shadowlin
67 Posts
Quote from shadowlin on April 1, 2018, 12:24 pm
There were some discussion about rbd over multipath iscsi on ceph maillist.
It seems Active/Active multipath can cause data corruption in some corner cases.
I am a newbie to iscsi so maybe my question is kinda stupid:P
1.Is that right the multipath can be achieved on one node with multiple NICs and multiple node with the same rbd?
2.Is Active/Active multipath with the same node with the same rbd perfectly safe?Is the the problem only Active/Active
3.Is Active/Passive multipath with different node and the same rbd safer than Active/Active multipath?Is rbd exclusive lock needed?Can petasan support acitve/passive multipath?
There were some discussion about rbd over multipath iscsi on ceph maillist.
It seems Active/Active multipath can cause data corruption in some corner cases.
I am a newbie to iscsi so maybe my question is kinda stupid:P
1.Is that right the multipath can be achieved on one node with multiple NICs and multiple node with the same rbd?
2.Is Active/Active multipath with the same node with the same rbd perfectly safe?Is the the problem only Active/Active
3.Is Active/Passive multipath with different node and the same rbd safer than Active/Active multipath?Is rbd exclusive lock needed?Can petasan support acitve/passive multipath?
admin
2,930 Posts
Quote from admin on April 1, 2018, 1:00 pm
If you wish to use active/passive, you just need to setup your multipath configuration on the client side to be active/passive. you do not need exclusive lock.
If you wish to use active/passive, you just need to setup your multipath configuration on the client side to be active/passive. you do not need exclusive lock.
shadowlin
67 Posts
Quote from shadowlin on April 1, 2018, 1:19 pm
I see.
Is it safe to use active/active multipath with paths on different nodes?
I see.
Is it safe to use active/active multipath with paths on different nodes?
shadowlin
67 Posts
Quote from shadowlin on April 1, 2018, 2:45 pm
Is it true the multipath patched kernel is only needed for iscsi node? Does the osd need the patched kernel to support multipath?
Is it true the multipath patched kernel is only needed for iscsi node? Does the osd need the patched kernel to support multipath?
admin
2,930 Posts
Quote from admin on April 1, 2018, 3:15 pm
Yes the kernel is only needed for iSCSI gateway. The osd does not need this kernel. We have a patched osd to support vmware commands sent from kernel client, but if you do no use vmware you do not need these changes either.
Yes the kernel is only needed for iSCSI gateway. The osd does not need this kernel. We have a patched osd to support vmware commands sent from kernel client, but if you do no use vmware you do not need these changes either.
shadowlin
67 Posts
Quote from shadowlin on April 1, 2018, 3:47 pm
I see.Thank you.
Another thing I can't figure out is to achieve active/active multipath across nodes. Do the targets need to communicate with echo other? How do they communicate?
I see.Thank you.
Another thing I can't figure out is to achieve active/active multipath across nodes. Do the targets need to communicate with echo other? How do they communicate?
admin
2,930 Posts
Quote from admin on April 1, 2018, 4:34 pm
Regular communication is from client initiator to targets, the client controls this. In few cases the target nodes communicate among themselves, this is done via ceph watch/notify notifications and writing attributes to the rbd object header.
Regular communication is from client initiator to targets, the client controls this. In few cases the target nodes communicate among themselves, this is done via ceph watch/notify notifications and writing attributes to the rbd object header.
shadowlin
67 Posts
Quote from shadowlin on April 2, 2018, 1:32 am
So targets themselves don't communicate directly.The communication is done via ceph.
Thank you, I now have more understanding about how petasan's multipath works now.
So targets themselves don't communicate directly.The communication is done via ceph.
Thank you, I now have more understanding about how petasan's multipath works now.
