mechanism native to Linux to advertise allocated addresses. While BGP support was possible, it wasn’t well integrated with Kubernetes.</p><p>There are three BGP implementation choices: FRR, BIRD, and GoBGP. For Kubernetes integration, a gRPC interface is essential. BIRD lacks one entirely, and FRR’s implementation is new and incomplete. However, both BIRD and FRR support redistribution — GoBGP does not. Another point in GoBGP’s favor: FRR and BIRD include many routing protocols designed for switches and routers, while cloud systems like Kubernetes only need BGP.</p><p>Redistribution moves routes between protocols — in routers, between BGP and an IGP like OSPF; on Linux hosts, between routing protocols and the kernel routing table. This is what we needed to add. We’d previously used a Linux Netlink library that could move routes between Linux and GoBGP routing tables.</p><h3>Objectives</h3><ul><li>Import routes added to Linux host interfaces into GoBGP’s routing table</li><li>Export routes from GoBGP into the Linux routing table</li><li>Filter exports by BGP communities for route control</li><li>Support virtual routing (VRF) for both import and export</li><li>Support IPv4 and IPv6</li></ul><h3>Import</h3><p>I started with Gemini 2.5, creating a GEMINI.md to track progress. Initially things went well — within a few days I had basic IPv4 import working. IPv6 proved trickier. The new Netlink code couldn’t populate the correct next-hop for IPv6, and Gemini couldn’t figure out why.</p><p>Manual debugging revealed the issue: the original GoBGP authors had taken a shortcut, reusing stored BGP peer addresses for next-hops. This worked for IPv4 but left IPv6 BGP updates without next-hop addresses. When I explained this to Gemini, its attempted fixes weren’t successful. The solution was to get next-hop addresses from the host’s exit interface — three addresses total: IPv4, IPv6 global, and IPv6 link-local — then use those when creating BGP updates.</p><p>This took a long time. I frequently had to debug and fix code myself. While debugging, I noticed Gemini had reused standard library names with aliases scattered throughout. When I asked it to refactor, it kept running out of tokens. (My request for increased tokens took over a month to approve.) Eventually I got it refactored and the next-hop processing working.</p><h3>Export</h3><p>By this point I’d had enough of Gemini and switched to Claude. The experience was far superior. Export development didn’t take long and generally worked the first time — the logic is significantly simpler than import.</p><h3>VRF Support</h3><p>I continued with Claude for VRF support on both import and export. This wasn’t simple. Claude would routinely add non-working logic and need multiple attempts to fix its mistakes. Code deemed complete would fail testing. As with Gemini, small, deliberate, supervised changes ultimately produced working code.</p><h3>Gemini 2.5</h3><p>Gemini made many mistakes. The most frustrating: attempting to modify code without reading the whole file. Despite using MCPs, it struggled to comprehend the complete codebase, partly due to token limitations. Hallucination was significant — describing changes it hadn’t made, getting BGP operation wrong on multiple occasions.</p><p>My favorite failure: no matter how many times it tried (with my help), it couldn’t compile the protobuf definitions. Even after I told it exactly how, it would fail again on subsequent attempts.</p><p>Finally, it was expensive. Gemini charges per token, ultimately costing many hundreds of dollars. Especially annoying were days when I’d spend three hours making no progress while Gemini apologized for its mistakes.</p><h3>Claude</h3><p>Switching to Claude was a significant improvement. While it still made mistakes, they weren’t repetitive like Gemini’s — once corrected, Claude followed instructions. I didn’t use any MCPs to help Claude understand the codebase structure, yet it performed much better.</p><p>Claude’s workflow suggests planning first, then implementation. Carefully reviewing generated plans results in fewer coding errors, ultimately saving time. Unlike Gemini, hallucination was rare. Another motivation for switching: Anthropic’s fixed-price plans proved much cheaper than Gemini.</p><h3>Key Learnings</h3><p>Issues common to both models:</p><p><strong>Don’t use automatic edits.</strong> All the talk about managing lots of agents doesn’t apply here. Allowing the agent to edit without review will likely create a significant mess that takes longer to untangle than reviewing each proposed change would have.</p><p><strong>They’re supremely confident.</strong> Both models declare success with great confidence. Independent testing is essential. Create unit tests, CI tests, and do manual testing.</p><p><strong>They forget the plan.</strong> Whether using GEMINI.md or planning mode, both consistently forgot steps — particularly updating the gRPC API and CLI.</p><p><strong>They make bad decisions.</strong> When moving the forked code to GitHub, I updated CI to add Netlink functionality, but somewhere along the line we broke previously working functionality. Instead of investigating, the models said the errors were unrelated to their code, so they didn’t need to look. The CI tests worked before the new code — it was clearly a problem. Another argument for not using automatic edits.</p><p><strong>Cheaper isn’t better.</strong> Pricing is unfortunately opaque. Gemini uses per-token pricing; Claude offers fixed monthly fees with caps. I used Claude’s lowest plan and never exceeded caps. Gemini’s costs were high by comparison, exacerbated by the model deciding token usage — you pay for its mistakes. This was the most frustrating part of using Gemini. On cost/outcome analysis, Claude was clearly better: better outcomes, lower cost. Fixed subscription pricing incentivizes providers to deliver better outcomes; per-token pricing does not.</p><h3>Conclusion</h3><p>It was faster. I estimate these AI tools reduced development time by 66–75%. I doubt I could have completed this with Gemini 2.5 alone; starting with Claude would have made the project much easier. That said, Gemini 3.0 might be a different story.</p><p><strong>Repository:</strong> <a href="https://github.com/purelb/gobgp-netlink">github.com/purelb/gobgp-netlink</a></p><p><em>I wrote the original document, Claude did a review to make it more concise and readable. Nano Banana created the image.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=854fc81c175d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[EVPN Route Reflector with SONiC using frr-mgmt-framework]]></title>
            <link>https://medium.com/sonic-nos/evpn-route-reflector-with-sonic-using-frr-mgmt-framework-db6d12b85ce7?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/db6d12b85ce7</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[routing]]></category>
            <category><![CDATA[networking]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Fri, 10 May 2024 15:01:18 GMT</pubDate>
            <atom:updated>2024-05-10T15:01:18.658Z</atom:updated>
            <content:encoded><![CDATA[<p>SONiC configuration can be confusing, using the default configuration mechanism its not possible to configure BGP EVPN. A search for tutorials will on cover two problems.</p><ol><li>Split-model configuration. Split-mode configuration is a useful workaround in the lab, using multiple configuration mechanisms and their associated files are a mess in an automated production environment. <a href="https://medium.com/sonic-nos/sonic-dont-use-split-mode-use-frr-mgmt-framework-a67ad76ec1a6">(Learn more about configuration modes)</a></li><li>No Route Reflector. A common topology is to use a route reflector to distribute L2 EVPN AFI prefixes.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*4zze_FFRDQ3mjpRC0Y40BQ.png" /></figure><p>This tutorial provides provides a real world foundation for deploying SONiC for BGP EVPN in production. It provides a sample BGP Route Reflector configuration and uses the option <em>frr-mgmt-framework </em>enabling the SONiC configuration to be maintained in a single file. <em>frr-mgmt-framework </em>has much broader configuration coverage for FRR features.</p><p>This tutorial was developed using version 202311</p><p><em>Note. Starting with a simple default configuration is often useful. Using sonic-cfggen for your switch type with the option — preset l1 will provide good starting point.