<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Rozoom News</title>
    <description>The latest articles on DEV Community by Rozoom News (rozoomnews).</description>
    <link>https://dev.to/rozoomnews</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Forganization%2Fprofile_image%2F15629%2F63606cd8-3812-460a-bb91-b333a347941b.png</url>
      <title>DEV Community: Rozoom News</title>
      <link>https://dev.to/rozoomnews</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/rozoomnews"/>
    <language>en</language>
    <item>
      <title>Rozoom in 60s, episode 2: read a Kubernetes cluster health card</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Fri, 13 Nov 2026 14:12:00 +0000</pubDate>
      <link>https://dev.to/rozoomnews/rozoom-in-60s-episode-2-read-a-kubernetes-cluster-health-card-251b</link>
      <guid>https://dev.to/rozoomnews/rozoom-in-60s-episode-2-read-a-kubernetes-cluster-health-card-251b</guid>
      <description>&lt;p&gt;You added a Kubernetes cluster to Rozoom. Is it healthy? This 55 second short shows how to read the health card.&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/7ls8HA4WTiY" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h2&gt;
  
  
  What the card tells you
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Where it lives.&lt;/strong&gt; Every cluster you add gets a card on Available Kubernetes Clusters. The checks start on their own.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The badge.&lt;/strong&gt; Look at it first: OK, Warning or Critical. Offline means Rozoom can't reach the cluster.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The top problem.&lt;/strong&gt; If the badge is not OK, the card names the most important problem right under it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Five areas.&lt;/strong&gt; Health, Capacity, HA, Security and Upgrades. The worst check in each area sets that area's badge.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Click to dig in.&lt;/strong&gt; Health opens Overview, Security opens Compliance Hub, Upgrades opens API Deprecation Scan.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Need the why?&lt;/strong&gt; Switch the card to Detailed. The Primary Alert box explains what the problem means.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Freshness.&lt;/strong&gt; Refresh runs every check now; otherwise they repeat every five minutes.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Get it
&lt;/h2&gt;

&lt;p&gt;Rozoom is a free, open source Kubernetes desktop app (Apache 2.0), no account needed.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Download: &lt;a href="https://rozoom.app/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=r60s-02-health-card" rel="noopener noreferrer"&gt;https://rozoom.app/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=r60s-02-health-card&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Docs: &lt;a href="https://rozoom.app/docs" rel="noopener noreferrer"&gt;https://rozoom.app/docs&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Episode 1, adding your first cluster: &lt;a href="https://www.youtube.com/watch?v=5Mjp4j82Bh0" rel="noopener noreferrer"&gt;https://www.youtube.com/watch?v=5Mjp4j82Bh0&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>kubernetes</category>
      <category>devops</category>
      <category>sre</category>
      <category>opensource</category>
    </item>
    <item>
      <title>Trending in self-hosting and Kubernetes: week of 10 November 2026</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Wed, 11 Nov 2026 07:03:07 +0000</pubDate>
      <link>https://dev.to/rozoomnews/trending-in-self-hosting-and-kubernetes-week-of-10-november-2026-2fdo</link>
      <guid>https://dev.to/rozoomnews/trending-in-self-hosting-and-kubernetes-week-of-10-november-2026-2fdo</guid>
      <description>&lt;p&gt;Four things that picked up attention this week, with what we think of each. The originals are linked; read them. This page only tells you why they matter for a small self-hosted setup.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. An empty &lt;code&gt;resources: {}&lt;/code&gt; quietly downgrades your pods on Kubernetes 1.34 to 1.36
&lt;/h2&gt;

&lt;p&gt;Original: &lt;a href="https://devops-daily.com/posts/kubernetes-empty-pod-resources-besteffort" rel="noopener noreferrer"&gt;An Empty resources: {} Makes Kubernetes 1.34 to 1.36 Pods BestEffort&lt;/a&gt; on devops-daily.com (also &lt;a href="https://dev.to/devopsdaily/an-empty-resources-makes-kubernetes-134-to-136-pods-besteffort-2p15"&gt;on dev.to&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;The claim: if a pod sets &lt;code&gt;spec.resources: {}&lt;/code&gt; at pod level, the kubelet classifies it as BestEffort even when every container has requests and limits. The authors measured it on k3s 1.34, 1.35 and 1.36: &lt;code&gt;oom_score_adj&lt;/code&gt; jumped to 1000, and next to a busy neighbour a pod that requested half a CPU got 0.06 to 0.08 cores instead of about 0.57. A namespace quota that blocks BestEffort pods rejected it. On 1.37.1 the classes were correct. GitLab Runner's Kubernetes executor generates that empty block from version 18.10.0 to 19.4.3.&lt;/p&gt;

&lt;p&gt;Our take: a useful correction to folklore as well. The article shows that the popular claim "these pods are evicted first" is wrong; the kubelet's eviction order did not change. Check your cluster in ten seconds:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;kubectl get pods -A -o custom-columns=NS:.metadata.namespace,POD:.metadata.name,QOS:.status.qosClass &lt;span class="o"&gt;|&lt;/span&gt; grep BestEffort
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If a workload that has limits shows up there, you have found it. If you prefer to see QoS class in a UI, our &lt;a href="https://news.rozoom.app/kubernetes/kubernetes-gui-2026/" rel="noopener noreferrer"&gt;Kubernetes GUI comparison&lt;/a&gt; covers the tools that show it at a glance.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. A 10 USD OpenStick as a home server
&lt;/h2&gt;

&lt;p&gt;Original: &lt;a href="https://blog.bosscher.ch/posts/openstick/" rel="noopener noreferrer"&gt;What I run on my $10 USD OpenStick homelab&lt;/a&gt; (825 points and 137 comments on &lt;a href="https://news.ycombinator.com/item?id=50406670" rel="noopener noreferrer"&gt;Hacker News&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;The hardware: four Cortex-A53 cores at 1 GHz, 512 MB of RAM (about 450 MB usable), 4 GB of eMMC, Debian with systemd and Podman, passively cooled. The author runs AdGuard Home (about 40 MB) and a handful of small utilities on it.&lt;/p&gt;

&lt;p&gt;Our take: the interesting part is not the gadget, it is the budget discipline. Services measured in tens of megabytes are the right default for a DNS blocker, an uptime monitor or a notifier. When a service outgrows 512 MB, the next honest step is a small VPS rather than a bigger box under the desk; our &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;VPS comparison&lt;/a&gt; shows what 4 to 6 EUR a month buys today.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. HolyArrs: a 15 MB self-hosted media stack
&lt;/h2&gt;

&lt;p&gt;Original: &lt;a href="https://github.com/HolyArrs/HolyArrs" rel="noopener noreferrer"&gt;HolyArrs on GitHub&lt;/a&gt;, posted as "Show HN: A 15 MB self-hosted Netflix and Spotify for your media server" (about 160 points and 125 comments on &lt;a href="https://news.ycombinator.com/item?id=50439000" rel="noopener noreferrer"&gt;Hacker News&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;The project replaces six separate media apps with one binary: about 15 MB to download, about 140 MB of RAM idle with 10,000 titles, builds for amd64, arm64 and armv7, AGPL-3.0.&lt;/p&gt;

&lt;p&gt;Our take: after the &lt;a href="https://news.rozoom.app/self-hosting/plex-exploited-hardening-checklist/" rel="noopener noreferrer"&gt;Plex exploitation&lt;/a&gt; earlier this season, a smaller attack surface is the selling point people are reacting to. One small binary is easier to audit than a stack of six apps, but a young project has no security track record yet. Note that 15 MB is the download, not the memory footprint: at 140 MB idle it would still fit on the OpenStick above, barely. Whatever you pick, keep it behind a reverse proxy with no directly published port; Docker bypasses ufw for published ports, which we measured in our &lt;a href="https://news.rozoom.app/self-hosting/docker-production-gotchas/" rel="noopener noreferrer"&gt;Docker production gotchas&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. NVIDIA publishes NVKE, a management plane for GPU Kubernetes
&lt;/h2&gt;

&lt;p&gt;Original: &lt;a href="https://github.com/dsx-ai-factory/nvke" rel="noopener noreferrer"&gt;dsx-ai-factory/nvke&lt;/a&gt; (124 points and 30 comments on &lt;a href="https://news.ycombinator.com/item?id=50387738" rel="noopener noreferrer"&gt;Hacker News&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;NVKE is a management plane in which every workload cluster gets its own control plane, etcd and node pools. The README labels it a preview and does not recommend it for production workloads. A local quickstart runs a management plane and a workload cluster on simulated machines, so you can try it without GPU hardware.&lt;/p&gt;

&lt;p&gt;Our take: irrelevant for a single server, relevant as a signal that cluster-per-customer is turning into a packaged product. If you run one or two clusters, keep using k3s and revisit NVKE when the preview label is gone.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we would do this week
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Run the BestEffort check above on any 1.34 to 1.36 cluster.&lt;/li&gt;
&lt;li&gt;If you use GitLab Runner's Kubernetes executor, upgrade it to 19.5.0 or later, as the original report recommends.&lt;/li&gt;
&lt;li&gt;Do not move anything production-grade onto preview software.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/trending-self-hosting-kubernetes-2026-11-10/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>kubernetes</category>
      <category>weekly</category>
    </item>
    <item>
      <title>After the eMudhra CA compromise: audit your own TLS stack in 30 minutes</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Mon, 02 Nov 2026 16:12:32 +0000</pubDate>
      <link>https://dev.to/rozoomnews/after-the-emudhra-ca-compromise-audit-your-own-tls-stack-in-30-minutes-1kc8</link>
      <guid>https://dev.to/rozoomnews/after-the-emudhra-ca-compromise-audit-your-own-tls-stack-in-30-minutes-1kc8</guid>
      <description>&lt;p&gt;On 17 October 2026 a researcher told the root programs that the private keys of eight eMudhra intermediate certificate authorities were compromised, and proved it the hard way: they signed a short message with each of the eight keys and attached the signatures. Anyone can verify those with OpenSSL. The keys had been recoverable since April 2025.&lt;/p&gt;

&lt;p&gt;Browsers were fixed within days. Firefox and Chrome blocked the intermediates on the 17th, Microsoft platforms on the 20th, Apple products on the 22nd. The thing you browse with is, by now, fine.&lt;/p&gt;

&lt;p&gt;The machines you administer are another matter. This is the checklist for those, plus the parts of the story that decide what you should check. Everything about the incident comes from the &lt;a href="https://emudhra.we-have-a-problem.com/" rel="noopener noreferrer"&gt;incident page&lt;/a&gt;, which we re-read the day this went out. We ran the four test endpoints from step 1 on one Ubuntu 26.04 box of ours, and all four failed with a certificate error, which is the result you want. We did not audit every runtime on it, and the steps below are not a record of us doing that.&lt;/p&gt;

&lt;h2&gt;
  
  
  What actually broke
&lt;/h2&gt;

&lt;p&gt;The eight intermediates chain to roots trusted by the Mozilla, Microsoft, Apple and Chrome root programs. Whoever holds one of the private keys can mint a certificate for any domain, and a client that still trusts the intermediate will accept it.&lt;/p&gt;

&lt;p&gt;The cause is worth remembering. The CA generated those keys incorrectly, so they can be recovered from the certificates themselves. That was true from the day the certificates were created, April 2025. Certificate Transparency logs hold no entries for these intermediates in that window, so clients that enforce CT, which means the major browsers, could not have been attacked through them. Clients that do not check CT are another story, and the incident page says plainly that nobody knows whether attackers used the flaw against them.&lt;/p&gt;

&lt;p&gt;The CA revoked the intermediates, and the root programs pushed that revocation through their own channels: OneCRL and CRLite for Mozilla, CRLSets for Chrome. That is how a browser stays protected without a new build. A great many clients never look at any of those channels.&lt;/p&gt;

&lt;h2&gt;
  
  
  Two layers of trust, and only one was hit
&lt;/h2&gt;

&lt;p&gt;Most self-hosting guides lump these together, so it helps to pull them apart.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your certificates are fine.&lt;/strong&gt; If Caddy, Traefik, acme.sh or certbot gets your certificates from Let's Encrypt, nothing about them changed. The eMudhra intermediates are not in that chain, and the ACME protocol and your account keys were not involved. Renewal keeps working.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What your tools trust is the exposed part.&lt;/strong&gt; Every outbound HTTPS request your server makes (&lt;code&gt;git pull&lt;/code&gt;, &lt;code&gt;curl&lt;/code&gt; to an API, a package install, a container pull, a webhook) is only as trustworthy as the certificate store on that machine. If the store still holds the bad intermediates, someone on the network path can pose as GitHub, your package mirror or any API, and the client will accept it without a murmur.&lt;/p&gt;

&lt;p&gt;That is the whole risk, and most of it sits in software that is not a browser.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why "update the OS" is not the whole fix
&lt;/h2&gt;

&lt;p&gt;Two things set browsers apart from everything else here.&lt;/p&gt;

&lt;p&gt;Revocation first. Browsers get these intermediates revoked through OneCRL, CRLite and CRLSets. &lt;code&gt;curl&lt;/code&gt;, &lt;code&gt;git&lt;/code&gt;, Go binaries and most language runtimes use none of those, so for them the only cure is a trust store that no longer contains the intermediates.&lt;/p&gt;

&lt;p&gt;Then Certificate Transparency. Browsers refuse certificates that were never logged. Most non-browser clients do not care. The incident page runs two extra test hosts for this: their certificates were never logged to CT, so a client that connects to them trusts the bad intermediates and does not enforce CT either. That combination is the one an attacker can use without anybody noticing.&lt;/p&gt;

&lt;p&gt;So the job is not one update but one per program that ships its own trust store or inherits one.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 30-minute audit
&lt;/h2&gt;

&lt;p&gt;Run it on every server, CI runner and dev machine you touch. Each step is a command or two.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Update the system CA bundle
&lt;/h3&gt;

&lt;p&gt;Debian and Ubuntu:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;sudo apt update &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; sudo apt upgrade -y ca-certificates
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;RHEL, Fedora, Alma, Rocky:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;sudo dnf upgrade -y ca-certificates
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Alpine:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;sudo apk upgrade ca-certificates
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The fixed package versions are listed on the incident page. For Debian 12 it is &lt;code&gt;20250419~deb12u2&lt;/code&gt;, for Ubuntu 24.04 &lt;code&gt;20260816ubuntu0.24.04.1&lt;/code&gt;, for Alpine &lt;code&gt;20261020-r0&lt;/code&gt;. Compare with &lt;code&gt;dpkg -l ca-certificates&lt;/code&gt; or &lt;code&gt;apk info ca-certificates&lt;/code&gt;, then run the real test. A certificate error means this client does not trust the compromised intermediates. A successful response means you are still exposed:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl https://dv-g3b-emudhra.we-have-a-problem.com/
curl https://dv-g3b-aaa-emudhra.we-have-a-problem.com/
curl https://dv-g3b-noct-emudhra.we-have-a-problem.com/
curl https://dv-g3b-aaa-noct-emudhra.we-have-a-problem.com/
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If you run Java from Debian or Ubuntu packages, do one more thing: the system update does not refresh the Java keystore, because of a long-standing bug in &lt;code&gt;ca-certificates-java&lt;/code&gt;. The incident page's advice is to recreate it as root:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;rm -f /etc/ssl/certs/java/cacerts &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; update-ca-certificates --fresh
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  2. Fix the bundles the system update never touched
&lt;/h3&gt;

&lt;p&gt;Several runtimes carry their own copy and will not notice your &lt;code&gt;ca-certificates&lt;/code&gt; upgrade.&lt;/p&gt;

&lt;p&gt;Python's &lt;code&gt;requests&lt;/code&gt; reads the &lt;code&gt;certifi&lt;/code&gt; bundle, not the system store. The fixed release is &lt;code&gt;certifi 2026.10.19&lt;/code&gt;, and &lt;code&gt;pip&lt;/code&gt; itself is fixed in 26.2.2:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;pip install --upgrade certifi pip
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Node.js ships its own root store rather than reading the system one by default. The patched releases are 26.12.1, 24.22.1 and 22.23.4, so a long-lived Node service needs a newer Node, not a newer &lt;code&gt;ca-certificates&lt;/code&gt;. Go on Linux does read the system store, but the project also shipped fixes in 1.27.3 and 1.26.10, and Rust programs that bundle &lt;code&gt;webpki-roots&lt;/code&gt; need a rebuild against a patched crate.&lt;/p&gt;

