On Mon, Mar 14, 2011 at 11:43:26AM -0400, Eric Schrock wrote: > On Mon, Mar 14, 2011 at 10:51 AM, Pasi Kärkkäinen <[1][email protected]> wrote: > > Hmm.. so does someone know the status of FMA regarding SES? > > Since I did the majority of that work, yes ;-) >
Thanks for the detailed answer! :) This would be an interesting Google Summer of Code project for someone.. -- Pasi > > [2]http://blogs.sun.com/eschrock/entry/zfs_fma_hotplug_and_fishworks > > [3]http://blogs.sun.com/eschrock/entry/solaris_platform_integration_generic_disk > [4]http://blogs.sun.com/robj/entry/a_sensor_abstraction_layer_for > [5]http://blogs.sun.com/eschrock/entry/ses_sensors > [6]http://prefetch.net/blog/index.php/2008/07/15/scsi-enclosure-services/ > > Was FMA made to work with SES only in fishworks/openstorage and not in > opensolaris? > > There are several parts to FMA and SES integration: > 1. Enumeration of SES devices and sensors > 2. Correlation of 'device-as-detector' SCSI faults (unrecoverable device > error) to physical devices > 3. Monitoring of fan and PSU sensors. > 4. Correlation of ZFS faults with physical devices > Of these, 1-3 are in illumos. So if you have a SES enclosure you will get > things like SMART failure that identify drives in a physically meaningful > way ("Bay 5", etc), provided your SES target is good about setting the > element names correctly. The basic sensor transport and monitoring > framework will also work, so you'll get faults for PSU and fan failure. > The vast majority of #4 is in illumos. ZFS vdevs can be associated with > FRUs (the physical manifestation of a device), and the diagnosis engine > knows how to generate faults against FRUs (look for ZPOOL_CONFIG_FRU). > But nothing ever sets that up in illumos. The reason this never made it > back completely was that the code I wrote didn't handle offline > reconfiguration of devices (powering off and moving devices around), > something that we could get away with in an appliance environment but not > acceptable for general use. > Also, the fishworks syseventd zfs module, which handled automatic > replacement of MPxIO devices, as well as a variety of other cirumstances > (log, cache, spare, etc) never made it back again because it had a lot of > appliance-specific assumptions. > Someone suitably motivated could pick up this work for illumos. It would > be a good opportunity to re-examine the ZFS/FMA model (something that was > talked about within Oracle but will probably never see the light of day). > Hope that helps, > - Eric > -- > Eric Schrock > Delphix > 275 Middlefield Road, Suite 50 > Menlo Park, CA 94025 > [7]http://www.delphix.com > > References > > Visible links > 1. mailto:[email protected] > 2. http://blogs.sun.com/eschrock/entry/zfs_fma_hotplug_and_fishworks > 3. > http://blogs.sun.com/eschrock/entry/solaris_platform_integration_generic_disk > 4. http://blogs.sun.com/robj/entry/a_sensor_abstraction_layer_for > 5. http://blogs.sun.com/eschrock/entry/ses_sensors > 6. http://prefetch.net/blog/index.php/2008/07/15/scsi-enclosure-services/ > 7. http://www.delphix.com/ _______________________________________________ OpenIndiana-discuss mailing list [email protected] http://openindiana.org/mailman/listinfo/openindiana-discuss