</em></p><h4>1. Enable frr-mgmt-framework on all switches</h4><pre>{<br>    &quot;DEVICE_METADATA&quot;: {<br>        &quot;localhost&quot;: {<br>            &quot;buffer_model&quot;: &quot;traditional&quot;,<br>            &quot;default_bgp_status&quot;: &quot;up&quot;,<br>            &quot;default_pfcwd_status&quot;: &quot;disable&quot;,<br>            &quot;frr_mgmt_framework_config&quot;: &quot;true&quot;,  <br>            &quot;hostname&quot;: &quot;sonic3-vs&quot;,<br>            &quot;hwsku&quot;: &quot;Force10-S6000&quot;,<br>            &quot;mac&quot;: &quot;52:54:00:19:82:2f&quot;,<br>            &quot;platform&quot;: &quot;x86_64-kvm_x86_64-r0&quot;,<br>            &quot;timezone&quot;: &quot;UTC&quot;<br>        }<br>    }<br>}</pre><p>Edit /etc/sonic/config_db to set “frr_mgmt_framework-config”: “true” in DEVICE_METADATA</p><h4>2. Configure IP Interfaces</h4><p>Configure the IP addresses on all three switches using either the CLI, by creating configuration files (snippets) and load them using <em>config load, </em>or by editing and reloading /etc/sonic/config_db.json</p><pre>root@sonic1-vs:~# config loopback add Loopback0<br>root@sonic1-vs:~# config interface ip add Loopback0 10.1.0.1/32<br>root@sonic1-vs:~# config interface ip add Ethernet4 10.0.0.8/31<br>root@sonic1-vs:~# config interface ip add Ethernet8 10.0.0.10/31<br>root@sonic1-vs:~# config save</pre><pre>{<br> &quot;LOOPBACK_INTERFACE&quot;: {<br>        &quot;Loopback0&quot;: {},<br>        &quot;Loopback0|10.1.0.1/32&quot;: {}<br>    },<br> &quot;INTERFACE&quot;: {<br>        &quot;Ethernet4&quot;: {},<br>        &quot;Ethernet4|10.0.0.8/31&quot;: {},<br>        &quot;Ethernet8&quot;: {},<br>        &quot;Ethernet8|10.0.0.10/31&quot;: {}<br>    }<br>}</pre><pre>root@sonic2-vs:~# config loopback add Loopback0<br>root@sonic2-vs:~# config interface ip add Loopback0 10.1.0.2/32<br>root@sonic2-vs:~# config interface ip add Ethernet4 10.0.0.9/31<br>root@sonic1-vs:~# config save</pre><pre>{<br> &quot;LOOPBACK_INTERFACE&quot;: {<br>        &quot;Loopback0&quot;: {},<br>        &quot;Loopback0|10.1.0.3/32&quot;: {}<br>    },<br> &quot;INTERFACE&quot;: {<br>        &quot;Ethernet4&quot;: {},<br>        &quot;Ethernet4|10.0.0.9/31&quot;: {}<br>    }<br>}</pre><pre>root@sonic3-vs:~# config loopback add Loopback0<br>root@sonic2-vs:~# config interface ip add Loopback0 10.1.0.3/32<br>root@sonic2-vs:~# config interface ip add Ethernet4 10.0.0.11/31<br>root@sonic1-vs:~# config save</pre><pre>{<br> &quot;LOOPBACK_INTERFACE&quot;: {<br>        &quot;Loopback0&quot;: {},<br>        &quot;Loopback0|10.1.0.3/32&quot;: {}<br>    },<br> &quot;INTERFACE&quot;: {<br>        &quot;Ethernet4&quot;: {},<br>        &quot;Ethernet4|10.0.0.11/31&quot;: {}<br>    }<br>}</pre><h4>3. Configure VLANs &amp; VXLANs</h4><p><em>sonic1-vs </em>is a route reflector for this configuration, VLAN and VXLAN is required on <em>sonic2-vs</em> and <em>sonic3-vs</em></p><pre>root@sonic2-vs:~# config vlan add 10<br>root@sonic2-vs:~# config vlan member add -u 10 Ethernet0<br>root@sonic2-vs:~# config vxlan add vtep 10.1.0.2<br>root@sonic2-vs:~# config vxlan evpn_nvo add nvo vtep<br>root@sonic2-vs:~# config vxlan map add vtep 10 1000</pre><pre>{<br>   &quot;VLAN&quot;: {<br>        &quot;Vlan10&quot;: {<br>            &quot;vlanid&quot;: &quot;10&quot;<br>        }<br>    },<br>    &quot;VLAN_MEMBER&quot;: {<br>        &quot;Vlan10|Ethernet0&quot;: {<br>            &quot;tagging_mode&quot;: &quot;untagged&quot;<br>        }<br>    },<br>    &quot;VXLAN_EVPN_NVO&quot;: {<br>        &quot;nvo&quot;: {<br>            &quot;source_vtep&quot;: &quot;vtep&quot;<br>        }<br>    },<br>    &quot;VXLAN_TUNNEL&quot;: {<br>        &quot;vtep&quot;: {<br>            &quot;src_ip&quot;: &quot;10.1.0.2&quot;<br>        }<br>    },<br>    &quot;VXLAN_TUNNEL_MAP&quot;: {<br>        &quot;vtep|map_1000_Vlan10&quot;: {<br>            &quot;vlan&quot;: &quot;Vlan10&quot;,<br>            &quot;vni&quot;: &quot;1000&quot;<br>        }<br>    }<br>}</pre><pre>root@sonic3-vs:~# config vlan add 10<br>root@sonic3-vs:~# config vlan member add -u 10 Ethernet0<br>root@sonic3-vs:~# config vxlan add vtep 10.1.0.3<br>root@sonic3-vs:~# config vxlan evpn_nvo add nvo vtep<br>root@sonic3-vs:~# config vxlan map add vtep 10 1000</pre><pre>{<br>   &quot;VLAN&quot;: {<br>        &quot;Vlan10&quot;: {<br>            &quot;vlanid&quot;: &quot;10&quot;<br>        }<br>    },<br>    &quot;VLAN_MEMBER&quot;: {<br>        &quot;Vlan10|Ethernet0&quot;: {<br>            &quot;tagging_mode&quot;: &quot;untagged&quot;<br>        }<br>    },<br>    &quot;VXLAN_EVPN_NVO&quot;: {<br>        &quot;nvo&quot;: {<br>            &quot;source_vtep&quot;: &quot;vtep&quot;<br>        }<br>    },<br>    &quot;VXLAN_TUNNEL&quot;: {<br>        &quot;vtep&quot;: {<br>            &quot;src_ip&quot;: &quot;10.1.0.3&quot;<br>        }<br>    },<br>    &quot;VXLAN_TUNNEL_MAP&quot;: {<br>        &quot;vtep|map_1000_Vlan10&quot;: {<br>            &quot;vlan&quot;: &quot;Vlan10&quot;,<br>            &quot;vni&quot;: &quot;1000&quot;<br>        }<br>    }<br>}</pre><h4>4. Configure BGP</h4><p>There is no cli support for <em>frr-management-framework</em> on SONiC, a CLI is somewhat redundant. The workflow used with this configuration model is to create the desired FRR configuration in a lab environment using vtysh and then map that configuration to config_db.json. Subsequent changes can be made using <em>config load</em> with configuration snippets.</p><p>This configuration uses BGP only with the switch sonic1-vs as route reflector. Using <em>frr-management-framewor</em>k OSPF could be configured as an IGP if desired.</p><p><strong>sonic1-vs config_db.json and resulting vtysh configuration.</strong></p><pre>{  <br>   &quot;BGP_GLOBALS&quot;: {<br>        &quot;default&quot;: {<br>            &quot;local_asn&quot;: &quot;65000&quot;,<br>            &quot;router_id&quot;: &quot;10.1.0.1&quot;<br>        }<br>    },<br>    &quot;BGP_GLOBALS_AF&quot;: {<br>         &quot;default|l2vpn_evpn&quot;: {<br>              &quot;advertise-all-vni&quot;: &quot;true&quot;<br>           }<br>    },<br>    &quot;BGP_GLOBALS_AF_NETWORK&quot;: {<br>         &quot;default|ipv4_unicast|10.1.0.1/32&quot;: {}<br>    },<br>    &quot;ROUTE_REDISTRIBUTE&quot;: {<br>         &quot;default|connected|bgp|ipv4&quot;: {}<br>    },<br>    &quot;BGP_NEIGHBOR&quot;: {<br>        &quot;default|10.0.0.9&quot;: {<br>              &quot;local_asn&quot;:  &quot;65000&quot;,<br>              &quot;asn&quot;: &quot;65000&quot;<br>         },<br>        &quot;default|10.0.0.11&quot;: {<br>               &quot;local_asn&quot;: &quot;65000&quot;,<br>               &quot;asn&quot;: &quot;65000&quot;<br>         }<br>    },<br>    &quot;BGP_NEIGHBOR_AF&quot;: {<br>         &quot;default|10.0.0.9|ipv4_unicast&quot;: {<br>              &quot;admin_status&quot;: &quot;true&quot;,<br>              &quot;rrclient&quot;: &quot;true&quot;,<br>              &quot;nhself&quot;: &quot;true&quot;,<br>              &quot;nexthop_self_force&quot;: &quot;true&quot;<br>          },<br>         &quot;default|10.0.0.9|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;,<br>             &quot;rrclient&quot;: &quot;true&quot;<br>          },<br>          &quot;default|10.0.0.11|ipv4_unicast&quot;: {<br>              &quot;admin_status&quot;: &quot;true&quot;,<br>              &quot;rrclient&quot;: &quot;true&quot;,<br>              &quot;nhself&quot;: &quot;true&quot;,<br>              &quot;nexthop_self_force&quot;: &quot;true&quot;<br>          },<br>         &quot;default|10.0.0.11|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;,<br>             &quot;rrclient&quot;: &quot;true&quot;<br>          }<br>    }<br>}</pre><pre>router bgp 65000<br> bgp router-id 10.1.0.1<br> no bgp default ipv4-unicast<br> neighbor 10.0.0.9 remote-as 65000<br> neighbor 10.0.0.11 remote-as 65000<br> !<br> address-family ipv4 unicast<br>  network 10.1.0.1/32<br>  redistribute connected<br>  neighbor 10.0.0.9 activate<br>  neighbor 10.0.0.9 route-reflector-client<br>  neighbor 10.0.0.9 next-hop-self force<br>  neighbor 10.0.0.11 activate<br>  neighbor 10.0.0.11 route-reflector-client<br>  neighbor 10.0.0.11 next-hop-self force<br> exit-address-family<br> !<br> address-family l2vpn evpn<br>  neighbor 10.0.0.9 activate<br>  neighbor 10.0.0.9 route-reflector-client<br>  neighbor 10.0.0.11 activate<br>  neighbor 10.0.0.11 route-reflector-client<br>  advertise-all-vni<br> exit-address-family<br>exit</pre><p><em>Note: Dont forget to configure next-hop-self force to override iBGPs default route reflector behavior.</em></p><p><strong>sonic2-vs config_db.json and resulting vtysh configuration.