&lt;p&gt;When we looked, certbot and containerd were still listed as pending on the incident page. Check the page again before assuming your version is covered.&lt;/p&gt;

&lt;p&gt;Whatever reads the system store still has to be restarted. A service that started before the update keeps the old store in memory until you restart it.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Pull or rebuild your container images
&lt;/h3&gt;

&lt;p&gt;An image freezes its CA bundle at build time. If it was built before the distro shipped the removal, it trusts the intermediates no matter how patched the host is.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker pull debian:bookworm   &lt;span class="c"&gt;# or your base image&lt;/span&gt;
docker compose build --pull   &lt;span class="c"&gt;# if you build your own&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The part that bites: an old tag never gets rebuilt. If your Dockerfile still says &lt;code&gt;node:20&lt;/code&gt; or &lt;code&gt;golang:1.23&lt;/code&gt; or &lt;code&gt;alpine:3.19&lt;/code&gt;, those lines have reached end of life, and pulling again returns the same stale bundle. Move to a supported tag. For images you do not build, check that the publisher has released a new tag after 20 October. This is the step most people skip and the one that protects the containers.&lt;/p&gt;

&lt;h3&gt;
  
  
  4. Update Electron apps wherever you develop
&lt;/h3&gt;

&lt;p&gt;Electron apps pick up the fix only when they are rebuilt on a patched Electron release (44.9.1, 43.8.1 and 42.11.16 are the fixed lines). Updating the OS or the browser does nothing for them. Slack, Discord and anything else on that runtime, on your desktop and on CI runners, needs its own update.&lt;/p&gt;

&lt;p&gt;Apple developer tools were a special case: the &lt;code&gt;git&lt;/code&gt;, &lt;code&gt;curl&lt;/code&gt; and &lt;code&gt;python3&lt;/code&gt; that ship with macOS still trusted the compromised certificates as of 28 October, and macOS 27.2 undid the fixes some people had applied by hand. On a Mac, test the tools directly with the curl commands above rather than trusting the OS version.&lt;/p&gt;

&lt;h3&gt;
  
  
  5. Look at the runners your CI really uses
&lt;/h3&gt;

&lt;p&gt;A hosted runner, a self-hosted agent or a build container is a supply-chain target here. If an attacker can pose as the package registry, whatever that runner builds is theirs to modify. Check that the runner image is current and that the pinned base image from step 3 has been refreshed. GitLab runner 19.5.1 is the fixed release if that is your runner.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this does not touch
&lt;/h2&gt;

&lt;p&gt;This topic was trending because of Let's Encrypt's multi-perspective validation, where domain control is checked from several network vantage points so that a BGP hijack on one path cannot fool the CA into issuing for a domain it should not. It is a real defence, and it lives at the issuance layer: it protects the moment a certificate is requested.&lt;/p&gt;

&lt;p&gt;The eMudhra problem sits at the trust layer. A CA's keys were compromised, which has nothing to do with how domains were validated, so multi-perspective validation would not have caught it. The two are easy to confuse. "Let's Encrypt validates from several places" is not a reason to skip the audit above.&lt;/p&gt;

&lt;p&gt;For your own ACME stack it comes down to three things. Your certificates are unaffected and renewals keep running. Your clients are the risk, and only a trust-store update fixes them. And Caddy or Traefik terminating TLS with your own certificate is fine, while the &lt;code&gt;git&lt;/code&gt;, &lt;code&gt;curl&lt;/code&gt;, &lt;code&gt;docker&lt;/code&gt; and language runtimes on the same server are the attack surface.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where to run the audit
&lt;/h2&gt;

&lt;p&gt;The stack you are checking is most likely a small VPS or a container host, and the steps above assume a shell on it. If you are still choosing the machine, we benchmarked the budget tier and wrote it up in &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;the VPS comparison&lt;/a&gt;. The ACME setup that serves this site, Caddy and Docker behind a single hostname, is in &lt;a href="https://news.rozoom.app/self-hosting/caddy-docker-production/" rel="noopener noreferrer"&gt;the production Caddy guide&lt;/a&gt;. The legacy-machine case, where &lt;code&gt;certbot&lt;/code&gt; will not install and you take renewal back with &lt;a href="https://news.rozoom.app/self-hosting/acme-sh-certificates-legacy-server/" rel="noopener noreferrer"&gt;acme.sh&lt;/a&gt;, is the one most self-hosters inherit sooner or later.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/emudhra-ca-compromise-acme-audit/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>tls</category>
      <category>security</category>
      <category>acme</category>
    </item>
    <item>
      <title>Blocking AI crawlers: robots.txt vs Cloudflare vs Anubis, and what each one actually does</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Fri, 30 Oct 2026 17:33:08 +0000</pubDate>
      <link>https://dev.to/rozoomnews/blocking-ai-crawlers-robotstxt-vs-cloudflare-vs-anubis-and-what-each-one-actually-does-4di3</link>
      <guid>https://dev.to/rozoomnews/blocking-ai-crawlers-robotstxt-vs-cloudflare-vs-anubis-and-what-each-one-actually-does-4di3</guid>
      <description>&lt;p&gt;A reader asked this under our analytics article, and it is the right question to ask in that order: &lt;code&gt;robots.txt&lt;/code&gt;, Cloudflare, Anubis. The three get discussed as alternatives. They are not. One publishes a preference, one enforces a decision at the edge, one makes the client pay for the page. Choosing between them means deciding which of those three things you actually want.&lt;/p&gt;

&lt;p&gt;We have fifteen days of access log from a site that went live in October, 9,996 lines of it, and a throwaway Anubis stack on a laptop. Everything below with a number attached comes from one of those two. Where we have no numbers of our own, it says so.&lt;/p&gt;

&lt;h2&gt;
  
  
  First, who even asks
&lt;/h2&gt;

&lt;p&gt;Before picking a tool, look at what the polite protocol is worth on your own log. Of 9,996 requests, 334 were for &lt;code&gt;robots.txt&lt;/code&gt;. Here is who sent them, against how much each crawler took:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Crawler&lt;/th&gt;
&lt;th&gt;Requests&lt;/th&gt;
&lt;th&gt;
&lt;code&gt;robots.txt&lt;/code&gt; fetches&lt;/th&gt;
&lt;th&gt;Distinct URLs&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;ClaudeBot&lt;/td&gt;
&lt;td&gt;484&lt;/td&gt;
&lt;td&gt;78&lt;/td&gt;
&lt;td&gt;330&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Googlebot&lt;/td&gt;
&lt;td&gt;541&lt;/td&gt;
&lt;td&gt;17&lt;/td&gt;
&lt;td&gt;333&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GoogleOther&lt;/td&gt;
&lt;td&gt;520&lt;/td&gt;
&lt;td&gt;0&lt;/td&gt;
&lt;td&gt;327&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPTBot&lt;/td&gt;
&lt;td&gt;391&lt;/td&gt;
&lt;td&gt;0&lt;/td&gt;
&lt;td&gt;387&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;meta-externalagent&lt;/td&gt;
&lt;td&gt;322&lt;/td&gt;
&lt;td&gt;0&lt;/td&gt;
&lt;td&gt;315&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AhrefsBot&lt;/td&gt;
&lt;td&gt;202&lt;/td&gt;
&lt;td&gt;21&lt;/td&gt;
&lt;td&gt;129&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;bingbot&lt;/td&gt;
&lt;td&gt;164&lt;/td&gt;
&lt;td&gt;18&lt;/td&gt;
&lt;td&gt;67&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OAI-SearchBot&lt;/td&gt;
&lt;td&gt;108&lt;/td&gt;
&lt;td&gt;35&lt;/td&gt;
&lt;td&gt;61&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Amazonbot&lt;/td&gt;
&lt;td&gt;96&lt;/td&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;td&gt;94&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;PerplexityBot&lt;/td&gt;
&lt;td&gt;63&lt;/td&gt;
&lt;td&gt;9&lt;/td&gt;
&lt;td&gt;21&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;YandexBot&lt;/td&gt;
&lt;td&gt;49&lt;/td&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;td&gt;38&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Applebot&lt;/td&gt;
&lt;td&gt;30&lt;/td&gt;
&lt;td&gt;5&lt;/td&gt;
&lt;td&gt;22&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The two zeroes in the middle of that table are the interesting part, and they are not what they look like at first glance.&lt;/p&gt;

&lt;p&gt;GPTBot fetched our &lt;code&gt;robots.txt&lt;/code&gt; zero times in 391 requests. OAI-SearchBot fetched it 35 times - and the two share IP ranges. The same applies at Meta: meta-externalagent, the crawler that collects training data, never asked for the file, while &lt;code&gt;facebookexternalhit&lt;/code&gt;, the link-preview fetcher, asked eight times. Google does it too: Googlebot fetched it 17 times, GoogleOther not once, and both run on the same infrastructure.&lt;/p&gt;

&lt;p&gt;So the honest reading is not "the training crawlers ignore &lt;code&gt;robots.txt&lt;/code&gt;". It is that &lt;strong&gt;the fetch happens somewhere else in the vendor's fleet, and the crawler you see in your log is downstream of a cache you cannot observe&lt;/strong&gt;. Which has a practical consequence: your log cannot tell you whether a &lt;code&gt;robots.txt&lt;/code&gt; rule was ever seen, let alone obeyed. If you want to know if a rule works, the only test is whether the disallowed paths stop being requested.&lt;/p&gt;

&lt;p&gt;Anthropic is the outlier in the other direction. 78 fetches across 484 requests is one &lt;code&gt;robots.txt&lt;/code&gt; check for every six content requests, and the first two requests ClaudeBot ever made to us were &lt;code&gt;/robots.txt&lt;/code&gt; and then &lt;code&gt;/&lt;/code&gt;, one second apart.&lt;/p&gt;

&lt;h2&gt;
  
  
  robots.txt: a request, with a thin layer of law behind it
&lt;/h2&gt;

&lt;p&gt;What it costs: nothing. What it does: nothing, mechanically. A disallowed path is still served to anyone who asks.&lt;/p&gt;

&lt;p&gt;What changed recently is the paperwork behind it. Article 53(1)(c) of the EU AI Act requires providers of general-purpose AI models to put in place a policy to identify and respect reservations of rights expressed under Article 4(3) of the copyright directive, "including through state of the art technologies". The accompanying &lt;a href="https://www.editionmultimedia.fr/wp-content/uploads/2025/07/Code-of-Practice-for-GPAI-Copyright-10-07-25.pdf" rel="noopener noreferrer"&gt;Code of Practice&lt;/a&gt; spells one of those technologies out: signatories commit to honour &lt;code&gt;robots.txt&lt;/code&gt; in line with RFC 9309.&lt;/p&gt;

&lt;p&gt;Read the scope carefully before you rely on it. That obligation lands on providers of general-purpose models, and the Code is voluntary and signed. A scraper that is not a model provider is outside it, and what counts as a valid reservation of rights is still a question for the courts. The &lt;a href="https://digital-strategy.ec.europa.eu/en/faqs/stakeholder-consultation-ai-and-copyright-compliance" rel="noopener noreferrer"&gt;Commission's own FAQ&lt;/a&gt; lists several other machine-readable mechanisms beside &lt;code&gt;robots.txt&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The practical value is real but narrow: a &lt;code&gt;robots.txt&lt;/code&gt; with explicit AI directives is cheap, and it is the artefact you point at if you ever need to show that you expressed a reservation. It is evidence, not a wall.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cloudflare: enforcement, if your DNS is already there
&lt;/h2&gt;

&lt;p&gt;This is the section where we have no first-party numbers, and we would rather say that than dress up the documentation. Our own blog does not sit behind Cloudflare, so nothing below was measured by us.&lt;/p&gt;

&lt;p&gt;From &lt;a href="https://developers.cloudflare.com/ai-crawl-control/" rel="noopener noreferrer"&gt;Cloudflare's docs&lt;/a&gt;, AI Crawl Control shows which AI services are crawling you, lets you set allow or block rules per crawler, tracks &lt;code&gt;robots.txt&lt;/code&gt; compliance, and exposes a pay-per-crawl option that is in private beta. The &lt;a href="https://developers.cloudflare.com/bots/additional-configurations/managed-robots-txt/" rel="noopener noreferrer"&gt;managed robots.txt&lt;/a&gt; feature generates the directives from your chosen policies and prepends them to whatever your origin already serves. Both are documented as available on all plans.&lt;/p&gt;

&lt;p&gt;The part that matters architecturally: this is the only one of the three options that blocks a request before it reaches your server, and the only one where somebody else maintains the list of who is who. That is the whole trade. You get enforcement without running anything, and in exchange the decision about what counts as an AI crawler is made by a third party, in front of your site, for every visitor.&lt;/p&gt;

&lt;h2&gt;
  
  
  Anubis: make the client pay
&lt;/h2&gt;

&lt;p&gt;Anubis is a reverse proxy that puts a proof-of-work challenge in front of the origin. No identity claim is involved: a client that cannot execute the challenge JavaScript does not get the page, whatever its user agent says.&lt;/p&gt;

&lt;p&gt;We ran v1.28.2 in front of an nginx origin on a laptop and probed it. With the bundled default policy:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Client&lt;/th&gt;
&lt;th&gt;Response&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;curl/8.18.0&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;200, straight through to the origin, 53 bytes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPTBot user agent&lt;/td&gt;
&lt;td&gt;deny page, 2,270 bytes, served with HTTP 200&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ClaudeBot user agent&lt;/td&gt;
&lt;td&gt;deny page, served with HTTP 200&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Googlebot user agent&lt;/td&gt;
&lt;td&gt;challenge page, ~7.1 KB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Desktop Chrome user agent&lt;/td&gt;
&lt;td&gt;challenge page, ~7.1 KB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Three things in that table are worth knowing before you deploy it.&lt;/p&gt;

&lt;p&gt;The deny page comes back with &lt;strong&gt;HTTP 200&lt;/strong&gt;, not 403. If you monitor your site by status code, a fully blocked crawler looks like a healthy one.&lt;/p&gt;

&lt;p&gt;A plain, honest non-browser client passes through untouched. That is deliberate - it keeps &lt;code&gt;curl&lt;/code&gt;, feed readers and your own monitoring working - but it also means Anubis is not a wall against scraping in general. Anything that does not pretend to be a browser is waved past by default.&lt;/p&gt;

&lt;p&gt;And the Googlebot row is a lab artefact worth flagging: Anubis verifies search engines by reverse DNS, which cannot succeed for a request arriving from &lt;code&gt;127.0.0.1&lt;/code&gt;. On a real deployment that row would be a pass, not a challenge. We are showing it because it is the kind of result that looks like a finding and is not one.&lt;/p&gt;

&lt;h3&gt;
  
  
  What the challenge costs the human
&lt;/h3&gt;

&lt;p&gt;This is the number that decides whether you can use Anubis on a site with real visitors. We drove the challenge with a headless Chromium and timed from navigation to the origin page rendering:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Difficulty&lt;/th&gt;
&lt;th&gt;Median solve&lt;/th&gt;
&lt;th&gt;Runs&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;2 (bundled default)&lt;/td&gt;
&lt;td&gt;0.08 s&lt;/td&gt;
&lt;td&gt;5&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;td&gt;0.27 s&lt;/td&gt;
&lt;td&gt;5&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;6&lt;/td&gt;
&lt;td&gt;50 to 75 s&lt;/td&gt;
&lt;td&gt;2 of 3&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;At the shipped default the gate is invisible. Two steps up it is a minute of a spinning page, because each level multiplies the work by sixteen. The third difficulty-6 run never finished inside our 90 s timeout at all, which is the honest shape of the tail: at that setting the cost is not just high, it is unbounded enough that some visitors simply give up. If you turn that dial because the default "feels low", you will not notice what you did - your own browser caches the clearance cookie for a week - while every first-time visitor pays.&lt;/p&gt;

&lt;p&gt;Two deployment traps we hit, both costing more time than the measurement itself:&lt;/p&gt;

&lt;p&gt;The &lt;code&gt;DIFFICULTY&lt;/code&gt; environment variable did not change the challenge. In v1.28.2 the startup log reports it, the served challenge ignores it, and the number that counts comes from the policy file. We only caught it by reading the difficulty out of the challenge page's own embedded JSON.&lt;/p&gt;

