Bug 956281 - Starting EAP 6.1 on windows with 32bit jvm can result in failure to start JVM
Summary: Starting EAP 6.1 on windows with 32bit jvm can result in failure to start JVM
Keywords:
Status: CLOSED CURRENTRELEASE
Alias: None
Product: JBoss Enterprise Application Platform 6
Classification: JBoss
Component: Scripts and Commands
Version: 6.1.0
Hardware: x86_64
OS: Windows
unspecified
unspecified
Target Milestone: DR1
: EAP 6.3.0
Assignee: Tomaz Cerar
QA Contact:
URL: https://community.jboss.org/message/8...
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2013-04-24 15:06 UTC by Tomaz Cerar
Modified: 2014-07-04 02:53 UTC (History)
12 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2014-06-28 15:26:21 UTC
Type: Bug
Embargoed:


Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Red Hat Knowledge Base (Solution) 22806 0 None None None Never

Internal Links: 1009916 1023444

Description Tomaz Cerar 2013-04-24 15:06:50 UTC
Description of problem:

Starting EAP 6.1.0.ER5 with 32bit Oracle JVM can result in

Error occurred during initialization of VM
Could not reserve enough space for object heap
Error: Could not create the Java Virtual Machine.
Error: A fatal exception has occurred. Program will exit.

Problem is that heap settings are too big.

Current memmory settings that we ship are:
-Xms1303M -Xmx1303M -XX:MaxPermSize=256M

if PermSize param is removed it starts.

Same problem occurs with domain mode where failure is not immediate but at starting servers, domain controller has lower memory settings.



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


How reproducible:
use 32but JDK7 and start EAP


Steps to Reproduce:
1.
2.
3.
  
Actual results:
JVM crashes

Expected results:
EAP startss


Additional info:

Comment 1 Andrig T Miller 2013-04-24 15:12:56 UTC
Which OS platform does it fail on?

Comment 2 Tomaz Cerar 2013-04-24 15:14:23 UTC
In my case it is Windows 8 64bit, with 32bit oracle jdk7

Comment 3 Petr Kremensky 2013-04-25 12:24:55 UTC
It seems more like an JVM issue. I've made some experiments on Windows Server 2008 R2 (amd64) machine in our lab.

JDKs:
1. Oracle JDK 64-bit
java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) 64-Bit Server VM (build 23.21-b01, mixed mode)

2. Oracle JDK 32-bit
java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) Client VM (build 23.21-b01, mixed mode, sharing)

initial state (values taken from windows task manager):
Physical memory (MB)
 | - Total      = 8181
 | - Cached     = 3508
 | - Available  = 5867
 | - Free       = 2374

Scenarios:
a) start standalone.bat with default JVM options (-Xms1303M -Xmx1303M -XX:MaxPermSize=256M)
b) start standalone.bat with doubled JVM options (-Xms2606M -Xmx2606M -XX:MaxPermSize=512M)

a1) default java_opts & 64b jdk
Server started without any exception.
Physical memory (MB)
 | - Cached     = 3485
 | - Available  = 5601
 | - Free       = 2131

a2) default java_opts & 32b jdk
Server started without any exception.
Physical memory (MB)
 | - Cached     = 3486
 | - Available  = 5611
 | - Free       = 2147

b1) doubled java_opts & 64b jdk
Server started without any exception.
Physical memory (MB)
 | - Cached     = 3485
 | - Available  = 5576
 | - Free       = 2107

b1) doubled java_opts & 32b jdk
Error occurred during initialization of VM
Could not reserve enough space for object heap
Error: Could not create the Java Virtual Machine.
Error: A fatal exception has occurred. Program will exit.

You can use easy test to determine, whether jvm should be able to start with given parametrs eg.:
W:\pkremens>java -Xms1303M -Xmx1303M -XX:MaxPermSize=256 -version
java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) Client VM (build 23.21-b01, mixed mode)

W:\pkremens>java -Xms2606M -Xmx2606M -XX:MaxPermSize=500 -version
Error occurred during initialization of VM
Could not reserve enough space for object heap
Error: Could not create the Java Virtual Machine.
Error: A fatal exception has occurred. Program will exit.

Comment 6 Andrig T Miller 2013-04-25 13:33:44 UTC
Updated the doc text with the workaround.

Comment 7 Rostislav Svoboda 2013-04-25 14:14:43 UTC
pkremens checked 'java -Xms2606M -Xmx2606M -version' command without -XX:MaxPermSize=500 on testing machine used in comment 3

64b - OK
32b - Error occurred during initialization of VM

Comment 8 Andrig T Miller 2013-04-25 14:17:24 UTC
Those parameters will never start on a 32-bit JVM.  The defaults at -Xms1303m -Xmx1303m, not -Xms2606m -Xmx2606m.

So, this is not a valid test.

