[LU-10959] Multirail remote peer issue

Fri, 12 Aug 2022 22:03:04 +0000

We are facing an issue with multirail remote peer declaration with Lustre 2.10.3.

Here is the description of the test cluster:
MGS:
eth0 10.128.11.155

MDS:
eth0 10.128.11.156
eth0:0 10.128.12.156

OSS:
eth0 10.128.11.157
eth0:0 10.128.12.157

Router:
eth0 10.128.11.158
eth0:0 10.128.12.158

Client:
eth0 10.128.11.159
eth0:0 10.128.12.159

Here is how I configure LNet on the different nodes:
On MGS node:

# modprobe lnet ; lnetctl lnet configure
# lnetctl net add --net tcp --if eth0
# lnetctl peer add --nid 10.128.11.156@tcp,10.128.12.156@tcp
# lnetctl peer add --nid 10.128.11.157@tcp,10.128.12.157@tcp
# lnetctl peer add --nid 10.128.11.158@tcp,10.128.12.158@tcp
# lnetctl route add --net tcp1 --gateway 10.128.11.158@tcp0

On MDS node:

# modprobe lnet ; lnetctl lnet configure
# lnetctl net add --net tcp --if eth0,eth0:0
# lnetctl peer add --nid 10.128.11.157@tcp,10.128.12.157@tcp
# lnetctl peer add --nid 10.128.11.158@tcp,10.128.12.158@tcp
# lnetctl route add --net tcp1 --gateway 10.128.11.158@tcp0

On OSS node:

# modprobe lnet ; lnetctl lnet configure
# lnetctl net add --net tcp --if eth0,eth0:0
# lnetctl peer add --nid 10.128.11.156@tcp,10.128.12.157@tcp
# lnetctl peer add --nid 10.128.11.158@tcp,10.128.12.158@tcp
# lnetctl route add --net tcp1 --gateway 10.128.11.158@tcp0

On Router node:

# modprobe lnet ; lnetctl lnet configure
# lnetctl net add --net tcp --if eth0,eth0:0
# lnetctl net add --net tcp1 --if eth0,eth0:0
# lnetctl peer add --nid 10.128.11.156@tcp,10.128.12.156@tcp
# lnetctl peer add --nid 10.128.11.157@tcp,10.128.12.157@tcp
# lnetctl peer add --nid 10.128.11.159@tcp1,10.128.12.159@tcp1
# lnetctl set routing 1

On client node:

# modprobe lnet ; lnetctl lnet configure
# lnetctl net add --net tcp1 --if eth0,eth0:0
# lnetctl peer add --nid 10.128.11.158@tcp1,10.128.12.158@tcp1
# lnetctl route add --net tcp0 --gateway 10.128.11.158@tcp1

After that, I can ‘lctl ping’ any node from any node on any NID, which is good. Then I mount the Lustre targets as I would usually do. But then, unfortunately I cannot mount Lustre from the client, it times out.

# mount -t lustre -o user_xattr 10.128.11.155@tcp:/lustre /mnt/lustre
^C

So, on the client I tried to declare as remote peers the servers that are on the other side of the router:

# lnetctl peer add --nid 10.128.11.156@tcp,10.128.12.156@tcp
# lnetctl peer add --nid 10.128.11.157@tcp,10.128.12.157@tcp

And now the mount works:

# mount -t lustre -o user_xattr 10.128.11.155@tcp:/lustre /mnt/lustre
# lfs df -h
UUID                       bytes        Used   Available Use% Mounted on
lustre-MDT0000_UUID           93.2M        1.6M       83.1M   2% /mnt/lustre[MDT:0]
lustre-MDT0001_UUID           93.2M        1.6M       83.1M   2% /mnt/lustre[MDT:1]
lustre-OST0000_UUID           14.0G       41.3M       13.3G   0% /mnt/lustre[OST:0]

filesystem_summary:        14.0G       41.3M       13.3G   0% /mnt/lustre

So it seems that without explicitly declaring the multirail capable nodes that are on the other side of the router, it would not work. But maybe I am doing something wrong in this configuration?

Thanks,
Sebastien.

Whamcloud Community JIRA

[LU-10959] Multirail remote peer issue