&lt;p&gt;Without &lt;code&gt;USE_REMOTE_ADDRESS=true&lt;/code&gt;, every single request returned HTTP 500, with &lt;code&gt;[misconfiguration] X-Real-IP header is not set&lt;/code&gt; in the log. Anubis assumes something in front of it sets that header. If nothing does, it fails closed on all traffic, not just bots.&lt;/p&gt;

&lt;h2&gt;
  
  
  Can you even tell who is who
&lt;/h2&gt;

&lt;p&gt;All three approaches eventually rest on identification, so we checked how well that works now. Every crawler request in our log, matched against the vendor's own published list or reverse DNS:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Crawler&lt;/th&gt;
&lt;th&gt;Requests&lt;/th&gt;
&lt;th&gt;Verified&lt;/th&gt;
&lt;th&gt;Against&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;ClaudeBot&lt;/td&gt;
&lt;td&gt;484&lt;/td&gt;
&lt;td&gt;484 / 484&lt;/td&gt;
&lt;td&gt;&lt;code&gt;claude.com/crawling/bots.json&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPTBot&lt;/td&gt;
&lt;td&gt;391&lt;/td&gt;
&lt;td&gt;391 / 391&lt;/td&gt;
&lt;td&gt;&lt;code&gt;openai.com/gptbot.json&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OAI-SearchBot&lt;/td&gt;
&lt;td&gt;108&lt;/td&gt;
&lt;td&gt;108 / 108&lt;/td&gt;
&lt;td&gt;&lt;code&gt;openai.com/searchbot.json&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;PerplexityBot&lt;/td&gt;
&lt;td&gt;63&lt;/td&gt;
&lt;td&gt;63 / 63&lt;/td&gt;
&lt;td&gt;&lt;code&gt;perplexity.ai/perplexitybot.json&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ChatGPT-User&lt;/td&gt;
&lt;td&gt;39&lt;/td&gt;
&lt;td&gt;37 / 39&lt;/td&gt;
&lt;td&gt;&lt;code&gt;openai.com/chatgpt-user.json&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Googlebot&lt;/td&gt;
&lt;td&gt;541&lt;/td&gt;
&lt;td&gt;541 / 541&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GoogleOther&lt;/td&gt;
&lt;td&gt;520&lt;/td&gt;
&lt;td&gt;520 / 520&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;bingbot&lt;/td&gt;
&lt;td&gt;164&lt;/td&gt;
&lt;td&gt;164 / 164&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Amazonbot&lt;/td&gt;
&lt;td&gt;96&lt;/td&gt;
&lt;td&gt;96 / 96&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;YandexBot&lt;/td&gt;
&lt;td&gt;49&lt;/td&gt;
&lt;td&gt;49 / 49&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Applebot&lt;/td&gt;
&lt;td&gt;30&lt;/td&gt;
&lt;td&gt;30 / 30&lt;/td&gt;
&lt;td&gt;reverse DNS&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;meta-externalagent&lt;/td&gt;
&lt;td&gt;322&lt;/td&gt;
&lt;td&gt;322 / 322&lt;/td&gt;
&lt;td&gt;AS32934 announced prefixes&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Better than the folklore suggests. The major AI crawlers now publish IP lists you can fetch and check, and ours matched to the request.&lt;/p&gt;

&lt;p&gt;Two exceptions. ChatGPT-User sent two requests from addresses outside the published list, which is the one user-triggered agent in the table and the one where a mismatch is least alarming. And meta-externalagent has no reverse DNS and no published crawler list we could find, so the only check available is whether the address sits in Meta's announced network prefixes, which is a much weaker claim: it says the packet came from that company's network, not that the crawler is the one it claims to be.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pacing, when identity is not available
&lt;/h2&gt;

&lt;p&gt;For anything that neither verifies nor identifies, the behaviour still gives it away. The interval between consecutive requests from the same crawler, over fifteen days:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Crawler&lt;/th&gt;
&lt;th&gt;Median gap&lt;/th&gt;
&lt;th&gt;Share of gaps under 1 s&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;GPTBot&lt;/td&gt;
&lt;td&gt;1 s&lt;/td&gt;
&lt;td&gt;92.1%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;meta-externalagent&lt;/td&gt;
&lt;td&gt;1 s&lt;/td&gt;
&lt;td&gt;88.2%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;YandexBot&lt;/td&gt;
&lt;td&gt;1 s&lt;/td&gt;
&lt;td&gt;75.0%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ClaudeBot&lt;/td&gt;
&lt;td&gt;1 s&lt;/td&gt;
&lt;td&gt;52.0%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AhrefsBot&lt;/td&gt;
&lt;td&gt;29 s&lt;/td&gt;
&lt;td&gt;31.8%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Googlebot&lt;/td&gt;
&lt;td&gt;158 s&lt;/td&gt;
&lt;td&gt;8.0%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GoogleOther&lt;/td&gt;
&lt;td&gt;137 s&lt;/td&gt;
&lt;td&gt;0.8%&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;That is a clean split, and it is not about politeness. Google fetched more pages than anyone (541 and 520 URLs) and did it with a median gap of over two minutes, because crawl pacing is a solved problem it has had decades to solve. The training crawlers took the site in one pass, as fast as it would serve.&lt;/p&gt;

&lt;p&gt;The practical use: on a quiet site, a histogram of inter-arrival gaps separates a crawl from an audience without reading a single user agent. A spike concentrated in one sub-second bucket is a machine. Humans do produce tight bursts - tab restores, prefetch - but they do not sustain them for 390 requests.&lt;/p&gt;

&lt;h2&gt;
  
  
  So what do you choose
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;If you want&lt;/th&gt;
&lt;th&gt;Use&lt;/th&gt;
&lt;th&gt;Cost&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;To state a preference and have the artefact on record&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;robots.txt&lt;/code&gt;, with explicit AI directives&lt;/td&gt;
&lt;td&gt;Nothing&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;To stop requests before your server sees them, without running anything&lt;/td&gt;
&lt;td&gt;Cloudflare AI Crawl Control&lt;/td&gt;
&lt;td&gt;Your DNS goes through them, and their list defines who is a bot&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;To make content unavailable to anything that cannot run a browser&lt;/td&gt;
&lt;td&gt;Anubis in front of the origin&lt;/td&gt;
&lt;td&gt;A gate every first-time human passes too&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;To know what is actually happening&lt;/td&gt;
&lt;td&gt;Your access log, parsed&lt;/td&gt;
&lt;td&gt;An afternoon, once&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;They stack, and that is the sane configuration: &lt;code&gt;robots.txt&lt;/code&gt; says what you want, one of the other two enforces whatever part of it you care about enough to enforce, and the log tells you whether any of it worked.&lt;/p&gt;

&lt;p&gt;What we run on this site: &lt;code&gt;robots.txt&lt;/code&gt;, and nothing else. We publish articles to be read, including by whatever summarises them later. The reason we built the Anubis stack at all was to be able to put a number on the alternative, and the number says the tool is fine and the difficulty knob is the hazard.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this cannot tell you
&lt;/h2&gt;

&lt;p&gt;One site, one niche, one fifteen-day window, 9,996 requests. A site with real commercial content attracts a different class of scraper, the kind that does not announce itself and does not appear in any of these tables, and against that population every number above is irrelevant.&lt;/p&gt;

&lt;p&gt;The Anubis measurements come from a laptop against localhost: no network latency, one client, and a CPU that is faster than the median phone. Treat the difficulty curve as the shape of the cost, not its absolute value.&lt;/p&gt;

&lt;p&gt;And the hole we cannot close with logs at all: fetching &lt;code&gt;robots.txt&lt;/code&gt; is not obeying it, and not fetching it is not ignoring it. Everything in the first table is a measurement of requests, not of consent.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/blocking-ai-crawlers/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>aicrawlers</category>
      <category>robotstxt</category>
      <category>cloudflare</category>
      <category>anubis</category>
    </item>
    <item>
      <title>Getting a brand new site crawled: who actually shows up, and in what order</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Fri, 30 Oct 2026 16:39:22 +0000</pubDate>
      <link>https://dev.to/rozoomnews/getting-a-brand-new-site-crawled-who-actually-shows-up-and-in-what-order-4fc</link>
      <guid>https://dev.to/rozoomnews/getting-a-brand-new-site-crawled-who-actually-shows-up-and-in-what-order-4fc</guid>
      <description>&lt;p&gt;We put a static site online on 15 October, published nothing for four days, then published three articles in one evening. The access log was recording the whole time. Fifteen days later it holds 9,987 requests, and the order in which the internet noticed us is not the order anybody plans for.&lt;/p&gt;

&lt;p&gt;No theory below. Every number is from that log.&lt;/p&gt;

&lt;h2&gt;
  
  
  The first visitor was an AI crawler, 65 minutes in
&lt;/h2&gt;

&lt;p&gt;The first page went live at 13:14 UTC. The log of what happened next, in order of first appearance:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Crawler&lt;/th&gt;
&lt;th&gt;First seen&lt;/th&gt;
&lt;th&gt;Requests, week 1&lt;/th&gt;
&lt;th&gt;Requests, 15 days&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;ClaudeBot&lt;/td&gt;
&lt;td&gt;15.10, 65 min after launch&lt;/td&gt;
&lt;td&gt;226&lt;/td&gt;
&lt;td&gt;484&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPTBot&lt;/td&gt;
&lt;td&gt;15.10, 70 min after launch&lt;/td&gt;
&lt;td&gt;210&lt;/td&gt;
&lt;td&gt;391&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OAI-SearchBot&lt;/td&gt;
&lt;td&gt;15.10, 70 min after launch&lt;/td&gt;
&lt;td&gt;33&lt;/td&gt;
&lt;td&gt;108&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;bingbot&lt;/td&gt;
&lt;td&gt;16.10&lt;/td&gt;
&lt;td&gt;70&lt;/td&gt;
&lt;td&gt;164&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Googlebot&lt;/td&gt;
&lt;td&gt;19.10, 19:41&lt;/td&gt;
&lt;td&gt;328&lt;/td&gt;
&lt;td&gt;541&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GoogleOther&lt;/td&gt;
&lt;td&gt;19.10, 23:01&lt;/td&gt;
&lt;td&gt;281&lt;/td&gt;
&lt;td&gt;520&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AhrefsBot&lt;/td&gt;
&lt;td&gt;20.10&lt;/td&gt;
&lt;td&gt;91&lt;/td&gt;
&lt;td&gt;202&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ChatGPT-User&lt;/td&gt;
&lt;td&gt;21.10&lt;/td&gt;
&lt;td&gt;4&lt;/td&gt;
&lt;td&gt;39&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Nothing linked to the site yet. No submission had been made anywhere. Anthropic's and OpenAI's crawlers found a brand new hostname on a small VPS within the hour, and started walking it.&lt;/p&gt;

&lt;p&gt;That is the practical point: &lt;strong&gt;if you publish technical writing today, your first readers are retrieval pipelines, and they arrive before any search engine knows you exist.&lt;/strong&gt; Decide your &lt;code&gt;robots.txt&lt;/code&gt; policy before you publish, not after you notice.&lt;/p&gt;

&lt;p&gt;Note what ClaudeBot did in its very first two requests, one second apart: &lt;code&gt;/robots.txt&lt;/code&gt;, then &lt;code&gt;/&lt;/code&gt;. It has asked for &lt;code&gt;robots.txt&lt;/code&gt; 78 times in fifteen days, once per crawl session. GPTBot has requested it zero times in the same window, while fetching 391 URLs. Same two seconds of effort, opposite behaviour. We will come back to that in a separate piece, because it decides which blocking method is worth deploying.&lt;/p&gt;

&lt;h2&gt;
  
  
  Google ignored us entirely until one step was done
&lt;/h2&gt;

&lt;p&gt;Zero Googlebot requests in the first four days. Not throttled, not slow: &lt;strong&gt;absent&lt;/strong&gt;. The site was reachable, had a sitemap, had &lt;code&gt;robots.txt&lt;/code&gt;, and was linked from an established domain we also run.&lt;/p&gt;

&lt;p&gt;What changed it was verifying the domain in Search Console and submitting the sitemap. From the log, that evening:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;19:35&lt;/strong&gt; - domain verified via a DNS TXT record.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;19:41:31&lt;/strong&gt; - first Googlebot request in the site's history: &lt;code&gt;/robots.txt&lt;/code&gt; and &lt;code&gt;/sitemap.xml&lt;/code&gt;, in the same second.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;19:42 to 19:46&lt;/strong&gt; - Googlebot walks every per-language sitemap in turn.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Six minutes.&lt;/strong&gt; You can argue about whether Search Console is formally required - it is not - but the difference between four days of silence and six minutes is hard to ignore when you are starting from nothing.&lt;/p&gt;

&lt;p&gt;Use a &lt;strong&gt;domain property&lt;/strong&gt; rather than a URL-prefix property if you can add a DNS record: one TXT at the apex covers every subdomain you will ever add, and it survives an http/https or www change that would otherwise split your data across properties.&lt;/p&gt;

&lt;p&gt;Then, three hours and twenty minutes after Googlebot, &lt;strong&gt;GoogleOther&lt;/strong&gt; showed up and has since fetched almost as much as Googlebot itself: 520 requests against 541. It is a separate crawler using the same infrastructure for non-search product work. If you ever decide to restrict Google, know that it is two names, not one.&lt;/p&gt;

&lt;h2&gt;
  
  
  Bing came through IndexNow in days, with no account at all
&lt;/h2&gt;

&lt;p&gt;IndexNow is a simple deal: you host a key file, then POST a list of URLs you changed, and participating engines (Bing, Yandex, Seznam, Naver) are told immediately instead of waiting for a crawl.&lt;/p&gt;

&lt;p&gt;The whole protocol is one request:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -X POST https://api.indexnow.org/indexnow \
  -H &lt;span class="s2"&gt;"Content-Type: application/json; charset=utf-8"&lt;/span&gt; \
  -d &lt;span class="s1"&gt;'{
    "host": "example.com",
    "key": "YOUR-KEY",
    "keyLocation": "https://example.com/YOUR-KEY.txt",
    "urlList": ["https://example.com/new-post/"]
  }'&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The key is a file at your domain root whose name is the key and whose content is the same string. That is the entire verification: if you can put a file there, you own the host.&lt;/p&gt;

&lt;p&gt;A &lt;code&gt;200&lt;/code&gt; means accepted, not indexed. But bingbot showed up on its own the day after our first submission, without an account, a console or a verification step - three days before Googlebot. For a new site that is a meaningful head start, and it costs one line in the deploy pipeline.&lt;/p&gt;

&lt;h2&gt;
  
  
  The index-file trap
&lt;/h2&gt;

&lt;p&gt;Here is the part that nearly fooled us, and it is the most useful paragraph in this post.&lt;/p&gt;

&lt;p&gt;Multilingual Hugo generates a &lt;strong&gt;sitemap index&lt;/strong&gt; at &lt;code&gt;/sitemap.xml&lt;/code&gt; - a list of other sitemaps, one per language - rather than a list of pages. We submitted that, and Search Console reported:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Status: Success. Discovered pages: 0.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Both halves are true and they look contradictory. The index was fetched and parsed correctly, so the status is a success; the index itself contains no page URLs, so the discovered count stays at zero until the children are fetched separately. A quick glance at "Success" and you walk away believing the job is done.&lt;/p&gt;

&lt;p&gt;We submitted the child sitemap for the main language as well:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Status: Success. Discovered pages: 23.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So: &lt;strong&gt;if your sitemap is an index, submit the per-language children too&lt;/strong&gt;, and treat the discovered-pages count, not the status word, as the thing that tells you it worked. The same applies to any generator that emits an index - large WordPress installs with Yoast do, and so does anything paginating its sitemaps past 50,000 URLs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Day 12: the long tail arrives all at once
&lt;/h2&gt;

&lt;p&gt;For eleven days the log had eight or nine distinct crawlers in it. Then, between 26 and 29 October, fourteen new ones appeared within four days:&lt;/p&gt;

&lt;p&gt;ElevenlabsCrawler (448 requests), meta-externalagent (322), MJ12bot (319), AwarioBot (308), SofyaBot (232), SemrushBot (213), Amazonbot (96), PerplexityBot (63), DataForSeoBot (62), YandexBot (49), Applebot (30), archive.org_bot (26), Baiduspider (14), DuckDuckBot (8).&lt;/p&gt;

&lt;p&gt;The pattern is clean enough to plan around. SEO backlink crawlers, brand-monitoring crawlers and the second tier of AI crawlers do not discover sites themselves - they read other people's indexes. Once Google and Bing have you, you are in a feed that resells discovery, and the rest follow within a fortnight whether you wanted them or not.&lt;/p&gt;