Comment 9 Petr Kremensky 2013-04-25 15:12:37 UTC
In our environment (Windows Server 2008 R2 (amd64) & Oracle JDK 32-bit) we are able to run both commands without any error 
> java -Xms1303M -Xmx1303M -XX:MaxPermSize=256M -version
> java -Xms1303M -Xmx1303M -version

Tomaz approved in description that removing MaxPermSize argument solves this issue, so it is good to have this documented.

Comment 10 Permaine Cheung 2013-04-25 16:41:15 UTC
This should go in AS first so that EAP will inherit the fix, right?
Brian, can you please reassign accordingly? Thanks!

Comment 11 Tomaz Cerar 2013-04-25 16:45:50 UTC
Andy when documenting the removal of PermGen there should be extra warning about domain mode, where we ship(domain.xml) with 256M for permgen for both servers that start, but removing it still results in failure as removal has to be done also in host.xml (on what it defaults if not present in domain.xml)

Also it looks like this problem only occurs on Windows8 and on Windows Server 2012. (They share same kernel)
I need to reproduce this on clean machines which i will have done later today.

Comment 12 Brian Stansberry 2013-04-25 16:54:26 UTC
Permaine, AIUI this is not an upstream problem, as upstream does not use such a large heap as it's default.

EAP uses its own heap configuration settings, different from upstream and tailored for different use cases. EAP should validate those every release. These settings shouldn't be coming from upstream, either in part or as a whole.

Tomaz, if there is an appropriate change to make to upstream, please file a WildFly JIRA.

Comment 14 Tomaz Cerar 2013-04-29 12:56:55 UTC
I have tested parameters on 
Windows Server 2008r2 64bit
Windows Server 2012 64bit
Windows 8 64bit (two different laptops)
Windows 7 64bit

And it fails only on Windows 8, even tried with different versions of jdk7.
Strange thing is that Windows Server 2012 works given that it shares kernel with windows 8.

The bizarre thing I found is that if run it with 
 -Xms1303M -Xmx1303M -XX:MaxPermSize=126M (maximum amount I could run it with)
it works in some cases and fails in some. so in 6 runs it would fail 2-3 times but work in other cases.

Comment 17 Scott Mumford 2013-05-06 01:08:22 UTC
Added draft Release Notes text.

Comment 18 Scott Mumford 2013-05-09 01:15:19 UTC
Adding NEEDINFO flag for verification of the Release Notes text above (since I forgot to when I initially added it)

Comment 19 Scott Mumford 2013-05-17 04:46:58 UTC
Removed NEEDINFO and set Doc Text flag from inclusion in 6.1.0 Release Notes.

Comment 20 Scott Mumford 2013-05-17 04:47:26 UTC
*for inclusion

Comment 21 Yukunix 2013-06-06 01:19:44 UTC
I wonder you tested 

java -Xms1303M -Xmx1303M -XX:MaxPermSize=256 -version

or

java -Xms1303M -Xmx1303M -XX:MaxPermSize=256M -version

I don't think -XX:MaxPermSize=256 is valid, it's probably just ignored.

With my Windows 7 64bit and JDK 32bit 1.7.0_21-b11, 

"java -Xms1303M -Xmx1303M -XX:MaxPermSize=187M -version" passed, but
"java -Xms1303M -Xmx1303M -XX:MaxPermSize=188M -version" failed.

So, not setting -XX:MaxPermSize is probably best hack so far.

(In reply to pkremens from comment #3)
> It seems more like an JVM issue. I've made some experiments on Windows
> Server 2008 R2 (amd64) machine in our lab.
> 
> JDKs:
> 1. Oracle JDK 64-bit
> java version "1.7.0_21"
> Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
> Java HotSpot(TM) 64-Bit Server VM (build 23.21-b01, mixed mode)
> 
> 2. Oracle JDK 32-bit
> java version "1.7.0_21"
> Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
> Java HotSpot(TM) Client VM (build 23.21-b01, mixed mode, sharing)
> 
> initial state (values taken from windows task manager):
> Physical memory (MB)
>  | - Total      = 8181
>  | - Cached     = 3508
>  | - Available  = 5867
>  | - Free       = 2374
> 
> Scenarios:
> a) start standalone.bat with default JVM options (-Xms1303M -Xmx1303M
> -XX:MaxPermSize=256M)
> b) start standalone.bat with doubled JVM options (-Xms2606M -Xmx2606M
> -XX:MaxPermSize=512M)
> 
> a1) default java_opts & 64b jdk
> Server started without any exception.
> Physical memory (MB)
>  | - Cached     = 3485
>  | - Available  = 5601
>  | - Free       = 2131
> 
> a2) default java_opts & 32b jdk
> Server started without any exception.
> Physical memory (MB)
>  | - Cached     = 3486
>  | - Available  = 5611
>  | - Free       = 2147
> 
> b1) doubled java_opts & 64b jdk
> Server started without any exception.
> Physical memory (MB)
>  | - Cached     = 3485
>  | - Available  = 5576
>  | - Free       = 2107
> 
> b1) doubled java_opts & 32b jdk
> Error occurred during initialization of VM
> Could not reserve enough space for object heap
> Error: Could not create the Java Virtual Machine.
> Error: A fatal exception has occurred. Program will exit.
> 
> You can use easy test to determine, whether jvm should be able to start with
> given parametrs eg.:
> W:\pkremens>java -Xms1303M -Xmx1303M -XX:MaxPermSize=256 -version
> java version "1.7.0_21"
> Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
> Java HotSpot(TM) Client VM (build 23.21-b01, mixed mode)
> 
> W:\pkremens>java -Xms2606M -Xmx2606M -XX:MaxPermSize=500 -version
> Error occurred during initialization of VM
> Could not reserve enough space for object heap
> Error: Could not create the Java Virtual Machine.
> Error: A fatal exception has occurred. Program will exit.

