<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Raymon Epping on Medium]]></title>
        <description><![CDATA[Stories by Raymon Epping on Medium]]></description>
        <link>https://medium.com/@raymonepping?source=rss-47ce9080309c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*GxqwuTNIO1w5GjMv</url>
            <title>Stories by Raymon Epping on Medium</title>
            <link>https://medium.com/@raymonepping?source=rss-47ce9080309c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 13:21:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@raymonepping/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I Took Terraform Out of My Vault Lab.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/i-took-terraform-out-of-my-vault-lab-78919b7d7595?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*16PG_CePnZJ7T65sd2st8A.png" width="1672"></a></p><p class="medium-feed-snippet">Then It Grew to Eight VMs.</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/i-took-terraform-out-of-my-vault-lab-78919b7d7595?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/i-took-terraform-out-of-my-vault-lab-78919b7d7595?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/78919b7d7595</guid>
            <category><![CDATA[redhat-openshift]]></category>
            <category><![CDATA[redhat-ansible]]></category>
            <category><![CDATA[hashicorp-terraform]]></category>
            <category><![CDATA[vault]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 12:16:01 GMT</pubDate>
            <atom:updated>2026-11-13T12:16:01.570Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[I Built the Same Vault Lab Twice.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/i-built-the-same-vault-lab-twice-fd026261a1a4?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*nRT9iAKeIQbtrnHu6sVNRg.png" width="1672"></a></p><p class="medium-feed-snippet">The Answer Was Both.</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/i-built-the-same-vault-lab-twice-fd026261a1a4?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/i-built-the-same-vault-lab-twice-fd026261a1a4?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/fd026261a1a4</guid>
            <category><![CDATA[hashicorp-terraform]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[redhat-ansible]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 12:16:01 GMT</pubDate>
            <atom:updated>2026-11-13T12:16:01.519Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[My HashiCorp Vault PKI Lab Serves Its Own Certificate]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/my-hashicorp-vault-pki-lab-serves-its-own-certificate-917b05322c4a?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*a3JO7ziYBlxWl8rGVv-FCg.png" width="1672"></a></p><p class="medium-feed-snippet">and Replaces It Every Ten Minutes&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/my-hashicorp-vault-pki-lab-serves-its-own-certificate-917b05322c4a?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/my-hashicorp-vault-pki-lab-serves-its-own-certificate-917b05322c4a?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/917b05322c4a</guid>
            <category><![CDATA[hashicorp-terraform]]></category>
            <category><![CDATA[pki-solution]]></category>
            <category><![CDATA[pki]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 15:31:01 GMT</pubDate>
            <atom:updated>2026-11-09T15:31:01.672Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[AirLock: The Hard Part of Vault Secret Migration]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/airlock-the-hard-part-of-vault-secret-migration-1120e95c8cc7?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*N7L3-vXsMjKRqO1yE9yn-A.png" width="1672"></a></p><p class="medium-feed-snippet">How I designed a controlled plaintext boundary for static secrets that genuinely have to move</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/airlock-the-hard-part-of-vault-secret-migration-1120e95c8cc7?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/airlock-the-hard-part-of-vault-secret-migration-1120e95c8cc7?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/1120e95c8cc7</guid>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[migration]]></category>
            <category><![CDATA[makefile]]></category>
            <category><![CDATA[secret-management-tools]]></category>
            <category><![CDATA[hashicorp-terraform]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 13:41:01 GMT</pubDate>
            <atom:updated>2026-11-09T13:41:01.567Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Ten Everyday Problems Where Vault Belongs in the Answer]]></title>
