[LU-7474] Using lnet init script and lnetctl brings up tcp interface Created: 24/Nov/15  Updated: 04/Dec/15  Resolved: 25/Nov/15

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: Lustre 2.7.0
Fix Version/s: Lustre 2.8.0

Type: Bug Priority: Minor
Reporter: Jesse Hanley Assignee: Doug Oucharek (Inactive)
Resolution: Duplicate Votes: 0
Labels: patch
Environment:

RHEL 6.6
lustre-client-2.7.1-2.6.32_504.30.3.el6.x86_64_g43f9fd9.x86_64


Issue Links:
Duplicate
duplicates LU-6340 DLC: /etc/rc.d/init.d/lnet doesn't co... Resolved
Severity: 3
Rank (Obsolete): 9223372036854775807

 Description   

We're attempting to migrate to lnetctl for some of our setup rather than using the lnet modprobe.conf file.

We removed the ip2nets parameter from the conf, and generated our /etc/sysconfig/lnet.conf file. Using the following method works (from an unconfigured system):

  • modprobe lnet
  • lnetctl lnet configure
  • lnetctl import < /etc/sysconfig/lnet.conf

The problem we're seeing is that the LNET init script calls lctl network up, which configures a default nid on the tcp interface (instead of the InfiniBand interface).

Is this indented behavior? Currently, we're planning to use the following in our LNET init script instead:

125,129c125
<       if [ -f $LUSTRE_LNET_CONFIG_UTILITY ] && [ -f "$LUSTRE_LNET_CONFIG_FILE" ]; then
<                 $LUSTRE_LNET_CONFIG_UTILITY lnet configure || exit 1
<         else
<                 lctl network up || exit 1
<         fi
---
>       lctl network up || exit 1

Thanks!

Jesse



 Comments   
Comment by Andreas Dilger [ 25/Nov/15 ]

Jesse, having this in the form of a patch in Gerrit would expedite landing.

Comment by Peter Jones [ 25/Nov/15 ]

Doug is looking into this one

Comment by Jesse Hanley [ 25/Nov/15 ]

Hey Andreas,

Here's the change in patch format, but I wasn't sure if someone would have a better approach.

diff --git a/lustre/scripts/lnet b/lustre/scripts/lnet
index 1a9d467..8decff1 100644
--- a/lustre/scripts/lnet
+++ b/lustre/scripts/lnet
@@ -122,7 +122,11 @@ case "$1" in
        run_preexec_check "start"
        touch /var/lock/subsys/lnet
        modprobe lnet || exit 1
-       lctl network up || exit 1
+       if [ -f $LUSTRE_LNET_CONFIG_UTILITY ] && [ -f "$LUSTRE_LNET_CONFIG_FILE" ]; then
+               $LUSTRE_LNET_CONFIG_UTILITY lnet configure || exit 1
+       else
+               lctl network up || exit 1
+       fi
        # if a routes config file is given then use it to configure the
        # routes if not then default to LUSTRE_ROUTES_CONFIG_FILE
        if [ -f "$2" ]; then

Thanks,

Jesse

Comment by Peter Jones [ 25/Nov/15 ]

Jesse

I think that Andreas has suggested pushing the patch to gerrit because that will be how we can easily get reviews of the suggested change. Maybe it will need to go through one or more iterations before it is ready to land, but perhaps it is ok as it is. Once we get it into the system we can move towards establishing that

Peter

Comment by Jesse Hanley [ 25/Nov/15 ]

Hey Peter and Andreas,

I should have a bit more of my homework before submitting this. I went to submit the change, and noticed the functionality was already on master. It looks like this was fixed in http://review.whamcloud.com/#/c/14000/ (as a part of https://jira.hpdd.intel.com/browse/LU-6340 ). This patch just wasn't in our 2.7 build. Maybe I should start my holiday a little early.

My apologies for submitting this.

Thanks for your help,

Jesse

Comment by Jian Yu [ 25/Nov/15 ]

Let me back-port the patch, Jesse, have a nice holiday!

Comment by Jian Yu [ 25/Nov/15 ]

The issue was resolved in LU-6340. Let's close this one as a duplicate.

Comment by James A Simmons [ 25/Nov/15 ]

I back ported LU-6340 just now. See http://review.whamcloud.com/#/c/17365.

Comment by Jian Yu [ 25/Nov/15 ]

Thank you, James.

Generated at Sat Feb 10 02:09:13 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.