From eead82d7137fd84d0c52fc5729bb0032279a87d1 Mon Sep 17 00:00:00 2001
From: ehila <ehila@redhat.com>
Date: Mon, 11 Nov 2024 21:46:12 -0500
Subject: [PATCH] feat: initial arbiter node addition

added support for creating install for clusters with arbiter nodes
added arbiter flavor to allow using smaller compute sizes for testing
---
 common.sh          |  5 +++++
 config_example.sh  | 21 +++++++++++++++++++++
 ocp_install_env.sh | 15 +++++++++++++++
 utils.sh           | 19 ++++++++++++++++---
 vm_setup_vars.yml  |  9 +++++++++
 5 files changed, 66 insertions(+), 3 deletions(-)

diff --git a/common.sh b/common.sh
index d7d909f55..110c56030 100644
--- a/common.sh
+++ b/common.sh
@@ -271,6 +271,7 @@ export CONTAINER_RUNTIME="podman"
 
 export NUM_MASTERS=${NUM_MASTERS:-"3"}
 export NUM_WORKERS=${NUM_WORKERS:-"2"}
+export ENABLE_ARBITER=${ENABLE_ARBITER:-}
 export NUM_EXTRA_WORKERS=${NUM_EXTRA_WORKERS:-"0"}
 export EXTRA_WORKERS_ONLINE_STATUS=${EXTRA_WORKERS_ONLINE_STATUS:-"true"}
 export EXTRA_WORKERS_NAMESPACE=${EXTRA_WORKERS_NAMESPACE:-"openshift-machine-api"}
@@ -285,6 +286,10 @@ export MASTER_MEMORY=${MASTER_MEMORY:-16384}
 export MASTER_DISK=${MASTER_DISK:-50}
 export MASTER_VCPU=${MASTER_VCPU:-8}
 
+export ARBITER_MEMORY=${ARBITER_MEMORY:-8192}
+export ARBITER_DISK=${ARBITER_DISK:-50}
+export ARBITER_VCPU=${ARBITER_VCPU:-4}
+
 export WORKER_MEMORY=${WORKER_MEMORY:-8192}
 export WORKER_DISK=${WORKER_DISK:-50}
 export WORKER_VCPU=${WORKER_VCPU:-4}
diff --git a/config_example.sh b/config_example.sh
index eefa0ff27..c6432e31e 100755
--- a/config_example.sh
+++ b/config_example.sh
@@ -452,6 +452,16 @@ set -x
 #
 #export NODES_PLATFORM=baremetal
 
+# ENABLE_ARBITER -
+# Set to any non zero length string value to enable the creation of an arbiter node.
+# Arbiter nodes take the place of a master node but run only critical containers to maintain HA for the cluster,
+# it is a TechPreview feature so `export FEATURE_SET="TechPreviewNoUpgrade"` must be set.
+#
+# Furthermore, since an Arbiter node takes the place of a master node the NUM_MASTERS count
+# must be set to `export NUM_MASTERS=2` and not the default `3` to avoid even number etcd members.
+#
+#export ENABLE_ARBITER=1
+
 # MASTER_HOSTNAME_FORMAT -
 # Set a custom hostname format for masters. This is a format string that should
 # include one %d field, which will be replaced with the number of the node.
@@ -476,6 +486,17 @@ set -x
 #export MASTER_DISK=40
 #export MASTER_VCPU=8
 
+# ARBITER_MEMORY, ARBITER_DISK, ARBITER_VCPU -
+# Change VM resources for arbiters
+## Defaults:
+## ARBITER_DISK=50
+## ARBITER_MEMORY=8192
+## ARBITER_VCPU=4
+#
+#export ARBITER_MEMORY=8192
+#export ARBITER_DISK=50
+#export ARBITER_VCPU=2
+
 # WORKER_HOSTNAME_FORMAT -
 # Set a custom hostname format for workers. This is a format string that should
 # include one %d field, which will be replaced with the number of the node.
diff --git a/ocp_install_env.sh b/ocp_install_env.sh
index 8682fbccd..0161eac56 100644
--- a/ocp_install_env.sh
+++ b/ocp_install_env.sh
@@ -181,6 +181,20 @@ EOF
     fi
 }
 
