Bug 1328913

Summary: Long running reliability tests show network errors on nodes
Product: OpenShift Container Platform Reporter: Vikas Laad <vlaad>
Component: NodeAssignee: Seth Jennings <sjenning>
Status: CLOSED ERRATA QA Contact: Vikas Laad <vlaad>
Severity: urgent Docs Contact:
Priority: urgent    
Version: 3.2.0CC: agoldste, aos-bugs, decarr, jkaur, jokerman, mmccomas, nbhatt, sjenning, sten, vlaad, xtian
Target Milestone: ---   
Target Release: 3.7.0   
Hardware: Unspecified   
OS: Unspecified   
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of:
: 1496245 (view as bug list) Environment:
Last Closed: 2017-11-28 21:51:43 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Bug Depends On: 1367141    
Bug Blocks: 1303130, 1496245    
Description Flags
dm-failure.log none

Description Vikas Laad 2016-04-20 14:33:11 UTC
Description of problem:

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

How reproducible:

Steps to Reproduce:

Actual results:

Expected results:

Additional info:

Comment 1 Vikas Laad 2016-04-20 14:44:21 UTC
Description of problem:
Long running reliability tests have run into this problem many times, these tests created few sample applications (ruby, cakephp, dancer and eap) applications in the beginning and keep rebuilding/accessing the apps for few days. I have the output if network debug script output but could not attach to the bug due to size. Please ping me on IRC vlaad is my nick.

Version-Release number of selected component (if applicable):
openshift v3.2.0.16

How reproducible:

Steps to Reproduce:
1. Please see the description

Actual results:
Logs are full of following errors 
Apr 20 10:42:31 ip-172-31-7-135 atomic-openshift-node: I0420 10:42:31.779154   63549 helpers.go:101] Unable to get network stats from pid 128265: couldn't read network stats: failure opening /proc/128265/net/dev: open /proc/128265/net/dev: no such file or directory

Expected results:

Additional info:

Comment 2 Vikas Laad 2016-04-20 14:45:22 UTC
I still have the environment running in case someone wants to look at it.

Comment 3 Ben Bennett 2016-04-20 15:56:31 UTC
These messages are usually from cadvisor / heapster trying to get the stats from a pod that's just gone away.  Are you getting messages repeated for the same pid for a long time?  Or are the pids changing.  If they are changing, I'm told that it's somewhat expected (but annoying) behavior.

Comment 4 Vikas Laad 2016-04-20 17:26:59 UTC
pids are changing.

Comment 5 dhodovsk 2016-06-03 14:34:17 UTC
Could you please try to reproduce the issue again with higher log level (5 could be fine) and send me the logs?

Comment 6 Vikas Laad 2016-06-06 15:24:04 UTC
I am running the tests, will update the bug with information when I have it reproduced.

Comment 8 Sten Turpin 2016-07-15 15:21:41 UTC
*** Bug 1357052 has been marked as a duplicate of this bug. ***

Comment 9 Seth Jennings 2016-07-15 21:39:32 UTC
The direct cause is that the container process has exited but cadvisor continues to try to monitor it.

This is because docker containers are removed from cadvisor monitoring implicitly by the removal of their cgroup, which cadvisor watches for with inotify.

A disconnect is occurring where the pid 1 of the docker container exits but the cgroup isn't removed, leading to a systemd docker container slice with no tasks:

# find /sys/fs/cgroup/systemd/system.slice/docker* -name tasks | xargs wc -l | sort -rn
0 /sys/fs/cgroup/systemd/system.slice/docker-4af4dc9c32a97a7a0bf0f26464898426389f18228dedf835e1ec8bad61d4c623.scope/tasks

Comment 10 Seth Jennings 2016-07-15 21:58:20 UTC
Created attachment 1180293 [details]

Attaching log with selected section regarding a container that resulted in an dead cgroup.  It is showing a device mapper removal failure that might be causing docker to bail out before the cgroup teardown.

Comment 11 Seth Jennings 2016-08-08 20:33:01 UTC
Upstream issue https://github.com/kubernetes/kubernetes/issues/30171

Comment 14 Derek Carr 2017-07-21 15:05:15 UTC
Opened a PR to cadvisor to reduce log spam:

Comment 17 Seth Jennings 2017-09-06 21:31:02 UTC
Origin PR:

This will reduce the spam but good log rotation practices must still be used to avoid filling the disk.

Comment 20 Neeraj 2017-09-26 09:15:03 UTC
Please let me know if this bug is backported in 3.2.1.

Comment 26 Vikas Laad 2017-10-26 17:15:53 UTC
I do not see these errors at logleve=2 in following version and later

openshift v3.7.0-0.143.2
kubernetes v1.7.0+80709908fd
etcd 3.2.1

Comment 30 errata-xmlrpc 2017-11-28 21:51:43 UTC
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.