&lt;p&gt;What that costs, measured: of 153.4 MB served in fifteen days, &lt;strong&gt;88.8 MB went to user agents that identify themselves as crawlers&lt;/strong&gt;. That is 58% of all bytes, against roughly 4,000 browser-like requests. On a static site served from a tiny VPS, nobody notices. On anything that renders pages per request, that is your capacity plan.&lt;/p&gt;

&lt;h2&gt;
  
  
  The order that actually works
&lt;/h2&gt;

&lt;p&gt;For a new site, in this order, with the reason each step exists:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Be linked from somewhere that already gets crawled.&lt;/strong&gt; One footer link from an existing domain is worth more than any submission form. Discovery is still mostly links.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ship a sitemap and keep &lt;code&gt;robots.txt&lt;/code&gt; honest.&lt;/strong&gt; Point to the sitemap from &lt;code&gt;robots.txt&lt;/code&gt;; make sure neither is left over from a staging config that disallows everything.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Send IndexNow on every publish, from the pipeline.&lt;/strong&gt; Cheap, instant, no account. It covers Bing and friends; it does nothing for Google, which does not participate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Verify the domain in Search Console and submit the sitemap - including the children.&lt;/strong&gt; This is the step that turned four days of nothing into six minutes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Then wait.&lt;/strong&gt; Being crawled is not being indexed, and being indexed is not ranking. The first two are infrastructure and take minutes; the third is content and takes months.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  What we would do differently
&lt;/h2&gt;

&lt;p&gt;Submit the sitemap to Search Console on day one, before any content exists. Verification is a DNS record and takes five minutes; doing it early means the crawler already knows the site when the first article appears, instead of starting the clock after it. We spent four days publishing into a void we could have avoided.&lt;/p&gt;

&lt;p&gt;The IndexNow ping we did fix: it now runs as a step in the deploy workflow, in the same place as the &lt;a href="https://news.rozoom.app/self-hosting/github-actions-rsync-deploy-no-shell/" rel="noopener noreferrer"&gt;post-deploy smoke test&lt;/a&gt;, so a publish that reaches the server always reaches the engines too. Doing it by hand means forgetting it exactly once, on the article that mattered.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this log cannot tell you
&lt;/h2&gt;

&lt;p&gt;It is one site, one niche, one fifteen-day window, on a hostname with no history. Crawl budget for an established domain behaves differently. We also cannot separate "arrived because of Search Console" from "arrived because it was the fourth day" with a sample of one - the six-minute gap is suggestive, not a controlled experiment. And &lt;code&gt;robots.txt&lt;/code&gt; fetch counts measure fetches, not obedience: a crawler that reads the file and ignores it looks identical to a polite one in this data.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/getting-a-new-site-crawled/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>seo</category>
      <category>indexnow</category>
      <category>sitemap</category>
      <category>crawlers</category>
    </item>
    <item>
      <title>Kubernetes Dashboard is archived and Lens is not free: what to use in 2026</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Fri, 30 Oct 2026 16:26:49 +0000</pubDate>
      <link>https://dev.to/rozoomnews/kubernetes-dashboard-is-archived-and-lens-is-not-free-what-to-use-in-2026-11i4</link>
      <guid>https://dev.to/rozoomnews/kubernetes-dashboard-is-archived-and-lens-is-not-free-what-to-use-in-2026-11i4</guid>
      <description>&lt;p&gt;For years there were two answers to "how do I look at my cluster without typing kubectl". You deployed &lt;strong&gt;kubernetes-dashboard&lt;/strong&gt;, or you installed &lt;strong&gt;Lens&lt;/strong&gt;. Both answers expired.&lt;/p&gt;

&lt;p&gt;The Dashboard repository was archived on 21 January 2026. The notice at the top of its README is short:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;This project is now archived and no longer maintained due to lack of active maintainers and contributors.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The same README names its successor: Headlamp, "an easy-to-use and extensible Kubernetes web UI". The last release, 7.14.0, is from October 2025 and still installs fine. It is also a piece of software with a service account in your cluster and nobody shipping security fixes for it.&lt;/p&gt;

&lt;p&gt;Lens did not die, it went up-market. The old &lt;code&gt;k8slens.dev&lt;/code&gt; now redirects to &lt;code&gt;lenshq.io&lt;/code&gt;, where the Personal tier is still free "for students, hobbyists, and engineers at small organizations" - with a footnote that organizations above $10 million in annual revenue or funding need a paid plan after a 30-day evaluation. Paid means $25 per seat per month on an annual subscription, $30 month-to-month, and $50 per seat for Enterprise. Lens Desktop also wants a Lens ID: it activates against an account, with a downloadable activation code as the documented path for air-gapped machines.&lt;/p&gt;

&lt;p&gt;So: one option is unmaintained, the other asks for a credit card and a login. Here is what is actually left, and what each one costs on your machine.&lt;/p&gt;

&lt;h2&gt;
  
  
  The shortlist
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Version&lt;/th&gt;
&lt;th&gt;License&lt;/th&gt;
&lt;th&gt;Who maintains it&lt;/th&gt;
&lt;th&gt;Price&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;k9s&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.51.0&lt;/td&gt;
&lt;td&gt;Apache-2.0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;derailed/k9s&lt;/code&gt;, ~35k stars&lt;/td&gt;
&lt;td&gt;free&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Headlamp&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.46.0&lt;/td&gt;
&lt;td&gt;Apache-2.0&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;kubernetes-sigs&lt;/code&gt;, SIG UI; in the CNCF since 2023 at Sandbox level&lt;/td&gt;
&lt;td&gt;free&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;FreeLens&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;2.0.0&lt;/td&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;freelensapp&lt;/code&gt; community project&lt;/td&gt;
&lt;td&gt;free&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Lens Desktop&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;current&lt;/td&gt;
&lt;td&gt;proprietary&lt;/td&gt;
&lt;td&gt;LensHQ&lt;/td&gt;
&lt;td&gt;free Personal with limits, $25-60/seat/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Dashboard&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;7.14.0&lt;/td&gt;
&lt;td&gt;Apache-2.0&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;archived 21.01.2026&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;free, unmaintained&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two of those are the interesting ones. Headlamp is the project the Dashboard itself points at, it lives under &lt;code&gt;kubernetes-sigs&lt;/code&gt;, and it runs either as a desktop app or inside the cluster behind a browser. FreeLens is the community continuation of OpenLens - the open-source core that Lens built on before the product moved behind an account; the name OpenLens still appears about three thousand times in its source. It ships as a standalone desktop app under MIT, with no account anywhere.&lt;/p&gt;

&lt;p&gt;And k9s, which is not a GUI at all. It is a terminal UI, which turns out to matter more than it sounds.&lt;/p&gt;

&lt;h2&gt;
  
  
  What they cost before they show you anything
&lt;/h2&gt;

&lt;p&gt;We installed all three on one Linux x86-64 machine, pointed them at the same cluster - Kubernetes 1.33.1, 47 pods across all namespaces - and measured.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Download&lt;/th&gt;
&lt;th&gt;On disk&lt;/th&gt;
&lt;th&gt;Processes&lt;/th&gt;
&lt;th&gt;RSS&lt;/th&gt;
&lt;th&gt;PSS&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;k9s&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;41 MB&lt;/td&gt;
&lt;td&gt;127 MB&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;105 MB&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;105 MB&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Headlamp desktop&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;150 MB&lt;/td&gt;
&lt;td&gt;404 MB&lt;/td&gt;
&lt;td&gt;7&lt;/td&gt;
&lt;td&gt;747 MB&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;379 MB&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;FreeLens&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;157 MB&lt;/td&gt;
&lt;td&gt;588 MB&lt;/td&gt;
&lt;td&gt;6&lt;/td&gt;
&lt;td&gt;765 MB&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;428 MB&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two numbers for memory, because for Electron apps only one of them is honest. &lt;strong&gt;RSS&lt;/strong&gt; counts every page resident for every process in the tree, so the shared Chromium libraries get counted six or seven times over. &lt;strong&gt;PSS&lt;/strong&gt; divides shared pages among the processes that use them, which is the number that actually reflects what the app takes away from the rest of your machine. For k9s the two are identical, because k9s is one process.&lt;/p&gt;

&lt;p&gt;Read the PSS column again: the two desktop apps cost &lt;strong&gt;380-430 MB before they display a single pod&lt;/strong&gt;. Those measurements were taken at the launch screen - Headlamp on its cluster list, FreeLens on its welcome page - with the kubeconfig loaded and nothing opened yet. k9s was measured with the full pod list of all 47 pods on screen and refreshing.&lt;/p&gt;

&lt;p&gt;That is the Electron baseline, and it is not a criticism of either project: it is the cost of shipping a browser engine with your application. It is simply a cost you should know about before you leave one open on a laptop all day next to your IDE.&lt;/p&gt;

&lt;h2&gt;
  
  
  The same tool, in the cluster, for 28 MB
&lt;/h2&gt;

&lt;p&gt;Headlamp has a second mode that the desktop comparison hides. The in-cluster deployment runs &lt;code&gt;headlamp-server&lt;/code&gt; as a normal pod and serves the identical UI to a browser:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Image&lt;/th&gt;
&lt;th&gt;Container RSS&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Headlamp in-cluster&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;259 MB&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;28.7 MB&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Twenty-nine megabytes, measured on the node while the deployment was running. The same project that costs 380 MB of PSS as a desktop app costs 29 MB of cluster memory as a server, and then every person on the team opens it in a tab they already have running.&lt;/p&gt;

&lt;p&gt;The trade is authentication. The desktop app reuses the kubeconfig that is already on your laptop. The in-cluster deployment needs a service account token - which means you decide what that account is allowed to touch, and you need a way to reach the service. That is a half hour of RBAC and ingress work you do not owe the desktop app. For a team it is a half hour well spent; for one person on one cluster it is pure overhead.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to choose
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;You live in a terminal, or your clusters are behind a bastion.&lt;/strong&gt; k9s. One static binary, one process, 105 MB, and it works over SSH where no desktop app can follow. It is also the only one here that costs nothing extra when you are already in a shell.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;You want a browser UI for a team.&lt;/strong&gt; Headlamp in the cluster. It is the project the Dashboard handed off to, it is in &lt;code&gt;kubernetes-sigs&lt;/code&gt;, and at 29 MB per cluster the resource argument is over before it starts. Budget the RBAC work.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;You want the Lens workflow and you are done paying for it.&lt;/strong&gt; FreeLens. MIT, no account, no activation, releases shipping steadily - 2.0.0 landed two days before we wrote this. It is the heaviest thing in the table, but it is heavy the way Lens was heavy, and that is what people are coming for.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;You are already a Lens customer with SSO and team features.&lt;/strong&gt; Stay. Nothing here replaces the paid feature set; the free tiers replace the free tier.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;You were about to deploy kubernetes-dashboard.&lt;/strong&gt; Don't. Not because it stopped working - because it is an unmaintained web UI holding a service account in your cluster, and that is a very specific kind of debt.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we did not measure
&lt;/h2&gt;

&lt;p&gt;Honest limits on the table above:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No startup timings.&lt;/strong&gt; Our test cluster flapped between Ready and NotReady during the session, which makes every time-to-first-render number meaningless. We dropped them rather than publish numbers we could not reproduce.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No long-run memory growth.&lt;/strong&gt; These are footprints at launch, not after eight hours with twenty tabs of logs open. The desktop apps will grow; how much is a different article.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Linux only.&lt;/strong&gt; macOS and Windows builds of the same versions are packaged differently and will not match these numbers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;We did not test Lens Desktop itself.&lt;/strong&gt; It requires an account to activate, and we were not going to create one to produce a memory figure. Its row above is its licensing, taken from its own pricing page.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;One disclosure, since this site belongs to a product in the same space: we build ROZOOM, a desktop Kubernetes IDE. It is not in the comparison table because the honest measurement we tried did not render in our headless test rig, and a number we cannot stand behind is worse than no number at all. If you want to look at it, it is at &lt;a href="https://rozoom.app/" rel="noopener noreferrer"&gt;rozoom.app&lt;/a&gt; - and the table above is the one we would hand a colleague regardless.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Is kubernetes-dashboard dead?&lt;/strong&gt;&lt;br&gt;
The repository is archived and read-only since 21 January 2026, and the maintainers cite a lack of active maintainers and contributors. Release 7.14.0 still installs and runs. It will not get security fixes, which is the part that matters for something holding cluster credentials.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What replaced the Kubernetes Dashboard?&lt;/strong&gt;&lt;br&gt;
Headlamp, by the Dashboard's own README. It lives in &lt;code&gt;kubernetes-sigs&lt;/code&gt; under SIG UI and joined the CNCF in 2023 at Sandbox level. It runs both as a desktop app and in-cluster behind a browser.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Is Lens still free?&lt;/strong&gt;&lt;br&gt;
There is still a free Personal tier, described as being for students, hobbyists and engineers at small organizations, with a footnote excluding organizations above $10 million in annual revenue or funding beyond a 30-day evaluation. Paid plans start at $25 per seat per month on annual billing. Lens Desktop activates against a Lens ID account.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What is FreeLens?&lt;/strong&gt;&lt;br&gt;
A free, MIT-licensed desktop Kubernetes UI, built by a community project on the OpenLens codebase - the open-source core Lens was built on. No account, no activation. Version 2.0.0 was released on 28 October 2026.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which Kubernetes GUI uses the least memory?&lt;/strong&gt;&lt;br&gt;
Of the three we measured, k9s at 105 MB - and it is a terminal UI, not a GUI. Among the graphical options, Headlamp deployed in-cluster is the lightest by an order of magnitude: 29 MB on the node, with the UI rendered by a browser you already have open.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Can I just use kubectl?&lt;/strong&gt;&lt;br&gt;
Yes, and for anything scripted you should. The UIs earn their keep for the other job: scanning thirty namespaces for the one pod that is unhappy, following logs across restarts, and giving people who do not live in a terminal a way to look without a cheat sheet.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's next
&lt;/h2&gt;

&lt;p&gt;If you are picking a UI, you are probably also picking what it connects to. The companion pieces: the &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;VPS comparison&lt;/a&gt; for the box underneath, and &lt;a href="https://news.rozoom.app/self-hosting/docker-production-gotchas/" rel="noopener noreferrer"&gt;Docker in production&lt;/a&gt; for the layer below that.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/kubernetes/kubernetes-gui-2026/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>kubernetes</category>
      <category>k9s</category>
      <category>headlamp</category>
      <category>freelens</category>
    </item>
    <item>
      <title>Plex Is Being Exploited. The Fix Shipped Eleven Weeks Earlier.</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 07:20:28 +0000</pubDate>
      <link>https://dev.to/rozoomnews/plex-is-being-exploited-the-fix-shipped-eleven-weeks-earlier-14b1</link>
      <guid>https://dev.to/rozoomnews/plex-is-being-exploited-the-fix-shipped-eleven-weeks-earlier-14b1</guid>
      <description>&lt;p&gt;A media server is the most forgivable thing to expose to the internet. It is also, this month, the most expensive.&lt;/p&gt;

&lt;p&gt;On 20 October 2026 CISA added four Plex Media Server vulnerabilities to its Known Exploited Vulnerabilities catalogue at once. Chained, they let an attacker with no credentials obtain an admin token and execute code as the Plex process. The researcher who found them scores the chain 9.8.&lt;/p&gt;

&lt;p&gt;The patched build, 1.43.3.10861, was released on &lt;strong&gt;4 August&lt;/strong&gt;. Everything that has happened since is about the gap between a fix existing and a fix being installed, and that gap is the only part of this story you control.&lt;/p&gt;

&lt;h2&gt;
  
  
  The timeline, because it is the argument
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Date&lt;/th&gt;
&lt;th&gt;Event&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;27 July 2026&lt;/td&gt;
&lt;td&gt;Researcher Zach Main reports the chain to Plex&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;3-4 August 2026&lt;/td&gt;
&lt;td&gt;Plex Media Server 1.43.3.10861 ships with the fix&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Early September 2026&lt;/td&gt;
&lt;td&gt;Shadowserver counts over 36,000 internet-exposed servers still unpatched&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;16 October 2026&lt;/td&gt;
&lt;td&gt;Full technical write-up published&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;17 October 2026&lt;/td&gt;
&lt;td&gt;First in-the-wild exploitation observed, against honeypots, dropping a cryptominer&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;19 October 2026&lt;/td&gt;
&lt;td&gt;Public proof-of-concept exploit on GitHub&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;20 October 2026&lt;/td&gt;
&lt;td&gt;CISA adds all four CVEs to the KEV catalogue&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;23 October 2026&lt;/td&gt;
&lt;td&gt;Federal remediation deadline for the two worst of them&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Eleven weeks of warning. Thirty-six thousand servers that did not use them.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the chain actually is
&lt;/h2&gt;

