From 243c1e94c33a6b0643e3d42d068320dfc045c479 Mon Sep 17 00:00:00 2001
From: Louis Cathala <delwiv@protonmail.com>
Date: Sat, 14 Mar 2026 17:22:24 +0100
Subject: [PATCH 1/3] fix: systemd ordering cycle in nvfd-fan-control.service

Remove After= and Wants= directives to avoid ordering cycles caused by
nvfd.service's After=multi-user.target.

Place StartLimitBurst/StartLimitIntervalSec in [Unit] section to rate
limit restarts: max 3 restarts in 400 seconds, with 90s between
attempts.

Changes:
- Removed After=nvfd.service, Wants=nvfd.service (causes cycles)
- Added StartLimitBurst=3, StartLimitIntervalSec=400 in [Unit] section
- RestartSec=90 (to fit within rate limit window)
- Script waits up to 30s for nvfd.service with 1s polling

This ensures:
- No systemd dependency cycles on start or stop
- Graceful waiting for nvfd without aggressive restart storms
- Rate-limited retries if nvfd fails to start
---
 utils/nvfd-fan-control.service |  8 ++++----
 utils/nvfd-fan-control.sh      | 10 ++++++++++
 2 files changed, 14 insertions(+), 4 deletions(-)

diff --git a/utils/nvfd-fan-control.service b/utils/nvfd-fan-control.service
index 8573d91..0d2adb3 100644
--- a/utils/nvfd-fan-control.service
+++ b/utils/nvfd-fan-control.service
@@ -1,13 +1,13 @@
 [Unit]
 Description=NVIDIA Fan Control - Temperature-aware per-GPU mode switching
-After=nvfd.service
-Wants=nvfd.service
+StartLimitIntervalSec=400
+StartLimitBurst=3
 
 [Service]
 Type=simple
 ExecStart=/usr/local/bin/nvfd-fan-control.sh --threshold-up 45 --threshold-down 35
-Restart=always
-RestartSec=5
+Restart=on-failure
+RestartSec=90
 User=root
 
 [Install]
diff --git a/utils/nvfd-fan-control.sh b/utils/nvfd-fan-control.sh
index 7a18b5d..1730504 100755
--- a/utils/nvfd-fan-control.sh
+++ b/utils/nvfd-fan-control.sh
@@ -64,6 +64,16 @@ done
 # Check for root privileges 
 [[ "$EUID" -ne 0 ]] && { echo "ERROR: This script must be run as root (use sudo)" >&2; exit 1; }
 
+# Wait for nvfd service to be active (up to 30 seconds)
+# Systemd handles restart rate limiting via StartLimitBurst/StartLimitIntervalSec
+echo "[INFO] Waiting for nvfd service to be ready..."
+for i in {1..30}; do
+  systemctl is-active --quiet nvfd.service && break
+  [[ $i -eq 30 ]] && exit 1
+  sleep 1
+done
+echo "[INFO] nvfd service is ready"
+
 # Acquire file lock to prevent multiple instances
 exec 200>"$LOCKFILE"
 if ! flock -n 200; then

From 2cca9118d8d35741297fdc6b39262c309112f912 Mon Sep 17 00:00:00 2001
From: Louis Cathala <delwiv@protonmail.com>
Date: Mon, 16 Mar 2026 13:38:42 +0100
Subject: [PATCH 2/3] Update utils/nvfd-fan-control.sh

Add error log if `nvfd.service` is not ready after 30s

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
---
 utils/nvfd-fan-control.sh | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/utils/nvfd-fan-control.sh b/utils/nvfd-fan-control.sh
index 1730504..e404891 100755
--- a/utils/nvfd-fan-control.sh
+++ b/utils/nvfd-fan-control.sh
@@ -69,7 +69,9 @@ done
 echo "[INFO] Waiting for nvfd service to be ready..."
 for i in {1..30}; do
   systemctl is-active --quiet nvfd.service && break
-  [[ $i -eq 30 ]] && exit 1
+  systemctl is-active --quiet nvfd.service && break
+  [[ $i -eq 30 ]] && { echo "[ERROR] nvfd service not ready after 30 seconds. Exiting." >&2; exit 1; }
+  sleep 1
   sleep 1
 done
 echo "[INFO] nvfd service is ready"

From 3182eba2be36b5cac3aa7c7de6018bf668a7e7d8 Mon Sep 17 00:00:00 2001
From: Louis Cathala <delwiv@protonmail.com>
Date: Mon, 16 Mar 2026 13:40:27 +0100
Subject: [PATCH 3/3] remove duplicate lines

---
 utils/nvfd-fan-control.sh | 2 --
 1 file changed, 2 deletions(-)

diff --git a/utils/nvfd-fan-control.sh b/utils/nvfd-fan-control.sh
index e404891..7712de0 100755
--- a/utils/nvfd-fan-control.sh
+++ b/utils/nvfd-fan-control.sh
@@ -68,11 +68,9 @@ done
 # Systemd handles restart rate limiting via StartLimitBurst/StartLimitIntervalSec
 echo "[INFO] Waiting for nvfd service to be ready..."
 for i in {1..30}; do
-  systemctl is-active --quiet nvfd.service && break
   systemctl is-active --quiet nvfd.service && break
   [[ $i -eq 30 ]] && { echo "[ERROR] nvfd service not ready after 30 seconds. Exiting." >&2; exit 1; }
   sleep 1
-  sleep 1
 done
 echo "[INFO] nvfd service is ready"