Bug 1349657 - process glusterd set TCP_USER_TIMEOUT failed
Summary: process glusterd set TCP_USER_TIMEOUT failed
Keywords:
Status: CLOSED CURRENTRELEASE
Alias: None
Product: GlusterFS
Classification: Community
Component: rpc
Version: mainline
Hardware: All
OS: Linux
unspecified
low
Target Milestone: ---
Assignee: Zhou Zhengping
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks: 1354404 1354405
TreeView+ depends on / blocked
 
Reported: 2016-06-23 22:21 UTC by Zhou Zhengping
Modified: 2017-01-03 11:22 UTC (History)
3 users (show)

Fixed In Version: 3.9.0
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
: 1354404 1354405 (view as bug list)
Environment:
Last Closed: 2017-01-03 11:22:02 UTC
Regression: ---
Mount Type: ---
Documentation: ---
CRM:
Verified Versions:
Embargoed:


Attachments (Terms of Use)

Description Zhou Zhengping 2016-06-23 22:21:50 UTC
Description of problem:


Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info:

Comment 1 Zhou Zhengping 2016-06-23 22:29:40 UTC
Description of problem:
We can see glusterd's error log like this:
2016-06-24 04:34:30.252200] I [MSGID: 106544] [glusterd.c:155:glusterd_uuid_init] 0-management: retrie
ved UUID: b49bd7b5-9cb5-4e87-8e23-8997bfb9c479
[2016-06-24 04:34:30.275351] I [MSGID: 106498] [glusterd-handler.c:3644:glusterd_friend_add_from_peerin
fo] 0-management: connect returned 0
[2016-06-24 04:34:30.275494] I [rpc-clnt.c:991:rpc_clnt_connection_init] 0-management: setting frame-ti
meout to 600
[2016-06-24 04:34:30.276462] W [socket.c:979:__socket_keepalive] 0-socket: failed to set TCP_USER_TIMEO
UT -1000 on socket 13, Invalid argument
[2016-06-24 04:34:30.276527] E [socket.c:3087:socket_connect] 0-management: Failed to set keep-alive: I
nvalid argument

Additional info:
in linux kernel function do_tcp_setsockopt ,we can see code like this:
       case TCP_USER_TIMEOUT:
                /* Cap the max timeout in ms TCP will retry/retrans
                 * before giving up and aborting (ETIMEDOUT) a connection.
                 */
                if (val < 0) 
                        err = -EINVAL;
                else 
                        icsk->icsk_user_timeout = msecs_to_jiffies(val);
                break;

this code means , the timeout should not be lower than 0.But in function glusterd_transport_keepalive_options_get, if option "transport.tcp-user-timeout" not set ,the priv->timout will be -1, it will cause socksetopt failed.

Comment 2 Vijay Bellur 2016-06-23 22:44:19 UTC
REVIEW: http://review.gluster.org/14785 (rpc: invalid argument when function setsockopt sets option TCP_USER_TIMEOUT) posted (#1) for review on master by Zhou Zhengping (johnzzpcrystal)

Comment 3 Vijay Bellur 2016-06-25 12:12:23 UTC
REVIEW: http://review.gluster.org/14785 (rpc: invalid argument when function setsockopt sets option TCP_USER_TIMEOUT) posted (#2) for review on master by Zhou Zhengping (johnzzpcrystal)

Comment 4 Vijay Bellur 2016-07-11 01:59:08 UTC
COMMIT: http://review.gluster.org/14785 committed in master by Jeff Darcy (jdarcy) 
------
commit b2c73cbf423de6201f956f522b7429615c88869d
Author: Zhou Zhengping <johnzzpcrystal>
Date:   Fri Jun 24 06:33:16 2016 +0800

    rpc: invalid argument when function setsockopt sets option TCP_USER_TIMEOUT
    
    If option "transport.tcp-user-timeout" hasn't been setted, glusterd's
    priv->timeout will be -1, which will cause invalid argument when
    set TCP_USER_TIMEOUT.
    
    Change-Id: Ibc16264ceac0e69ab4a217ffa27c549b9fa21df9
    BUG: 1349657
    Signed-off-by: Zhou Zhengping <johnzzpcrystal>
    Reviewed-on: http://review.gluster.org/14785
    CentOS-regression: Gluster Build System <jenkins.org>
    Smoke: Gluster Build System <jenkins.org>
    NetBSD-regression: NetBSD Build System <jenkins.org>
    Reviewed-by: Jeff Darcy <jdarcy>

Comment 5 Oleksandr Natalenko 2016-07-11 09:10:29 UTC
I believe, this should be backported to 3.7 branch as we observe same warnings with 3.7.11/12/13.


Note You need to log in before you can comment on or make changes to this bug.