From 09f1ccc64ca07f27666e4b949cca032148d5d13a Mon Sep 17 00:00:00 2001
From: "SPEARS, DUSTIN (ds443n)" <ds443n@att.com>
Date: Tue, 5 Nov 2024 16:16:48 -0500
Subject: [PATCH] Update ceph-osd to be able to use tini

Sometimes the pod fails to terminate correctly,
leaving zombie processes. Add option to use tini
to handle processes correctly. Additionally update
log-tail script to handle sigterm and sigint.

Change-Id: I96af2f3bef5f6c48858f1248ba85abdf7740279c
---
 ceph-osd/Chart.yaml                         |  2 +-
 ceph-osd/templates/bin/osd/_log-tail.sh.tpl | 14 +++++++++++---
 ceph-osd/templates/daemonset-osd.yaml       | 14 ++++++++++----
 ceph-osd/values.yaml                        |  5 +++++
 releasenotes/notes/ceph-osd.yaml            |  1 +
 5 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/ceph-osd/Chart.yaml b/ceph-osd/Chart.yaml
index d03d42b85..8b1687ea9 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: 0.1.52
+version: 0.1.53
 home: https://github.com/ceph/ceph
 ...
diff --git a/ceph-osd/templates/bin/osd/_log-tail.sh.tpl b/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
index a6b3edd10..301259142 100644
--- a/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
+++ b/ceph-osd/templates/bin/osd/_log-tail.sh.tpl
@@ -18,14 +18,20 @@ wait_for_file "${osd_id_file}" "${WAIT_FOR_OSD_ID_TIMEOUT}"
 log_file="/var/log/ceph/${DAEMON_NAME}.$(cat "${osd_id_file}").log"
 wait_for_file "${log_file}" "${WAIT_FOR_OSD_ID_TIMEOUT}"
 
+trap "exit" SIGTERM SIGINT
+keep_running=true
+
 function tail_file () {
-  while true; do
-    tail --retry -f "${log_file}"
+  while $keep_running; do
+    tail --retry -f "${log_file}" &
+    tail_pid=$!
+    wait $tail_pid
+    sleep 1
   done
 }
 
 function truncate_log () {
-  while true; do
+  while $keep_running; do
     sleep ${TRUNCATE_PERIOD}
     if [[ -f ${log_file} ]] ; then
       truncate -s "${TRUNCATE_SIZE}" "${log_file}"
@@ -37,3 +43,5 @@ tail_file &
 truncate_log &
 
 wait -n
+keep_running=false
+wait
diff --git a/ceph-osd/templates/daemonset-osd.yaml b/ceph-osd/templates/daemonset-osd.yaml
index 3ba2ce7e9..47f801418 100644
--- a/ceph-osd/templates/daemonset-osd.yaml
+++ b/ceph-osd/templates/daemonset-osd.yaml
@@ -315,8 +315,11 @@ spec:
               value: {{ .Values.logging.truncate.period | quote }}
             - name: WAIT_FOR_OSD_ID_TIMEOUT
               value: {{ .Values.logging.osd_id.timeout | quote }}
-          command:
-            - /tmp/log-tail.sh
+          {{- if .Values.conf.tini.log_runner.enabled }}
+          command: ["/usr/local/bin/tini", "--", "/tmp/log-tail.sh"]
+          {{- else }}
+          command: ["/tmp/log-tail.sh"]
+          {{- end }}
           volumeMounts:
             - name: pod-tmp
               mountPath: /tmp
@@ -357,8 +360,11 @@ spec:
               value: {{ tuple "ceph_mon" "internal" "mon" $envAll | include "helm-toolkit.endpoints.endpoint_port_lookup" | quote }}
             - name: MON_PORT_V2
               value: {{ tuple "ceph_mon" "internal" "mon_msgr2" $envAll | include "helm-toolkit.endpoints.endpoint_port_lookup" | quote }}
-          command:
-            - /tmp/osd-start.sh
+          {{- if .Values.conf.tini.ceph_osd_default.enabled }}
+          command: ["/usr/local/bin/tini", "--", "/tmp/osd-start.sh"]
+          {{- else }}
+          command: ["/tmp/osd-start.sh"]
+          {{- end }}
           lifecycle:
             preStop:
               exec:
diff --git a/ceph-osd/values.yaml b/ceph-osd/values.yaml
index 27df42d1a..8d6c8e86e 100644
--- a/ceph-osd/values.yaml
+++ b/ceph-osd/values.yaml
@@ -179,6 +179,11 @@ jobs:
     startingDeadlineSecs: 60
 
 conf:
+  tini:
+    log_runner:
+      enabled: false
+    ceph_osd_default:
+      enabled: false
   ceph:
     global:
       # auth
diff --git a/releasenotes/notes/ceph-osd.yaml b/releasenotes/notes/ceph-osd.yaml
index a90aa5730..f7c077829 100644
--- a/releasenotes/notes/ceph-osd.yaml
+++ b/releasenotes/notes/ceph-osd.yaml
@@ -53,4 +53,5 @@ ceph-osd:
   - 0.1.50 Allow lvcreate to wipe existing LV metadata
   - 0.1.51 Update Ceph images to patched 18.2.2 and restore debian-reef repo
   - 0.1.52 Use quay.io/airshipit/kubernetes-entrypoint:latest-ubuntu_focal by default
+  - 0.1.53 Update ceph-daemon to be able to use tini init system
 ...