From 76b447a7f4b9651463e98d739ffb4e5976352db5 Mon Sep 17 00:00:00 2001
From: Ewout Prangsma <ewout@prangsma.net>
Date: Thu, 8 Feb 2018 11:20:00 +0100
Subject: [PATCH] Improving constraints

---
 docs/constraints.md | 59 ++++++++++++++++++++-------------------------
 1 file changed, 26 insertions(+), 33 deletions(-)

diff --git a/docs/constraints.md b/docs/constraints.md
index 1cc78d822..1ca3aae25 100644
--- a/docs/constraints.md
+++ b/docs/constraints.md
@@ -3,10 +3,13 @@
 The ArangoDB operator tries to honor various constraints to support high availability of
 the ArangoDB cluster.
 
-## Run agents on separate machines
+## Run agents and dbservers on separate machines
 
-It is essential for HA that agents are running on separate nodes.
-To ensure this, the agent Pods are configured with pod-anti-affinity.
+It is essential for resilience and high availability that no two agents
+are running on the same node and no two dbservers are running running
+on the same node.
+
+To ensure this, the agent and dbserver Pods are configured with pod-anti-affinity.
 
 ```yaml
 kind: Pod
@@ -29,38 +32,28 @@ spec:
             - key: role
               operator: In
               values:
-              - agent
+              - agent (or dbserver)
 ```
 
-## Run dbservers on separate machines
+The settings used for pod affinity are based on the `spec.environment` setting.
 
-It is needed for HA that dbservers are running on separate nodes.
-To ensure this, the dbserver Pods are configured with pod-anti-affinity.
+For a `development` environment we use `preferredDuringSchedulingIgnoredDuringExecution`
+so deployments are still possible on tiny clusters like `minikube`.
 
-```yaml
-kind: Pod
-spec:
-  affinity:
-    podAntiAffinity:
-      requiredDuringSchedulingIgnoredDuringExecution:
-      - weight: 100
-        podAffinityTerm:
-          labelSelector:
-            matchExpressions:
-            - key: app
-              operator: In
-              values:
-              - arangodb
-            - key: arangodb_cluster_name
-              operator: In
-              values:
-              - <cluster-name>
-            - key: role
-              operator: In
-              values:
-              - dbserver
-```
+For `production` environments we enforce (anti-)affinity using
+`requiredDuringSchedulingIgnoredDuringExecution`.
 
-Q: Do we want to allow multiple dbservers on a single node?
-   If so, we should use `preferredDuringSchedulingIgnoredDuringExecution`
-   antiy-affinity.
+## Run coordinators on separate machines
+
+It is preferred to run coordinators of separate machines.
+
+To achieve this, the coordinator Pods are configured with pod-anti-affinity
+using the `preferredDuringSchedulingIgnoredDuringExecution` setting.
+
+## Run syncworkers on same machine as dbserver
+
+It is preferred to run syncworkers on the same machine as
+dbservers.
+
+To achieve this, the syncworker Pods are configured with pod-affinity
+using the `preferredDuringSchedulingIgnoredDuringExecution` setting.