[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] Re: User specific priorities on for iscsi paths doesn't seem to work



On Fri, 2009-08-07 at 17:21 -0400, Mike Snitzer wrote:
> On Fri, Aug 07 2009 at  5:07pm -0400,
> Akshay Lal <alal cleversafe com> wrote:
> 
> > Mike Snitzer wrote:
> >> On Fri, Aug 07 2009 at  4:25pm -0400,
> >> Akshay Lal <alal cleversafe com> wrote:
> >>
> >>   
> >>> I'm having a few issues with path priorities. It seems that the 
> >>> choice  of path to use during I/O is independent of the user defined 
> >>> priorities  for each path.
> >>>
> >>> I am setting the priorities by executing writing a script that is 
> >>> used  by prio_callout. This seems to work when I execute multipath 
> >>> -ll since  all the specified priorities show up correctly. (the   
> >>> path_grouping_policy being used is  failover)
> >>>     
> >>
> >> ...
> >>
> >>   
> >>> Is there something I'm doing wrong? I would like to be able to define 
> >>>  the priorities per device, and ensure that data only traverses on 
> >>> the  lower priority path when
> >>> a) a failure to the first path (path with a higher priority) occures
> >>> b) no other path with a higher priority exists
> >>>     
> >>
> >> Do things behave as you'd like if you change path_grouping_policy to
> >> 'group_by_prio'?
> >>
> >> Mike
> >>   
> 
> 
> > Mike:
> >
> > It seems that if I were to set the path_grouping_policy to  
> > "group_by_prio", then it seems to be working similar to a multibus  
> > configuration. What I would like is for within a single multipath group,  
> > (say mpath1), specify a primary path and an alternate/failover path. If  
> > I can make this configurable via user land tool that'd be great. In this  
> > vein, I had considered priorities with hope that if I can set the  
> > priority of a certain path within a group then the path with the highest  
> > priority will always be chosen and the other path (with the lower  
> > priority) will only come into play when the primary goes down.
> >
> > Below is the output of the multipath -ll & conf file when setting the  
> > path_grouping_policy to group_by_prio.
> >
> >
> > multipath -ll:
> > --------------
> > mpath2 (244534e3833623961) dm-1 DSNET,Dispersed Store
> > [size=47G][features=1 queue_if_no_path][hwhandler=0][rw]
> > \_ round-robin 0 [prio=15][enabled]
> > \_ 85:0:0:0 sdc 8:32  [active][ready]
> > \_ 87:0:0:0 sde 8:64  [active][ready]
> > mpath1 (244534e3266616134) dm-0 DSNET,Dispersed Store
> > [size=47G][features=1 queue_if_no_path][hwhandler=0][rw]
> > \_ round-robin 0 [prio=15][active]
> > \_ 84:0:0:0 sdb 8:16  [active][ready]
> > \_ 86:0:0:0 sdd 8:48  [active][ready]
> >
> >
> > /etc/multipath.conf:
> > --------------------
> > defaults {
> >        udev_dir                /dev
> >        polling_interval        1
> >        selector                "round-robin 0"
> >        path_grouping_policy    group_by_prio
> >        getuid_callout          "/sbin/scsi_id -g -u -s /block/%n"
> >        prio_callout            "/bin/bash  
> > /root/MultipathScripts/mpath_prio_alt %n"
> >        path_checker            tur
> >        rr_min_io               128
> >        max_fds                 8192
> >        rr_weight               priorities
> >        failback                immediate
> >        no_path_retry           queue
> >        user_friendly_names     yes
> > }
> 
> 
> Please don't top-post.
> 
> I'm pretty sure John meant to say "group_by_prio" rather than "failover"
> in his initial reply to this thread.  John originally got this insight
> (dummy device section et. al. applies to RHEL 5.3) back in April:
> 
> https://www.redhat.com/archives/dm-devel/2009-April/msg00157.html
> 
> Which multipath/distro are you using?
> 
> Mike
<snip>
Actually, we are using failover in our environment.  Perhaps I have
missed something but it is working well for us as far as I can tell.  We
have a single path unless it fails in which case we go to the next path
in priority order.  We are handling load balancing across paths in a
different way as we found the performance of multibus was less than we
could achieve otherwise.  Then again, this is not an area of expertise
for me.  Thanks - John
-- 
John A. Sullivan III
Open Source Development Corporation
+1 207-985-7880
jsullivan opensourcedevel com

http://www.spiritualoutreach.com
Making Christianity intelligible to secular society


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]