[ 
https://issues.apache.org/jira/browse/CASSANDRA-7353?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Joshua McKenzie updated CASSANDRA-7353:
---------------------------------------

    Attachment: 7353_v3.txt

Slipped my mind - we have MAX_HEAP_SIZE and HEAP_NEWSIZE available to be tuned 
in conf/cassandra-env.ps1, so adding more logic on top of that to facilitate an 
edge-case like this is unnecessary.

That said, I've tidied up a few things, tossed a 100 ms sleep in the launch 
process to catch JVM init errors (catches it at 10 on my machine but figured a 
little breathing room wouldn't hurt), and changed the logic we use to check to 
see if a process started correctly (v3 patch attached).

Philip, will this launch delay potentially cause us headaches in ccm or dtests?

> Java heap being set too large on Windows with 32-bit JVM
> --------------------------------------------------------
>
>                 Key: CASSANDRA-7353
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7353
>             Project: Cassandra
>          Issue Type: Bug
>         Environment: Windows Server 2008, 8G RAM, 32-bit JVM
>            Reporter: Philip Thompson
>            Assignee: Joshua McKenzie
>            Priority: Minor
>              Labels: Windows
>         Attachments: 7353_v1.txt, 7353_v2.txt, 7353_v3.txt
>
>
> On windows, the JVM settings for max heap size and new gen heap size are set 
> based on the total system memory. When the system has 8G of RAM, the max heap 
> size is set to 2048M. However, according to http://goo.gl/1ElbLm, the 
> recommended max heap for a 32 bit JVM on Windows is 1.8G.
> When cassandra is started on Windows under these conditions, the following 
> error is seen:
> 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.
> Switching to a 64-bit JVM on the same machine solves the issue. If a 32-bit 
> JVM is being used, cassandra should be started up with a smaller heap than 
> would be normally used to prevent the error.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to