isi_for_array will send a shutdown -p command to all the nodes, but it's possible that the node where you initiate the isi_for_array starts to shut down before all nodes have received the shutdown -p. Thus, you have nodes that are left online that don't shutdown. isi config is designed with that scenario in mind.
/sbin/shutdown on OneFS is a wrapper for the real shutdown command and it does handle things like flushing caches, writing a backup journal, and unmounting /ifs. But, I agree with others here - isi config > shutdown all is the safest way to shutdown the cluster plus manual verification that all nodes are shut down.
The proper procedure is to use isi config. There is more to gracefully shutting down a cluster than just halting the nodes. The proper procedure takes into account properly handling the journal as well as cluster quorum.
In addition, you need to manually verify that the nodes really have shut down. There's at least one nasty bug out there that reboots a node instead of shutting it down so it takes a second shutdown to finish it (I've never seen it fail to shutdown on the 2nd attempt). This bug makes it difficult to remotely shut down a cluster - you need to be prepared to connect directly to the node, either via the serial port or via a static IP address.
BernieC
76 Posts
2
November 10th, 2015 07:00
isi_for_array will send a shutdown -p command to all the nodes, but it's possible that the node where you initiate the isi_for_array starts to shut down before all nodes have received the shutdown -p. Thus, you have nodes that are left online that don't shutdown. isi config is designed with that scenario in mind.
/sbin/shutdown on OneFS is a wrapper for the real shutdown command and it does handle things like flushing caches, writing a backup journal, and unmounting /ifs. But, I agree with others here - isi config > shutdown all is the safest way to shutdown the cluster plus manual verification that all nodes are shut down.
AdamFox
254 Posts
1
November 10th, 2015 06:00
The proper procedure is to use isi config. There is more to gracefully shutting down a cluster than just halting the nodes. The proper procedure takes into account properly handling the journal as well as cluster quorum.
Anonymous User
170 Posts
1
November 10th, 2015 06:00
In addition, you need to manually verify that the nodes really have shut down. There's at least one nasty bug out there that reboots a node instead of shutting it down so it takes a second shutdown to finish it (I've never seen it fail to shutdown on the 2nd attempt). This bug makes it difficult to remotely shut down a cluster - you need to be prepared to connect directly to the node, either via the serial port or via a static IP address.
carlilek
2 Intern
•
205 Posts
0
November 10th, 2015 12:00
Oh... I can never get the things to actually shut down, they always reboot for me. I'll try the isi config next time.
lucky-7
6 Posts
0
November 10th, 2015 16:00
Hi AdamFox and ed.wilts and Bernie Case
Thanks so much for the answers!
Thanks,