</strong></p><pre>{   <br>   &quot;BGP_GLOBALS&quot;: {<br>        &quot;default&quot;: {<br>            &quot;local_asn&quot;: &quot;65000&quot;,<br>            &quot;router_id&quot;: &quot;10.1.0.2&quot;,<br>            &quot;default_ipv4_unicast&quot;: &quot;true&quot;<br>        }<br>    },<br>    &quot;BGP_GLOBALS_AF&quot;: {<br>         &quot;default|l2vpn_evpn&quot;: {<br>              &quot;advertise-all-vni&quot;: &quot;true&quot;<br>           }<br>    },<br>    &quot;BGP_GLOBALS_AF_NETWORK&quot;: {<br>         &quot;default|ipv4_unicast|10.1.0.2/32&quot;: {}<br>    },<br>    &quot;BGP_NEIGHBOR&quot;: {<br>        &quot;default|10.0.0.8&quot;: {<br>              &quot;local_asn&quot;:  &quot;65000&quot;,<br>              &quot;asn&quot;: &quot;65000&quot;<br>         }<br>    },<br>    &quot;BGP_NEIGHBOR_AF&quot;: {<br>         &quot;default|10.0.0.8|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;<br>           }<br>    }<br>}</pre><pre>router bgp 65000<br> bgp router-id 10.1.0.2<br> neighbor 10.0.0.8 remote-as 65000<br> !<br> address-family ipv4 unicast<br>  network 10.1.0.2/32<br> exit-address-family<br> !<br> address-family l2vpn evpn<br>  neighbor 10.0.0.8 activate<br>  advertise-all-vni<br> exit-address-family<br>exit</pre><p><strong>sonic3-vs config_db.json and resulting vtysh configuration.</strong></p><pre>{<br>    &quot;BGP_GLOBALS&quot;: {<br>        &quot;default&quot;: {<br>            &quot;local_asn&quot;: &quot;65000&quot;,<br>            &quot;router_id&quot;: &quot;10.1.0.3&quot;,<br>            &quot;default_ipv4_unicast&quot;: &quot;true&quot;<br>        }<br>    },<br>    &quot;BGP_GLOBALS_AF&quot;: {<br>         &quot;default|l2vpn_evpn&quot;: {<br>              &quot;advertise-all-vni&quot;: &quot;true&quot;<br>           }<br>    },<br>    &quot;BGP_GLOBALS_AF_NETWORK&quot;: {<br>         &quot;default|ipv4_unicast|10.1.0.3/32&quot;: {}<br>    },<br>    &quot;BGP_NEIGHBOR&quot;: {<br>        &quot;default|10.0.0.10&quot;: {<br>              &quot;local_asn&quot;:  &quot;65000&quot;,<br>              &quot;asn&quot;: &quot;65000&quot;<br>         }<br>    },<br>    &quot;BGP_NEIGHBOR_AF&quot;: {<br>         &quot;default|10.0.0.10|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;<br>           }<br>    }<br>}</pre><pre>router bgp 65000<br> bgp router-id 10.1.0.3<br> neighbor 10.0.0.10 remote-as 65000<br> !<br> address-family ipv4 unicast<br>  network 10.1.0.3/32<br> exit-address-family<br> !<br> address-family l2vpn evpn<br>  neighbor 10.0.0.10 activate<br>  advertise-all-vni<br> exit-address-family<br>exit</pre><h4>5. Confirm Operation</h4><p>Check the state of BGP neighbors.</p><pre>sonic1-vs# show bgp summary<br><br>IPv4 Unicast Summary (VRF default):<br>BGP router identifier 10.1.0.1, local AS number 65000 vrf-id 0<br>BGP table version 8<br>RIB entries 11, using 2112 bytes of memory<br>Peers 2, using 1449 KiB of memory<br><br>Neighbor        V         AS   MsgRcvd   MsgSent   TblVer  InQ OutQ  Up/Down State/PfxRcd   PfxSnt Desc<br>10.0.0.9        4      65000       161       176        0    0    0 01:04:47            1        6 N/A<br>10.0.0.11       4      65000       143       168        0    0    0 02:16:27            1        6 N/A<br><br>Total number of neighbors 2<br><br>L2VPN EVPN Summary (VRF default):<br>BGP router identifier 10.1.0.1, local AS number 65000 vrf-id 0<br>BGP table version 0<br>RIB entries 3, using 576 bytes of memory<br>Peers 2, using 1449 KiB of memory<br><br>Neighbor        V         AS   MsgRcvd   MsgSent   TblVer  InQ OutQ  Up/Down State/PfxRcd   PfxSnt Desc<br>10.0.0.9        4      65000       161       176        0    0    0 01:04:47            4        6 N/A<br>10.0.0.11       4      65000       143       168        0    0    0 02:16:27            2        6 N/A<br><br>Total number of neighbors 2</pre><p>Inspecting bgp on sonic1-vs shows that there neighbors for the IPv4 and L2VPN EVPN AFIs.</p><pre>sonic1-vs# show bgp l2vpn evpn <br>BGP table version is 20, local router ID is 10.1.0.1<br>Status codes: s suppressed, d damped, h history, * valid, &gt; best, i - internal<br>Origin codes: i - IGP, e - EGP, ? - incomplete<br>EVPN type-1 prefix: [1]:[EthTag]:[ESI]:[IPlen]:[VTEP-IP]:[Frag-id]<br>EVPN type-2 prefix: [2]:[EthTag]:[MAClen]:[MAC]:[IPlen]:[IP]<br>EVPN type-3 prefix: [3]:[EthTag]:[IPlen]:[OrigIP]<br>EVPN type-4 prefix: [4]:[ESI]:[IPlen]:[OrigIP]<br>EVPN type-5 prefix: [5]:[EthTag]:[IPlen]:[IP]<br><br>   Network          Next Hop            Metric LocPrf Weight Path<br>Route Distinguisher: 10.1.0.2:2<br> *&gt;i[2]:[0]:[48]:[52:54:00:4d:94:f5]<br>                    10.1.0.2                      100      0 i<br>                    RT:65000:1000 ET:8<br> <strong>*&gt;i[2]:[0]:[48]:[52:54:00:4d:94:f5]:[128]:[fe80::5054:ff:fe4d:94f5]<br>                    10.1.0.2                      100      0 i<br>                    RT:65000:1000 ET:8</strong><br> *&gt;i[2]:[0]:[48]:[fe:54:00:d3:73:c6]<br>                    10.1.0.2                      100      0 i<br>                    RT:65000:1000 ET:8<br> *&gt;i[3]:[0]:[32]:[10.1.0.2]<br>                    10.1.0.2                      100      0 i<br>                    RT:65000:1000 ET:8<br>Route Distinguisher: 10.1.0.3:2<br> *&gt;i[2]:[0]:[48]:[fe:54:00:f3:b1:32]<br>                    10.1.0.3                      100      0 i<br>                    RT:65000:1000 ET:8<br> *&gt;i[3]:[0]:[32]:[10.1.0.3]<br>                    10.1.0.3                      100      0 i<br>                    RT:65000:1000 ET:8<br><br>Displayed 6 out of 6 total prefixes</pre><p>Inspecting the bgp l2vpn evpn AFI shows that <em>sonic2-vs</em> in advertising the mac address <em>52:54:00:4d:94:f5</em></p><p>Confirm that this mac address is being advertised (reflected) to <em>sonic3-vs</em>.</p><pre>sonic3-vs# show bgp l2vpn evpn<br>BGP table version is 21, local router ID is 10.1.0.3<br>Status codes: s suppressed, d damped, h history, * valid, &gt; best, i - internal<br>Origin codes: i - IGP, e - EGP, ? - incomplete<br>EVPN type-1 prefix: [1]:[EthTag]:[ESI]:[IPlen]:[VTEP-IP]:[Frag-id]<br>EVPN type-2 prefix: [2]:[EthTag]:[MAClen]:[MAC]:[IPlen]:[IP]<br>EVPN type-3 prefix: [3]:[EthTag]:[IPlen]:[OrigIP]<br>EVPN type-4 prefix: [4]:[ESI]:[IPlen]:[OrigIP]<br>EVPN type-5 prefix: [5]:[EthTag]:[IPlen]:[IP]<br><br>   Network          Next Hop            Metric LocPrf Weight Path<br>Route Distinguisher: 10.1.0.2:2<br> *&gt;i[2]:[0]:[48]:[52:54:00:4d:94:f5]<br>                    10.1.0.2                 0    100      0 i<br>                    RT:65000:1000 ET:8<br> *&gt;<strong>i[2]:[0]:[48]:[52:54:00:4d:94:f5]:[128]:[fe80::5054:ff:fe4d:94f5]<br>                    10.1.0.2                 0    100      0 i<br>                    RT:65000:1000 ET:8</strong><br> *&gt;i[3]:[0]:[32]:[10.1.0.2]<br>                    10.1.0.2                 0    100      0 i<br>                    RT:65000:1000 ET:8<br>Route Distinguisher: 10.1.0.3:2<br> *&gt; [2]:[0]:[48]:[fe:54:00:f3:b1:32]<br>                    10.1.0.3                           32768 i<br>                    ET:8 RT:65000:1000<br> *&gt; [3]:[0]:[32]:[10.1.0.3]<br>                    10.1.0.3                           32768 i<br>                    ET:8 RT:65000:1000<br><br>Displayed 5 out of 5 total prefixes</pre><p>Confirm using the SONiC CLI that the remote address is resolvable from sonic3-vs.</p><pre>root@sonic3-vs:/etc/sonic# show vxlan remotemac 10.1.0.2<br>+--------+-------------------+--------------+-------+---------+<br>| VLAN   | MAC               | RemoteVTEP   |   VNI | Type    |<br>+========+===================+==============+=======+=========+<br>| Vlan10 | 52:54:00:4d:94:f5 | 10.1.0.2     |  1000 | dynamic |<br>+--------+-------------------+--------------+-------+---------+<br>Total count : 1</pre><h4>Conclusion</h4><p>This relatively simple frr-management-framework can be used as a foundation to add additional functionality such EVPN multihoming. To learn more about using frr-management-framework, take a look at <a href="https://www.sonicnos.net/content/guides/config-frr">https://www.sonicnos.net/content/guides/config-frr</a> and <a href="https://medium.com/sonic-nos/sonic-dont-use-split-mode-use-frr-mgmt-framework-a67ad76ec1a6">SONiC — Don’t use split-mode, use frr-mgmt-framework!!</a>!</p><p>The configuration files used to create this tutorial can be found on<a href="https://github.com/adamdunstan/sonic-nos-bgp-evpn-lab"> github.</a></p><p>Also take a look at <a href="https://www.sonicnos.net/content/guides/config-frr">https://www.sonicnos.net</a> for more SONiC guides and tutorial focused on the publicly available Open Source version of SONiC.</p><p>If I can help you with your SONiC journey, please reach out…..</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db6d12b85ce7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/sonic-nos/evpn-route-reflector-with-sonic-using-frr-mgmt-framework-db6d12b85ce7">EVPN Route Reflector with SONiC using frr-mgmt-framework</a> was originally published in <a href="https://medium.com/sonic-nos">SoNIC NOS</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SONiC —  Don’t use split-mode, use frr-mgmt-framework!!]]></title>