            <link>https://medium.com/@raymonepping/ten-everyday-problems-where-vault-belongs-in-the-answer-aed6fe33d4f0?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/aed6fe33d4f0</guid>
            <category><![CDATA[vault]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[hashicorp]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[enterprise-software]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Sat, 03 Oct 2026 10:08:14 GMT</pubDate>
            <atom:updated>2026-10-03T10:08:14.446Z</atom:updated>
            <content:encoded><![CDATA[<h4>I built one Vault platform on a laptop to answer every question at once.</h4><figure><img alt="At dusk, an operator with a mug in hand looks out from a calm operations room over a server hall where every rack carries a small amber token light, with the glass-walled issuing room glowing amber at the centre and an empty key board with dust outlines in the foreground." src="https://cdn-images-1.medium.com/max/1024/1*9jT_mDkLPQlejDs9tG5n_g.jpeg" /></figure><blockquote>Here are ten situations from ordinary IT life… <br>…and the part of that platform that deals with each</blockquote><p>In my <a href="https://medium.com/@raymonepping/one-hashicorp-vault-demo-to-rule-them-all-0837919d0303?sk=1afccb854edd8e00fc902f92e83c3dc3"><strong>previous article</strong></a><strong> </strong>I described a HashiCorp Vault Enterprise platform that runs on one laptop:</p><ul><li>ten Vault nodes in four clusters,</li><li>a load balancer,</li><li>identity providers,</li><li>a certificate hierarchy, monitoring,</li><li>a Kubernetes cluster</li><li>and a portal that shows all of it.</li></ul><blockquote>That article was organised around the platform.</blockquote><p><strong>This one is organised around problems.</strong></p><p>Most people do not wake up wanting a secrets platform. They have a password in six configuration files, a certificate that expired on a Sunday, or an auditor asking who read something.</p><p>Below are ten situations I keep meeting, what Vault does about each, and the command or portal page in the platform that shows it working. Every case ends with what a real deployment adds on top of what a laptop can show.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CrqDiVPqEfRoASTN9oGzRg.png" /><figcaption>The portal’s home page: primary, performance and DR clusters with live node state</figcaption></figure><p><strong>The platform all ten cases run on.</strong></p><h3>1. The database password that is in six config files</h3><figure><img alt="A database engineer at a rack in a server hall clips a glowing amber token into a slot on a database server, while six yellowed printouts of the same configuration lie in a recycling tray in the foreground." src="https://cdn-images-1.medium.com/max/1024/1*_At4nIVGLso4n76reXZuNw.jpeg" /></figure><p>An application was given a database account three years ago. The password is in a configuration file, in the deployment pipeline, in a backup of both, and in the head of whoever set it up.</p><blockquote>Nobody rotates it, because nobody knows what breaks when they do.</blockquote><p>Vault’s database engine turns the question around. The application asks Vault for a credential when it needs one. Vault creates a database user on the spot, with a lifetime measured in minutes or hours, and removes it when the lease ends. There is nothing to rotate, because nothing lives long enough to need it. A leaked credential is useless by the time anyone tries it, and the audit log says which application asked for it.</p><p>For accounts that cannot be created on demand, Vault keeps a fixed account and changes its password on a schedule. Vault’s own management account is rotated once at setup, after which only Vault knows its password.</p><p><strong>You have this problem when:</strong> the same database password appears in more than one repository, or nobody can say when it was last changed.</p><p><strong>In the platform:</strong> `make db-test` requests a dynamic credential, logs in to PostgreSQL with it, revokes the lease and shows the login failing. It does the same for a static role, and then proves that a credential issued for one tenant cannot open another tenant’s database.</p><pre># the static role of the tenant &quot;builders&quot;: read it, rotate it, read it again<br>export VAULT_NAMESPACE=builders<br>vault read -field=password database/static-creds/builder-static<br>vault write -f database/rotate-role/builder-static<br>vault read -field=password database/static-creds/builder-static # a different password</pre><p><strong>What production adds:</strong> a conversation with each application team about connection pools, which must cope with credentials that change.