Comment 22 Petr Kremensky 2013-06-12 05:44:36 UTC
Hi  Yukunix,
I've tested
> java -Xms1303M -Xmx1303M -XX:MaxPermSize=256M -version

Comment 23 Tomaz Cerar 2013-07-17 10:27:30 UTC
I would like to add what users on community forum confirmed that same problem can be reproduced on 64-bit Windows 7, 32-bit JDK 7u25, 8 gig RAM, JBoss EAP 6.1.

Comment 32 Tomaz Cerar 2013-09-23 22:27:13 UTC
Yes it also works without PermGen attribute.
so just having -Xmx1g -Xms1G also works in 6 scenarios I tested.

Comment 36 Rostislav Svoboda 2013-09-27 13:16:51 UTC
https://github.com/jbossas/jboss-eap/pull/440

There are several reason why PR 440 shouldn't be merged:
 - devel_ack is -
 - we are changing defaults because of unsupported platform
 - this is fix for Windows 8 which is not supported in EAP
   -- see comment 14 whre Tomaz confirms it's just issue in Windows 8
 - there is inconsistency between JVM option for Windows and other platforms 
   -- "-Xms1G -Xmx1G" vs. "-Xms1303M -Xmx1303M -XX:MaxPermSize=256M"


Changing qa_ack to -.

Please revert PR 440.

I think this issue should be addressed in WildFly and this BZ should be closed as Won't fix.

Comment 37 Tomaz Cerar 2013-09-27 13:25:01 UTC
Rostislav, WildFly does not have this issue, as it ships with much much lower memory settings.

current setting in wildfly:
set "JAVA_OPTS=-Xms64M -Xmx512M -XX:MaxPermSize=256M"

It is issue only in EAP.
I did change Devel_ack back to + but it was not recorded, setting it again.

Comment 39 Rostislav Svoboda 2013-09-30 12:09:17 UTC
qa_ack was granted before we knew it's Windows 8 only problem - Comment 14.

I would prefer to document this as known issue - we are changing defaults because of unsupported platform from final product PoV and there is inconsistency between JVM option for Windows and other platforms.

NOTE: I don't think 75% of devs are on Windows 8 64bit with 32bit JVM.

Comment 41 Max Rydahl Andersen 2013-09-30 12:43:42 UTC
Hi,

I can't speak for all users of EAP, but I can speak based on our usage statistics for JBoss Developer Studio.

Afaic this issue is *not* limited to Windows 8, but to Windows JVM's.
Thus it also affects Windows XP and Windows 7 users.

That is ~73% of our users, of which half are using JDK 32-bit.

Even if it is *just* Windows 8 then the number of users here exceeds Mac OSX users and is ~65% of what total Linux installation base is.

Here are the numbers from last monday:
Windows 327573
Linux   3955
Mac     1495

Of those 2559 was Windows 8 users.

Thus please don't under estimate the bad impact it has by the defaults preventing startup of EAP.

Comment 42 Max Rydahl Andersen 2013-09-30 12:53:13 UTC
p.s. we had similar issue with Windows XP in the past - yes, it is not a production supported runtime but it is a developer supported one thus it is relevant to fix it.

Comment 48 Max Rydahl Andersen 2014-01-09 14:33:46 UTC
Does this pm_ack+ and devel_ack+ means it will be fixed in next update to EAP 6.1 ?

Comment 49 Brian Stansberry 2014-01-09 16:49:37 UTC
Those acks are for EAP 6.3.

Comment 51 Kabir Khan 2014-05-20 17:20:04 UTC
https://github.com/jbossas/jboss-eap/pull/925

Comment 52 Petr Kremensky 2014-05-23 09:45:30 UTC
Verified on EAP 6.3.0.ER4.


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