&lt;p&gt;Four issues, each unremarkable alone, which is exactly why they sat unnoticed:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-96652&lt;/strong&gt; (CVSS 5.3), server-side request forgery: Plex can be made to send POST requests to arbitrary destinations, including its own internal APIs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-96654&lt;/strong&gt; (CVSS 6.9), URL injection: attacker-controlled parameters reach other plugins' functions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-96655&lt;/strong&gt; (CVSS 5.3), another SSRF, this one in the transcoder endpoint.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-96656&lt;/strong&gt; (CVSS 8.6), arbitrary file write in the transcoder settings: a shared library can be planted where Plex will load it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Read the last one again, because it is the pattern worth carrying to every other service you run: &lt;strong&gt;a process that writes files, and later loads code from the directory it writes to, is one file-write bug away from code execution.&lt;/strong&gt; Codecs, plugins, themes, extensions, uploads with an include path. The individual scores are medium. The chain is 9.8.&lt;/p&gt;

&lt;p&gt;Exploitation needs network access to TCP port 32400. Plex's Remote Access feature, or a manual port forward, is what turns a living-room service into an internet-facing one.&lt;/p&gt;

&lt;h2&gt;
  
  
  Do these three things first
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;1. Check the full build number, not the marketing version.&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Debian or Ubuntu package install&lt;/span&gt;
dpkg -s plexmediaserver &lt;span class="o"&gt;|&lt;/span&gt; grep &lt;span class="s1"&gt;'^Version'&lt;/span&gt;
&lt;span class="c"&gt;# official Docker image&lt;/span&gt;
docker &lt;span class="nb"&gt;exec&lt;/span&gt; plex dpkg -s plexmediaserver &lt;span class="o"&gt;|&lt;/span&gt; grep &lt;span class="s1"&gt;'^Version'&lt;/span&gt;
&lt;span class="c"&gt;# any install, from the box itself: the identity endpoint answers without a token&lt;/span&gt;
curl -s http://127.0.0.1:32400/identity
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Anything below &lt;code&gt;1.43.3.10861&lt;/code&gt; is affected. An early 1.43.3 build is still affected: the last five digits are the ones that matter.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Update, then confirm the running process is the new one.&lt;/strong&gt; A container that was never re-pulled, or a package held back by a pin, will happily report a new version in the UI while the old binary is still serving.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker compose pull plex &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; docker compose up -d plex
docker &lt;span class="nb"&gt;exec&lt;/span&gt; plex dpkg -s plexmediaserver &lt;span class="o"&gt;|&lt;/span&gt; grep &lt;span class="s1"&gt;'^Version'&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;3. If the server was internet-reachable on an old build, assume it was reached.&lt;/strong&gt; Plex's own advisory says to treat such servers as potentially compromised, and gives indicators: unfamiliar &lt;code&gt;.so&lt;/code&gt; or &lt;code&gt;.dll&lt;/code&gt; files in the Codecs directory, and outbound connections to &lt;code&gt;proxystream.st&lt;/code&gt;. Check both before you call it done:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# unexpected shared libraries in the codec directory&lt;/span&gt;
find ~/Library/Application\ Support/Plex\ Media\ Server/Codecs \
     /var/lib/plexmediaserver/Library/Application\ Support/Plex\ Media\ Server/Codecs \
     -name &lt;span class="s1"&gt;'*.so'&lt;/span&gt; -o -name &lt;span class="s1"&gt;'*.dll'&lt;/span&gt; 2&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/dev/null &lt;span class="o"&gt;|&lt;/span&gt; sort

&lt;span class="c"&gt;# anything the box is talking to that you cannot name&lt;/span&gt;
ss -tn state established &lt;span class="o"&gt;|&lt;/span&gt; awk &lt;span class="s1"&gt;'NR&amp;gt;1{print $4}'&lt;/span&gt; &lt;span class="o"&gt;|&lt;/span&gt; sort &lt;span class="o"&gt;|&lt;/span&gt; uniq -c &lt;span class="o"&gt;|&lt;/span&gt; sort -rn &lt;span class="o"&gt;|&lt;/span&gt; head -20
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A cryptominer announces itself in CPU. Proxyware does not: it sells your address as residential bandwidth and stays quiet. Absence of load is not absence of compromise.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checklist that would have saved you anyway
&lt;/h2&gt;

&lt;p&gt;None of the following requires knowing about a CVE in advance. That is the entire point: these controls pay off during the weeks when you have not patched yet, which is every service you run, always.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Do not port-forward a service to the internet to use it yourself.&lt;/strong&gt; Remote Access and a forwarded port exist to serve people who are not on your network. If the only remote user is you, a WireGuard tunnel or a mesh VPN removes the attack surface completely rather than defending it. Exposure is a decision, and most of the time nobody ever makes it deliberately.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Default-deny inbound, with the open ports written down.&lt;/strong&gt; On a box that runs several things, the firewall is the only place where "what is reachable" is answerable in one screen:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ufw default deny incoming
ufw allow 443/tcp
ufw status numbered
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Read the output top to bottom and ask what matches first. On a legacy box we audited this month, a careful three-subnet allowlist sat four lines below a blanket &lt;code&gt;ACCEPT&lt;/code&gt; for the same port from &lt;code&gt;0.0.0.0/0&lt;/code&gt;, which meant it had never restricted anything. &lt;a href="https://news.rozoom.app/self-hosting/inherited-server-security-audit/" rel="noopener noreferrer"&gt;The full audit is here&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Put HTTP services behind one reverse proxy, never directly on the internet.&lt;/strong&gt; One TLS configuration, one access log, one place to add authentication. The pattern we run on this very site, in full, is in &lt;a href="https://news.rozoom.app/self-hosting/caddy-docker-production/" rel="noopener noreferrer"&gt;Caddy and Docker in production&lt;/a&gt;. The short version for a service that needs no public access at all:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;media.example.com {
    @notlocal not remote_ip 10.0.0.0/8 100.64.0.0/10 192.168.0.0/16
    respond @notlocal 403
    reverse_proxy 127.0.0.1:32400
}
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Bind the service to localhost and let the proxy be the only door.&lt;/strong&gt; A container published as &lt;code&gt;-p 32400:32400&lt;/code&gt; is on every interface the host owns, including the public one; &lt;code&gt;-p 127.0.0.1:32400:32400&lt;/code&gt; is not. This single character difference is the most common accidental exposure we find.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Treat outbound traffic as something you are allowed to care about.&lt;/strong&gt; Both observed payloads here, the miner and the proxyware, are useless without egress. A home or small-office box has no business opening arbitrary outbound connections, and a baseline you can diff is worth more than any IOC list published after the fact.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Know how you would rebuild, and have tested it.&lt;/strong&gt; After this class of compromise the honest advice is a clean install, which is only survivable if your data is somewhere else and restoring is something you have done before, not something you plan to learn under pressure. Where that somewhere else should be, with real prices on real data, is in our &lt;a href="https://news.rozoom.app/cloud/best-backup-storage-for-self-hosters-2026/" rel="noopener noreferrer"&gt;backup storage comparison&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Subscribe to the release notes of anything you expose.&lt;/strong&gt; Eleven weeks is not an unlucky interval, it is an unread mailing list. For a self-hosted service the vendor has no way to reach you, and no obligation to try.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we do not know
&lt;/h2&gt;

&lt;p&gt;We have not seen the inside of a compromised host ourselves, and we are not going to pretend otherwise. The indicators above are Plex's, the exposure counts are Shadowserver's, the CVSS numbers are the researcher's, and the exploitation dates come from the vendors and CISA. Our contribution is the part we can stand behind: what the pattern means for the rest of your stack, and the configuration we actually run.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Which Plex version fixes Plex2Shell?&lt;/strong&gt;&lt;br&gt;
1.43.3.10861 and later, released on 4 August 2026. Builds before that are affected, including earlier 1.43.3 builds.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Is my Plex server at risk if it is not exposed to the internet?&lt;/strong&gt;&lt;br&gt;
The chain needs network access to TCP port 32400. A server reachable only from your LAN is still exploitable from that LAN, so a compromised laptop or an untrusted guest device is enough. It is a smaller risk, not a theoretical one.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How do I know if my Plex server was already compromised?&lt;/strong&gt;&lt;br&gt;
Plex's advisory points at two indicators: unfamiliar &lt;code&gt;.so&lt;/code&gt; or &lt;code&gt;.dll&lt;/code&gt; files in the Codecs directory, and outbound connections to &lt;code&gt;proxystream.st&lt;/code&gt;. Observed payloads have been cryptominers and proxy malware. If the server was internet-facing on an old build, the safe assumption is compromise.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does updating remove an existing compromise?&lt;/strong&gt;&lt;br&gt;
No. Patching closes the door; it does not remove anything already inside. After a confirmed compromise the only clean path is a rebuild from known-good media plus a restore of data you can verify.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Should I just turn off Plex Remote Access?&lt;/strong&gt;&lt;br&gt;
If the only person streaming from outside your home is you, yes, and reach it over a VPN instead. Remote Access is the feature that puts port 32400 on the internet, and the 36,000 exposed servers counted in September were mostly not deliberate decisions.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's next
&lt;/h2&gt;

&lt;p&gt;If this prompted a look at the rest of your stack, the &lt;a href="https://news.rozoom.app/self-hosting/inherited-server-security-audit/" rel="noopener noreferrer"&gt;one-hour audit for a server you inherited&lt;/a&gt; is the checklist we use on an unknown box, and the &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;VPS comparison&lt;/a&gt; covers where to put a service you would rather not run at home at all.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/plex-exploited-hardening-checklist/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>security</category>
      <category>plex</category>
      <category>firewall</category>
    </item>
    <item>
      <title>Best VPS for Self-Hosting in 2026: Hetzner vs DigitalOcean vs Vultr</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:20:01 +0000</pubDate>
      <link>https://dev.to/rozoomnews/best-vps-for-self-hosting-in-2026-hetzner-vs-digitalocean-vs-vultr-4op6</link>
      <guid>https://dev.to/rozoomnews/best-vps-for-self-hosting-in-2026-hetzner-vs-digitalocean-vs-vultr-4op6</guid>
      <description>&lt;p&gt;We queried the Hetzner API twice in one day while writing this. At midday, the cheap plan every comparison recommends was unavailable in all three European locations. Five hours later, the same plan was orderable again in two of them. Nobody who copies a price out of another comparison can tell you that, because it is not in the price.&lt;/p&gt;

&lt;p&gt;So this one is built differently. The performance numbers come from a box we pay for and benchmarked ourselves. The prices come from the vendors' own APIs, read on the day of writing, with the plan ID next to every figure so you can check them yourself.&lt;/p&gt;

&lt;h2&gt;
  
  
  TL;DR
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hetzner wins on price for self-hosting&lt;/strong&gt;: &lt;strong&gt;CX23&lt;/strong&gt; (x86, 2 vCPU / 4 GB / 40 GB) at €5.49/mo, about &lt;strong&gt;4x cheaper&lt;/strong&gt; than the equivalent Droplet, with 5x the included traffic.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Check availability before you plan around it.&lt;/strong&gt; CX23 is not orderable everywhere, and whether it is orderable at all changes during the day. The always-available fallback is &lt;strong&gt;CAX11&lt;/strong&gt; (Arm, €5.99/mo); the x86 fallback is &lt;strong&gt;CPX22 at €19.49/mo&lt;/strong&gt;, which is only about 20% under DigitalOcean.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;DigitalOcean&lt;/strong&gt; ($24/mo) buys you the best API and docs, per-second billing since January 2026, and 4 TB of transfer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Vultr&lt;/strong&gt; ($20/mo) buys you reach: 31 locations, plus a genuinely free 512 MB tier most comparisons claim doesn't exist.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Live prices, read today
&lt;/h2&gt;

&lt;p&gt;The 2 vCPU / 4 GB tier is where most self-hosters land: a few containers, a database, a reverse proxy.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Plan&lt;/th&gt;
&lt;th&gt;vCPU / RAM / disk&lt;/th&gt;
&lt;th&gt;Price&lt;/th&gt;
&lt;th&gt;Included traffic&lt;/th&gt;
&lt;th&gt;Over-usage&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Hetzner (x86)&lt;/td&gt;
&lt;td&gt;CX23&lt;/td&gt;
&lt;td&gt;2 / 4 GB / 40 GB&lt;/td&gt;
&lt;td&gt;€5.49/mo&lt;/td&gt;
&lt;td&gt;20 TB&lt;/td&gt;
&lt;td&gt;€1.00/TB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hetzner (Arm)&lt;/td&gt;
&lt;td&gt;CAX11&lt;/td&gt;
&lt;td&gt;2 / 4 GB / 40 GB&lt;/td&gt;
&lt;td&gt;€5.99/mo&lt;/td&gt;
&lt;td&gt;20 TB&lt;/td&gt;
&lt;td&gt;€1.00/TB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hetzner (x86)&lt;/td&gt;
&lt;td&gt;CPX22&lt;/td&gt;
&lt;td&gt;2 / 4 GB / 80 GB&lt;/td&gt;
&lt;td&gt;€19.49/mo&lt;/td&gt;
&lt;td&gt;20 TB&lt;/td&gt;
&lt;td&gt;€1.00/TB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;DigitalOcean&lt;/td&gt;
&lt;td&gt;Basic&lt;/td&gt;
&lt;td&gt;2 / 4 GB / 80 GB&lt;/td&gt;
&lt;td&gt;$24.00/mo&lt;/td&gt;
&lt;td&gt;4 TB&lt;/td&gt;
&lt;td&gt;~$0.01/GB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vultr&lt;/td&gt;
&lt;td&gt;vc2-2c-4gb&lt;/td&gt;
&lt;td&gt;2 / 4 GB / 80 GB&lt;/td&gt;
&lt;td&gt;$20.00/mo&lt;/td&gt;
&lt;td&gt;3 TB&lt;/td&gt;
&lt;td&gt;~$0.01/GB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Hetzner bills a public IPv4 separately at €0.50/mo. IPv6 is free. Add it to the Hetzner rows if you need a v4 address, which you probably do.&lt;/p&gt;

&lt;p&gt;Prices change. Every number above was read from the vendors' own pricing APIs on the date of this post, not copied from another comparison.&lt;/p&gt;

&lt;h2&gt;
  
  
  The availability trap nobody mentions
&lt;/h2&gt;

&lt;p&gt;Search for a VPS recommendation and you will be told to buy the cheapest Hetzner CX box. The price is right. Whether you can buy it is a different question, and the answer changes.&lt;/p&gt;

&lt;p&gt;Here are our two reads of the Hetzner API, the same endpoint, the same day:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Midday.&lt;/strong&gt; CX23 (2 vCPU / 4 GB / 40 GB, €5.49) listed a price for Falkenstein, Nuremberg and Helsinki, and an &lt;strong&gt;empty availability list&lt;/strong&gt;. Price visible, server not orderable anywhere in Europe.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Five hours later.&lt;/strong&gt; The same plan came back as orderable in Nuremberg and Helsinki. Falkenstein still not.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Neither read was wrong. Capacity moves. What matters is that the number everyone quotes is not the number that tells you whether you can deploy:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The &lt;code&gt;prices[]&lt;/code&gt; array still lists locations you cannot order in, at their old prices. It is not an availability signal.&lt;/li&gt;
&lt;li&gt;The signal is &lt;code&gt;server_types[].locations[].available&lt;/code&gt;, per location, and it is live.&lt;/li&gt;
&lt;li&gt;If you order anyway, the API fails with &lt;code&gt;invalid_input: unsupported location for server type&lt;/code&gt;, which reads like you passed a bad region rather than like the plan is gone.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;GET /v1/datacenters&lt;/code&gt; was removed in June 2026 and returns HTTP 410, so there is no second place to look.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Practical advice: check availability for your target location immediately before you order, try a second European location before giving up, and keep a fallback plan in mind. Today that fallback is CAX11, which has been orderable in all three locations every time we looked.&lt;/p&gt;

&lt;h2&gt;
  
