[LU-4397] Permanently disabled OST causes clients to hang on df (statfs)

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.2
• Affects Version/s: Lustre 2.4.1
• Environment: RHEL6 2.6.32-358.18.1.el6.x86_64

Epic/Theme

OST client

• Severity: 4

Epic

client hang

• Assignee: WC Triage
• Reporter: Wolfgang Baudler

Description

A no longer existing OST has been permanently disabled on the MGS using

lctl conf_param vegas-OST0059.osc.active=0

After this, clients hang on df (after lustre is remounted the next time). strace shows it hanging on a statfs call.

Tried with the lustre mount option lazystatfs (undocumented?) which is supposed to work around this issue, but it did not help with lustre 2.4.1, clients still hang on df. This is the standard df, lfs df seems to work OK.

Issue Links

• duplicates: LU-4010 Don't wait for active target if OBD_STATFS_NODELAY flag is set.
• is related to: LU-7668 permanently remove deactivated OSTs from configuration log

Comments

[Wolfgang Baudler]

This seems to work. The syntax is

lctl conf_param <fsname>.llite.lazystafs=1

So, is this expected behaviour and setting this option is required on any lustre filesystem with permanently deactivated OSTs? Or is it still a bug? The conf_param option does not seem to be documented.

[Andreas Dilger]

Note that it is also possible to set lazystatfs permanently on all clients using:

lctl conf_param llite.lazystatfs=1

[Wolfgang Baudler]

I can confirm the behaviour described by Eric Kolb above. Problem exists on 2.5.3. I did not get a chance to test it with 2.5.2, so not sure if it was re-introduced or if it was never fixed.

[Eric Kolb]

Hello,

We recently upgraded out clients to 2.5.3 and this very issue appears to have manifested itself again?

Lustre: setting import RSF1-OST0007_UUID INACTIVE by administrator request
Lustre: Layout lock feature supported.
Lustre: Mounted RSF1-client

$ strace -v -f df /RSF1
statfs("/var/lib/nfs/rpc_pipefs", {f_type=0x67596969, f_bsize=4096, f_blocks=0, f_bfree=0, f_bavail=0, f_files=0, f_ffree=0, f_fsid=

{0, 0}

, f_namelen=255, f_frsize=4096}) = 0
statfs("/RSF1",

At which point the df hangs until reboot.

Because some of these client re-export Lustre via NFS and Samba the exportfs etc. hand on boot. For now we put a hack in /etc/rc.local as work around

/etc/init.d/samba stop
/etc/init.d/nfs stop
mount -t lustre 10.82.0.15@tcp1:/RSF1 /RSF1
/usr/sbin/lctl set_param llite.*.lazystatfs=1
/etc/init.d/nfs start
/etc/init.d/samba start

Perhaps we missed something but this seem fit our experience.

–
Eric Kolb
Data Centre Services
University of Victoria
Office: 250-721-7658

[Peter Jones]

Landed for 2.5.2 and 2.6

[Andreas Dilger]

The change 7762 was landed for 2.5.0, but I found another problem related to this on my test system: http://review.whamcloud.com/8949

[Jodi Levi (Inactive)]

Was this problem resolved with Change, 7762?
Let us know if this ticket can be closed.

[Andreas Dilger]

Seems this may be the same as LU-4010 - patch is http://review.whamcloud.com/7762