            <link>https://medium.com/sonic-nos/sonic-dont-use-split-mode-use-frr-mgmt-framework-a67ad76ec1a6?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/a67ad76ec1a6</guid>
            <category><![CDATA[routing]]></category>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Mon, 01 Apr 2024 17:28:09 GMT</pubDate>
            <atom:updated>2024-04-01T17:39:42.569Z</atom:updated>
            <content:encoded><![CDATA[<h3>SONiC — Don’t use split-mode, use frr-mgmt-framework!!!</h3><p>Searching for a tutorial on how to configure SONiC switches for a specific use case will undoubtedly lead you to a guide that uses split-mode configuration. SONiC uses FRR as its routing stack and in split-mode the switch is configured using the SONiC configuration mechanism and routing is configured using FRR’s mechanism, simple in the Lab but a mess in an automated production deployment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nKSsm7UOGvHgZsPEBx8MwQ.jpeg" /><figcaption>Created by Gemini</figcaption></figure><p>In many cases the default integrated configuration mechanism, <em>bgpcfgd</em>, lacks the capabilities required to build the necessary configuration, therefore resorting to split configuration.</p><p>There is better alternative, <strong><em>frr-mgmt-framework</em></strong>. This integrated configuration mechanism supports BGP, OSPF, STATIC, IGMP, PIM as well as VRFs and BGP EVPN supporting significantly more configuration variables than <em>bgpcfgd</em> while maintaining a single configuration mechanism.</p><p>To use <strong><em>frr-mgmt-framework</em></strong> add “frr_mgmt_framework_config”: “true” to device DEVICE_METADATA</p><pre>{<br>&quot;DEVICE_METADATA&quot;: {<br>        &quot;localhost&quot;: {<br>            &quot;buffer_model&quot;: &quot;traditional&quot;,<br>            &quot;default_pfcwd_status&quot;: &quot;disable&quot;,<br>            &quot;frr_mgmt_framework_config&quot;: &quot;true&quot;,<br>            &quot;hostname&quot;: &quot;sonic4-vs&quot;,<br>            &quot;hwsku&quot;: &quot;Force10-S6000&quot;,<br>            &quot;mac&quot;: &quot;52:54:00:37:93:82&quot;,<br>            &quot;platform&quot;: &quot;x86_64-kvm_x86_64-r0&quot;,<br>            &quot;synchronous_mode&quot;: &quot;enable&quot;,<br>        }<br>    }<br>}</pre><p>Now you can configure switches using this model. For example the configuration below enables BGP in the default VRF and includes BGP EVPN using route reflector without an IGP. (you cannot configure this with <em>bgpcfgd)</em></p><pre>&quot;BGP_GLOBALS&quot;: {<br>        &quot;default&quot;: {<br>            &quot;local_asn&quot;: &quot;65000&quot;,<br>            &quot;router_id&quot;: &quot;10.1.0.1&quot;<br>        }<br>    },<br>    &quot;BGP_GLOBALS_AF&quot;: {<br>         &quot;default|l2vpn_evpn&quot;: {<br>              &quot;advertise-all-vni&quot;: &quot;true&quot;<br>           }<br>    },<br>    &quot;BGP_GLOBALS_AF_NETWORK&quot;: {<br>         &quot;default|ipv4_unicast|10.1.0.1/32&quot;: {}<br>    },<br>    &quot;ROUTE_REDISTRIBUTE&quot;: {<br>         &quot;default|connected|bgp|ipv4&quot;: {}<br>    },<br>    &quot;BGP_NEIGHBOR&quot;: {<br>        &quot;default|10.0.0.9&quot;: {<br>              &quot;local_asn&quot;:  &quot;65000&quot;,<br>              &quot;asn&quot;: &quot;65000&quot;<br>         },<br>        &quot;default|10.0.0.11&quot;: {<br>               &quot;local_asn&quot;: &quot;65000&quot;,<br>               &quot;asn&quot;: &quot;65000&quot;<br>         }<br>    },<br>    &quot;BGP_NEIGHBOR_AF&quot;: {<br>         &quot;default|10.0.0.9|ipv4_unicast&quot;: {<br>              &quot;admin_status&quot;: &quot;true&quot;,<br>              &quot;rrclient&quot;: &quot;true&quot;,<br>              &quot;nhself&quot;: &quot;true&quot;,<br>              &quot;nexthop_self_force&quot;: &quot;true&quot;<br>          },<br>         &quot;default|10.0.0.9|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;,<br>             &quot;rrclient&quot;: &quot;true&quot;<br>          },<br>          &quot;default|10.0.0.11|ipv4_unicast&quot;: {<br>              &quot;admin_status&quot;: &quot;true&quot;,<br>              &quot;rrclient&quot;: &quot;true&quot;,<br>              &quot;nhself&quot;: &quot;true&quot;,<br>              &quot;nexthop_self_force&quot;: &quot;true&quot;<br>          },<br>         &quot;default|10.0.0.11|l2vpn_evpn&quot;: {<br>             &quot;admin_status&quot;: &quot;true&quot;,<br>             &quot;rrclient&quot;: &quot;true&quot;<br>          }<br>    }</pre><p>This SONiC <em>config_db.json</em> will generate the following FRR configuration dynamically.</p><pre>sonic1-vs# show run<br>Building configuration...<br><br>Current configuration:<br>!<br>frr version 8.5.1<br>frr defaults traditional<br>hostname sonic1-vs<br>log syslog informational<br>log facility local4<br>no zebra nexthop kernel enable<br>agentx<br>no service integrated-vtysh-config<br>!<br>password zebra<br>enable password zebra<br>!<br>router bgp 65000<br> bgp router-id 10.1.0.1<br> no bgp default ipv4-unicast<br> neighbor 10.0.0.9 remote-as 65000<br> neighbor 10.0.0.11 remote-as 65000<br> !<br> address-family ipv4 unicast<br>  network 10.1.0.1/32<br>  redistribute connected<br>  neighbor 10.0.0.9 activate<br>  neighbor 10.0.0.9 route-reflector-client<br>  neighbor 10.0.0.9 next-hop-self force<br>  neighbor 10.0.0.11 activate<br>  neighbor 10.0.0.11 route-reflector-client<br>  neighbor 10.0.0.11 next-hop-self force<br> exit-address-family<br> !<br> address-family l2vpn evpn<br>  neighbor 10.0.0.9 activate<br>  neighbor 10.0.0.9 route-reflector-client<br>  neighbor 10.0.0.11 activate<br>  neighbor 10.0.0.11 route-reflector-client<br>  advertise-all-vni<br> exit-address-family<br>exit<br>!<br>ip nht resolve-via-default<br>!<br>ipv6 nht resolve-via-default<br>!<br>end</pre><p>Another attribute of this mechanism is that it is dynamic for all of its configuration operations. Loading and updated <em>config_db.json</em> will result in reconfiguration of FRR. Watch /var/log/frr/zebra.log on your switch to see the operation of the configuration script.</p><p>One of the reasons that so many “online experts” write guides and create youtube videos using split-mode is documentation. There is lots of existing documentation (and guides) for FRR that can be repurposed in addition to the official FRR documentation. SONiC however has very sparse documentation. This is true for both <em>bgpcfgd</em> and <em>frr-mgmt-framework.</em></p><p>The documentation for <em>frr-mgmt-framework</em> on the SONiC github repo is limited to the original <a href="https://github.com/sonic-net/SONiC/blob/master/doc/mgmt/SONiC_Design_Doc_Unified_FRR_Mgmt_Interface.md">High Level Design document</a> but its worth a look to understand how the mechanism works and understand how to structure your configuration.</p><p>However the code for <em>frr-mgmt-framework</em> is much easier to understand as the configuration is generated from a single python program called <a href="https://github.com/sonic-net/sonic-buildimage/blob/master/src/sonic-frr-mgmt-framework/frrcfgd/frrcfgd.py">frrcfgd.py</a> in the SONiC buildimage repo. By inspecting the code its easy to figure out the syntax used in <em>config_db.json</em> and in particular the configuration commands that are supported.</p><p>When I create a new switch configuration, I usually start using “split-mode” to get an FRR configuration that meets my needs. Once its verified I move the configuration into <em>config_db.json </em>using <em>frr-mgmt-framework</em> so that there is only one configuration file that needs to be managed on the switch.</p><p>If you want to learn more about using this configuration mode, take a look at <a href="https://www.sonicnos.net/content/guides/config-frr">https://www.sonicnos.net/content/guides/config-frr</a></p><p>In the near future Ill write a guide on configuring BGP EVPN using <em>frr-mgmt-framework </em>and post it on <a href="https://medium.com/sonic-nos">https://medium.com/sonic-nos</a> as well as <a href="https://www.sonicnos.net/content/guides/config-frr">https://www.sonicnos.net</a></p><p>If I can help you with your SONiC journey, please reach out…..</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a67ad76ec1a6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/sonic-nos/sonic-dont-use-split-mode-use-frr-mgmt-framework-a67ad76ec1a6">SONiC —  Don’t use split-mode, use frr-mgmt-framework!!</a> was originally published in <a href="https://medium.com/sonic-nos">SoNIC NOS</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Declarative Linux — NixOS. The server management solution you have been waiting for!]]></title>
            <link>https://medium.com/thermokline/declarative-linux-nixos-the-server-management-solution-you-have-been-waiting-for-75eb742cfa31?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/75eb742cfa31</guid>