</p><h3>2. The certificate that expired on a Sunday</h3><figure><img alt="An empty operations room on a quiet Sunday morning: an unoccupied chair, a wall calendar, and through the glass a small issuing unit pushing out a fresh amber token into a server’s holder while an old framed certificate gathers dust on the desk." src="https://cdn-images-1.medium.com/max/1024/1*9cyadWFdMKJC8XhiRGfk0A.jpeg" /></figure><p>Certificates valid for a year or two are renewed by hand, by someone who has done it once before. The reminder goes to a mailbox of a colleague who left. The outage arrives on a weekend.</p><p>The way out is to make certificates so short-lived that renewal has to be automatic. Vault’s PKI engine issues a certificate in milliseconds through an API call, so a lifetime of days or hours becomes practical. Devices and servers that cannot call an API use the enrolment protocol they already speak: ACME, EST, SCEP and CMPv2 are all enabled.</p><p>The hierarchy has three levels. A root that stays offline and signs one thing. An intermediate inside Vault. Leaf certificates that expire before anyone could misuse them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s9RkZGdYNKD53-C8oAvOHQ.png" /><figcaption>The keys page: protocols, the certificate chain and revocation state</figcaption></figure><blockquote><strong>The keys page: enrolment protocols, the active chain and the revocation list.</strong></blockquote><p><strong>You have this problem when:</strong> certificate expiry dates live in a spreadsheet or a calendar.</p><p><strong>In the platform:</strong> the portal’s own certificate comes from this hierarchy and `make portal-cert-renew` replaces it. The Keys page issues and revokes certificates against the live intermediate. `make certs-status` shows issuer, expiry and names of the listener certificate.</p><p><strong>What production adds:</strong> the root key in an HSM, and a signing ceremony with more than one person in the room.</p><h3>3. Kubernetes workloads with secrets in their manifests</h3><figure><img alt="A container terminal at dusk where each shipping container receives an amber token at the gate after its identity is scanned, while a clipboard manifest with a taped-on password slip hangs forgotten on a fence post." src="https://cdn-images-1.medium.com/max/1024/1*Vppy04KAI8beVAtzlr1YIw.jpeg" /></figure><p>A Kubernetes Secret is base64, which is an encoding and not encryption. Teams put database passwords in them, commit the manifests, and the password is now in Git history for good.</p><p>With Vault, the pod proves who it is with the service-account token Kubernetes already gave it. Vault checks that token with the cluster and hands back only what that service account is allowed to have. No secret is stored in the manifest, the image or the pipeline.</p><p>There are three common ways to deliver the secret to the pod, and teams argue about which is best. The platform runs all three next to each other: the External Secrets Operator, which syncs into a Kubernetes Secret; the Secrets Store CSI driver, which mounts it as a file; and a Vault Agent sidecar, which renders it into the pod.</p><p><strong>You have this problem when:</strong> `kubectl get secret -o yaml` shows something you would not want in a chat message.</p><p><strong>In the platform:</strong> `make k8s-status` shows the three paths, called doors in the demo. Each can be closed from the demo’s page, which revokes the lease in Vault; the pod loses its database access within seconds. All three paths verify Vault’s certificate.</p><p><strong>What production adds:</strong> a choice. Run one delivery method per cluster, not three.</p><h3>4. The application that must encrypt data but should not hold the key</h3><figure><img alt="At a glass counter, an employee slides an open document through a narrow slot into a sealed room where a machine returns it sealed in an amber-lit envelope; the key never leaves the room, while an old key taped under the desk lies in the foreground." src="https://cdn-images-1.medium.com/max/1024/1*cnweHa128gabyhWrOn-6Kg.jpeg" /></figure><p>A team has to encrypt personal data in its database. They pick a library, generate a key, and then face the real question: where does the key go? In practice it ends up next to the data it protects.</p><p>Vault’s Transit engine offers encryption as a service. The application sends plaintext and gets ciphertext back, or the other way around. The key never leaves Vault. Rotating it is one call, old ciphertext stays readable, and it can be rewrapped to the newest key version without the application ever seeing plaintext. Who may encrypt and who may decrypt are separate permissions, so a service that writes records does not have to be able to read them.