  
  What actually matters when you self-host
&lt;/h2&gt;

&lt;h3&gt;
  
  
  1. Egress is the silent budget killer
&lt;/h3&gt;

&lt;p&gt;Nextcloud, photo sync, backups: traffic adds up faster than compute. This is where the three split hardest.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hetzner&lt;/strong&gt; includes &lt;strong&gt;20 TB&lt;/strong&gt; on every Cloud plan, then bills &lt;strong&gt;€1.00/TB&lt;/strong&gt;, in 100 MB blocks. Not free beyond the cap - but ten times cheaper than the other two, and the cap is high enough that most self-hosters never reach it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;DigitalOcean&lt;/strong&gt; includes 0.5-6 TB depending on plan (4 TB at the tier above), then about &lt;strong&gt;$0.01/GB&lt;/strong&gt; - roughly &lt;strong&gt;$10/TB&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Vultr&lt;/strong&gt; includes 0.5-5 TB (3 TB at this tier), then about &lt;strong&gt;$0.01/GB&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Hetzner notifies you at 75% and 100% of the included traffic. Those are notifications, not a cap: usage keeps billing either way.&lt;/p&gt;

&lt;h3&gt;
  
  
  2. vCPU is marketing, not performance
&lt;/h3&gt;

&lt;p&gt;All three sell "vCPU", but a shared Hetzner core and a dedicated DigitalOcean core are not the same unit. The only way to know is to measure - which we did, for the one box we own.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Snapshots and backups are not free anywhere
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hetzner&lt;/strong&gt;: snapshots at &lt;strong&gt;€0.0143/GB per month&lt;/strong&gt; on the &lt;em&gt;compressed&lt;/em&gt; size; backups are a flat &lt;strong&gt;20% of the server price&lt;/strong&gt; and give you 7 slots.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;DigitalOcean&lt;/strong&gt;: snapshots &lt;strong&gt;$0.06/GB/month&lt;/strong&gt;; backups 20% (weekly) or 30% (daily) of Droplet cost.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Vultr&lt;/strong&gt;: snapshots billed per GB of stored data.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Nobody in this bracket gives you free snapshot storage. Budget it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The benchmark: a real Hetzner box, measured
&lt;/h2&gt;

&lt;p&gt;We took a Hetzner Cloud server we run in production - 2 vCPU shared x86 (Intel Xeon Skylake), 4 GB RAM, 40 GB SSD, Nuremberg - and ran standard, reproducible tests. These numbers are ours.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;CPU&lt;/strong&gt; (OpenSSL speed, single thread):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Test&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;AES-256-CBC (16 KB blocks)&lt;/td&gt;
&lt;td&gt;~630 MB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SHA-256 (16 KB blocks)&lt;/td&gt;
&lt;td&gt;~255 MB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Disk&lt;/strong&gt; (40 GB SSD):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Test&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Sequential write (fdatasync)&lt;/td&gt;
&lt;td&gt;468 MB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sequential read, cold&lt;/td&gt;
&lt;td&gt;1.1 GB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sequential read, warm (cached)&lt;/td&gt;
&lt;td&gt;6.0 GB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Network&lt;/strong&gt; (single stream):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Test&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Download from OVH mirror&lt;/td&gt;
&lt;td&gt;175 MB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Download from Ubuntu mirror&lt;/td&gt;
&lt;td&gt;82 MB/s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The box sits on a 1 Gbit/s port with IPv6. For a shared-vCPU plan at this price the disk is the quiet star: 468 MB/s sustained write is plenty for a database and a media stack on the same host.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Three honest caveats.&lt;/strong&gt; This is a CX-line x86 box, the same 2 vCPU / 4 GB / 40 GB spec as the CX23 we recommend, but not the Arm CAX11. We do not rent from DigitalOcean or Vultr, so their figures here are vendor-published, not measured. And a single-thread OpenSSL run on a shared core is a floor, not a guarantee: noisy neighbours are the trade-off of the shared tier.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the prices actually compare
&lt;/h2&gt;

&lt;p&gt;Against DigitalOcean's $24 for 2 vCPU / 4 GB:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;CX23 (x86) at €5.49&lt;/strong&gt; - roughly &lt;strong&gt;4x cheaper&lt;/strong&gt;, with 5x the included traffic. Half the disk (40 GB vs 80 GB), and subject to the availability caveat above.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CAX11 (Arm) at €5.99&lt;/strong&gt; - the same gap, always orderable so far, if your stack runs on arm64. Most upstream container images ship arm64 builds now; check the ones you depend on.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CPX22 (x86) at €19.49&lt;/strong&gt; - about &lt;strong&gt;20% cheaper&lt;/strong&gt;, with 5x the traffic and the same 80 GB disk. This is your number if CX23 is unavailable and you need x86.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That last line is the one missing from every comparison we read. The Hetzner price advantage is 4x when the cheap line is available and roughly a fifth off when it is not, and which of those you get depends on the hour you sign up. Add €0.50/mo for IPv4 to any of them.&lt;/p&gt;

&lt;h2&gt;
  
  
  When to pick which
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Pick Hetzner if:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;You're in or near Europe, or your users are.&lt;/li&gt;
&lt;li&gt;You run Nextcloud, a media stack, or backups - anything traffic-heavy.&lt;/li&gt;
&lt;li&gt;You want the most compute per euro, and you can check plan availability before you commit.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Pick DigitalOcean if:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;You want the smoothest API and the best docs.&lt;/li&gt;
&lt;li&gt;You need per-second billing for short-lived workloads.&lt;/li&gt;
&lt;li&gt;You want one-click apps instead of configuring a stack by hand.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Pick Vultr if:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;You need a region the others don't cover - 31 locations against Hetzner's five.&lt;/li&gt;
&lt;li&gt;You want to test on the free tier before paying.&lt;/li&gt;
&lt;li&gt;You run bursty workloads and spin instances up and down often.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Our recommendation
&lt;/h2&gt;

&lt;p&gt;For European self-hosting, &lt;strong&gt;Hetzner Cloud&lt;/strong&gt; is the pick, and the benchmark above shows the cheap shared tier is not slow. Order &lt;strong&gt;CX23&lt;/strong&gt; at €5.49/mo if it is available in your location; take &lt;strong&gt;CAX11&lt;/strong&gt; at €5.99/mo if it is not and your stack runs on Arm; fall back to &lt;strong&gt;CPX22&lt;/strong&gt; at €19.49/mo if you need x86 either way. Check availability first, in that order, and the decision takes a minute.&lt;/p&gt;

&lt;p&gt;If your users are in the US or Asia, DigitalOcean and Vultr are both solid - just model the egress before you sign up.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Is Hetzner good for self-hosting?&lt;/strong&gt;&lt;br&gt;
Yes, and it is still the price leader in Europe. Two caveats: the identity verification step at signup, and plan availability that changes without notice - check your target location before ordering.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Why can't I order the cheap Hetzner plan everyone recommends?&lt;/strong&gt;&lt;br&gt;
Because CX-line availability varies by location and over time, while the price stays visible on comparison sites and in the API's own price list. Availability is a separate field and it flips: we saw CX23 unavailable across all of Europe and orderable in two locations five hours later. Try another European location, then fall back to CAX11.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does DigitalOcean charge for egress?&lt;/strong&gt;&lt;br&gt;
Yes. Beyond the included transfer (4 TB at the 4 GB tier) it's about $0.01/GB, roughly $10 per TB. Model this before you sign up if you serve media or move backups.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does Vultr have a free tier?&lt;/strong&gt;&lt;br&gt;
Yes - a 1 vCPU / 512 MB / 10 GB instance in a small number of locations, with no included bandwidth. There is also a $2.50/mo plan, but it is IPv6-only and limited to two locations. The cheapest generally useful plan is $5/mo.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which VPS has the best price-to-performance?&lt;/strong&gt;&lt;br&gt;
At 2 vCPU / 4 GB, Hetzner leads on both price and included traffic by a wide margin when its cheap line (CX23, CAX11) is available. If you are pushed onto CPX22, the gap narrows to roughly 20%, which is close enough that DigitalOcean's tooling or Vultr's regions can be the better buy.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Can I self-host on a $6 VPS?&lt;/strong&gt;&lt;br&gt;
Yes, for a reverse proxy, a couple of containers and a small database. The moment you add media or heavy backups, the thing that breaks your budget is egress, not CPU.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's next
&lt;/h2&gt;

&lt;p&gt;This is the first in a series. Coming up: &lt;strong&gt;running Nextcloud AIO on Hetzner&lt;/strong&gt; and &lt;strong&gt;Caddy + Docker in production&lt;/strong&gt; - both built on the same box we benchmarked here.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>vps</category>
      <category>hetzner</category>
      <category>digitalocean</category>
    </item>
    <item>
      <title>Deploy From GitHub Actions Without Giving CI a Shell on Your Server</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:19:25 +0000</pubDate>
      <link>https://dev.to/rozoomnews/deploy-from-github-actions-without-giving-ci-a-shell-on-your-server-47bh</link>
      <guid>https://dev.to/rozoomnews/deploy-from-github-actions-without-giving-ci-a-shell-on-your-server-47bh</guid>
      <description>&lt;p&gt;Every "deploy a static site with GitHub Actions" tutorial ends the same way: generate an SSH key, paste the private half into repository secrets, rsync the build to your server. It works on the first try, which is the problem. What you just created is an interactive shell account on your server, held by a system that runs code from pull requests.&lt;/p&gt;

&lt;p&gt;This post is the version with the guard rails: a key that can do exactly one thing, a host key you pinned yourself, and three checks that stop a broken build from replacing a working site. It is the pipeline that ships this blog, and it is maybe fifteen extra minutes of work.&lt;/p&gt;

&lt;p&gt;The server side assumes you have a box. If you don't, we compared the cheap tiers in &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;Best VPS for self-hosting in 2026&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the naive deploy key actually grants
&lt;/h2&gt;

&lt;p&gt;An entry in &lt;code&gt;~/.ssh/authorized_keys&lt;/code&gt; with no options is a login. Anything holding that private key can:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;open an interactive shell as that user,&lt;/li&gt;
&lt;li&gt;read every file that user can read, including other projects on the same box and any &lt;code&gt;.env&lt;/code&gt; lying around,&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;sudo&lt;/code&gt; if the user has it,&lt;/li&gt;
&lt;li&gt;forward ports and use your server as a jump host into whatever it can reach.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;None of that is needed to copy a directory of HTML. And the private key lives in a CI system where a workflow change, a compromised third-party action, or a misconfigured &lt;code&gt;pull_request_target&lt;/code&gt; trigger can read it.&lt;/p&gt;

&lt;p&gt;We ran exactly this setup for months without thinking about it, which is the honest reason this post exists.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lock the key to one command
&lt;/h2&gt;

&lt;p&gt;OpenSSH lets you attach options to each key in &lt;code&gt;authorized_keys&lt;/code&gt;. The one that matters is &lt;code&gt;command=&lt;/code&gt;, which replaces whatever the client asked to run with what you specify:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight ssh"&gt;&lt;code&gt;&lt;span class="na"&gt;command&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"/usr/bin/rrsync -wo /home/deploy/site"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="n"&gt;restrict&lt;/span&gt; &lt;span class="n"&gt;ssh&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;ed25519&lt;/span&gt; &lt;span class="no"&gt;AAAA&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt; &lt;span class="n"&gt;github&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;actions&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Piece by piece:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;rrsync&lt;/code&gt;&lt;/strong&gt; ships with rsync (&lt;code&gt;/usr/bin/rrsync&lt;/code&gt; on Debian and Ubuntu; it may be a gzipped script under &lt;code&gt;/usr/share/doc/rsync/scripts/&lt;/code&gt; on older releases). It is a wrapper that accepts only rsync's server-side invocation and refuses everything else.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;-wo&lt;/code&gt;&lt;/strong&gt; means write-only: the key can upload into the directory and cannot download from it. If your deploy only pushes, take the stricter option.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The path argument&lt;/strong&gt; is a jail. Paths outside it are rejected, including anything clever with &lt;code&gt;..&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;restrict&lt;/code&gt;&lt;/strong&gt; disables port forwarding, agent forwarding, X11 and PTY allocation in one word. On any OpenSSH from the last several years, prefer it over listing &lt;code&gt;no-port-forwarding,no-agent-forwarding,...&lt;/code&gt; by hand.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Test it the direct way, from your laptop with that private key:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ssh -i deploy_key deploy@your-server &lt;span class="s1"&gt;'id'&lt;/span&gt;      &lt;span class="c"&gt;# must fail&lt;/span&gt;
rsync -rltz -e &lt;span class="s2"&gt;"ssh -i deploy_key"&lt;/span&gt; ./public/ deploy@your-server:/   &lt;span class="c"&gt;# must succeed&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The first command failing is the whole point. If it prints a user id, the options are not being applied - check that the whole thing is on one line and that the line starts with &lt;code&gt;command=&lt;/code&gt;, not with the key type.&lt;/p&gt;

&lt;p&gt;Note the destination path in the rsync call: it is relative to the jail, so &lt;code&gt;/&lt;/code&gt; means the directory you locked the key into.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pin the host key, don't disable the check
&lt;/h2&gt;

&lt;p&gt;The other half of the tutorials' shortcut is &lt;code&gt;StrictHostKeyChecking=no&lt;/code&gt;, which turns off the check that the server you are shipping to is the server you think it is. Pin it instead. On the server:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ssh-keyscan -t ed25519 your-server.example
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Put that line into a repository secret and write it to &lt;code&gt;known_hosts&lt;/code&gt; in the job. Now a changed host key fails the deploy loudly instead of silently shipping your site to whoever answers on that address.&lt;/p&gt;

&lt;h2&gt;
  
  
  The workflow
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Build and deploy&lt;/span&gt;

&lt;span class="na"&gt;on&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;push&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;branches&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s"&gt;main&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
  &lt;span class="na"&gt;workflow_dispatch&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;

&lt;span class="na"&gt;concurrency&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;group&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;deploy&lt;/span&gt;
  &lt;span class="na"&gt;cancel-in-progress&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;

&lt;span class="na"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;deploy&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;runs-on&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ubuntu-latest&lt;/span&gt;
    &lt;span class="na"&gt;steps&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;actions/checkout@v5&lt;/span&gt;
        &lt;span class="na"&gt;with&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
          &lt;span class="na"&gt;submodules&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;recursive&lt;/span&gt;

      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Build&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;your-build-command&lt;/span&gt;

      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Guard against an empty build&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;set -euo pipefail&lt;/span&gt;
          &lt;span class="s"&gt;test -s public/index.html&lt;/span&gt;
          &lt;span class="s"&gt;test -s public/sitemap.xml&lt;/span&gt;

      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Deploy over rsync&lt;/span&gt;
        &lt;span class="na"&gt;env&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
          &lt;span class="na"&gt;DEPLOY_HOST&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;${{ secrets.DEPLOY_HOST }}&lt;/span&gt;
          &lt;span class="na"&gt;DEPLOY_USER&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;${{ secrets.DEPLOY_USER }}&lt;/span&gt;
          &lt;span class="na"&gt;DEPLOY_SSH_KEY&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;${{ secrets.DEPLOY_SSH_KEY }}&lt;/span&gt;
          &lt;span class="na"&gt;DEPLOY_KNOWN_HOSTS&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;${{ secrets.DEPLOY_KNOWN_HOSTS }}&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;set -euo pipefail&lt;/span&gt;
          &lt;span class="s"&gt;mkdir -p ~/.ssh&lt;/span&gt;
          &lt;span class="s"&gt;printf '%s\n' "$DEPLOY_SSH_KEY" &amp;gt; ~/.ssh/deploy_key&lt;/span&gt;
          &lt;span class="s"&gt;chmod 600 ~/.ssh/deploy_key&lt;/span&gt;
          &lt;span class="s"&gt;printf '%s\n' "$DEPLOY_KNOWN_HOSTS" &amp;gt; ~/.ssh/known_hosts&lt;/span&gt;
          &lt;span class="s"&gt;rsync -rltz --delete --omit-dir-times \&lt;/span&gt;
            &lt;span class="s"&gt;-e "ssh -i ~/.ssh/deploy_key -o IdentitiesOnly=yes -o UserKnownHostsFile=~/.ssh/known_hosts" \&lt;/span&gt;
            &lt;span class="s"&gt;public/ "$DEPLOY_USER@$DEPLOY_HOST:/"&lt;/span&gt;

