From d08a7f74427ea2cf7d355a0f7f6d8f583e2d0cba Mon Sep 17 00:00:00 2001
From: Carlo Lobrano <c.lobrano@gmail.com>
Date: Thu, 3 Jul 2025 12:22:12 +0200
Subject: [PATCH] OCPBUGS-58324: podman-etcd Add OOM score adjustment for etcd
 containers

This change introduces a new `oom` parameter to the `podman-etcd` OCF
agent. This allows tuning the Out-Of-Memory (OOM) score adjustment for
the etcd container.

The `oom` parameter accepts integer values from -1000 to 1000,
defaulting to -997 (system-node-critical equivalent).

see https://kubernetes.io/docs/concepts/scheduling-eviction/node-pressure-eviction/#node-out-of-memory-behavior

Key changes:
- Added `OCF_RESKEY_oom` parameter to agent definition (`content type="integer"`).
- Integrated `--oom-score-adj` option into `podman_start()`.
- Implemented input validation for `oom` in `podman_validate()`,
  ensuring values are within the [-1000:1000] range.
---
 heartbeat/podman-etcd | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/heartbeat/podman-etcd b/heartbeat/podman-etcd
index 6762112ec..884b7c579 100755
--- a/heartbeat/podman-etcd
+++ b/heartbeat/podman-etcd
@@ -45,6 +45,7 @@ OCF_RESKEY_nic_default="br-ex"
 OCF_RESKEY_authfile_default="/var/lib/kubelet/config.json"
 OCF_RESKEY_allow_pull_default="1"
 OCF_RESKEY_reuse_default="0"
+OCF_RESKEY_oom_default="-997"
 
 : ${OCF_RESKEY_image=${OCF_RESKEY_image_default}}
 : ${OCF_RESKEY_pod_manifest=${OCF_RESKEY_pod_manifest_default}}
@@ -53,6 +54,7 @@ OCF_RESKEY_reuse_default="0"
 : ${OCF_RESKEY_authfile=${OCF_RESKEY_authfile_default}}
 : ${OCF_RESKEY_allow_pull=${OCF_RESKEY_allow_pull_default}}
 : ${OCF_RESKEY_reuse=${OCF_RESKEY_reuse_default}}
+: ${OCF_RESKEY_oom=${OCF_RESKEY_oom_default}}
 
 #######################################################################
 
@@ -230,6 +232,16 @@ to stop the container before pacemaker.
 <shortdesc lang="en">drop-in dependency</shortdesc>
 <content type="boolean"/>
 </parameter>
+
+<parameter name="oom" required="0" unique="0">
+<longdesc lang="en">
+Tune the host's Out-Of-Memory (OOM) preferences for containers (accepts values from -1000 to 1000).
+Default to same OOM score as system-node-critical
+https://kubernetes.io/docs/concepts/scheduling-eviction/node-pressure-eviction/#node-out-of-memory-behavior
+</longdesc>
+<shortdesc lang="en">OOM for container</shortdesc>
+<content type="integer" default="${OCF_RESKEY_oom_default}"/>
+</parameter>
 </parameters>
 
 <actions>
@@ -1226,7 +1238,10 @@ podman_start()
 	fi
 
 	podman_create_mounts
-	local run_opts="-d --name=${CONTAINER}"
+	local run_opts="--detach --name=${CONTAINER}"
+
+	run_opts="$run_opts --oom-score-adj=${OCF_RESKEY_oom}"
+
 	# check to see if the container has already started
 	podman_simple_status
 	if [ $? -eq $OCF_SUCCESS ]; then
@@ -1513,6 +1528,11 @@ podman_validate()
 		exit $OCF_ERR_CONFIGURED
 	fi
 
+	if [ "$OCF_RESKEY_oom" -lt -1000 ] || [ "$OCF_RESKEY_oom" -gt 1000 ]; then
+		ocf_exit_reason "'oom' value ${OCF_RESKEY_oom} is out of range [-1000:1000]"
+		exit $OCF_ERR_CONFIGURED
+	fi
+
 	return $OCF_SUCCESS
 }