</p><p>For data that must keep its shape, such as a card number that still has to pass validation, the Transform engine does format-preserving encryption. For large payloads, the application asks Vault for a data key and encrypts locally: envelope encryption.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tHSVrm-sj7fbOV2NiqD6Lg.png" /><figcaption>The engines page: dynamic secrets, PKI and key-value, data protection</figcaption></figure><blockquote><strong>One card per engine. Each runs its operation against the live mount.</strong></blockquote><p><strong>You have this problem when:</strong> the encryption key is in the same configuration as the database connection string.</p><p><strong>In the platform:</strong> the Engines page encrypts, decrypts and rotates against the live Transit mount, and each tenant has its own keys. One mount also has seal wrap switched on, which adds a second layer of encryption from the seal to everything stored there.</p><pre># encrypt, rotate the key, decrypt: the old ciphertext still opens<br>vault write -field=ciphertext demo_transit/encrypt/my-key plaintext=$(printf &#39;IBAN NL00 0000&#39; | base64)<br>vault write -f demo_transit/keys/my-key/rotate<br>vault write -field=plaintext demo_transit/decrypt/my-key ciphertext=&#39;vault:v1:…&#39; | base64 -d</pre><p><strong>What production adds:</strong> an HSM or cloud key service behind the seal. The platform uses SoftHSM to show the same interface in software.</p><h3>5. The secret you need in order to fetch your secrets</h3><figure><img alt="At a loading-dock gatehouse a courier is handed an amber token after the gate checks the van itself, while a laminated card with a written password lies in the gatehouse bin." src="https://cdn-images-1.medium.com/max/1024/1*lFVt9P3cNoAyGofvfXyjaA.jpeg" /></figure><p>Once secrets are in Vault, the application needs a way to log in to Vault. That takes a credential, and the temptation is to put a long-lived token in an environment file.</p><blockquote>The problem has moved one step and is otherwise unchanged.</blockquote><p>The platform has nine machine identities and none of them holds a long-lived token. Each is an AppRole whose logins are accepted from the platform network only. A rotator replaces the secret-id before it expires. A Vault Agent logs in and hands its consumer a token that lives for a short time and is renewed for it. A watchdog restarts the agent when Vault rejects its token.</p><p>One consumer goes further. The metrics exporter has no token at all: it talks to a Vault Proxy on its own loopback address, and the proxy adds the token on the way out.</p><p><strong>You have this problem when:</strong> a Vault token in a pipeline variable is older than the pipeline’s last change.</p><p><strong>In the platform:</strong> `make agents-status` lists every machine identity with its token lifetime, secret-id expiry and last rotation. `make rotation-test` breaks the pattern on purpose: it revokes a token, destroys a secret-id and revokes the token the nodes unseal with, then waits for the platform to repair itself.</p><p><strong>What production adds:</strong> platform identities where they exist. On a cloud or on Kubernetes the workload already has an identity, and Vault can trust that instead of an AppRole.</p><h3>6. The service account nobody dares to rotate</h3><figure><img alt="In a basement plant room a facilities technician replaces the cylinder of an old lock while the machinery keeps running, a new amber token ready in the other hand, and an old brass key hangs on a dusty hook with a faded tag." src="https://cdn-images-1.medium.com/max/1024/1*poSwCHPYzvlPsbJJlKlbzQ.jpeg" /></figure><p>Every directory has them: accounts created for a system years ago, with a password that never expires and membership of more groups than anyone can explain.</p><blockquote>Changing the password means finding every place it is used, so it is not changed.</blockquote><p>Vault’s LDAP secrets engine takes over those accounts. For a fixed account, Vault sets a new password on a schedule and hands the current one to whoever is entitled to it. For temporary needs, Vault creates a directory account with a lifetime and deletes it afterwards. A password policy in Vault defines what a generated password looks like, so it satisfies the directory’s rules.</p><p><strong>You have this problem when:</strong> a directory account has “<em>password never expires</em>” ticked and <em>nobody</em> remembers why.