      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Smoke test&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;set -euo pipefail&lt;/span&gt;
          &lt;span class="s"&gt;for url in https://example.com/ https://example.com/sitemap.xml; do&lt;/span&gt;
            &lt;span class="s"&gt;code=$(curl -s -o /dev/null -w '%{http_code}' --retry 3 --retry-delay 5 "$url")&lt;/span&gt;
            &lt;span class="s"&gt;echo "$url -&amp;gt; $code"&lt;/span&gt;
            &lt;span class="s"&gt;test "$code" = "200"&lt;/span&gt;
          &lt;span class="s"&gt;done&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The parts that are not decoration:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;code&gt;concurrency&lt;/code&gt; without &lt;code&gt;cancel-in-progress&lt;/code&gt;.&lt;/strong&gt; Two deploys running at once against the same directory with &lt;code&gt;--delete&lt;/code&gt; is how you serve a half-copied site. Queueing them costs nothing.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The empty-build guard.&lt;/strong&gt; &lt;code&gt;--delete&lt;/code&gt; is what makes rsync a deploy rather than an upload, and it is also what turns a build that silently produced nothing into a deleted website. Three &lt;code&gt;test -s&lt;/code&gt; lines are cheaper than that outage. Assert on files that must exist and on at least one that only appears when the build really ran.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;code&gt;IdentitiesOnly=yes&lt;/code&gt;.&lt;/strong&gt; Without it, ssh may offer other keys from the agent first and get refused before it reaches yours.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;code&gt;--omit-dir-times&lt;/code&gt;.&lt;/strong&gt; With a write-only jail, rsync's attempts to set directory timestamps can fail noisily. This keeps the output clean without weakening anything.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The smoke test runs against the public URL&lt;/strong&gt;, not against the server over ssh. It is the only step that proves a visitor can see the result: it catches a dead container, an expired certificate and a proxy pointing at the wrong directory, none of which rsync knows about.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this does and does not protect you from
&lt;/h2&gt;

&lt;p&gt;It does protect you from the realistic failure: a leaked CI secret turning into shell access on your server. The key can write one directory and nothing else, cannot read back, cannot forward, cannot open a terminal.&lt;/p&gt;

&lt;p&gt;It does not protect you from someone who can change the workflow file, because they can replace the content you ship. If the site is the thing you care about, protect the branch: require review, and keep the deploy on &lt;code&gt;push&lt;/code&gt; to the default branch only - never on &lt;code&gt;pull_request_target&lt;/code&gt;, which runs with your secrets available to a fork's code.&lt;/p&gt;

&lt;p&gt;It also does not protect against the directory itself. Whatever is served from that path can be overwritten by the key. Keep the deploy target separate from config, data volumes and anything else the server holds.&lt;/p&gt;

&lt;h2&gt;
  
  
  The retrofit, in order
&lt;/h2&gt;

&lt;p&gt;If you already have an unrestricted deploy key, fixing it takes a few minutes:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Confirm &lt;code&gt;rrsync&lt;/code&gt; exists on the server: &lt;code&gt;ls -l /usr/bin/rrsync&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Edit &lt;code&gt;authorized_keys&lt;/code&gt; and add &lt;code&gt;command="/usr/bin/rrsync -wo /path/to/site",restrict&lt;/code&gt; in front of the existing key line.&lt;/li&gt;
&lt;li&gt;From your laptop with that key, check that &lt;code&gt;ssh ... 'id'&lt;/code&gt; fails and a test rsync still succeeds.&lt;/li&gt;
&lt;li&gt;Re-run the workflow with &lt;code&gt;workflow_dispatch&lt;/code&gt; and watch the smoke test pass.&lt;/li&gt;
&lt;li&gt;Only then delete the backup of the old line.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Rotating the key at the same time is a reasonable habit, especially if it was ever printed in a log.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's next
&lt;/h2&gt;

&lt;p&gt;The server side of the same setup: &lt;a href="https://news.rozoom.app/self-hosting/caddy-docker-production/" rel="noopener noreferrer"&gt;Caddy and Docker in production&lt;/a&gt; - automatic HTTPS, several sites on one box, and the cache headers that decide whether your readers see the page you just published.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/github-actions-rsync-deploy-no-shell/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>githubactions</category>
      <category>ssh</category>
      <category>deployment</category>
    </item>
    <item>
      <title>The One-Hour Audit for a Server You Inherited</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:18:49 +0000</pubDate>
      <link>https://dev.to/rozoomnews/the-one-hour-audit-for-a-server-you-inherited-2f2l</link>
      <guid>https://dev.to/rozoomnews/the-one-hour-audit-for-a-server-you-inherited-2f2l</guid>
      <description>&lt;p&gt;Somebody hands you the root password to a machine that has been running for years. It serves real sites, nobody remembers how it was set up, and the person who did set it up is gone.&lt;/p&gt;

&lt;p&gt;The temptation is to do the big thing first: rebuild it, migrate it, containerise it. Do the small thing first. One hour, five questions, and you will know whether the machine is an inconvenience or an emergency.&lt;/p&gt;

&lt;p&gt;This is the audit we ran on exactly such a box last week, with what it actually turned up. Every finding below is real, and not one of them was visible from the outside.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. What is listening, and what does the firewall think?
&lt;/h2&gt;

&lt;p&gt;These two answers are always compared, never read separately. The first is the truth about the machine; the second is what somebody once intended.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ss -tlnp                     &lt;span class="c"&gt;# every listening TCP socket, with the process&lt;/span&gt;
iptables -L INPUT -n --line-numbers
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Our finding, and it is the one to steal from this article: the firewall had a careful allowlist for the admin panel port, three trusted subnets, written by somebody who knew what they were doing. Four lines &lt;strong&gt;above&lt;/strong&gt; it sat a blanket &lt;code&gt;ACCEPT&lt;/code&gt; for the same port from &lt;code&gt;0.0.0.0/0&lt;/code&gt;, added later.&lt;/p&gt;

&lt;p&gt;iptables evaluates in order and stops at the first match. The allowlist had never, from the moment that line was added, allowed anything less than everybody. The admin login had been open to the internet for years behind a rule that read, to a casual glance, like a restriction.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Read your rules top to bottom, with line numbers, and ask what matches first.&lt;/strong&gt; A rule list is a program, not a set.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Who is actually using the thing you are about to close?
&lt;/h2&gt;

&lt;p&gt;The audit's single most useful habit. Before closing any path, read its access log and count real users.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# top talkers against one path, last 10k lines&lt;/span&gt;
grep -h &lt;span class="s1"&gt;'phpmyadmin'&lt;/span&gt; /var/log/nginx/*access* &lt;span class="o"&gt;|&lt;/span&gt; awk &lt;span class="s1"&gt;'{print $1}'&lt;/span&gt; \
  &lt;span class="o"&gt;|&lt;/span&gt; sort &lt;span class="o"&gt;|&lt;/span&gt; uniq -c &lt;span class="o"&gt;|&lt;/span&gt; sort -rn &lt;span class="o"&gt;|&lt;/span&gt; head
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Two of our findings came straight out of this:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A &lt;strong&gt;phpMyAdmin from 2017&lt;/strong&gt;, reachable from the internet, on a web server whose PHP had no upstream patches left. The log showed 201 requests from one address and 38 from another, both scanners walking a wordlist. Not one human session in the window.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;proftpd running&lt;/strong&gt;, in plaintext, no TLS. &lt;code&gt;xferlog&lt;/code&gt; was empty since June. The only connections were &lt;code&gt;anonymous&lt;/code&gt; login attempts from a botnet.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Both were closed in minutes, with no risk of breaking anyone's workflow, because the logs proved there was no workflow. Had the logs shown daily sessions from one office IP, the answer would have been an allowlist and a conversation rather than a shutdown.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The rule: never close a service on the belief that nobody uses it. Close it on evidence that nobody uses it.&lt;/strong&gt; Those feel the same right up until the invoicing department calls.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Do not lock yourself out
&lt;/h2&gt;

&lt;p&gt;Everything in this section is firewall work on a machine you reach only through that firewall. Two precautions, both cheap.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Arm an auto-revert before you commit anything.&lt;/strong&gt; A timer that restores the known-good ruleset in ten minutes, cancelled by hand once you have confirmed you can still get in:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;iptables-save &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; /root/iptables.good
&lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s1"&gt;'iptables-restore &amp;lt; /root/iptables.good'&lt;/span&gt; &lt;span class="o"&gt;|&lt;/span&gt; at now + 10 minutes
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Apply your change, open a &lt;strong&gt;brand new SSH session&lt;/strong&gt; from somewhere else, and only then cancel the job with &lt;code&gt;atrm&lt;/code&gt;. Testing on the session you already have proves nothing: an established connection survives rules that block new ones.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Then make it survive a reboot,&lt;/strong&gt; which is a separate step and the one people forget:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;service iptables save          &lt;span class="c"&gt;# or: netfilter-persistent save&lt;/span&gt;
systemctl is-enabled iptables
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Ours rebooted on a weekly schedule we did not know about. The hardening was still there afterwards, because the rules had been saved, and we knew it was still there because we checked after the reboot rather than assuming.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. What renews itself, and what only appears to?
&lt;/h2&gt;

&lt;p&gt;The quiet killer. On this box, TLS renewal was a job inside a control panel whose licence had expired two months earlier. The renewal cron died with the licence. Four sites had already been failing their TLS handshake for five days before the audit found it, and seven more were scheduled to expire within three weeks.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# every scheduled job on the box, in one pass&lt;/span&gt;
crontab -l&lt;span class="o"&gt;;&lt;/span&gt; ls -l /etc/cron.*/ /etc/cron.d/&lt;span class="o"&gt;;&lt;/span&gt; systemctl list-timers --all
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Read all of it, including the commented lines, because they tell you what used to run. And look for duplicates: we disabled one renewal cron and only on a second pass found a &lt;strong&gt;second one&lt;/strong&gt;, same job, different hour, four lines further down.&lt;/p&gt;

&lt;p&gt;Taking renewal back is a short job even on an end-of-life distro - we wrote it up in &lt;a href="https://news.rozoom.app/self-hosting/acme-sh-certificates-legacy-server/" rel="noopener noreferrer"&gt;acme.sh on a server where certbot will not install&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. What is installed that nothing uses?
&lt;/h2&gt;

&lt;p&gt;Legacy boxes accumulate runtimes. Ours had eleven separate PHP builds, going back to 5.2. The interesting question is not what is installed but what is &lt;strong&gt;wired up&lt;/strong&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;grep -rhoE &lt;span class="s1"&gt;'php[0-9]+'&lt;/span&gt; /etc/nginx/conf.d/*.conf /etc/httpd/conf.d/*.conf \
  &lt;span class="o"&gt;|&lt;/span&gt; sort &lt;span class="o"&gt;|&lt;/span&gt; uniq -c &lt;span class="o"&gt;|&lt;/span&gt; sort -rn
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Eight sites were on 7.4, three on 8.0, and every build older than that was dead weight. That single command changed the migration plan: a machine that looks like it needs a hosting platform supporting PHP 5.2 actually needs 7.4 and 8.x, which is to say anything modern.&lt;/p&gt;

&lt;p&gt;An audit is not only about what to close. It is about discovering that the scary part of the job does not exist.&lt;/p&gt;

&lt;h2&gt;
  
  
  What an hour bought
&lt;/h2&gt;

&lt;p&gt;Four findings on one machine: an allowlist that allowed everyone, a nine-year-old admin tool exposed to scanners, an FTP daemon serving nobody, and certificate renewal that had been dead for two months with four sites already down.&lt;/p&gt;

&lt;p&gt;None of it required a rebuild. None of it required a maintenance window. Three of the four were fixed the same hour, and the fourth - the distro itself being past end of life - is the one honest finding you cannot patch, which is exactly why it belongs in the migration plan rather than in today's work.&lt;/p&gt;

&lt;p&gt;If the audit sends you shopping for a replacement box, we compared the options with live prices in &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;Best VPS for self-hosting in 2026&lt;/a&gt;, and wrote up four years on one provider in &lt;a href="https://news.rozoom.app/cloud/hetzner-cloud-review-2026/" rel="noopener noreferrer"&gt;our Hetzner Cloud review&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/inherited-server-security-audit/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>security</category>
      <category>iptables</category>
      <category>audit</category>
    </item>
    <item>
      <title>Hetzner Cloud Review 2026: Four Years, Three Servers, and Every Bill Explained</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:18:13 +0000</pubDate>
      <link>https://dev.to/rozoomnews/hetzner-cloud-review-2026-four-years-three-servers-and-every-bill-explained-26kn</link>
      <guid>https://dev.to/rozoomnews/hetzner-cloud-review-2026-four-years-three-servers-and-every-bill-explained-26kn</guid>
      <description>&lt;p&gt;The oldest server in our account was created on 29 September 2022 and is still running. It has never been migrated, rebuilt or restored. Two more boxes joined it since, the newest three weeks ago. That is the entire basis of this review: three machines, four years, one invoice a month, and a list of things that went wrong that is almost entirely our own fault.&lt;/p&gt;

&lt;p&gt;Most Hetzner reviews are a pricing table with adjectives. This one is the bill, itemised, plus the parts of the bill that are easy to miss, and the honest account of what we broke.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we actually run
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Box&lt;/th&gt;
&lt;th&gt;Type&lt;/th&gt;
&lt;th&gt;Created&lt;/th&gt;
&lt;th&gt;Job&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;td&gt;CX32, 4 vCPU / 8 GB&lt;/td&gt;
&lt;td&gt;Sep 2022&lt;/td&gt;
&lt;td&gt;Legacy multi-site host, eleven small sites, mail&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;td&gt;CX22, 2 vCPU / 4 GB&lt;/td&gt;
&lt;td&gt;Nov 2023&lt;/td&gt;
&lt;td&gt;Three containerised storefronts plus three CI runners&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;td&gt;CX23, 2 vCPU / 4 GB&lt;/td&gt;
&lt;td&gt;Oct 2026&lt;/td&gt;
&lt;td&gt;Static blog, analytics, a relay service&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;No load balancers, no managed anything, no Kubernetes. Three Linux boxes, Docker, and a reverse proxy. Four years of uptime that we did not have to think about is the headline finding, and it is worth saying clearly before the complaints start: in four years, not one of our outages was caused by Hetzner.&lt;/p&gt;

&lt;h2&gt;
  
  
  The bill, every line
&lt;/h2&gt;

&lt;p&gt;These figures come from the Hetzner pricing API, read the day this was published, in EUR excluding VAT.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Line item&lt;/th&gt;
&lt;th&gt;Price&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;CX23, 2 vCPU / 4 GB / 40 GB&lt;/td&gt;
&lt;td&gt;€5.49/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CAX11, Arm, 2 vCPU / 4 GB / 40 GB&lt;/td&gt;
&lt;td&gt;€5.99/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CPX22, 2 vCPU / 4 GB / 80 GB&lt;/td&gt;
&lt;td&gt;€19.49/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Primary IPv4 address&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;€0.50/mo&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;IPv6&lt;/td&gt;
&lt;td&gt;free&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Included traffic, EU locations&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;20 TB/mo&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Traffic over the allowance&lt;/td&gt;
&lt;td&gt;€1.00/TB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Automated backups&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;+20% of the server price&lt;/strong&gt;, 7 slots&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Snapshots and custom images&lt;/td&gt;
&lt;td&gt;€0.0143/GB/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Block storage volumes&lt;/td&gt;
&lt;td&gt;€0.0572/GB/mo&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Floating IP&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;€3.00/mo&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two of those lines deserve a paragraph each, because they are the two that quietly take your money.&lt;/p&gt;

&lt;h3&gt;
  
  
  The floating IP line
&lt;/h3&gt;

&lt;p&gt;A floating IP costs &lt;strong&gt;€3.00 a month, six times the price of the primary IPv4 on the server&lt;/strong&gt;, and it is billed whether or not anything is attached to it. We found two of them on our account left over from a migration that had finished months earlier, routing nothing. Deleting them cut €6 a month, which on a fleet whose cheapest box is €5.49 is a whole extra server.&lt;/p&gt;

&lt;p&gt;Go and look at your floating IPs now. If one is unattached, it is not a spare, it is a subscription.&lt;/p&gt;

&lt;h3&gt;
  
  
  The traffic line is genuinely generous
&lt;/h3&gt;

