From 089ab7c1a8490384de4c5e6e7aa313763840dfbf Mon Sep 17 00:00:00 2001
From: Felipe Sanches Zanoni <Felipe.SanchesZanoni@windriver.com>
Date: Thu, 22 Aug 2024 16:00:51 -0300
Subject: [PATCH] Do not stop Ceph fixed monitor when disabling Controller

The fixed Ceph monitor for AIO-DX should never be disabled even if
the controller is Locked. This way the Ceph cluster will have a higher
availability.

This also blocks upgrade rollback when the from-release has only the
floating monitor. When the active controller is in from-release and
the standby controller is in to-release, Ceph will get unresponsive
when the standby controller gets locked and the mon.controller-X
stops.

This change was required by commit:
https://review.opendev.org/c/starlingx/update/+/926971

Test-Plan:
  PASS: AIO-DX: Lock standby controller and verify ceph has three
  monitors quorum and one host is down with HEALTH_WARN status and
  Degraded data redundancy alarm.

Story: 2011122
Task: 50899

Signed-off-by: Felipe Sanches Zanoni <Felipe.SanchesZanoni@windriver.com>
Change-Id: Ie9c9ad800649f1aca65e837acc2e867c575554e6
---
 ceph/ceph/files/ceph.sh | 47 ++++++++++++++++++++++++++++++-----------
 1 file changed, 35 insertions(+), 12 deletions(-)

diff --git a/ceph/ceph/files/ceph.sh b/ceph/ceph/files/ceph.sh
index d9d1806e8..d427f40f3 100644
--- a/ceph/ceph/files/ceph.sh
+++ b/ceph/ceph/files/ceph.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #
-# Copyright (c) 2023 Wind River Systems, Inc.
+# Copyright (c) 2023-2024 Wind River Systems, Inc.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -37,19 +37,42 @@ start ()
 
 stop ()
 {
-    if [[ "$system_type" == "All-in-one" ]] && [[ "$system_mode" == "simplex" ]]; then
+    if [ "${system_type}" == "All-in-one" ] && [ "${system_mode}" == "simplex" ]; then
+        # AIO-SX
         logecho "Ceph services will continue to run on node"
-        exit 0
+        RC=0
+    elif [ "$system_type" == "All-in-one" ] && [ "${system_mode}" != "simplex" ]; then
+        # AIO-DX and AIO-DX+
+        # Will stop OSDs and MDS processes only.
+        # mon.controller will be already stopped on standby controllers.
+        # mon.${hostname} must be running.
+        logecho "Ceph services will be stopped, except local ceph monitor"
+
+        if [ -f ${CEPH_FILE} ]; then
+            rm -f ${CEPH_FILE}
+        fi
+
+        ${INITDIR}/ceph-init-wrapper stop osd >> ${LOGFILE} 2>&1
+        local rc_osd=$?
+        logecho "rc_osd=${rc_osd}"
+
+        ${INITDIR}/ceph-init-wrapper stop mds >> ${LOGFILE} 2>&1
+        local rc_mds=$?
+        logecho "rc_mds=${rc_mds}"
+
+        RC=0
+        [ ${rc_osd} -ne 0 ] || [ ${rc_mds} -ne 0 ] && RC=1
+    else
+        # Standard and Standard Dedicated Storage
+        logecho "Stopping ceph services..."
+
+        if [ -f ${CEPH_FILE} ]; then
+            rm -f ${CEPH_FILE}
+        fi
+
+        ${INITDIR}/ceph-init-wrapper stop >> ${LOGFILE} 2>&1
+        RC=$?
     fi
-
-    logecho "Stopping ceph services..."
-
-    if [ -f ${CEPH_FILE} ]; then
-        rm -f ${CEPH_FILE}
-    fi
-
-    ${INITDIR}/ceph-init-wrapper stop >> ${LOGFILE} 2>&1
-    RC=$?
 }
 
 # If system is an AIO the mtcClient will run this script twice