</p><p><strong>In the platform:</strong> `scripts/test-ldap-secrets.sh` runs seven checks, and it checks them against the directory itself: it binds to OpenLDAP with the password Vault issued, rotates, and confirms that the old password is refused.</p><pre>vault read -field=password ldap/static-cred/svc-reports # the current password of a fixed account<br>vault write -f ldap/rotate-role/svc-reports # a new one, set in the directory by Vault<br>vault read ldap/creds/temp-operator # a temporary account with a lease</pre><p><strong>What production adds:</strong> an inventory. Before Vault can rotate an account, someone has to find out what uses it.</p><h3>7. Administrators who share one password</h3><figure><img alt="Three administrators at an operations desk each hold their own amber token and look at their own phone for a code, while one of them peels a faded sticky note off the edge of a shared monitor." src="https://cdn-images-1.medium.com/max/1024/1*XJgL5S_s-QyTQo8axpN9Mw.jpeg" /></figure><p>A small team shares the admin login for a critical system. It is in a password manager, or on a note. When someone leaves, the password stays the same, and the log says “admin” did it.</p><p>In the platform, people log in as themselves. The directory knows people and groups. Keycloak turns the directory into OIDC and SAML. Vault maps a group to a policy, so one group membership gives a person the same rights whichever way they arrive.</p><p>Three controls sit on top:</p><ol><li><strong>A second factor without a third party.</strong> <br>Vault has its own TOTP method. <br>Administrators must present a code from any authenticator app; nothing outside the platform is involved.</li><li><strong>Lockout.</strong><br>Five wrong attempts lock an account for fifteen minutes.</li><li><strong>Vault as an identity provider.</strong><br>Grafana has a “<em>Sign in with Vault</em>” button, so an application behind Vault learns who someone is without seeing a password.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gVriIZUvcOsLE3oVZS2_HQ.png" /><figcaption>The identity page: every enabled auth method with a live login</figcaption></figure><blockquote>Each card performs a real login against the cluster.</blockquote><p><strong>You have this problem when:</strong> an offboarding checklist contains the line “<em>change the shared passwords</em>”.</p><p><strong>In the platform:</strong> `make users` lists the directory users, their groups and whether MFA applies. `scripts/test-mfa.sh` checks five things, among them that the right password without a code yields a challenge and no token. `make identity-verify` logs in as each demo user.</p><pre># a directory user in the MFA group: password alone gives a challenge, not a token<br>vault login -method=ldap username=repping<br># Vault asks for the code from the authenticator before it issues anything</pre><p><strong>What production adds:</strong> MFA for everyone, enforced at the identity provider the organisation already has.</p><h3>8. “Who read that secret, and when?”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kJ-glvW71TVuEKni-zJVqw.jpeg" /></figure><p>An auditor asks who accessed the production database credentials last quarter. With passwords in files, the true answer is: anyone who could read the file, and we cannot tell.</p><p>Vault writes every request and every response to its audit log, with the identity, the path and the result. Secret values are hashed in the log, so the log itself is not a leak, and a hash can still be matched against a known value.</p><p>How the log reaches the SIEM matters more than it seems. Vault refuses to complete a request when no audit device can record it. A device that writes straight to the SIEM therefore turns a SIEM outage into a Vault outage. In the platform, each node writes to a file, a sidecar rotates it, and forwarders translate events to CEF and send them on. The SIEM can be down for a day without Vault noticing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MAXW_EzwdIYyu3Ho1BB4Ig.png" /><figcaption>The audit page: identity, path and result of every request</figcaption></figure><blockquote><strong>The same files feed Loki and the SIEM.</strong></blockquote><p><strong>You have this problem when:</strong> answering the auditor takes a week and ends with “probably”.</p><p><strong>In the platform:</strong> `make audit-tail NODE=vault-1` follows a node’s log, and the Audit page shows identity, path and result per request. `make siem-up` starts Splunk when a demo needs it. Because Vault Enterprise is licensed by clients, an exporter publishes the client count and Grafana charts it.</p><p><strong>What production adds:</strong> a second audit device of a different type, and retention that matches the regulation you answer to.