Mesos is a cluster manager that provides efficient resource isolation and sharing across distributed applications, or frameworks. See Mesos website for more details.
NOTE: if you build Mesos on system with e.g. libcurl4-nss-dev it will be hardlinked to this implementation library. On Debian libcurl-dev is the virtual package for:
libcurl-devlibcurl4-nss-devlibcurl4-openssl-dev
| Mesos version \ GCC | gcc 4.4 | gcc 4.6 | gcc 4.7 | gcc 4.8 | gcc 4.9 |
|---|---|---|---|---|---|
| 0.20 | ✔ | ✔ | ✔ | ✔ | ✔ |
| 0.21 | ✔ | ✔ | ✖ | ✔ | ✔ |
| 0.22 | ✖ | ✖ | ✖ | ✔ | ✔ |
Mesos is compiled with C++11 support.
Mesos arguments could be specified by creating files structure in /etc/mesos-slave or in /etc/mesos-master.
For specifing e.g. --isolation=cgroups you would create
/etc
/mesos-slave
isolation # with contents 'cgroups'
In similar manner you can restrict hardware resources used by mesos-slave:
/etc
/mesos-slave
/resources
cpu # with contents e.g. '5'
Build deb package for Debian/Ubuntu with following:
./build_mesos --repo https://git-wip-us.apache.org/repos/asf/mesos.git?ref=0.15.0 --version 0.15.0
or supply just ref to tag in the (default) repo:
./build_mesos --ref 0.22.1 --build-version p1
If you want to activate GPU support (Mesos >= 0.29.0), you'll need to first have a local copy of bleeding edge of Mesos (as there is no rev yet). You'll also have installed the nVidia GDK as instructed in the nVidia Guide
./build_mesos --src-dir=/path/to/mesos --enable-gpu
Then don't forget to look at the nVidia doc to run Mesos Agent with the proper tags
Default gcc on Wheezy is 4.7 which isn't compatible with Mesos >= 0.21. A workaround is using gcc 4.6 which could be used along with 4.7:
apt-get install gcc-4.6 g++-4.6
just specify cxx and cc flags:
./build_mesos --ref 0.21.1-rc2 --build-version 1 --cxx "g++-4.6" --cc "gcc-4.6"
See gcc instructions.
export LD_LIBRARY_PATH="/root/gcc-4.8.4/lib64"
export LD_RUN_PATH="/root/gcc-4.8.4/lib64"
export PATH="/root/gcc-4.8.4/bin:$PATH"
- you'll definitely need ~2GB RAM for compilation
- if fact due to
make -j $(($(num_cores)*2))you need cca 2GB per core
- if fact due to
- Ruby (build scripts uses FPM)
- If using GPU, you must have followed the nVidia Guide and installed requirements
-
install following packages
$ sudo apt-get install build-essential ruby2.1 ruby2.1-dev rubygems $ gem install fpm $ sudo apt-get install build-essential python-dev autoconf automake git make libssl-dev libtool libsasl2-dev -
some version of
libcurl-dev(provided by multiple packages) -
Mesos >= 0.21
- Debian:
libapr1-dev libsvn-dev
- Debian:
```
$ sudo apt-get install ruby1.9.1 ruby1.9.1-dev build-essential
$ gem install fpm
$ sudo apt-get install build-essential python-dev autoconf automake git make libssl-dev libtool libsasl2-dev
```
-
Java support
- e.g.
openjdk-7-jre-headless,openjdk-7-jdk,maven
- e.g.
-
Python packages
- Debian:
python-setuptools
- Debian:
-
Mesos >= 0.14
- Debian:
libsasl2-dev
- Debian:
-
Mesos >= 0.21
- Debian:
libapr1-dev libsvn-dev
- Debian:
- If you're using Power architecture (ppc64le) to compile, you may need to compile from source FPM.
sudo apt-get update -qq
sudo apt-get install -y --force-yes \
tar \
wget \
tree \
autoconf \
automake \
git \
make \
gcc \
libcurl4-nss-dev \
libsasl2-dev \
libsasl2-modules \
libssl-dev \
libapr1-dev \
libsvn-dev \
libffi-dev \
openjdk-8-jdk \
maven \
autoconf libtool \
build-essential \
python-setuptools \
python-dev \
python-boto \
ruby \
ruby-dev \
rubygems-integration \
rake-compiler \
bsdtar
# This is to build Marathon
echo "deb https://dl.bintray.com/sbt/debian /" | sudo tee -a /etc/apt/sources.list.d/sbt.list
apt-key adv --keyserver hkp://keyserver.ubuntu.com:80 --recv 642AC823
sudo apt-get update -yqq
sudo apt-get install -yqq --force-yes sbt
# Some gem dependencies
sudo gem install \
yard \
rspec \
rubygems-tasks \
bundler \
tasks \
task \
custom_require
# We need to install FPM manually as it doesn't run on Xenial on Power
cd ~
git clone https://github.com/jordansissel/fpm.git
cd ~/fpm
bundle install
make
sudo make install
# This is hacky as the default nvidia script doesn't check for arch
case "$(arch)" in
"x86_64" | "amd64" )
ARCH="amd64"
;;
"ppc64le" | "ppc64el" )
ARCH="ppc64le"
;;
esac
wget http://developer.download.nvidia.com/compute/cuda/7.5/Prod/gdk/gdk_linux_${ARCH}_352_79_release.run
chmod +x gdk_linux_${ARCH}_352_79_release.run
sudo ./gdk_linux_${ARCH}_352_79_release.run --installdir=/opt/nvidia-gdk --silent
echo "/opt/nvidia-gdk/usr/src/gdk/nvml/lib" | sudo tee /etc/ld.so.conf.d/nvidia-gdk.conf
sudo ldconfig
# Using trunk
cd ~
git clone https://git-wip-us.apache.org/repos/asf/mesos.git
git clone https://github.com/deric/mesos-deb-packaging.git
cd mesos-deb-packaging
./build-mesos --src-dir=${HOME}/mesos --enable-gpu
Package could be automatically configured by a Puppet module
- Tomas Barton
- Jason Dusek
- Jeremy Lingmann
- Chris Buben