Bug 1335951

Summary:	heavy logging leads to Docker daemon OOM-ing
Product:	Red Hat Enterprise Linux 7	Reporter:	Jay Vyas <jvyas>
Component:	docker	Assignee:	Nalin Dahyabhai <nalin>
Status:	CLOSED ERRATA	QA Contact:	Luwen Su <lsu>
Severity:	medium	Docs Contact:
Priority:	unspecified
Version:	7.2	CC:	amurdaca, aos-bugs, dwalsh, gouyang, jokerman, lsm5, lsu, mmccomas, rmeggins, tstclair
Target Milestone:	rc	Keywords:	Extras
Target Release:	7.3
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:	docker-1.10.3-46.el7.4	Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2016-11-04 09:08:30 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 Jay Vyas 2016-05-13 15:22:13 UTC

Description of problem:

A simple container that is very verbose can cause docker's daemon to die if it doesnt take a break between logging events. 

This appears to be related to https://github.com/docker/docker/issues/18057.

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

Docker version 1.9.1, build b795b73/1.9.1
atomic-enterprise v3.2.0.16

How reproducible:

Always, especially on machines with small free memory < 2G.


Steps to Reproduce:
1. export SPAM="ASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDFASDF"
2. docker run -d gcr.io/google_containers/busybox:1.24 '/bin/sh' '-c' 'while true ; do echo $SPAM$SPAM$SPAM ; done'

Actual results:

docker daemon dies within about a minute due to memory overheating

Expected results:

machine should just die if disk fills up (not that this is a noble goal, but in our case, thats what we expected!)

Additional info:

Comment 2 Jay Vyas 2016-05-27 16:27:04 UTC

More details for folks doing scale testing of logging resilience: 

To generate the cluster wide soak test for this feature (basically, its a way to rapidly destroy a cluster if logging isnt tuned correctly, or else, to fill up ELK with logging if you are evaluating production quality performance of ELK logging stack for openshift), you can run : 

https://github.com/kubernetes/kubernetes/pull/24536 

(--ginkgo.focus="Logging soak" --scale=5) to generate 5 noisy pods per node.

Comment 3 Nalin Dahyabhai 2016-05-31 17:36:35 UTC

I think a combination of 1.10's changes to limit how much of the container's data it buffers, along with https://github.com/docker/docker/pull/22982 (currently in design review), should keep us from hitting OOM.

Comment 4 Rich Megginson 2016-05-31 17:46:32 UTC

So does that mean with docker 1.10 we have at least 2 options - 1) use docker journald log driver and avoid json file logs completely, or 2) use the docker json log file driver and configure it not to fill up the container file system?

Comment 5 Timothy St. Clair 2016-05-31 17:55:44 UTC

@Nalin have you verified the memory limit reached with your PR against the repro example listed above?

Comment 6 Nalin Dahyabhai 2016-05-31 19:06:26 UTC

(In reply to Timothy St. Clair from comment #5)
> @Nalin have you verified the memory limit reached with your PR against the
> repro example listed above?

The original description's case has its memory usage limited by 1.10, as starting with that version the container blocks trying to write to stdout/stderr until the daemon is ready to read that data, presumably after flushing whatever it's already done to disk.

The upstream issue 18057 also mentions OOMing when echo -n is used, so that the output doesn't include newlines.  That's something PR 22982 aims to fix, and it does when I run it on my system.

In both cases we grow until we hit a plateau.  Reading the logs seems to require allocating more memory than just writing them, though that now plateaus, too.

Comment 7 Nalin Dahyabhai 2016-08-05 15:45:40 UTC

This should be fixed by a backport that landed in docker-1.10.3-46.el7.4 and later.

Comment 11 errata-xmlrpc 2016-11-04 09:08:30 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.

https://rhn.redhat.com/errata/RHSA-2016-2634.html