search

SUSE / lrbd

GNU Lesser General Public License v2.1
24 stars 19 forks source link

iSCSI multipath, Can it possibly lead to data inconsistency? #10

Closed phucvinh52 closed 7 years ago

phucvinh52 commented 7 years ago

Hello everybody, I using lrbd deploy 2 gateways iscsi, 1 client iscsi connect to 2 gateways (multipath gateway A and gateway B, use alua active/enable). I have a question, when i am writing data to iscsi A, A fail, data continue write to iscsi B. Can it possibly lead to data inconsistency? Sorry i'm bad english. Thanks.

swiftgist commented 7 years ago

Hello, Are you using SUSE kernel for your iSCSI gateways? The included kernel module handles this. BTW, upstream has chosen a different path regarding this solution. I expect the userspace https://github.com/open-iscsi/tcmu-runner to become the preferred solution, but it's currently in progress. I do plan to adapt lrbd if it is still needed. No worries about your English. :)

Eric

phucvinh52 commented 7 years ago

Hi Eric, I do not use SUSE kernel for my iSCSI gateways :'(. But thank for your information, It is helpful for me.

phucvinh52 commented 7 years ago

Hi Eric, Can I use openSUSE (not SUSE enterprise) for gateway?

ddiss commented 7 years ago

Yes, openSUSE Leap 42.2 includes the same kernel. IIUC, www.petasan.org have also picked up the kernel patches. That said, as Eric mentioned, tcmu-runner will likely become the preferred upstream solution.

phucvinh52 commented 7 years ago

Thank you very much @ddiss and i have some question but maybe i should open another title.

squarejester commented 6 years ago

Isn't it true that using TCMU instead of LRBD has the limitation that is DOES NOT support persistent reservations (like the kind REQUIRED for Windows cluster)? Is so, why is OpenSUSE trying to migrate to TCMU in userspace if it is basically useless for deploying block storage for Windows clusters?

ddiss commented 6 years ago

The purpose of migrating to tcmu is to remain consistent with upstream. The tcmu-runner backend is capable of deferring to the kernel for local reservation handling, which uses a local file for storage of PR state.

squarejester commented 6 years ago

In plain English does that mean TCMU instead of lrbd now supports persistent reservation required for Windows clustering or not?