cloudfoundry/java-buildpack-memory-calculator

Name: java-buildpack-memory-calculator

Owner: Cloud Foundry

Description: Cloud Foundry JVM Memory Calculator

Created: 2015-04-24 09:20:03.0

Updated: 2018-05-16 21:14:05.0

Pushed: 2018-05-08 16:00:00.0

Homepage: null

Size: 442

Language: Go

GitHub Committers

UserMost Recent Commit# Commits

Other Committers

UserEmailMost Recent Commit# Commits

README

Java Buildpack Memory Calculator

The Java buildpack memory calculator determines values for JVM memory options with the goal of enabling an application to perform well while not exceeding the total memory available in a container (which results in the application being killed).

The buildpack provides the following inputs to the memory calculator:

The memory calculator prints the JVM memory option settings described below, excluding any the user has specified, which are assumed to be correct.

For Java 8 and later, the memory calculator sets the maximum metaspace size (-XX:MaxMetaspaceSize) based on the number of classes that will be loaded and sets the reserved code cache size (-XX:ReservedCodeCacheSize) to 240 Mb.

For Java 7, it sets the maximum permanent generation size (-XX:MaxPermSize) based on the number of classes that will be loaded and sets the reserved code cache size (-XX:ReservedCodeCacheSize) to 48 Mb.

It sets the maximum direct memory size (-XX:MaxDirectMemorySize) to 10 Mb.

It sets the stack size (-Xss) to a default value (unless the user has specified the stack size) and then calculates the amount of memory that will be consumed by the application's thread stacks.

Finally, it sets the heap size (-Xmx) to total memory (after any head room has been subtracted) minus the above values.

If the values need to be adjusted, the user can either increase the total memory available or set one or more JVM memory options to suitable values. Unless the user specifies the heap size (-Xmx), increasing the total memory available results in the heap size setting increasing by the additional total memory. Similarly, changing the value of other options affects the heap size. For example, if the user increases the maximum direct memory size from its default value of 10 Mb to 20 Mb, then this will reduce the calculated heap size by 10 Mb.

If the application container hits its memory limit, typically resulting in the JVM process being killed, a head room percentage may be specified to leave some unallocated memory for the JVM's own use.

If the estimated number of threads or loaded classes needs to be modified, this can be achieved by configuring the buildpack. For example, when the OpenJDK JRE is used, the number of threads can be modified as in the following example:

 set-env my-application JBP_CONFIG_OPEN_JDK_JRE '{ memory_calculator: { stack_threads: 200 } }'

and the number of loaded classes can be modified as in the following example:

 set-env my-application JBP_CONFIG_OPEN_JDK_JRE '{ memory_calculator: { class_count: 1000 } }'

Please consult the Java Buildpack documentation for up to date configuration information.

The document Java Buildpack Memory Calculator v3 provides some rationale for the memory calculator externals.

Compressed class space size

According to the HotSpot GC Tuning Guide:

The MaxMetaspaceSize applies to the sum of the committed compressed class space and the space for the other class metadata.

so the memory calculator does not set the compressed class space size (-XX:CompressedClassSpaceSize) since the memory for the compressed class space is bounded by the maximum metaspace size (-XX:MaxMetaspaceSize).

Command line options

The Java buildpack memory calculator documentation can be generated by running the following command:

 docs
generate-docs-from-help.sh

The generated docs may be viewed here.

Getting started

Install Go and then get the memory calculator (in the Go source tree).

We run our tests with Ginkgo/Gomega and manage dependencies with Godep. Ginkgo is one of the dependencies we manage, so get Godep before starting work:

et -v github.com/cloudfoundry/java-buildpack-memory-calculator
rc/github.com/cloudfoundry/java-buildpack-memory-calculator

et -v github.com/tools/godep

(The -v options on go get are there so you can see what packages are compiled under the covers.)

The (bash) script ci/test.sh uses (the correct version of) Ginkgo to run the tests (using the correct versions of the dependencies). test.sh will recompile Ginkgo if necessary.

The parameters to runTests are passed directly to Ginkgo. For example:

est.sh -r=false memory

will run the tests in the memory subdirectory without recursion into lower subdirectories (which is the default).

The current Go environment is not modified by test.sh.

Development

To develop against the code, you should issue:

p restore

in the project directory before building or running tests directly from the command line.

If you wish to develop against a particular tagged version then, in the project directory, you need to checkout this version (using git checkout <tag>) and re-issue godep restore before proceeding.

If godep restore fails, it is because one of the dependencies cannot be obtained, or else it cannot be (re)set to the version this project depends on. Normally go get -u <project> for the dependency in error will then allow godep restore to complete normally.

Release binaries

The executables are built for more than one platform, so the Go compiler must exist for the target platforms we need (currently linux and darwin). The shell script (ci/build.sh) will use the Go compiler with the GOOS environment variable to generate the executables.

This will not work if the Go installation doesn't support all these platforms, so you may have to ensure Go is installed with cross-compiler support.

License

The Java Buildpack Memory Calculator is Open Source software released under the Apache 2.0 license.

Contributing

Please refer to the Contributors' Guide.

Community

Others involved in Java buildpack memory calculator development use the #java-buildpack channel of the Cloud Foundry slack organisation for discussion.


This work is supported by the National Institutes of Health's National Center for Advancing Translational Sciences, Grant Number U24TR002306. This work is solely the responsibility of the creators and does not necessarily represent the official views of the National Institutes of Health.