From baf6c3db32f1f3ef942117212534a731537f933e Mon Sep 17 00:00:00 2001
From: jenkins <brad@bstein.dev>
Date: Sun, 4 Oct 2026 07:46:47 -0500
Subject: [PATCH] maintenance: leave runtime restarts to planned maintenance

---
 services/maintenance/node-ops/node-nofile-daemonset.yaml | 3 +++
 services/maintenance/node-ops/scripts/node_nofile.sh     | 8 ++------
 2 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/services/maintenance/node-ops/node-nofile-daemonset.yaml b/services/maintenance/node-ops/node-nofile-daemonset.yaml
index f259dbfc..f28d75e8 100644
--- a/services/maintenance/node-ops/node-nofile-daemonset.yaml
+++ b/services/maintenance/node-ops/node-nofile-daemonset.yaml
@@ -10,6 +10,9 @@ spec:
       app: node-nofile
   updateStrategy:
     type: RollingUpdate
+    rollingUpdate:
+      # Host settings persist while these setup-only helpers are replaced.
+      maxUnavailable: 25%
   template:
     metadata:
       labels:
diff --git a/services/maintenance/node-ops/scripts/node_nofile.sh b/services/maintenance/node-ops/scripts/node_nofile.sh
index dfeed339..000c960b 100644
--- a/services/maintenance/node-ops/scripts/node_nofile.sh
+++ b/services/maintenance/node-ops/scripts/node_nofile.sh
@@ -30,13 +30,9 @@ printf "524288" > /host/proc/sys/fs/inotify/max_user_watches
 printf "32768" > /host/proc/sys/fs/inotify/max_queued_events
 
 if [ "${changed}" -eq 1 ]; then
-  sleep "$(( (RANDOM % 300) + 10 ))"
+  # Load unit definitions now; activate new limits at a planned K3s restart.
+  # A setup-helper rollout must not restart the node's entire pod runtime.
   chroot /host /bin/systemctl daemon-reload
-  for unit in k3s k3s-agent; do
-    if [ -f "/host/etc/systemd/system/${unit}.service" ]; then
-      chroot /host /bin/systemctl restart "${unit}"
-    fi
-  done
 fi
 
 sleep infinity