            <category><![CDATA[vmware]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[virtualization]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Tue, 23 Jan 2024 17:45:37 GMT</pubDate>
            <atom:updated>2024-01-23T17:45:37.093Z</atom:updated>
            <content:encoded><![CDATA[<h3>Declarative Linux — NixOS. The server management solution you have been waiting for!</h3><p>If you have been faced with building a system to fully automate the installation of bare metal servers in bulk and providing ongoing infrastructure management, you will know it’s challenging. Finding a single solution for full automation usually requires adopting specific vendor solutions and/or adapting your plans to their infrastructure architecture. The emergence of Cloud providers has reduced the development of platforms in this complex area as vendors have focused on automating on-top of the Cloud.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lLcAQa9dq2fo07iMQD9XDA.png" /></figure><p>The process usually looks something like this. The server is booted using PXE boot, either from a simple DHCP server or from a server manager (I have used a variety of tools for this but none are great). Using the booted OS image, configure the hardware which primarily consists of setting up storage. Once this task is complete the production operating system is installed and the server rebooted to the target OS. Most OS have a way to configure the OS that is specific to the OS, I have never bothered to use them preferring to configure the OS and application environment, such as kubernetes, in the final installation step. There are lots of choices for the final step once the target OS is installed, tools such as Ansible. Finally with the application environment installed, there is another set of tools. A critical challenge is making the servers configuration idempotent and consistent across the fleet when using a chain of loosely coupled tools. Over the years I have thought about taking the software we built for the projects where we had to install lots of servers (I have done a few) and create a generic solution, but in each case decided the permutation complexity was too high.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Kaq10H7gHwScrBOVoZXq4Q.png" /></figure><p><a href="https://nixos.org/">NixOS </a>offers a different solution to Server Management simplifying the process of installation and ongoing operations. Instead of progressing through a chain of tools, NixOS provides a single, relatively simple Linux OS and package management environment that ensures that every server is installed identically and remains that way.</p><p>NixOS has three attributes that make it ideally suited to use as the underlying Linux for a large scale application platform.</p><ol><li>It’s easy to build custom Linux installation images.</li><li>Package management is integrated with configuration and is declarative and idempotent.</li><li>Reproducibility.</li></ol><p>The complete Linux configuration, including packages, users, networking, storage is contained in one or more configuration files. These configuration files build the operating system and its application environment. This declarative mechanism ensures that the configuration matches the operating environment. The server can be updated by modifying the configuration and switching to it. Importantly the configuration can also be rolled back.</p><p>Often a day 2 operational problem requires loading addition debugging tools and in many cases these tools get left behind making some servers different to others. Over time the configuration and package installation can skew increasing the probability of inconsistent behavior across the server fleet. NixOS addresses this debugging problem with Ad hoc shell environments that create a shell with the programs needed for debugging without installing the package permanently.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rhs5AlDCzd4JUdkhEK8AqA.png" /></figure><p>Key to NixOS’s operation is its package manager which can be installed independently on other Linux derivatives, as well as other tools for build automation and Home Manager for managing user environments in a declarative manner. Next time I upgrade my Linux desktop machine, I’m serious considering switching to NixOS to help keep track of all of the packages installed on the system and ensure that I can create reproducible environments for each of my projects.</p><p>Next time I get involved in a project where we need to install lots of bare metal servers, we are going with NixOS.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=75eb742cfa31" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thermokline/declarative-linux-nixos-the-server-management-solution-you-have-been-waiting-for-75eb742cfa31">Declarative Linux — NixOS. The server management solution you have been waiting for!</a> was originally published in <a href="https://medium.com/thermokline">ThermoKline</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SONiC, FRR split configuration, a step backwards?]]></title>
            <link>https://medium.com/sonic-nos/sonic-frr-split-configuration-a-step-backwards-fb4db0e181a1?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb4db0e181a1</guid>
            <category><![CDATA[router]]></category>
            <category><![CDATA[cisco]]></category>
            <category><![CDATA[networking-tips]]></category>
            <category><![CDATA[networking]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Mon, 08 Jan 2024 20:17:27 GMT</pubDate>
            <atom:updated>2024-01-09T21:39:49.136Z</atom:updated>
            <content:encoded><![CDATA[<p>Browsing the YANG files in the SONiC source code I found that it is now possible to enable a new split configuration mode.</p><pre>leaf docker_routing_config_mode {<br>                    description &quot;This leaf allows different configuration modes for FRR:<br>                                - separated: FRR config generated from ConfigDB, each FRR daemon has its own config file<br>                                - unified: FRR config generated from ConfigDB, single FRR config file<br>                                - split: FRR config not generated from ConfigDB, each FRR daemon has its own config file<br>                                - split-unified: FRR config not generated from ConfigDB, single FRR config file&quot;;<br>                    type string {<br>                        pattern &quot;separated|unified|split|split-unified&quot;;<br>                    }<br>                    default &quot;unified&quot;;</pre><p>Instead of using the integrated SONiC configuration, this mode splits the configuration of SONiC and the routing component FRR into two configuration mechanism on the same switch.</p><p>SONiC’s integrated mechanism parses the SONiC configuration via a configuration model written in Python using Jinja templates into the loaded FRR configuration. The native SONiC cli does not provide any routing functionality, depending upon the FRR VTYSH CLI. This seems reasonable but the routing configuration in the default model is not generalized, more use case focused and combine that with sparse documentation makes using SONiC integrated and default routing configuration model challenging.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/665/1*-1VVxfkYOGYgb_85iKBVUA.png" /></figure><p>Why not just use the FRR VTYSH command set, it’s just like the Cisco CLI? The problem is the default configuration model does not allow a saved VTYSH configuration to be persisted between reboots. Enter split configuration model.</p><p><strong>Split</strong> configuration allows the use of FRR configuration mechanisms such as VTYSH but still requires a SONiC configuration resulting in <strong>two configuration</strong> mechanisms.</p><p><strong>While this seems like a simple solution </strong>(and one that has been attractive to the commercial SONiC providers), s<strong>plit mode creates more problems than it solves.</strong></p><p><strong>Instead of a single tool chain for switch configuration and management, split mode requires a tool chain for SONiC and another for FRR that are interdependent.</strong></p><p><strong><em>Every other switch provider uses a single configuration model.</em></strong></p><p>Any new switching infrastructure should be automated and managed programmatically, with few uses for an interactive CLI. This is how SONiC is designed, the CLI is rudimentary and there are an abundance of programmatic mechanisms to manage the switches.</p><p>SONiC presents a significant challenge for new users, delve into the source code to understand how to configure routing using a single automation tool chain in integrated mode or use split mode and develop an automation tool chain for SONiC and FRR that accommodates their interdependencies. Neither are good solutions.</p><p>I think Split mode and its use by commercial SONiC vendors is a step backwards, postponing the need to address SONiC routing configuration and documentation issues.</p><p><strong><em>A more technical discussion on SONiCs FRR configuration mechanisms is located at </em></strong><a href="https://www.sonicnos.net/content/guides/config-frr"><strong><em>sonicnos.net</em></strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb4db0e181a1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/sonic-nos/sonic-frr-split-configuration-a-step-backwards-fb4db0e181a1">SONiC, FRR split configuration, a step backwards?</a> was originally published in <a href="https://medium.com/sonic-nos">SoNIC NOS</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[EPIC Gateway is Open Source — Not what we wanted, but…]]></title>