+function arbiterStanza() {
+    if [[ ! -z "${ENABLE_ARBITER:-}" ]]; then
+cat <<EOF
+arbiter:
+  name: arbiter
+  replicas: 1
+  hyperthreading: Enabled
+  architecture: $(get_arch install_config)
+  platform:
+    baremetal: {}
+EOF
+    fi
+}
+
 function libvirturi() {
     if [[ "$REMOTE_LIBVIRT" -ne 0 ]]; then
 cat <<EOF
@@ -314,6 +328,7 @@ controlPlane:
   architecture: $(get_arch install_config)
   platform:
     baremetal: {}
+$(arbiterStanza)
 $(featureSet)
 platform:
   baremetal:
diff --git a/utils.sh b/utils.sh
index 33a6304f7..bd72086ac 100755
--- a/utils.sh
+++ b/utils.sh
@@ -224,9 +224,22 @@ function node_map_to_install_config_hosts() {
     start_idx="$2"
     role="$3"
 
-    for ((idx=$start_idx;idx<$(($1 + $start_idx));idx++)); do
+    # If arbiter is enabled, an arbiter node will be created so we increase the number of hosts by 1
+    # when the role is for master to capture the arbiter position.
+    # If the role is for a worker, we increment the index since the worker position has moved by 1.
+    if [[ ! -z "${ENABLE_ARBITER:-}" && "$role" == "master" ]]; then
+      num_hosts=$((num_hosts + 1))
+    elif [[ ! -z "${ENABLE_ARBITER:-}" && "$role" == "worker" ]]; then
+      start_idx=$((start_idx + 1))
+    fi
+
+    for ((idx=$start_idx;idx<$(($num_hosts + $start_idx));idx++)); do
       name=$(node_val ${idx} "name")
       mac=$(node_val ${idx} "ports[0].address")
+      local node_role=$role
+      if [[ ! -z "${ENABLE_ARBITER:-}" && $name =~ "arbiter" && "$role" == "master" ]]; then
+        node_role=arbiter
+      fi
 
       driver=$(node_val ${idx} "driver")
       if [ $driver == "ipmi" ] ; then
@@ -247,7 +260,7 @@ function node_map_to_install_config_hosts() {
 
       cat << EOF
       - name: ${name}
-        role: ${role}
+        role: ${node_role}
         bmc:
           address: ${address}
           username: ${username}
@@ -266,7 +279,7 @@ EOF
         # FIXME(stbenjam) Worker code in installer should accept
         # "default" as well -- currently the mapping doesn't work,
         # so we use the raw value for BMO's default which is "unknown"
-        if [[ "$role" == "master" ]]; then
+        if [[ "$node_role" == "master" ]] || [[ "$node_role" == "arbiter" ]] ; then
             if [ -z "${MASTER_HARDWARE_PROFILE:-}" ]; then
                 cat <<EOF
         rootDeviceHints:
diff --git a/vm_setup_vars.yml b/vm_setup_vars.yml
index 761ce65d0..b6e97b778 100644
--- a/vm_setup_vars.yml
+++ b/vm_setup_vars.yml
@@ -5,10 +5,13 @@
 # openshift-installer terraform templates
 ironic_prefix: "{{ ironic_prefix }}"
 
+arbiter_enabled: "{{ lookup('env', 'ENABLE_ARBITER') | default(false) }}"
+
 # We enable more memory and masters in dev-scripts compared to the minimal setup
 # in metal3-dev-env
 num_masters: 3
 num_workers: 1
+num_arbiters: "{{ 1 if arbiter_enabled else 0 }}"
 num_extraworkers: 0
 extradisks_list: "{{ lookup('env', 'VM_EXTRADISKS_LIST').split(' ') | default(['vdb']) }}"
 extradisks_size: "{{ lookup('env', 'VM_EXTRADISKS_SIZE') | default(['8G']) }}"
@@ -19,6 +22,12 @@ flavors:
     vcpu: "{{ lookup('env', 'MASTER_VCPU') }}"
     extradisks: "{{ lookup('env', 'VM_EXTRADISKS') | default(false) }}"
 
+  arbiter:
+    memory: "{{ lookup('env', 'ARBITER_MEMORY') }}"
+    disk: "{{ lookup('env', 'ARBITER_DISK') }}"
+    vcpu: "{{ lookup('env', 'ARBITER_VCPU') }}"
+    extradisks: "{{ lookup('env', 'VM_EXTRADISKS') | default(false) }}"
+
   worker:
     memory: "{{ lookup('env', 'WORKER_MEMORY') }}"
     disk: "{{ lookup('env', 'WORKER_DISK') }}"