[LU-16026] Proper ROOT dir handles in remote directory

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Lai Siyao
• Reporter: Shuichi Ihara

Description

Sub directory mount allows mounting a sub directory of filesytem on different mount point on client.
If that directory is not MDT0 and assigned with other MDT index (e.g. MDT1) as a root DIR, that remote directory is mounted on client as "ROOT", but this is actual "ROOT" of filesystem.

It's a problem in this case and cuases too many unnecessary lock grant/cancel when if application traverse the lustre tree (e.g. permission check) many times from same client.
In general, when client send lookup requests with locks to server and it can be kept until it cancels. However, if remote directory is mounted on client as "ROOT", that locks will be canceled and MDS needs to revalidate all time.

Here is a simple reproducer.

Create /exafs/mdt0 with MDT0 and /exafs/mdt1 with MDT1.

[root@ec01 ~]# lfs mkdir -i 0 /exafs/mdt0
[root@ec01 ~]# lfs mkdir -i 1 /exafs/mdt1

[root@ec01 ~]# lfs setdirstripe -D -c 1 -i 0 --max-inherit=-1 /exafs/mdt0
[root@ec01 ~]# lfs setdirstripe -D -c 1 -i 1 --max-inherit=-1 /exafs/mdt1

[root@ec01 ~]# mkdir /mnt/exafs/mdt0 -p
[root@ec01 ~]# mkdir /mnt/exafs/mdt1 -p

mount /exafs/mdt0 and /exafs/mdt1 on client as a sub dir.
[root@ec01 ~]# mount -t lustre 10.0.11.208@o2ib12:10.0.11.209@o2ib12:/exafs/mdt0 /mnt/exafs/mdt0
[root@ec01 ~]# mount -t lustre 10.0.11.208@o2ib12:10.0.11.209@o2ib12:/exafs/mdt1 /mnt/exafs/mdt1

[root@ec01 ~]# df -t lustre
Filesystem                                          1K-blocks    Used   Available Use% Mounted on
10.0.11.208@o2ib12:10.0.11.209@o2ib12:/exafs      56230674432 2433444 55659026780   1% /exafs
10.0.11.208@o2ib12:10.0.11.209@o2ib12:/exafs/mdt0 56230674432 2433444 55659026780   1% /mnt/exafs/mdt0
10.0.11.208@o2ib12:10.0.11.209@o2ib12:/exafs/mdt1 56230674432 2433444 55659026780   1% /mnt/exafs/mdt1

[root@ec01 ~]# lctl set_param ldlm.namespaces.*.lru_size=1000000

 

1st case (mdtest to MDT0 with sub dir mount)

[root@ec01 ~]# lctl set_param mdc.*.stats=clear
[root@ec01 ~]# mpirun -np 16 --allow-run-as-root /work/tools/bin/mdtest -n 10000 -C -E -r -F -u -d /mnt/exafs/mdt0

SUMMARY rate: (of 1 iterations)
   Operation                     Max            Min           Mean        Std Dev
   ---------                     ---            ---           ----        -------
   File creation               17871.676      17871.676      17871.676          0.000
   File stat                       0.000          0.000          0.000          0.000
   File read                   27612.328      27612.328      27612.328          0.000
   File removal                24975.434      24975.434      24975.434          0.000
   Tree creation                 373.126        373.126        373.126          0.000
   Tree removal                   81.089         81.089         81.089          0.000

[root@ec01 ~]# lctl get_param mdc.*MDT0000*.stats
mdc.exafs-MDT0000-mdc-ffff8cedc293e800.stats=
snapshot_time             1147669.149937342 secs.nsecs
start_time                0.000000000 secs.nsecs
elapsed_time              1147669.149937342 secs.nsecs
req_waittime              960054 samples [usec] 22 638501 146216676 449927615916
req_active                960054 samples [reqs] 1 19 7969292 69391178
ldlm_ibits_enqueue        320018 samples [reqs] 1 1 320018 320018
mds_close                 320000 samples [usec] 27 20104 35563148 6982285958
ldlm_cancel               160000 samples [usec] 22 19544 14834323 3810788927
obd_ping                  1 samples [usec] 64 64 64 4096
seq_query                 1 samples [usec] 143 143 143 20449

2nd case (mdtest to remote direcotry with sub dir mount)

[root@ec01 ~]# lctl set_param mdc.*.stats=clear
[root@ec01 ~]# mpirun -np 16 --allow-run-as-root /work/tools/bin/mdtest -n 10000 -C -E -r -F -u -d /mnt/exafs/mdt1

SUMMARY rate: (of 1 iterations)
   Operation                     Max            Min           Mean        Std Dev
   ---------                     ---            ---           ----        -------
   File creation               12583.511      12583.511      12583.511          0.000
   File stat                       0.000          0.000          0.000          0.000
   File read                   17565.493      17565.493      17565.493          0.000
   File removal                15153.482      15153.482      15153.482          0.000
   Tree creation                 393.056        393.056        393.056          0.000
   Tree removal                  235.953        235.953        235.953          0.000


[root@ec01 ~]# lctl get_param mdc.*MDT0001*.stats
mdc.exafs-MDT0001-mdc-ffff8cee701cf800.stats=
snapshot_time             1147812.226056545 secs.nsecs
start_time                0.000000000 secs.nsecs
elapsed_time              1147812.226056545 secs.nsecs
req_waittime              1920132 samples [usec] 23 36225 460375703 180834738289
req_active                1920132 samples [reqs] 1 58 28583891 443285545
ldlm_ibits_enqueue        800057 samples [reqs] 1 1 800057 800057
mds_close                 320000 samples [usec] 31 19646 57650756 14258855424
ldlm_cancel               640037 samples [usec] 23 36225 131679890 44212017584
obd_ping                  2 samples [usec] 89 115 204 21146
seq_query                 2 samples [usec] 69 179 248 36802

There are 2.5x ldlm_ibits_enqueue and 4 x ldlm_cancel in 2nd case(mounted remote directory as "ROOT").