Stopping a DAOS system¶
Follow these steps to completely stop a running DAOS system.
This procedure assumes a standard DAOS RPM installation where
daos_server and daos_agent are managed by systemd.
All commands should be run as root.
-
Check the DAOS system health. At a minimum, run the
dmg system query -vanddmg pool listcommands on an admin node and check that the reported status is as expected. If any issues are found, it is advisable to address those issues before stopping the whole DAOS system. -
Ensure that there are no client applications with open pool connections. In particular, terminate any user jobs that are using DAOS and terminate any
dfusemounts that may still exist. If necessary, thedmg pool evictcommand can be used on an admin node to disconnect any active pool connections. -
Stop the
daos_agentdaemon on all client nodes by runningsystemctl stop daos_agentin a parallel shell. -
Check that the
daos_agentservice is stopped on all client nodes by runningsystemctl status daos_agentin a parallel shell. Resolve any issues when DAOS agent services are still running. -
Stop the DAOS engines by running
dmg system stopon one admin node. -
Verify that all engines are stopped by running the
dmg system query -vcommand on an admin node. If any engines are not in a "Stopped" state, address those issues. -
Stop the
daos_serverdaemon on all server nodes by runningsystemctl stop daos_serverin a parallel shell. -
Check that the
daos_serverservice is stopped on all server nodes by runningsystemctl status daos_serverin a parallel shell. Resolve any issues when DAOS server services are still running. -
In scenarios where an automatic restart of the
daos_serverdaemon after a server reboot is undesirable (for example, during a maintenance where node status needs to be checked before restarting DAOS), disable the systemd service on the affected servers by runningsystemctl disable daos_serverin a parallel shell. -
Similarly, to prevent an automatic restart of the
daos_agentdaemon after client reboots, disable the systemd service on the affected clients by runningsystemctl disable daos_agentin a parallel shell.