            <link>https://medium.com/thermokline/epic-gateway-is-open-source-not-what-we-wanted-but-ad5ef7d23169?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/ad5ef7d23169</guid>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Wed, 13 Dec 2023 18:58:58 GMT</pubDate>
            <atom:updated>2023-12-22T15:31:45.795Z</atom:updated>
            <content:encoded><![CDATA[<h3>EPIC API Gateway is Open Source — Not what we wanted, but…</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zSOfM7EoWIMoDIQUqU2eXg.png" /><figcaption>Generated by Bing Chat &amp; DALL-E 3</figcaption></figure><p>In 2020 we began the development of an external Gateway for Kubernetes. We had just finished working on building a public k8s, namespace as a service platform for a large enterprise. As the platform was namespace-as-a-service, not virtual clusters, attempting to share ingresses between tenants did not provide the control or automation required. Additionally, there were multiple regional complexes and there was a need to distribute applications, in a controlled manner between regions. Application access was a difficult problem to solve.</p><p>The EPIC Gateway project was born. We chose to trade higher initial complexity for lower operating complexity. The story of packaging begins.</p><h4>Enterprise Software</h4><p>We developed the first version of the platform and began demonstrating it to interested users in 2021. We had packaged the first version as Enterprise software with automated installation procedure for bare metal servers. Additionally the first implementation was prior to the development of the Kubernetes Gateway API and used an adapted version of the Service API. This version had two problems. First the implementation using the Service API did not have a simple workflow making gateways was clunky, but the more significant problem was installation. By definition, the gateway needed to interact with the network and required allocation of IP addresses beyond those automatically provided by DHCP in the datacenter. Further there were often slight differences in how the customer infrastructure handed networking, in some cases making it impossible to deploy the software. Often we would get asked if we could run it at a Cloud Provider, the answer was no. Cloud providers provide limited ability to control networking, specifically allocate and advertise addresses. Many of the bare metal providers have implemented automation for adding hosts or creating VMs that have the same problems, restricting allocation and advertising of addresses. Instead of using robust BGP based host route ECMP load balancing we would get the question, why can’t you just use Linux HA…</p><p>The technical challenges were compounded with organizational challenges. In the founder&#39;s recent projects, there was a single infrastructure team, we quickly learned that this is not the case in most Enterprises. The deployment of the EPIC software requiring approval and assistance from lots of different teams. The result was that it just ended up to be easier to use the cloud providers less functional solution or continue using the <strong>home grown solution thats primary design goal was avoiding other departments.</strong></p><h4>Gateway-as-a-Service</h4><p>The next packaging was Gateway-as-a-Service released in 2022. We already had a Colo data center where we kept our servers so we added the platform there. We were able to configure the network as we needed. This avoided the problem of Gateway installation but required we develop a self service user interface. Implementing the User Interface was a significant effort especially if you add billing. We didn’t add billing to the UI because our purpose was to gather feedback and test users for the purpose of addition fundraising. During this time, we had also adopted the new Kubernetes Gateway API making the workload workflow much better. The Gateway Service worked well, we had many users try the platform however it did not result in the ability to raise funds. There is a very simple problem with these types of Services, it takes a long time to get sufficient users to figure out if the Service will generate enough revenue to be a suitable investment. Investors don’t want to fund this <em>flatline</em> period, but they are ready to invest when the company as $1M of ARR. Not only would have getting to $1M of ARR taken additional development, but it would also have significant marketing costs. During 2023 we shutdown the EPIC Gateway Service Preview.</p><h4>Open Source</h4><p>So during the summer of 2023 began the effort to Open Source the EPIC platform and it was published on GitHUB in November. Often overlooked, packaging takes significant effort. In addition to getting the source code organized, documentation and a new set of installation procedures need to be created. From experience we knew this had to include something very easy to try. If your interested, we encourage you to try the platform at <a href="https://github.com/epic-gateway">https://github.com/epic-gateway</a></p><p>So we end up Open Sourcing the platform. Since the Open Source industry started, much has changed. The most profound change is the emergence of Cloud Providers. While many would like to believe that they market is growing fast enough for all other sectors of the market not to be impacted, this is simply not true. Cloud Providers have consumed a significant amount of the IT spend impacting other supplier categories, this includes Open Source infrastructure. Infrastructure for Cloud provider users gets broad support, but those that provide alternatives, even if they are better, have to overcome the “why don’t I just use the cloud”. Open Source also narrows the addressable market, its focus is a very technical users and in many cases they use the software without financially contributing. Even in the most successful Open Source projects, a small percentage of users financially contribute, this is a problem well know and well reported without any resolution.</p><p>We are proud of the platform we created, it rivals any API Gateway platform built by the Cloud providers and provides features that are unparalleled by any product in the marketplace. We focused on building a complete solution, not just components that seem easy at first but end up with a bespoke solution with higher complexity for the complete lifecycle.</p><p>Is this a new chapter of EPIC, that remains to be seen. Perhaps Open Source will result in the adoption and financial contribution that will enable us to continue to develop the platform.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ad5ef7d23169" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thermokline/epic-gateway-is-open-source-not-what-we-wanted-but-ad5ef7d23169">EPIC Gateway is Open Source — Not what we wanted, but…</a> was originally published in <a href="https://medium.com/thermokline">ThermoKline</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Did you overlook platform when selecting API Security & Management?]]></title>
            <link>https://medium.com/thermokline/did-you-overlook-platform-when-selecting-api-security-management-9121725740f5?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/9121725740f5</guid>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[app-development]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[network-security]]></category>
            <category><![CDATA[api-gateway]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Mon, 14 Aug 2023 16:53:48 GMT</pubDate>
            <atom:updated>2023-12-11T15:28:36.866Z</atom:updated>
