ISCSI Path recovery
ufm
46 Posts
April 23, 2020, 12:45 pmQuote from ufm on April 23, 2020, 12:45 pmBut shouldn't the IP address of the fallen ISCSI Target some time after he returns to work go back to him?
But shouldn't the IP address of the fallen ISCSI Target some time after he returns to work go back to him?
admin
2,930 Posts
April 23, 2020, 3:54 pmQuote from admin on April 23, 2020, 3:54 pmMost failover systems do not do this, in PetaSAN you can still manually move the paths at any time, we also have an automatic balancing of paths, but you have to click to do it.
The main reason is it takes a short time to move the path, current io will block during this, it may be better not to do this automatically. Another reason in PetaSAN, there is no concept of an owning node, all paths can be served by any node.
Most failover systems do not do this, in PetaSAN you can still manually move the paths at any time, we also have an automatic balancing of paths, but you have to click to do it.
The main reason is it takes a short time to move the path, current io will block during this, it may be better not to do this automatically. Another reason in PetaSAN, there is no concept of an owning node, all paths can be served by any node.
Last edited on April 23, 2020, 3:57 pm by admin · #2
ufm
46 Posts
April 23, 2020, 4:34 pmQuote from ufm on April 23, 2020, 4:34 pmI am not ready to say whose approach is more correct. But it seems to me that the ability to specify "enable automatic balancing for this drive" is a good idea.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
I am not ready to say whose approach is more correct. But it seems to me that the ability to specify "enable automatic balancing for this drive" is a good idea.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
admin
2,930 Posts
April 23, 2020, 4:58 pmQuote from admin on April 23, 2020, 4:58 pmIt is only a mouse click, the user is in control of when to do it. Currently our auto rebalance is quite simple, try to average the path count on all nodes and minimize number of moves. We have longer plans to allow different logic for this, for example based on load/traffic historic data (we store chart data for 1 year), we even think there could be some AI that could suggest things, the idea is to propose a more intelligent re-balance instead of pure path count, then suggest this to the admin, but the admin must click to apply.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
Ok i got it 🙂
It is only a mouse click, the user is in control of when to do it. Currently our auto rebalance is quite simple, try to average the path count on all nodes and minimize number of moves. We have longer plans to allow different logic for this, for example based on load/traffic historic data (we store chart data for 1 year), we even think there could be some AI that could suggest things, the idea is to propose a more intelligent re-balance instead of pure path count, then suggest this to the admin, but the admin must click to apply.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
Ok i got it 🙂
Last edited on April 23, 2020, 5:02 pm by admin · #4
ISCSI Path recovery
ufm
46 Posts
Quote from ufm on April 23, 2020, 12:45 pmBut shouldn't the IP address of the fallen ISCSI Target some time after he returns to work go back to him?
But shouldn't the IP address of the fallen ISCSI Target some time after he returns to work go back to him?
admin
2,930 Posts
Quote from admin on April 23, 2020, 3:54 pmMost failover systems do not do this, in PetaSAN you can still manually move the paths at any time, we also have an automatic balancing of paths, but you have to click to do it.
The main reason is it takes a short time to move the path, current io will block during this, it may be better not to do this automatically. Another reason in PetaSAN, there is no concept of an owning node, all paths can be served by any node.
Most failover systems do not do this, in PetaSAN you can still manually move the paths at any time, we also have an automatic balancing of paths, but you have to click to do it.
The main reason is it takes a short time to move the path, current io will block during this, it may be better not to do this automatically. Another reason in PetaSAN, there is no concept of an owning node, all paths can be served by any node.
ufm
46 Posts
Quote from ufm on April 23, 2020, 4:34 pmI am not ready to say whose approach is more correct. But it seems to me that the ability to specify "enable automatic balancing for this drive" is a good idea.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
I am not ready to say whose approach is more correct. But it seems to me that the ability to specify "enable automatic balancing for this drive" is a good idea.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
admin
2,930 Posts
Quote from admin on April 23, 2020, 4:58 pmIt is only a mouse click, the user is in control of when to do it. Currently our auto rebalance is quite simple, try to average the path count on all nodes and minimize number of moves. We have longer plans to allow different logic for this, for example based on load/traffic historic data (we store chart data for 1 year), we even think there could be some AI that could suggest things, the idea is to propose a more intelligent re-balance instead of pure path count, then suggest this to the admin, but the admin must click to apply.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
Ok i got it 🙂
It is only a mouse click, the user is in control of when to do it. Currently our auto rebalance is quite simple, try to average the path count on all nodes and minimize number of moves. We have longer plans to allow different logic for this, for example based on load/traffic historic data (we store chart data for 1 year), we even think there could be some AI that could suggest things, the idea is to propose a more intelligent re-balance instead of pure path count, then suggest this to the admin, but the admin must click to apply.
P.S. After rebooting the nodes with OSD, you do not need to turn on each OSD with your hands, right? 🙂
Ok i got it 🙂