[ Top posting ]
Hi Andy,
We where looking for this while designing a gluster block utility. But
after staring at your valuable points in reply, we have tweaked the
transaction logic (to send only one command at a time).
Of course it is good to have the mutual exclusion for accessing
configfs in the future.
Please let us know your plans in case if rtslib targets this feature
in the upcoming near releases.
Thanks a lot !
--
Prasanna
On Thu, Jan 12, 2017 at 11:38 PM, Andy Grover <agrover(a)redhat.com> wrote:
> On 01/11/2017 10:48 PM, Prasanna Kalever wrote:
>>
>> Hi,
>>
>> I have observed targetcli fails when we try to configure multiple
>> backstores in parallel
>>
>>
>> # for i in {1..30}; do targetcli /backstores/fileio create ${i}
>> /mnt/file${i} 10M& done
>>
>> Created fileio 3 with size 10485760
>> This _Backstore already exists in configFS
>> This _Backstore already exists in configFS
>>
>> Version Info:
>> targetcli-2.1.fb43
>> python-rtslib-2.1.fb59-2.fc24.noarch
>>
>> Is this a known issue ?
>
>
> Yes, there is no mutual exclusion between different processes accessing
> configfs. We've had an open issue for this for a while:
>
>
https://github.com/open-iscsi/rtslib-fb/issues/36
>
> We could change rtslib to acquire/release some sort of filesystem-based
> lock, but this would also likely require API changes as well.
>
> Was your test case meant just to highlight the issue, or is this an actual
> use case?
>
> There are also base LIO configfs issues that prevent correct creation of
> backstores without doing a recursive scan of /sys/kernel/config/target/core
> first. rtslib performs best with a single instance doing multiple
> operations.
>
> -- Andy
> _______________________________________________
> targetcli-fb-devel mailing list -- targetcli-fb-devel(a)lists.fedorahosted.org
> To unsubscribe send an email to
> targetcli-fb-devel-leave(a)lists.fedorahosted.org