</p><h3>9. Rules that live in a wiki</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y4oKkTCs89IJjo_JaBiUYA.jpeg" /></figure><p>“Production secrets must have an owner.” “Regulated keys are never exportable.” “No changes outside the change window.” These sentences are in a policy document. Whether anyone follows them is checked once a year, by sampling.</p><p>Access-control policies in Vault say which identity may touch which path. Rules about the request itself are a job for Sentinel, and the platform has four, at three enforcement levels on purpose:</p><ul><li>A secret under a regulated path must carry an owner and a classification. Hard: there is no override.</li><li>A regulated Transit key can never be made exportable or deletable. Soft: an operator can override, and the override is in the audit log.</li><li>Changes outside the change window are allowed and logged. Advisory.</li><li>One group of people can reach Vault through the load balancer only.</li></ul><p>The configuration itself follows the same idea. Every mount, policy, role and namespace is in Terraform, in seven roots, so a change is a reviewed diff.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a9TRpZ2JN0xpWvQoJgOpfA.png" /><figcaption>The control page: namespaces, Control Groups and separation of duties</figcaption></figure><blockquote><strong>A Control Group holds a secret until a second identity approves the request.</strong></blockquote><p><strong>You have this problem when:</strong> a rule is checked by a yearly sample instead of at the moment someone breaks it.</p><p><strong>In the platform:</strong> `scripts/test-sentinel.sh` runs twelve checks with a token whose access-control policies would allow every request, covering both the refusals and the requests that must still pass. `make tf-plan-all` fails when the running clusters differ from the code.</p><pre># refused, and the answer names the policy: regulated-secrets-need-owner<br>vault kv put secret/regulated/payments value=x<br># accepted<br>vault kv put secret/regulated/payments value=x owner=platform-team classification=confidential</pre><p><strong>What production adds:</strong> the plan running in a pipeline, so drift is found the day it appears.</p><h3>10. Losing a data centre, or a bad upgrade</h3><figure><img alt="At dusk two identical data-centre buildings stand across a still river; one has gone dark and the other lights up floor by floor in amber and white, a bridge between them carrying traffic towards the lit one." src="https://cdn-images-1.medium.com/max/1024/1*18Ap653E14NCAMiUOPWPuA.jpeg" /></figure><p>Once everything depends on Vault, Vault is the system that may not go down. Two days matter most: the day a site is lost and the day an upgrade goes wrong.</p><p>For the first, a disaster-recovery cluster mirrors the primary, tokens and leases included. Failing over is a procedure, and the portal runs it step by step: fence the primary in the load balancer, promote the DR cluster, route clients to it. The last time I ran it from the portal, the failover took 11 seconds and the failback 26.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8p4D02BywPfqRDU3BrtBxQ.png" /><figcaption>The availability page: primary and DR state with the failover procedure</figcaption></figure><blockquote><strong>The red button starts a real promotion.</strong></blockquote><p>Rehearsing it taught me two things that no diagram shows. The token that authorises the promotion has to be a batch token minted on the primary beforehand, because it cannot be minted once the primary is gone. And when the DR cluster is promoted, the performance secondary follows it without being asked, but it does not come back on its own. My failback worked for the DR cluster and left the performance secondary pointing at the wrong place, while a status check still reported it as streaming. Failback now brings it home as its last step, and the check looks at the connection instead of the label.</p><p>For the second day there are backups that are proven. `make backup` writes snapshots together with the unseal material that opens them, encrypted. `make restore-drill` restores that set into throwaway containers and reads a known value back. Upgrades go one node at a time, secondaries first, while a client keeps reading. Vault has no in-place downgrade, so going back means restoring the backup taken just before.</p><p><strong>You have this problem when:</strong> the failover procedure exists as a document and was last run by someone who has since changed jobs.</p><p><strong>In the platform:</strong> `make dr-token`, `make dr-failover`, `make dr-failback`, `make replication-status`, `make restore-drill`, `make vault-upgrade TAG=…` and `make vault-rollback`.