[LU-17889] Autotest nodes should not use 'fe80:' link local (loopback) IPv6 address

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Affects Version/s: Lustre 2.16.0
• Environment: Maloo test bed nodes
• Severity: 3
• Assignee: Michael Kvardakov
• Reporter: James A Simmons

Description

Pushing patches to run sanity-lnet test with IPv6 which are working locally for me to maloo revealed that the IPv6 address setup is only local link (fe80:.....). These addresses are only visible to the local node so the more complex LNet test can't run. Also LNet ignores these addresses. Proper IPv6 addresses need to be setup.

Issue Links

• is related to: LU-17367 Failover on master: Invalid NID string
• is related to: LU-16822 test IPv6 NIDs in sanity-lnet.sh

Comments

[Michael Kvardakov]

simmonsja cannot reproduce it, getting different lnectl ping error. If I understand correctly, --source should take as local ipv6 address which starts with fd33:3981:3213:f010 at onyx and fd33:3981:3213:f020 at trevis. Could we review how this parameter is generated?

[Michael Kvardakov]

/usr/sbin/lnetctl net add --net tcp --if test1pg
default via 10.240.16.1 dev eth0
NI 0@lo expect status "up" found "up"
NI 10.240.25.232@tcp expect status "up" found "up"
NI 2001:db8:0:f101::1@tcp expect status "up" found "up"
/usr/sbin/lnetctl ping --source 2001:db8:0:f101::1@tcp 10.240.25.232@tcp
manage:
- ping:
  errno: -5
  descr: ! 'failed to ping 10.240.25.232@tcp: Input/output error'
 sanity-lnet test_214: @@@@@@ FAIL: /usr/sbin/lnetctl ping --source 2001:db8:0:f101::1@tcp 10.240.25.232@tcp failed 

[James A Simmons]

we shall see what fails

[Gerrit Updater]

"James Simmons <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/55435
Subject: LU-17889 lnet: test IPv6
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: bbacbd65ed4cf72d676aedd6f74e0fde65ae361e

[Michael Kvardakov]

simmonsja which distro are affected?

[Charlie Olmstead]

mkvardakov please take a look. The IPV6 addresses are only available to the local node.

[Peter Jones]

Thanks Lee. I assigned this to you last week because Charlie was on PTO. I'll reassign to him to provide further updates as things progress. It looks like LU-16822 should have fixes merging shortly.

[Lee Ochoa (Inactive)]

pjones I haven't been involved with this effort so I'm only mildly in tune with the details, Charlie and Minh can provide more insight but as far as I know the AT-side development is blocked by LU-16822

[Peter Jones]

Lee

Could you please see what is needed here on our side?

Thanks

Peter