When dealing with legacy daemons that don't support graceful configuration reloads (via SIGHUP), we often need a complete process restart. The standard systemctl reload behavior doesn't accommodate this workflow out of the box.
The naive approach of combining kill and restart in ExecReload fails because:
[Service]
ExecReload=/bin/kill -9 $MAINPID && /usr/bin/MYSERVICE
This creates multiple issues:
- Race conditions between kill and restart
- Systemd loses process tracking
- No proper service state transitions
We need to leverage systemd's restart mechanisms properly:
[Service]
ExecStart=/usr/bin/MYSERVICE
ExecReload=/bin/systemctl kill --signal=KILL MYSERVICE.service
Restart=on-failure
RestartSec=5s
Here's a full service file that handles this properly:
[Unit]
Description=Legacy Daemon with Kill-Restart Reload
[Service]
Type=simple
ExecStart=/usr/local/bin/legacy-daemon --config /etc/legacy.conf
ExecReload=/bin/systemctl kill --signal=KILL legacy-daemon.service
Restart=always
RestartSec=5s
TimeoutStopSec=30
KillMode=process
[Install]
WantedBy=multi-user.target
After implementing:
# systemctl daemon-reload
# systemctl start legacy-daemon
# systemctl reload legacy-daemon # Should kill and restart
# journalctl -u legacy-daemon -f # Verify restart occurred
For services needing pre-restart cleanup:
[Service]
ExecStart=/usr/bin/complex-daemon
ExecReload=/usr/local/bin/daemon-cleanup ; /bin/systemctl restart complex-daemon.service
Restart=no # Since we handle it explicitly
- Ensure your application can handle sudden termination
- Consider file descriptor leaks during hard kills
- Monitor for restart loops in production
- For stateful services, implement proper shutdown handlers
When managing traditional UNIX daemons with systemd, we often encounter scenarios where a simple configuration reload isn't sufficient - sometimes the service needs complete termination and restart. The standard ExecReload directive falls short here as it typically just sends SIGHUP.
With a basic service definition:
[Service]
ExecStart=/usr/bin/MYSERVICE
Type=simple
The systemctl reload command fails because:
- No reload mechanism is defined
- systemd doesn't assume kill/restart behavior by default
The proper approach involves creating a custom reload handler:
[Service]
ExecStart=/usr/bin/MYSERVICE
ExecStop=/bin/kill -TERM $MAINPID
ExecReload=/bin/sh -c "/bin/kill -TERM $MAINPID && /usr/bin/MYSERVICE"
Restart=on-failure
Type=simple
Key components:
- Explicit termination command (
ExecStop) - Compound reload command using shell
- Restart policy for robustness
For complex services, consider using a reload wrapper script:
#!/bin/bash
# /usr/libexec/MYSERVICE-reload
systemctl stop MYSERVICE.service
systemctl start MYSERVICE.service
Then reference it in your unit file:
[Service]
ExecReload=/usr/libexec/MYSERVICE-reload
After implementation:
# Validate the unit file
systemd-analyze verify /etc/systemd/system/MYSERVICE.service
# Test the reload behavior
systemctl daemon-reload
systemctl start MYSERVICE
systemctl reload MYSERVICE
# Check status
journalctl -u MYSERVICE -f
For configuration file changes specifically, consider using path triggers:
[Unit]
Description=My Service
After=network.target
[Path]
PathModified=/etc/MYSERVICE.conf
[Service]
ExecStart=/usr/bin/MYSERVICE
Type=simple
This automatically restarts when the config file changes.