[chef] Re: Re: Re: Re: [ANN] jvmargs, a sanity checker for JVM options

["John E. Vincent (lusis)" < >]

Here's the logic I've always used:

- Set an Xmx almost always. I would rather the process take a dump
than take the system down with it.

- If you know you average a certain amount of memory usage after
running for a while, set that to your Xms. Reasoning being that the
startup hit is less painful than memory allocation on demand. Priming
the pump and all that.

- If you have competing jvms, all bets are off. Just size them as best you 
can.

As a general practice I usually set Xmx and Xms the same. There are
valid reasons, however ,that you might not want to do that such as GC
pressure and abuse of ProcessBuilder but like I said above, I'd rather
give the process the memory I know it will need at peak (if possible)
and avoid allocation issues.

JVM tuning lends itself heavily to cargo culting and really the only
safe thing to do out of the gate is set an Xmx so that you don't tank
the system and profile (JMX for instance) while you learn the
operating pattern.

On Thu, Mar 21, 2013 at 1:44 PM, Jesse Campbell 
<
 >
 wrote:
> I've read plenty that suggests that depending on the situation, having
> different Xms and Xmx values can be beneficial.
>
> i've seen things like this:
> http://java-monitor.com/forum/showthread.php?t=427
> which suggest that setting them the same has no particularly beneficial
> effect.
>
> and I've seen things like this:
> http://weblogs.java.net/blog/kcpeppe/archive/2011/12/19/setting-xmx-xms-still-considered-harmful
> which suggest that the ideal settings are for xms to be your 99% case, and
> xmx allow for some wiggle room
>
> in fact, i haven't seen any really good recent analyses that support setting
> them the same, just lots of rhetoric
>
> -jesse
>
>
> On Thu, Mar 21, 2013 at 9:31 AM, Bryan Berry 
> <
 >
>  wrote:
>>
>> min and max should be the same because there is no benefit to having a
>> lower min according to everything I have read. It can slow down your
>> application startup however.
>>
>> it will not support parsing a space delimited string, only the options
>> specifically for the JVM, for example
>>
>> "java  -Djava.io.tmpdir=/tmp/foo -server -Xmx256m -Xint
>> -XX:MaxPermSize=512M -classpath /tmp/foo.jar:/tmp/baz.jar
>> org.apache.catalina.Bootstrap start "
>>
>> jvmargs doesn't add the java binary to the beginning nor the
>> "org.apache.catalina.Bootstrap start"
>>
>> I do want to handle the classpath but i have not figured out a clean
>> way to handle that yet :(
>>
>> i would very much appreciate feedback and testing. this afternoon i
>> hope to add support for arbitrary rules like the following
>>
>> "if Max heap is greater than available RAM, raise an error"
>> "if Max heap is changed, also set Min heap to same value"
>>
>> On Thu, Mar 21, 2013 at 2:18 PM, Jesse Campbell 
>> <
 >
>>  wrote:
>> > hey, been looking for something like this to come along, looks like a
>> > very
>> > clean, sane way to handle it :D
>> >
>> > in your todo you have ensure min and max heap should be the same. why is
>> > this?
>> >
>> > i didn't dig to far into the code, but will it support parsing a space
>> > delimited string, as the current tomcat and tomcat6 cookbooks expect, or
>> > will the cookbook be expected to handle that?
>> >
>> > thanks!
>> >
>> >
>> > On Thu, Mar 21, 2013 at 4:53 AM, Bryan Berry 
>> > <
 >
>> > wrote:
>> >>
>> >> hey folks,
>> >>
>> >> i have written a little gem to parse JVM options and would love some
>> >> feedback on it. Managing JVM options in a cookbooks is a total PITA
>> >> and I hope this makes it easier. This is less than alpha quality, i am
>> >> posting it here to get some feedback, not to notify others of a
>> >> production-ready tool. I foresee it being used in Java-related
>> >> cookbooks to provide a set of application-specific defaults and then
>> >> make it easy to override or add to them in recipe code.
>> >>
>> >> https://github.com/bryanwb/jvmargs
>> >>
>> >> This library parses JVM options so that any given option is only
>> >> stored once. It does this by breaking each option into one of the
>> >> three option categories:
>> >>
>> >> 1. standard, ex: -server, -enablesystemassertions
>> >> 2. nonstandard, ex: -Xmx128M, -Xint
>> >> 3. unstable, ex: -XX:-DisableExplicitGC, -XX:AltStack=128M
>> >> 4. directive, ex: -Dcom.sun.management.jmxremote
>> >>
>> >> JVMArgs will ensure that only one value is stored for any given
>> >> option. Here is a quick example
>> >>
>> >> args = JVMArgs::Args.new("-XX:-DisableExplicitGC", "-Xmx256M")
>> >> # "-XX:-DisableExplicitGC" is now stored in args
>> >> args.add("-XX:+DisableExplicitGC")
>> >> args.add("-Xmx2G")
>> >> # the settings are now "-XX:+DisableExplicitGC" and "-Xmx2G"
>> >
>> >
>
>