Whamcloud - gitweb
LU-17306 ofd: return error for reconnection 95/53195/5
authorAlexander Boyko <alexander.boyko@hpe.com>
Thu, 16 Nov 2023 22:57:24 +0000 (17:57 -0500)
committerOleg Drokin <green@whamcloud.com>
Wed, 20 Dec 2023 01:44:15 +0000 (01:44 +0000)
commitfe098e400458e650998f64c6babba41a6751cef7
tree01e90911405726b2c9a802414b58e863908adc80
parent07e02a600e5707de30e1441ce56b68b0cbc3c260
LU-17306 ofd: return error for reconnection

During the cleanup orphan phase, reconnection leads to unsynchronized
last id between MDT and OST. This means that MDT could assign non
existing objects to a client for a file create operation.

ofd_create_hdl()) capstor-OST0087: dropping old orphan cleanup request
MDS LAST_ID [0x2540000400:0xb6941:0x0] (747841) is 352 behind OST
    LAST_ID [0x2540000400:0xb6aa1:0x0] (748193), trust the OST

recovery-small 144c reproduce bug where MDT lost synchronization
with OST.

Fixes: 63e17799a3 ("LU-8367 osp: enable replay for precreation request")
HPE-bug-id: LUS-11969
Signed-off-by: Alexander Boyko <alexander.boyko@hpe.com>
Change-Id: I22c3d3b3db2acc9ad8f1b978b234afe7d3eef51d
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/53195
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Andriy Skulysh <andriy.skulysh@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
lustre/include/obd_support.h
lustre/ofd/ofd_dev.c
lustre/tests/recovery-small.sh