On 03/23/2012 02:41 PM, Hart, Brian R. wrote: (https://www.redhat.com/archives/dm-devel/2012-March/msg00155.html) > We have a couple of Dell MD storage arrays that when installed > setup multipath.conf to use this line: prio_callout > "/sbin/mpath_prio_rdac /dev/%n" On 03/23/2012 02:41 PM, Bryn M. Reeves wrote: (https://www.redhat.com/archives/dm-devel/2012-March/msg00155.html) >Callouts make life difficult when file systems go away so like the >path checkers before them they were merged into the libmultipath >shared library. This means that the daemon can lock them into memory >via mlock(2)/mlockall(2) and not have to worry about being able to >load a binary from disk when the paths to that storage have failed - >this is very useful if your root file system is on multipath and you >need to recover from a failure. > >In RHEL5 this is dealt with using a complex and fragile private >namespace and RAM-backed file system - the required binaries are >copied into this ramfs at daemon startup and the daemon unmounts >unnecessary file systems to avoid blocking when failures occur. > >For this reason the parameter is now just "prio", e.g.: > >device { … > prio rdac >} I can see that using shared libraries to handle priorities makes handling failed paths easier in many situations. But how to configure active/backup multipath in the following case, for example - 1st path uses iSCSI over Infiniband - 2nd path used iSCSI over Gigabit Ethernet - iSCSI target does not support ALUA - we want Infiniband path to have higher priority Will it be enough to set path_grouping_policy = multibus and path_selector = “service-time 0”? Will Gigabit Ethernet path be never used as long as Infiniband path is working? I appreciate your help, Alexander Murashkin Email: AlexanderMurashkin@xxxxxxx |
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
