Rebooting server results in shutdown of said server
Steven_1980
1 Post
April 18, 2018, 6:20 pmQuote from Steven_1980 on April 18, 2018, 6:20 pmHello all,
I am new to PetaSAN and using version 2.0.0 and so far I like the software. My current (test) setup at home is 3 times Dell Optiplex 990 with 8 GB memory, 120GB SSD for OS,1 2TB for storage and 2 extra network cards plus a home build system with 6 2TB hard drives, 24GB of memory, 64GB USB stick for OS and in total 3 network cards.
Today I logged on by SSH (putty) to one of the management nodes and gave the reboot command, the server reboots and after a short time the server is being switched off.
When in Cluster maintenance the option Fencing is switched to off the server nicely reboots and stays online.
Is this a bug or is this behavior as designed? I have checked the admin guide but it doesn't say anything about it
Kind regards
Steven
Hello all,
I am new to PetaSAN and using version 2.0.0 and so far I like the software. My current (test) setup at home is 3 times Dell Optiplex 990 with 8 GB memory, 120GB SSD for OS,1 2TB for storage and 2 extra network cards plus a home build system with 6 2TB hard drives, 24GB of memory, 64GB USB stick for OS and in total 3 network cards.
Today I logged on by SSH (putty) to one of the management nodes and gave the reboot command, the server reboots and after a short time the server is being switched off.
When in Cluster maintenance the option Fencing is switched to off the server nicely reboots and stays online.
Is this a bug or is this behavior as designed? I have checked the admin guide but it doesn't say anything about it
Kind regards
Steven
admin
2,930 Posts
April 18, 2018, 7:32 pmQuote from admin on April 18, 2018, 7:32 pmYes this is working as expected. If you know you will be rebooting a system you should turn fencing off or re-assign all iSCSI paths to other nodes before rebooting or wait a couple of minutes before restarting.
Fencing is a safety measure that when a node does not respond to the consul cluster via heartbeats and other cluster nodes decide to take over its resources, they try to kill the original node first. This is to make sure the original node is not half dead and still thinks it owns the iSCSI disks and can write to them. The existing cluster nodes have no idea you purposely switched the node off yourself and assume it is going bad.
A side observation regarding putting the os on USB stick, this is not recommended since during cluster recovery the monitors keep issuing and updating cluster maps and need to have decent write speed os disk.
Yes this is working as expected. If you know you will be rebooting a system you should turn fencing off or re-assign all iSCSI paths to other nodes before rebooting or wait a couple of minutes before restarting.
Fencing is a safety measure that when a node does not respond to the consul cluster via heartbeats and other cluster nodes decide to take over its resources, they try to kill the original node first. This is to make sure the original node is not half dead and still thinks it owns the iSCSI disks and can write to them. The existing cluster nodes have no idea you purposely switched the node off yourself and assume it is going bad.
A side observation regarding putting the os on USB stick, this is not recommended since during cluster recovery the monitors keep issuing and updating cluster maps and need to have decent write speed os disk.
Last edited on April 18, 2018, 7:48 pm by admin · #2
Rebooting server results in shutdown of said server
Steven_1980
1 Post
Quote from Steven_1980 on April 18, 2018, 6:20 pmHello all,
I am new to PetaSAN and using version 2.0.0 and so far I like the software. My current (test) setup at home is 3 times Dell Optiplex 990 with 8 GB memory, 120GB SSD for OS,1 2TB for storage and 2 extra network cards plus a home build system with 6 2TB hard drives, 24GB of memory, 64GB USB stick for OS and in total 3 network cards.
Today I logged on by SSH (putty) to one of the management nodes and gave the reboot command, the server reboots and after a short time the server is being switched off.
When in Cluster maintenance the option Fencing is switched to off the server nicely reboots and stays online.Is this a bug or is this behavior as designed? I have checked the admin guide but it doesn't say anything about it
Kind regards
Steven
Hello all,
I am new to PetaSAN and using version 2.0.0 and so far I like the software. My current (test) setup at home is 3 times Dell Optiplex 990 with 8 GB memory, 120GB SSD for OS,1 2TB for storage and 2 extra network cards plus a home build system with 6 2TB hard drives, 24GB of memory, 64GB USB stick for OS and in total 3 network cards.
Today I logged on by SSH (putty) to one of the management nodes and gave the reboot command, the server reboots and after a short time the server is being switched off.
When in Cluster maintenance the option Fencing is switched to off the server nicely reboots and stays online.
Is this a bug or is this behavior as designed? I have checked the admin guide but it doesn't say anything about it
Kind regards
Steven
admin
2,930 Posts
Quote from admin on April 18, 2018, 7:32 pmYes this is working as expected. If you know you will be rebooting a system you should turn fencing off or re-assign all iSCSI paths to other nodes before rebooting or wait a couple of minutes before restarting.
Fencing is a safety measure that when a node does not respond to the consul cluster via heartbeats and other cluster nodes decide to take over its resources, they try to kill the original node first. This is to make sure the original node is not half dead and still thinks it owns the iSCSI disks and can write to them. The existing cluster nodes have no idea you purposely switched the node off yourself and assume it is going bad.
A side observation regarding putting the os on USB stick, this is not recommended since during cluster recovery the monitors keep issuing and updating cluster maps and need to have decent write speed os disk.
Yes this is working as expected. If you know you will be rebooting a system you should turn fencing off or re-assign all iSCSI paths to other nodes before rebooting or wait a couple of minutes before restarting.
Fencing is a safety measure that when a node does not respond to the consul cluster via heartbeats and other cluster nodes decide to take over its resources, they try to kill the original node first. This is to make sure the original node is not half dead and still thinks it owns the iSCSI disks and can write to them. The existing cluster nodes have no idea you purposely switched the node off yourself and assume it is going bad.
A side observation regarding putting the os on USB stick, this is not recommended since during cluster recovery the monitors keep issuing and updating cluster maps and need to have decent write speed os disk.