Bug 1487339

Summary:	Single pod is able to use enough resources to make a node go NotReady
Product:	OpenShift Container Platform	Reporter:	Sten Turpin <sten>
Component:	Node	Assignee:	Seth Jennings <sjenning>
Status:	CLOSED CURRENTRELEASE	QA Contact:	DeShuai Ma <dma>
Severity:	unspecified	Docs Contact:
Priority:	unspecified
Version:	3.6.1	CC:	aos-bugs, decarr, jokerman, mmccomas, mwhittin, sten
Target Milestone:	---	Keywords:	OpsBlocker
Target Release:	---
Hardware:	x86_64
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2017-09-12 19:29:06 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:
Embargoed:

Description Sten Turpin 2017-08-31 16:24:26 UTC

Description of problem: A pod can create enough of a workload that the node underneath becomes unresponsive 


Version-Release number of selected component (if applicable): atomic-openshift-3.6.173.0.5-1.git.0.f30b99e.el7.x86_64


How reproducible: Sometimes


Steps to Reproduce:
1. Run a pod with a particular workload. 

Actual results:
Node stops responding via SSH, reports NotReady in k8s


Expected results:
CGroups should mitigate resource use to ensure all pods can run and administrative access is maintained. 

Additional info:
We're observing this on AWS nodes, m4.xlarge with 250GB GP2 EBS. This is relevant because we have seen issues wit nodes being both CPU-bound and, spearately, io token bound.

Comment 2 Sten Turpin 2017-08-31 16:27:51 UTC

often shows https://bugzilla.redhat.com/show_bug.cgi?id=1459589 "iptables soft lockup" as a symptom.