</p><pre>make dr-token # before the bad day, not during it<br>make dr-failover # fence, promote, route; asks you to type FAILOVER<br>make dr-failback # and back; asks you to type FAILBACK<br>make replication-status # fails unless both secondaries are connected to the primary</pre><p><strong>What production adds:</strong> a second site. On one laptop the “lost” primary is still running, fenced. A real loss is a procedure in the runbook and not a button.</p><h3>What the ten have in common</h3><p>In every case the starting point is a long-lived secret that people have to manage: a password, a key, a certificate, a shared login. Vault’s answer is the same each time. Make the credential short-lived, issue it to an identity that has proven itself, and record that it happened. The ten cases differ in what is being issued and to whom.</p><p>That is also why they fit on one platform. The database credential in case 1 is requested by the machine identity from case 5, over TLS from case 2, inside a namespace guarded by the rules from case 9, written to the audit log from case 8, and still there after the failover in case 10. Separate demos show each part on its own. Running them together shows whether they still work when they depend on each other.</p><h4>Where to start</h4><p>Nobody solves ten problems at once. If I had to order them for an organisation that is starting out, it would be this:</p><ol><li><strong>Machine identities (5) and audit (8) first. </strong><br>Everything else stands on these two. Without a sound way for workloads to log in, the other cases reintroduce the long-lived secret they were meant to remove. Without the audit log you cannot show that anything improved.</li><li><strong>Then the case that hurts most.</strong><br>For most teams that is database credentials (1) or certificates (2), because those cause the outages people remember.</li><li><strong>Rules as code (9) before the second team arrives.</strong><br>One team can keep its own house in order. With two teams the conventions drift, and the rules have to be enforced by the platform.</li><li><strong>Failover and restore (10) before anything critical depends on it.</strong> Rehearse it while a failed rehearsal is still cheap. <br><em>Mine failed twice, on a laptop, with nobody waiting.</em></li></ol><h4>Try it yourself</h4><p>The platform lives in a <a href="https://github.com/raymonepping/vault_reference"><strong>private repository</strong></a> on <a href="https://github.com/raymonepping/vault_reference"><strong>GitHub</strong></a>. <br>Access is on request: leave a comment on this article or send me a message with your GitHub username, and I will add you.</p><p>The <a href="https://medium.com/@raymonepping/one-hashicorp-vault-demo-to-rule-them-all-0837919d0303?sharedUserId=raymonepping"><strong>previous article</strong></a> covers what you need on the host and how the platform is built up.</p><p>Once it runs, each case above is a command:</p><ol><li>Database credentials | `make db-test` |</li><li>Certificates | Keys page, `make certs-status`, `make portal-cert-renew` |</li><li>Kubernetes | `make k8s-status` |</li><li>Encryption as a service | Engines page |</li><li>Machine identities | `make agents-status`, `make rotation-test` |</li><li>Directory accounts | `scripts/test-ldap-secrets.sh` |</li><li>People and MFA | `make users`, `scripts/test-mfa.sh` |</li><li>Audit | Audit page, `make audit-tail NODE=vault-1` |</li><li>Rules as code | `scripts/test-sentinel.sh`, `make tf-plan-all` |</li><li>Failover, backup, upgrade | `make dr-failover`, `make restore-drill`, `make vault-upgrade` |</li></ol><p>If one of the ten is your problem this quarter, start with that command and read the script behind it.</p><blockquote><strong>Questions and corrections are welcome here.</strong></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aed6fe33d4f0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[One HashiCorp Vault Demo to Rule Them All]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/one-hashicorp-vault-demo-to-rule-them-all-0837919d0303?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1440/1*CrqDiVPqEfRoASTN9oGzRg.png" width="1440"></a></p><p class="medium-feed-snippet">An Enterprise-Grade HashiCorp Vault Platform on One Laptop</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/one-hashicorp-vault-demo-to-rule-them-all-0837919d0303?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/one-hashicorp-vault-demo-to-rule-them-all-0837919d0303?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/0837919d0303</guid>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[podman]]></category>
