Skip to content
Jeff LeFevre edited this page Jan 15, 2020 · 51 revisions

Directions to clone and build. Tested on clean install of 64-bit Ubuntu 18.04 and Centos7. NOTE: requires at least 30GB disk space to build.

Be sure you have enough disk space to build.

On Cloudlab machines, the ${HOME} dir is not large enough, so format and mount one of the larger disks. NOTE: do not wipe your primary disk!.

lsblk;      # show available devices.
df -h;      # show mounted devices.
$disk=sda4; # choose a device with enough space.
sudo mkfs -t ext4 /dev/$disk;   # USE WITH CAUTION
sudo mkdir -p /mnt/$disk;
sudo mount /dev/$disk /mnt/$disk;
sudo chmod a+rw /mnt/$disk;
df -h;

Install dependencies and clone repo

PKG_MGR=apt-get; # for Ubuntu
PKG_MGR=yum;     # for CentOS
sudo ${PKG_MGR} update;
sudo ${PKG_MGR} install wget cmake git gnupg dstat python-pip -y; 
git clone https://github.com/uccross/skyhookdm-ceph.git;
cd skyhookdm-ceph;

Checkout latest and verify branch before running submodule update.

git pull;
git checkout skyhook-luminous;
git branch -a;
git submodule update --init --recursive;
sudo ./install-deps.sh;
./do_cmake.sh;

BUILD Ceph with Skyhook

cd build;

make -jN cls_tabular run-query sky_tabular_flatflex_writer ceph_test_skyhook_query vstart;

  • Add -jN to create n jobs i.e., if you have 12 cores use -j12 to compile with 12 cores
  • Takes about 13 min with 3.2 GHz 12 core CPU
  • All is not required, but make -j12 all takes about 25 min with 3.2 GHz 12 core CPU
  • To save time, just make cls_tabular run-query for repeat builds, most Skyhook functionality is in there

Start a virtual cluster for dev testing

After compiling vstart above, from the build dir, stop any previously running vstart and then start a new one.

../src/stop.sh; MGR=1 MDS=0 MON=1 OSD=3 ../src/vstart.sh -d -n -x

  • IMPORTANT: anytime you recompile Skyhook you should also recompile vstart and stop/start the virtual cluster again.

Get running cluster information

  • vstart will show a cluster status url at the end such as

dashboard urls: http://abc.123.xy.z:41000/

  • Similar cluster status info via cmd line: bin/rados df

Run some test queries

You can now go to the Run test queries page.

Clone this wiki locally