From 9a37183b26526086ef577acc365b59e1b54471fc Mon Sep 17 00:00:00 2001
From: Stephen Taylor <stephen.taylor.1@att.com>
Date: Fri, 6 May 2022 10:11:31 -0600
Subject: [PATCH] [ceph-osd] Remove ceph-mon dependency in ceph-osd liveness
 probe

It is possible for misbehaving ceph-mon pods to cause the ceph-osd
liveness probe to fail for healthy ceph-osd pods, which can cause
healthy pods to get restarted unnecessarily. This change removes
the ceph-mon query from the ceph-osd liveness probe so the probe
is only dependent on ceph-osd state.

Change-Id: I9e1846cfdc5783dbb261583e04ea19df81d143f4
---
 ceph-osd/Chart.yaml                      |  2 +-
 ceph-osd/templates/bin/osd/_check.sh.tpl | 12 ++++--------
 releasenotes/notes/ceph-osd.yaml         |  1 +
 3 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/ceph-osd/Chart.yaml b/ceph-osd/Chart.yaml
index 54ff58f8e..f5bd86bb4 100644
--- a/ceph-osd/Chart.yaml
+++ b/ceph-osd/Chart.yaml
@@ -15,6 +15,6 @@ apiVersion: v1
 appVersion: v1.0.0
 description: OpenStack-Helm Ceph OSD
 name: ceph-osd
-version: 0.1.40
+version: 0.1.41
 home: https://github.com/ceph/ceph
 ...
diff --git a/ceph-osd/templates/bin/osd/_check.sh.tpl b/ceph-osd/templates/bin/osd/_check.sh.tpl
index dc321806f..3ed90d01a 100644
--- a/ceph-osd/templates/bin/osd/_check.sh.tpl
+++ b/ceph-osd/templates/bin/osd/_check.sh.tpl
@@ -25,17 +25,13 @@ cond=1
 for sock in $SOCKDIR/$SBASE.*.$SSUFFIX; do
  if [ -S $sock ]; then
   OSD_ID=$(echo $sock | awk -F. '{print $2}')
-  OSD_STATE=$(ceph -f json-pretty --connect-timeout 1 --admin-daemon "${sock}" status|grep state|sed 's/.*://;s/[^a-z]//g')
-  NOUP_FLAG=$(ceph --name client.bootstrap-osd --keyring /var/lib/ceph/bootstrap-osd/ceph.keyring status | awk '/flags/{print $2}' | grep noup)
+  OSD_STATE=$(ceph -f json --connect-timeout 1 --admin-daemon "${sock}" status|jq -r '.state')
   echo "OSD ${OSD_ID} ${OSD_STATE}";
-  # this might be a stricter check than we actually want.  what are the
-  # other values for the "state" field?
-  if [ "x${OSD_STATE}x" = 'xactivex' ]; then
-   cond=0
-  elif [ "${NOUP_FLAG}" ] && [ "x${OSD_STATE}x" = 'xprebootx' ]; then
+  # Succeed if the OSD state is active (running) or preboot (starting)
+  if [ "${OSD_STATE}" = "active" ] || [ "${OSD_STATE}" = "preboot" ]; then
    cond=0
   else
-   # one's not ready, so the whole pod's not ready.
+   # Any other state is unexpected and the probe fails
    exit 1
   fi
  else
diff --git a/releasenotes/notes/ceph-osd.yaml b/releasenotes/notes/ceph-osd.yaml
index 296651613..913a16d4f 100644
--- a/releasenotes/notes/ceph-osd.yaml
+++ b/releasenotes/notes/ceph-osd.yaml
@@ -41,4 +41,5 @@ ceph-osd:
   - 0.1.38 Skip pod wait in post-apply job when disruptive
   - 0.1.39 Allow for unconditional OSD restart
   - 0.1.40 Remove udev interactions from osd-init
+  - 0.1.41 Remove ceph-mon dependency in ceph-osd liveness probe
 ...