            <content:encoded><![CDATA[<p>The platform used to implement API Security and Management is arguably as important as the features Security and Management features offered. The platform provides the dataplane, managing requests is the primary purpose for implementing an API Security or Management system. The features that are being developed or implemented depend on the platform, it&#39;s key to not lose sight of this.</p><p><strong><em>We developed the </em></strong><a href="https://www.epic-gateway.org/"><strong><em>EPIC platform</em></strong></a><strong><em> as an ideal execution environment for Proxy engines implementing API Security and Management. It provides the configuration and automation required by network and security engineers to enable developers to dynamically create and manage their applications. The list of required functionality is extensive and covered in this post.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/919/1*KgE_hXyFg7qob2ZZVfAgEQ.png" /></figure><p>There are two key components in the platform: the Proxy engine and the Proxy execution environment that hosts the Proxy engine.</p><h4>Proxy Engine</h4><p>The Security and Management components configure, program and manage Proxy engines therefore there is a tight coupling between the API product and the Proxy engine. In many solutions this results in the Proxy engine being bundled with the API Security or Management tooling.</p><p><strong><em>Understanding the specifics of the Proxy engines used to implement API Security &amp; Management is critical to understanding the solution capabilities.</em></strong></p><p>There is a significant amount of variability in how Proxy engines are delivered with API Security &amp; Management solutions.</p><ul><li>Bundled Proxy Engine. In these cases there is often little visibility to the Proxy Engine. It could be a custom built using libraries from a popular frameworks (node.js, rust, python, go) or a fork of a popular Open Source Proxy. There are so many Open Source Proxy engines that it doesn&#39;t make sense to build a new one.</li><li>Supported Proxy Engines. Some solutions support commonly available Proxy Engines, these are usually one of the more popular Open Source engines such as Envoy, HAProxy or Nginx.</li></ul><p>There are so many Proxy engines that creating a comprehensive list would be almost impossible and many Proxy engines are use case specific, such as Varnish for Content Distribution, and are not suited to the API Security and Management use case.</p><ul><li>Development Framework Proxies. Almost every modern development language has libraries that can be used to create a Proxy engine. The functionality available in those libraries defines the base functionality and additional features can be developed. However no two proxies developed using a single language/framework are the same. With the exception of very specific scale or use cases, I don’t see the need to develop another Proxy engine. This type of solution is often used to shortcut initial development at the expense of longer term support and stability.</li><li>Complete Proxy Engines. There are lots of these as well, both Open Source and proprietary. From a functionality perspective there are differences but in general from a Request processing perspective their functionality is largely the same. Performance varies often based upon when development began and development language used but it’s usually not that important.</li></ul><p>The important differences between Proxy Engines are how they are programmatically configured and how telemetry data is returned. In the case of Development Framework Proxy, any mechanism can be used as long as it can be developed, however complete Proxy engines have predefined configuration and telemetry interfaces. Two examples:</p><ul><li>Envoy Proxy. The programmatic configuration interface Envoy uses is called xDS and supports REST or gRPC. Most tooling that configures Envoy uses the gRPC interface as it provides more functionality than REST (including Acnodal EPIC). Go has libraries for gRPC and Envoy is widely used in Kubernetes. Considering its popularity in Kubernetes, it&#39;s not surprising that statistics are scraped from a REST endpoint.</li><li>Nginx. In its Open Source form, Nginx is configured using a configuration file and that configuration reloaded from a CLI. (There is a commercial version with a REST interface, as with many of these Open Source/Commercial products it is hard to figure out what you have to pay for!). The security and management functionality are loaded as modules and where these modules are provided as part of the solution, in practise additional configuration mechanisms can be added by the module provider. By default telemetry is provided by logging however there are other modules such as OTEL that can be used.</li></ul><p>Each Proxy engine has its own unique management interfaces and supporting each of them is a unique development for the API Security and Management solution, supporting multiple Proxy engines creates a feature matrix requiring additional attention.</p><h4>Proxy Execution Environment</h4><p>Proxy engines are executables that terminate and establish sessions, acting as a “middle-man” between the client and the server applications. In most cases with API Security and Management solutions, how and where the Proxy engine runs is left to the individual deploying the solution.</p><p><em>Proxy Engines need to run somewhere, the selection and configuration of the execution environment is critical to the security, scale and reliability of the solution.</em></p><p>The execution environment for the Proxy engine or engines is critical to security, scale and operation. The Proxy engine depends on the underlying infrastructure for operation and security. In the Security context, if the Proxy engine is executed in an insecure execution environment, the API Security platform is insecure. Put simply the execution environment is a key component of API security and Management.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/717/1*pbZtLD2eSGnmMfEwix2EQg.png" /></figure><p>API Security and Management provides access from untrusted clients to controlled resources. Therefore one side of the API Security and Management system is exposed to the untrusted Internet. This exposure creates an attack surface. Therefore the configuration of the execution environment is necessarily an extension of the API Security and Management system. For example, one or more public exposures (IP Address/DNS) are required for the API system and a limited set of open protocol ports.</p><p>Procedurally, and in a statically configured system, creating an execution environment in a Linux host consists of exposing at least one interface to the public internet, configuring security rules in the form of IP tables and matching the Proxy Engine listener port configuration. However such a solution is not practical for production use. Only one IP address/port tuple is available limiting the configuration of multiple upstream endpoints to DNS configuration.</p><p>Being limited to a small number of shared Proxy engines is in itself an operational and security problem. A shared Proxy engine results in a more complex Proxy engine configuration and in some cases necessitates Proxy and execution environment configuration specific to one of the shared Proxy use case potentially creating additional attack surface for others.</p><p>Scaling and operation for production is also limited with a simplistic static execution environment. Adding Proxy engines from new applications requires operational intervention. Scaling to meet demands that cannot be served by a execution environment creates even more complexity as the system will likely need to be redesigned to integrate with routings ECMP to provide distribution among execution environments.</p><h4>The problem with Kubernetes Ingresses</h4><p>If your environment has Kubernetes, it’s temping to believe that this problem can be solved with in-cluster Kubernetes specific solutions.<strong> There is however a major flaw with this assumption. This solution puts the untrusted access surface, the downstream side of the Proxy engine, inside the Kubernetes cluster. Compromising the Proxy engine POD will expose the cluster to lateral movement into application PODs.</strong></p><h4>Creating a platform for API Security and Management</h4><p>The implementation of an API Security and Management platform is a complex task taking significant effort. Consider the following attributes that we have implemented in our platform and their need in a reliable, secure execution environment.</p><ul><li>uService Proxy Engines. Containerizing the Proxy Engines allows the platform to create as many individual Proxy engines as needed for each use case. These individual Proxy engines can each be configured to provide maximum security for that use case which avoid the need for merging use case configurations. Many Proxy engines have plugins for specific protocols or functions, using a uService platform further enhances security by limiting the plugin use to specific instances. Testing or using newer versions of the Proxy engine can be limited to specific use cases. uService Proxy engines when managed by the platform can also scale easily with each Gateway consisting of multiple instances of the Proxy engine as demanded by workload.