When a disk starts to die slowly multipath starts to Failing & Reinstating paths and this keeps forever.. (I'm using LSI-3008HBA card with SAS-JBOD not FC-Network)


Sep 13 11:20:17 DEV2 kernel: sd 0:0:190:0: attempting task abort! scmd(ffff88110e632948)
Sep 13 11:20:17 DEV2 kernel: sd 0:0:190:0: [sdft] tag#3 CDB: opcode=0x0 00 00 00 00 00 00
Sep 13 11:20:17 DEV2 kernel: scsi target0:0:190: handle(0x0037), sas_address(0x5000c50093d4e7c6), phy(38)
Sep 13 11:20:17 DEV2 kernel: scsi target0:0:190: enclosure_logical_id(0x500304800929ec7f), slot(37)
Sep 13 11:20:17 DEV2 kernel: scsi target0:0:190: enclosure level(0x0001),connector name(1   )
Sep 13 11:20:17 DEV2 kernel: sd 0:0:190:0: task abort: SUCCESS scmd(ffff88110e632948)
Sep 13 11:20:18 DEV2 kernel: device-mapper: multipath: Failing path 130:240.
Sep 13 11:25:34 DEV2 kernel: device-mapper: multipath: Reinstating path 130:240.

As you can see kernel aborted the mission and after that multipath failed. So I want to get rid of this problem via telling Multipath "do not Reinstate the path".
This method will keep dead the zombie disk.

How can I do that?


We would need to know how multipath is determining a path is alive. If you look at multipath.conf you will see there's an optional path_checker option you can see a list of them (with the default being readsector0). Another thing to do is have a path priority ordering such that your path grouping such that disabling failback will prevent it being used unless there's an emergency. According to https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/storage_administration_guide/removing-path-to-storage-device after removing manual references to the sdft you could use the following (but substitute sdft for sda to match the case you gave):

Take the path offline using echo offline > /sys/block/sda/device/state.

This will cause any subsequent I/O sent to the device on this path to be failed immediately. Device-mapper-multipath will continue to use the remaining paths to the device.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.