[LU-14708] updates to last_rcvd on new client connect and disconnect should be synchronous

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Vladimir Saveliev
• Reporter: Vladimir Saveliev

Description

Otherwise, clients may get evicted if failover happens before update commit.

test_149() {
        remount_client $MOUNT

	replay_barrier_nosync mds1

        fail_nodf mds1

	ls $MOUNT || error "ls failed"
}
run_test 149 "failover after client remount"

ends with

== recovery-small test 149: failover after client remount ============================================ 13:43:02 (1622025782)
Stopping client new4 /mnt/lustre (opts:)
Starting client: new4:  -o user_xattr,flock new4@tcp:/lustre /mnt/lustre
Replay barrier on lustre-MDT0000
Failing mds1 on new4
Stopping /mnt/lustre-mds1 (opts:) on new4
reboot facets: mds1
Failover mds1 to new4
mount facets: mds1
Starting mds1: -o localrecov  /dev/mapper/mds1_flakey /mnt/lustre-mds1
Started lustre-MDT0000
ls: cannot access /mnt/lustre: Input/output error
 recovery-small test_149: @@@@@@ FAIL: ls failed

Issue Links

• is related to: LU-15588 osd_trans_stop() misses th_sync flag set in top thandle