</li><li>Combine proxy, network &amp; execution environment management. When a new Proxy is created or its configuration changed, that event should trigger the necessary configuration of the execution environment. While it may seem simple; by automating the allocation of IP addresses, advertising and exposing those addresses, generation of DNS records and opening only the ports necessary for the proxy listeners enables Secure APIs to be exposed on-demand in a manner consistent with predefined policies. If any of these steps is manual, adhoc or distributed among multiple organizations or functions they because a dangerous source of configuration error.</li><li>Separate from Application resources. It&#39;s tempting to place the API functionality next to the applications in Kubernetes clusters. While many will argue that a Cluster can be made sufficiently secure by configuration, correct configuration over Clusters with the amount of complexity and possible variance of tooling makes this a difficult objective to achieve every-time at scale. Therefore the simpler solution of providing a high level of isolation between the execution environment and application resources simplifies this risk completely avoiding any possibility of lateral movement.</li><li>Access Endpoints behind Firewalls. Isolating application resources also results in putting them behind additional security, in particular firewalls. In the case of Kubernetes, almost all POD resources are on networks where PODs are not externally addressable.</li><li>Multicluster, multiple Endpoint. The API Security and Management environment needs to be able to communicate with all of the upstream application endpoints irrespective of location and operating environment. Tight integration with kubernetes workflow is required to simplify application operators. Not all applications are located in Kubernetes clusters, the ability to add application endpoints running on dedicated hosts, VMs or other environments is also required with a similar workflow.</li><li>Scales across execution environment nodes. A proxy consisting of multipleproxy execution instances should scale across multiple compute instances. This required the implementation and integration with networking infrastructure to leverage network hardware capabilites to ensure performance. The addition and removal of Proxy engines should dynamically trigger network distribution of packets.</li><li>Limit Proxy hops. Often the solution to scale Proxies is the addition of more Proxies in the request flow. Each proxy terminating either a TCP or HTTP session adds latency and configuration complexity. The fewer Proxies, ideally one, in the path the better the performance and simpler the configuration. In practise this takes attention to detail.</li><li>The right Proxy Engine. The Proxy Engine used in API Security and Management defines the solution capabilities. How the Proxy engine manages configuration directly impacts scale and operations. For example, updating configuration files vs. a dynamic programmable API dramatically increases the functionality and security of the overall solution. Implementing security functionality often involves additional programming of the Proxy, a choice of programing interface entry points and languages makes this effort significantly easier. For example, the Envoy proxy engine enables dynamically loadable programs as part of the configuration written using WebAssembly or LUA to interact with the request flow. Where more low level functionality is required, it provides the ability to develop plugin modules to be added to the execution binary.</li></ul><h4>Conclusion</h4><p>I could have written this as a list of things that the platform we developed provides, however I think it’s more valuable to provide insight into the requirements and why they are necessary in implementing API Security in particular and also API Management when Public APIs are exposed to the Internet. A detailed understanding of API technology is required to implement an API Security platform, I hope you agree that in addition a detailed understanding of how to create the Execution environment is equally important.</p><p>If you would like to leverage the EPIC platform in your API Security deployment or integrate it with your existing solution, please let me know.</p><p>You can find the documentation and source code at <a href="https://www.epic-gateway.org/">https://www.epic-gateway.org/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9121725740f5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thermokline/did-you-overlook-platform-when-selecting-api-security-management-9121725740f5">Did you overlook platform when selecting API Security &amp; Management?</a> was originally published in <a href="https://medium.com/thermokline">ThermoKline</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Visualize the k8s Gateway API]]></title>
            <link>https://medium.com/thermokline/visualize-the-k8s-gateway-api-246722cac00f?source=rss-f8c5a88f31d4------2</link>
            <guid isPermaLink="false">https://medium.com/p/246722cac00f</guid>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[api-gateway]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Adam Dunstan]]></dc:creator>
            <pubDate>Thu, 10 Aug 2023 13:03:05 GMT</pubDate>
            <atom:updated>2023-12-11T15:35:43.302Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://www.epic-gateway.org/">Try the EPIC platform its Open Source</a></p><p>Creating a working API Gateway using the <a href="https://gateway-api.sigs.k8s.io/">k8s Gateway API</a> requires more configuration resources than the simpler ingress/loadbalancer alternative. A gateway consists of gatewayclass, gateway, routes, services and endpoints as well as any implementation specific resources. While it provides significantly more flexibility than the simpler alternatives, these resources require correct configuration (except endpoint which are dynamically created) to define the desired state of the Gateway.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iiLGm34vLM_h0YwiMRwyrw.png" /><figcaption>Gateway Service Manager — note the target POD information</figcaption></figure><p>The Gateway API does not specify how the Gateway is implemented, therefore figuring out how the resulting gateway is configured and if it is operating as desired can be difficult. One of the most common requests we received is a better way to view the resulting gateway including target cluster context.</p><p>In our implementation each Gateway consists of a configurable number of envoy proxy instances and these instances are running in a purpose built cluster separate from the application clusters. Installed on the application cluster is a controller for the Gateway API that configures the application cluster’s networking and sends information necessary to create the desired gateway to the Gateway Cluster (EPIC).</p><p>The controller in the application cluster sends all of the gateway information to the Gateway cluster for use in creating the Gateway and its proxy instances. Using this information, the Gateway’s configuration and operational state is mapped visually providing a simple and complete view of how the requested gateway is configured and how requests will be handled. The visualization also includes the nodes, PODs and namespaces upstream from the gateway.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*e8WhhTrrqaZas-JLNiPO-A.png" /></figure><p>Comparing the Route object in the Gateway Cluster with the application cluster shows the commonality in syntax and therefore the simplicity in debugging possible routing problems.</p><p>Gateways require proxy engines to distribute requests, in our implementation Envoy. It is useful to view the native Envoy configuration resulting from the Gateway API resources.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6mJ8UWs5Z276sOAAX0jI3A.png" /></figure><p>More information can be found at <a href="https://www.epic-gateway.org/">https://www.epic-gateway.org/</a> including a quick start guide that can be used to setup a simple test environment.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=246722cac00f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thermokline/visualize-the-k8s-gateway-api-246722cac00f">Visualize the k8s Gateway API</a> was originally published in <a href="https://medium.com/thermokline">ThermoKline</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>