            <category><![CDATA[vault]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[devsecops]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Fri, 02 Oct 2026 13:10:21 GMT</pubDate>
            <atom:updated>2026-10-02T13:11:39.837Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Building a Multi-Tenant Cryptographic Control Plane with HashiCorp Vault]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/building-a-multi-tenant-cryptographic-control-plane-with-hashicorp-vault-77ceb7176b5e?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/2600/1*V9H95gdxQ30KgttWa0HFpg.png" width="2688"></a></p><p class="medium-feed-snippet">Implementing Enterprise Key Management, Transit, PKI, and HSM Custody</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/building-a-multi-tenant-cryptographic-control-plane-with-hashicorp-vault-77ceb7176b5e?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/building-a-multi-tenant-cryptographic-control-plane-with-hashicorp-vault-77ceb7176b5e?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/77ceb7176b5e</guid>
            <category><![CDATA[vault]]></category>
            <category><![CDATA[cryptography]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[devsecops]]></category>
            <category><![CDATA[multitenancy]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 28 Sep 2026 13:38:46 GMT</pubDate>
            <atom:updated>2026-10-06T08:16:20.106Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Permission That Expires the Moment It’s Used]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/the-permission-that-expires-the-moment-its-used-df4b3e324d65?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*EOHLgyvq9JhytY41FCnVCw.png" width="1672"></a></p><p class="medium-feed-snippet">Why an AI agent should never hold more authority than the task in front of it, not even on a second try</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/the-permission-that-expires-the-moment-its-used-df4b3e324d65?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/the-permission-that-expires-the-moment-its-used-df4b3e324d65?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/df4b3e324d65</guid>
            <category><![CDATA[devsecops]]></category>
            <category><![CDATA[agentic-ai-architecture]]></category>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 28 Sep 2026 12:24:27 GMT</pubDate>
            <atom:updated>2026-09-28T12:30:47.211Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Vault Behind the Permission]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/the-vault-behind-the-permission-8c5a14690d5c?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*Cz5M6Hh-YmLEyi2pRx628A.png" width="1672"></a></p><p class="medium-feed-snippet">What building governed AI agent authority twice, on HashiCorp Vault Enterprise, actually taught me</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/the-vault-behind-the-permission-8c5a14690d5c?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/the-vault-behind-the-permission-8c5a14690d5c?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c5a14690d5c</guid>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[agentic-ai-security]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[enterprise-vault]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 28 Sep 2026 11:49:13 GMT</pubDate>
            <atom:updated>2026-09-28T12:29:43.550Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Permission Vault Has to Prove]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@raymonepping/the-permission-vault-has-to-prove-a2e90f937424?source=rss-47ce9080309c------2"><img src="https://cdn-images-1.medium.com/max/1672/1*tv03T8FPkg5Di_gVBUp02g.png" width="1672"></a></p><p class="medium-feed-snippet">What HashiCorp Vault enforces for AI agents, what I added around it, and what changed when I put native Agentic IAM to the test</p><p class="medium-feed-link"><a href="https://medium.com/@raymonepping/the-permission-vault-has-to-prove-a2e90f937424?source=rss-47ce9080309c------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@raymonepping/the-permission-vault-has-to-prove-a2e90f937424?source=rss-47ce9080309c------2</link>
            <guid isPermaLink="false">https://medium.com/p/a2e90f937424</guid>
            <category><![CDATA[hashicorp-vault]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[directed-acyclic-graph]]></category>
            <category><![CDATA[devops]]></category>
            <dc:creator><![CDATA[Raymon Epping]]></dc:creator>
            <pubDate>Mon, 28 Sep 2026 10:39:01 GMT</pubDate>
            <atom:updated>2026-09-28T12:27:03.300Z</atom:updated>
        </item>
    </channel>
</rss>