From 2b500465dae3f77cbffcd80fc2c65444f0bc28e6 Mon Sep 17 00:00:00 2001
From: Sergiy Markin <smarkin@mirantis.com>
Date: Thu, 12 Dec 2024 00:52:33 +0000
Subject: [PATCH] [ceph] Fix for ceph-osd pods restart

This PS updates ceph-osd pod containers making
sure that osd pods are not stuck at deletion. In
this PS we are taking care of another background
process that has to be terminated by preStop
script.

Change-Id: Icebb6119225b4b88fb213932cc3bcf78d650848f
---
 ceph-osd/Chart.yaml                                | 2 +-
 ceph-osd/templates/bin/osd/_log-runner-stop.sh.tpl | 7 +++++++
 ceph-osd/templates/bin/osd/_log-tail.sh.tpl        | 4 +++-
 releasenotes/notes/ceph-osd.yaml                   | 1 +
 4 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/ceph-osd/Chart.yaml b/ceph-osd/Chart.yaml
index 0d52d074ba..41c556e273 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: 0.1.56
+version: 0.1.57
 home: https://github.com/ceph/ceph
 ...
diff --git a/ceph-osd/templates/bin/osd/_log-runner-stop.sh.tpl b/ceph-osd/templates/bin/osd/_log-runner-stop.sh.tpl
index eed9dbb755..4658c9855c 100644
--- a/ceph-osd/templates/bin/osd/_log-runner-stop.sh.tpl
+++ b/ceph-osd/templates/bin/osd/_log-runner-stop.sh.tpl
@@ -26,3 +26,10 @@ do
   kill -9 ${TAIL_PID};
   sleep 1;
 done
+
+SLEEP_PID="$(cat /tmp/ceph-log-runner-sleep.pid)"
+while kill -0 ${SLEEP_PID} >/dev/null 2>&1;
+do
+  kill -9 ${SLEEP_PID};
+  sleep 1;
+done
diff --git a/ceph-osd/templates/bin/osd/_log-tail.sh.tpl b/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
index 541aa5fbfa..e903760fee 100644
--- a/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
+++ b/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
@@ -25,7 +25,7 @@ function tail_file () {
   while $keep_running; do
     tail --retry -f "${log_file}" &
     tail_pid=$!
-    echo $tail_pid > /tmp/ceph-log-runner.pid
+    echo $tail_pid > /tmp/ceph-log-runner-tail.pid
     wait $tail_pid
     if [ -f /tmp/ceph-log-runner.stop ]; then
       keep_running=false
@@ -37,6 +37,8 @@ function tail_file () {
 function truncate_log () {
   while $keep_running; do
     sleep ${TRUNCATE_PERIOD}
+    sleep_pid=$!
+    echo $sleep_pid > /tmp/ceph-log-runner-sleep.pid
     if [[ -f ${log_file} ]] ; then
       truncate -s "${TRUNCATE_SIZE}" "${log_file}"
     fi
diff --git a/releasenotes/notes/ceph-osd.yaml b/releasenotes/notes/ceph-osd.yaml
index 165a9c522e..2fc30d8429 100644
--- a/releasenotes/notes/ceph-osd.yaml
+++ b/releasenotes/notes/ceph-osd.yaml
@@ -57,4 +57,5 @@ ceph-osd:
   - 0.1.54 Remove use of tini for ceph-daemon
   - 0.1.55 Update ceph-osd pod containers to make sure OSD pods are properly terminated at restart
   - 0.1.56 Add preStop lifecycle script to log-runner
+  - 0.1.57 Added code to kill another background process in log-runner at restart
 ...