&lt;p&gt;20 TB per month included, per server, in the European locations, and €1.00 per TB beyond it. For comparison, the equivalent allowance at the big US clouds is measured in hundreds of gigabytes and the overage is priced per gigabyte, typically around €0.09. At a terabyte of egress a month, the difference is between a rounding error and a bill that doubles.&lt;/p&gt;

&lt;p&gt;This is the single strongest reason to run anything bandwidth-shaped - media, backups, a package mirror, an analytics endpoint - on Hetzner rather than on a hyperscaler. We looked at the alternatives in &lt;a href="https://news.rozoom.app/self-hosting/best-vps-self-hosting-2026/" rel="noopener noreferrer"&gt;Best VPS for self-hosting in 2026&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What went wrong in four years
&lt;/h2&gt;

&lt;p&gt;Every incident below was ours. We are listing them because they are the ones you will hit too, and none of them appear in a feature comparison.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. The images ship with no swap
&lt;/h3&gt;

&lt;p&gt;A CI runner on box 2 went offline and stayed offline for &lt;strong&gt;ten days&lt;/strong&gt;. The symptom read like an unreachable machine. The cause was a build that allocated more than the 4 GB available, the kernel OOM-killer taking the runner process, and systemd not retrying. The box had been up for 85 days with &lt;strong&gt;zero swap&lt;/strong&gt;, because the stock image does not configure any.&lt;/p&gt;

&lt;p&gt;A 4 GB swapfile in &lt;code&gt;/etc/fstab&lt;/code&gt; fixed it permanently, and the same build then completed:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;fallocate -l 4G /swapfile &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; chmod 600 /swapfile
mkswap /swapfile &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; swapon /swapfile
&lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s1"&gt;'/swapfile none swap sw 0 0'&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&amp;gt;&lt;/span&gt; /etc/fstab
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If you run builds on a 4 GB box, do this on day one. Treat "service has been down for days" on a Hetzner box as a possible OOM before you treat it as a network problem.&lt;/p&gt;

&lt;h3&gt;
  
  
  2. Docker build cache ate two thirds of the disk
&lt;/h3&gt;

&lt;p&gt;On a 38 GB disk, 24 GB was Docker build cache. Nothing warns you; the box simply starts failing writes. &lt;code&gt;docker system prune -af --volumes&lt;/code&gt; on a schedule, or at minimum a disk alert, is not optional on a 40 GB plan.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. fail2ban banned our own bridge
&lt;/h3&gt;

&lt;p&gt;A host started answering 503 to everything. fail2ban had matched traffic coming from the Docker bridge, which is to say from our own reverse proxy, and banned the gateway address. Whitelist your bridge networks in &lt;code&gt;ignoreip&lt;/code&gt; before you deploy fail2ban next to Docker, not after.&lt;/p&gt;

&lt;h3&gt;
  
  
  4. The plan you budgeted for may not be orderable
&lt;/h3&gt;

&lt;p&gt;The CX line's availability flips through the day, per location. We read the API at midday and the €5.49 plan was unavailable across all three European locations; five hours later it was orderable in two of them. The &lt;code&gt;prices[]&lt;/code&gt; array lists locations you cannot order in, which makes a price look like an offer when it is only a record. The live signal is &lt;code&gt;server_types[].locations[].available&lt;/code&gt;, and &lt;code&gt;GET /v1/datacenters&lt;/code&gt; has returned HTTP 410 since June 2026, so there is no second source to check.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Hetzner still does not sell
&lt;/h2&gt;

&lt;p&gt;As of this month's changelog, and this is the honest reason not to choose them:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No managed database.&lt;/strong&gt; No equivalent of RDS or managed Postgres. You run the database, you handle the failover, you own the backups.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No managed Kubernetes.&lt;/strong&gt; There is an official cloud controller manager and a CSI driver, which make a self-managed cluster pleasant, but the control plane is yours to operate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Thin managed everything else.&lt;/strong&gt; Object storage and storage boxes exist; load balancers, networks and firewalls exist and work. Beyond that the catalogue is deliberately short, and 2026 added features, not products.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fewer regions.&lt;/strong&gt; Three European locations plus a handful elsewhere. If your users are in Asia or South America, latency is a real argument against.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Support is a ticket, not a phone call.&lt;/strong&gt; In four years we have opened almost none, which is its own kind of answer, but do not expect an account manager.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Backups: the checkbox we leave off, and why
&lt;/h2&gt;

&lt;p&gt;Hetzner's automated backups cost &lt;strong&gt;20% of the server price&lt;/strong&gt; and keep 7 slots. On a €5.49 box that is €1.10 a month, which is cheap enough that the default advice should be to turn it on.&lt;/p&gt;

&lt;p&gt;We leave it off, and the reasoning is worth copying only if the second half applies to you: snapshots of a whole server restore a whole server, which is the wrong unit when what you need back is one database and one config directory. Our backups go to S3-compatible object storage on a schedule, versioned, and they get restore-tested. If you are &lt;strong&gt;not&lt;/strong&gt; going to build that, turn the checkbox on today. An untested €1.10 backup beats a disciplined plan you never wrote.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should use Hetzner, and who should not
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Use it if&lt;/strong&gt; you run your own stack, your traffic is bandwidth-heavy, your users are in Europe, and you would rather spend an hour on &lt;code&gt;apt&lt;/code&gt; than €200 a month on managed services. The price-to-performance is not a marginal win, it is a factor of three or four, and the hardware has been boring in the best sense.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Do not use it if&lt;/strong&gt; you need a managed database with a support contract behind it, a control plane somebody else gets paged for, or a region in Asia. The money you save will go straight back out as your own time, and if your time is the scarce resource, that trade is bad.&lt;/p&gt;

&lt;p&gt;For us, across four years and three machines, the verdict is simple: the infrastructure has never been the problem. Everything on the incident list above we did to ourselves, and every one of those mistakes would have cost more to make somewhere else.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/cloud/hetzner-cloud-review-2026/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>cloud</category>
      <category>hetzner</category>
      <category>vps</category>
      <category>review</category>
    </item>
    <item>
      <title>acme.sh on a Server Where certbot Will Not Install</title>
      <dc:creator>Rozoom News Editor</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:17:37 +0000</pubDate>
      <link>https://dev.to/rozoomnews/acmesh-on-a-server-where-certbot-will-not-install-f60</link>
      <guid>https://dev.to/rozoomnews/acmesh-on-a-server-where-certbot-will-not-install-f60</guid>
      <description>&lt;p&gt;Four sites on one legacy box went dark. Not slow, not 500 - dark, as in the TLS handshake failed before a request existed. They had been down for five days before anyone noticed, because nothing was watching the one thing that mattered.&lt;/p&gt;

&lt;p&gt;The cause was mundane: certificate renewal lived inside a control panel, the panel's licence expired, and its daily renewal cron died with it. Seven more certificates on the same machine were scheduled to expire over the following three weeks.&lt;/p&gt;

&lt;p&gt;The distro was years past end of life, so &lt;code&gt;certbot&lt;/code&gt; would not install cleanly and the vendor packages were long gone. This is the recipe that fixed it in under thirty minutes without touching a single vhost file, and the monitoring that makes sure it cannot happen quietly again.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why acme.sh and not certbot
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;certbot&lt;/code&gt; is a Python application with a dependency tree. On a current distro that is fine. On an end-of-life one it is the whole problem: the system Python is too old for modern &lt;code&gt;cryptography&lt;/code&gt; wheels, pip wants a compiler you do not want on a production box, and the snap route needs a daemon the distro never shipped.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;acme.sh&lt;/code&gt; is &lt;strong&gt;a shell script&lt;/strong&gt;. It needs &lt;code&gt;bash&lt;/code&gt; and either &lt;code&gt;curl&lt;/code&gt; or &lt;code&gt;wget&lt;/code&gt;. That is the entire dependency list. It talks to the ACME API over HTTPS via curl, so even an ancient OpenSSL is not in the signing path, and &lt;code&gt;bash 4.2&lt;/code&gt; is new enough.&lt;/p&gt;

&lt;p&gt;There is also a pleasant irony worth knowing if you are escaping a panel: several commercial panels use acme.sh under the hood. You are not replacing the mechanism, you are taking it out from under something that stopped running.&lt;/p&gt;

&lt;h2&gt;
  
  
  Install, and fix the default CA
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl https://get.acme.sh &lt;span class="o"&gt;|&lt;/span&gt; sh -s &lt;span class="nv"&gt;email&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;you@yourdomain.dev
&lt;span class="nb"&gt;source&lt;/span&gt; ~/.bashrc
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then immediately do the step most guides skip. Since version 3, &lt;strong&gt;acme.sh defaults to ZeroSSL, not Let's Encrypt&lt;/strong&gt;, and ZeroSSL requires account registration with EAB credentials. If you wanted Let's Encrypt, say so once:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;acme.sh --set-default-ca --server letsencrypt
acme.sh --register-account -m you@yourdomain.dev
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Skip this and your first issue attempt fails with a registration error that looks like a network problem.&lt;/p&gt;

&lt;h2&gt;
  
  
  Issue without touching the vhosts
&lt;/h2&gt;

&lt;p&gt;The goal is zero changes to the web server configuration, because on a box like this the configs are generated by something you no longer control and a hand edit will be overwritten.&lt;/p&gt;

&lt;p&gt;Two things make that possible.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;One: use webroot validation into the directory the server already serves.&lt;/strong&gt; Check first whether a path is already wired up - panels usually leave one behind:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -s -o /dev/null -w &lt;span class="s1"&gt;'%{http_code}\n'&lt;/span&gt; \
  http://yourdomain.dev/.well-known/acme-challenge/test
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Anything other than a connection error means the path is reachable and you only need to find which directory on disk it maps to. Then:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;acme.sh --issue -d yourdomain.dev -d www.yourdomain.dev \
  -w /var/www/letsencrypt
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Two: write the certificate to the filenames the server already expects.&lt;/strong&gt; Do not point the config at acme.sh's directory; point acme.sh at the config's filenames. &lt;code&gt;--install-cert&lt;/code&gt; does exactly that, and the reload hook makes it live:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;acme.sh --install-cert -d yourdomain.dev \
  --fullchain-file /etc/ssl/sites/yourdomain.dev.crt \
  --key-file      /etc/ssl/sites/yourdomain.dev.key \
  --reloadcmd     &lt;span class="s2"&gt;"nginx -s reload &amp;amp;&amp;amp; apachectl graceful"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The content of the files changes, the paths never do, and nothing in the web server configuration is edited.&lt;/p&gt;

&lt;h3&gt;
  
  
  Expired certificates do not block renewal
&lt;/h3&gt;

&lt;p&gt;The natural worry: every site redirects HTTP to HTTPS, and HTTPS is broken, so how does validation reach the challenge file?&lt;/p&gt;

&lt;p&gt;It works. Let's Encrypt follows the redirect to HTTPS &lt;strong&gt;and does not validate the certificate&lt;/strong&gt; during the HTTP-01 challenge. A long-expired certificate, a self-signed one, a hostname mismatch - none of it stops issuance. You do not need to unwind the redirects first, which is good, because on a generated config you could not anyway.&lt;/p&gt;

&lt;h2&gt;
  
  
  Automate it, then check for a rival
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;acme.sh --install-cronjob
crontab -l &lt;span class="o"&gt;|&lt;/span&gt; grep acme
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That second line is the important one, and it is the mistake we made. The panel's renewal cron was disabled - and a &lt;strong&gt;second&lt;/strong&gt; panel cron doing the same job at a different hour was sitting four lines further down, missed on the first pass. Two renewal mechanisms racing over the same files is a problem that surfaces weeks later as a certificate that is valid on disk and stale in the server's memory.&lt;/p&gt;

&lt;p&gt;Read the whole crontab. Comment out, do not delete, so you can see what used to be there:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;crontab -l &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; /root/crontab.bak-&lt;span class="nv"&gt;$(&lt;/span&gt;date +%Y%m%d&lt;span class="o"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;acme.sh renews at 60 days by default, so a failed run has 30 days of retries before anything is visible to users. That is a generous margin and exactly why a silent failure can run for months.&lt;/p&gt;

&lt;h2&gt;
  
  
  Monitor the certificate that is served, not the file on disk
&lt;/h2&gt;

&lt;p&gt;This is the part that would have caught the original outage, and almost nobody sets it up.&lt;/p&gt;

&lt;p&gt;Checking the file is not enough: it tells you that renewal wrote something, not that the running process picked it up. A reload that silently failed leaves a fresh file on disk and an expired certificate in memory. Check the handshake:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;#!/usr/bin/env bash&lt;/span&gt;
&lt;span class="c"&gt;# days until the served certificate expires, per domain&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; d &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="s2"&gt;"$@"&lt;/span&gt;&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="k"&gt;do&lt;/span&gt;
  &lt;span class="nv"&gt;end&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="nv"&gt;$(&lt;/span&gt;&lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="o"&gt;|&lt;/span&gt; openssl s_client -servername &lt;span class="s2"&gt;"$d"&lt;/span&gt; -connect &lt;span class="s2"&gt;"$d"&lt;/span&gt;:443 2&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/dev/null \
        &lt;span class="o"&gt;|&lt;/span&gt; openssl x509 -noout -enddate 2&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/dev/null &lt;span class="o"&gt;|&lt;/span&gt; cut -d= -f2&lt;span class="o"&gt;)&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="o"&gt;[&lt;/span&gt; -z &lt;span class="s2"&gt;"$end"&lt;/span&gt; &lt;span class="o"&gt;]&lt;/span&gt;&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt; &lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s2"&gt;"CRITICAL $d: no TLS handshake"&lt;/span&gt;&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="nb"&gt;continue&lt;/span&gt;&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="k"&gt;fi&lt;/span&gt;
  &lt;span class="nv"&gt;days&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="nv"&gt;$(&lt;/span&gt;&lt;span class="o"&gt;(&lt;/span&gt; &lt;span class="o"&gt;(&lt;/span&gt; &lt;span class="nv"&gt;$(&lt;/span&gt;date -d &lt;span class="s2"&gt;"$end"&lt;/span&gt; +%s&lt;span class="o"&gt;)&lt;/span&gt; - &lt;span class="nv"&gt;$(&lt;/span&gt;date +%s&lt;span class="o"&gt;)&lt;/span&gt; &lt;span class="o"&gt;)&lt;/span&gt; / 86400 &lt;span class="o"&gt;)&lt;/span&gt;&lt;span class="o"&gt;)&lt;/span&gt;
  &lt;span class="o"&gt;[&lt;/span&gt; &lt;span class="s2"&gt;"$days"&lt;/span&gt; -lt 25 &lt;span class="o"&gt;]&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s2"&gt;"WARN $d: $days days left"&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s2"&gt;"OK $d: $days days"&lt;/span&gt;
&lt;span class="k"&gt;done&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Run it daily across every hostname the box serves and alert under 25 days. Two properties make it the right check: it fails loudly when the handshake fails at all, which is the real outage, and because it reads what the server hands a client, a reload that did not happen cannot hide behind a renewed file.&lt;/p&gt;

&lt;p&gt;Add a tail of the renewal log to the same job and you have the whole story in one alert:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;tail -n 5 /var/log/acme-renew.log
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  The result
&lt;/h2&gt;

&lt;p&gt;Eleven certificates reissued in one pass, all valid for the next three months, four dead sites back up, renewal owned by a 200 KB shell script with a cron entry we can read. No vhost file was modified; only the bytes inside the certificate files changed.&lt;/p&gt;

&lt;p&gt;The wider lesson is not about ACME clients. It is that &lt;strong&gt;the expiry of a thing is not the failure; the silence is.&lt;/strong&gt; The renewal had been broken for roughly two months before the first site went down, and the only reason anyone found out was a site going dark in front of users. A twenty-line handshake check would have raised it the first week.&lt;/p&gt;

&lt;p&gt;If you are putting a new box together rather than rescuing an old one, Caddy does all of this with no configuration at all - we wrote that stack up in &lt;a href="https://news.rozoom.app/self-hosting/caddy-docker-production/" rel="noopener noreferrer"&gt;Caddy and Docker in production&lt;/a&gt;. This recipe is for the machine you inherited, the one where the config is generated, the distro is frozen, and the only acceptable change is the smallest one.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://news.rozoom.app/self-hosting/acme-sh-certificates-legacy-server/" rel="noopener noreferrer"&gt;news.rozoom.app&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosting</category>
      <category>tls</category>
      <category>acme</category>
      <category>letsencrypt</category>
    </item>
  </channel>
</rss>
