From bf2ffca5fe11715fbda8eba025bd4eb499f55682 Mon Sep 17 00:00:00 2001
From: Nico Schottelius <nico@nico-notebook.schottelius.org>
Date: Sun, 8 Aug 2021 13:33:53 +0200
Subject: [PATCH] ++knotdns readme

---
 apps/knotdns/README.md | 76 ++++++++++++++++++++++++++++++++++++++----
 1 file changed, 70 insertions(+), 6 deletions(-)

diff --git a/apps/knotdns/README.md b/apps/knotdns/README.md
index 5a88545..6b89ed0 100644
--- a/apps/knotdns/README.md
+++ b/apps/knotdns/README.md
@@ -1,27 +1,91 @@
 ## Authoritative DNS for ungleich
 
-* Zone are stored in git
+* DNS zones are stored in git repository
+* All zones are Bind/Knot compatible below zones/
+* Filenames starting with a dot should be ignored
+  * They are symlinked by some zones, as zones are the same
 
-## Reload mechansim
-
-### Constraints
+## Reload mechansim constraints
 
 * If possible stay with the regular/upstream container
   * Rebuilding causes a delay and extra work
+  * We want to base the work on czniz/knot image
+* Need to generate config file from zones
+  * Very easy to generate
+  * However needs to include synthrecord directives
+  * This step *might* also use kustomize edit?
+* "Double commits" are somewhat ugly
+  * App 1 commits a zone file change
+  * App 2 / CI/CD modifies the configuration file - commits again
+  * The pipeline needs to analyse *what* changed to prevent a circle
+    of commits
+    * git might be smart enough already and failing to commit again,
+      as there is no change
+    * Then we need to
+* Time to deploy should be low
+  * Seconds, not minutes
+  * Rebuilding containers seems to be excessive
+  * Flux might need to get triggered instead of relying only on
+    periodic updates
+    * Might be possible with flux using
+      https://fluxcd.io/docs/components/notification/
+      * Might require https://github.com/fluxcd/notification-controller/issues/230
+
+## Reload using CI/CD pipeline
+
+Theorethical flow:
+
+* Git repository is pushed to CI/CD
+* We need to select a CI/CD system first
+  * Ongoing work in https://redmine.ungleich.ch/issues/9565
+  * Might be bit overkill "just for DNS"
+  * Might be usable for other workflows, too
+* CI/CD "builds" on trigger "something"
+  * A helm chart
+  * A container
+    * Jenkins would be suited for this
+  * A configmap
+    * This overlaps 80% of flux/kustomize
 
 
+## Reload using helm / configmap
 
-## Git cloning inside the pod
+Theorethical flow:
+
+* git push triggers creating a new helm chart
+  * Might need a CI pipeline in between
+  * Might be Jenkins/Buildbot/etc.
+* helm chart is uploaded to a (local) chartmuseum
+* flux updates itself to the latest chart using semversion constraints
+* Might be easy to include a webhook
+
+## Reload using git cloning inside the pod
 
 * It's easy to write a shell script that does git pull && checkzone &&
   reload
 * Needs ssh keys or token inside the pods
+  * Could be injected via env
+* Could use a *git-hook* to reload knot, if the zone files are working
+* Needs git inside the container
+* additional files could be injected via configmap
+* No direct webhook for trigger support
+  * Might have a webhook pod that triggers reload in one or the other way
 
+Sample git
 ```
 git clone https://nico:<TOKEN>@gitea.default.svc.c2.k8s.ooo/nico/ungleich-k8s.git
 ```
 
-### Flux/git repository
+Sketch shell script:
+
+
+```
+#!/bin/sh
+
+
+
+
+### Relooad using Flux/git repository
 
 **TL;DR**