Skip to content

Latest commit

 

History

History
605 lines (450 loc) · 20.4 KB

README.md

File metadata and controls

605 lines (450 loc) · 20.4 KB
Raw

Build and Test Environment for Motr

Quick Start (MacOS)

  • Install

    • Homebrew

      /usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

    • bash 4.x

      brew install bash

    • GNU readlink

      brew install coreutils

    • VMware Fusion or VirtualBox (VMware is recommended for better experience)

    • Vagrant (and Vagrant VMware Utility in case of VMware)

    • Vagrant plugins (for VMware the license needs to be purchased)

      vagrant plugin install vagrant-{env,hostmanager,scp}
vagrant plugin install vagrant-vmware-desktop # for VMware or

vagrant plugin install vagrant-vbguest # for VirtualBox

    • Ansible

      brew install ansible # on Linux or macOS hosts

  • Configure

    • m0vg script (make sure you have $HOME/bin in the $PATH)

      MOTR_SRC=~/src/motr  # use the actual Motr location on your host system
ln -s $MOTR_SRC/scripts/m0vg $HOME/bin/

    • VMs

      # open virtual cluster configuration file in default editor
m0vg env edit

      paste the following template updating parameters as desired:

      # a host directory to share among all VMs,
# by default it's a parent of $MOTR_SRC dir
#M0_VM_SHARED_DIR=~

# default sharing mechanism is NFS, which is recommended,
# but in case it doesn't work for some reason, a virtual
# provider specific sharing mechanism can be used (e.g.
# VirtualBox shared folder)
#M0_VM_SHARE_TYPE=provider

# a comma-separated list of additional packages to be installed
# on each VM (they must be available in default `yum` repositories
# or EPEL)
#M0_VM_EXTRA_PKGS=python36,python36-pip

# a script executed on CMU node after provisioning is finished,
# a `sudo` can be used in the script to gain root privileges
#M0_VM_PROVISION_SCRIPT=~/vagrant-postinstall.sh

# amount of RAM available on CMU node
#M0_VM_CMU_MEM_MB=4096

# number of client VMs
#M0_VM_CLIENT_NR=2
# amount of RAM available on every client node
#M0_VM_CLIENT_MEM_MB=3072

# number of ssu VMs
#M0_VM_SSU_NR=3
# amount of RAM available on every ssu node
#M0_VM_SSU_MEM_MB=2048
# number of data drives on every ssu node
#M0_VM_SSU_DISKS=12
# size of each data drive on ssu node
#M0_VM_SSU_DISK_SIZE_GB=8

      see m0vg params output for the full list of supported configuration parameters

  • Run

    • check VMs state

      m0vg status

    • create cmu VM (this can take ~30 minutes depending on the internet connection, CPU and system disk speed)

      m0vg up cmu

    • restart cmu VM in order to activate shared folder

      m0vg reload cmu

    • logon on cmu and check contents of /data dir

      m0vg tmux
ls /data

    • create ssu and client VMs (can take about ~40 minutes depending on the number of configured ssu and client nodes)

      m0vg up /ssu/ /client/
m0vg reload /ssu/ /client/

    • stop all nodes when they're not needed to be running

      m0vg halt

    • if a node hangs (e.g. Motr crash in kernel or deadlock) it can be forced to shutdown using -f option for halt command, for example:

      m0vg halt -f client1

Quick Start (Windows)

  • Install

    • VMware Workstation or VirtualBox (VMware is recommended for better experience)

    • Vagrant (and Vagrant VMware Utility in case of VMware)

    • Vagrant plugins (for VMware the license needs to be purchased)

      vagrant plugin install vagrant-{env,hostmanager,scp}
vagrant plugin install vagrant-vmware-desktop # for VMware or

vagrant plugin install vagrant-vbguest # for VirtualBox

    • Git for Windows During installation, when asked, choose the following options (keep other options to their default setting):

      • Use Git and optional Unix tools from the Command Prompt
      • Checkout as-is, commit Unix-style line ending
      • Enable symbolic links

  • Configure

    • Open Git Bash terminal, add CRLF configuration option to make sure that Motr/Hare scripts can work on VM

      git config --global core.autocrlf input

    • Clone Motr repository somewhere, just as an example let's say it's in $HOME/src/motr:

      mkdir -p src
cd src
git clone --recursive git@github.com:Seagate/cortx-motr.git motr

    • Create a persistent alias for m0vg script:

      cat <<EOF >> $HOME/.bash_profile
# use the actual Motr location on your host system
export MOTR_SRC=$HOME/src/motr
alias m0vg="\$MOTR_SRC/scripts/m0vg"
EOF

      Exit and re-launch Git Bash terminal. At this point the setup should be complete.

  • Run

    • Follow the steps from Run section under Quick Start (MacOS) above.

      NOTE: during m0vg up <node> command execution you may be asked to enter your Windows username and password, and then grant permissions for creating Windows shared directory. To avoid manually entering the credentials for every node, set SMB_USERNAME/SMB_PASSWORD environment variables with the correspondent values. Note: make sure SMB_PASSWORD is not saved in your bash history (for security reasons).

Overview

This directory contains scripts for quick deployment of a "devvm" virtual machine (based on a stock Centos7 by default), prepared for Motr development and testing on a local desktop or a laptop.

The virtual machine is automatically created from the official Centos7 base image, which is downloaded from the vagrantcloud repository. After provisioning and installation of the required rpm packages, including build tools and latest Lustre from Whamcloud's repository, it takes about 2.5GB of extra disk space per VM.

Besides main virtual machine, which can be used as a build node, additional machines can be provisioned as well to provide a cluster-like environment for debugging and testing motr on multiple nodes. The main machine is named cmu, machines with attached disks are named ssuN, and "client" machines are named clientN, where N is a natural number.

All machines are accessible by names (with .local suffix) within their private network, with password-less ssh access from the main node to other test nodes. A directory containing motr source code is shared with each machine over NFS. This should provide a short-enough "prepare/build/test" cycle for efficient development workflow.

Depending on the host OS, different virtualization providers are supported: on Linux those are Libvirt/KVM and VirtualBox, on macOS - VMware Fusion and VirtualBox, on Windows - VMware Workstation and VirtualBox.

Requirements

In order to run these scripts, additional tools have to be installed first. It's assumed that either macOS, Windows or Linux is used as a host operating system.

On Ubuntu Linux all of the above prerequisites can be installed with a single command:

sudo apt install qemu-kvm libvirt-bin vagrant ansible

Though, it's actually better to get a more up-to-day versions of Vagrant and Ansible than those provided by a distribution. The procedure is same as described below for macOS.

On macOS the easiest way to install those tools is to download VirtualBox/VMware Fusion and Vagrant packages from their official web-sites (refer to the links above).

And install Ansible using Python package manager pip, which is available on macOS "out of the box":

# install for current user only
# make sure that '$HOME/.local/bin' is in your PATH
pip install --user ansible

# install system-wide
sudo pip install ansible

Another popular alternative is to use MacPorts or Homebrew package managers:

# install Ansible using MacPorts
sudo port install py36-ansible

# install Ansible using Homebrew
brew install ansible

After Vagrant is installed, a couple of plugins need to be installed also. On Linux it is vagrant-libvirt (for kvm support), and on macOS/Windows it's vagrant-vbguest, when using VirtualBox and vagrant-vmware-desktop, when using VMware Fusion/Workstation:

# linux with Qemu/KVM
vagrant plugin install vagrant-libvirt

# macOS with VirtualBox
vagrant plugin install vagrant-vbguest

# macOS/Windows with VMware Fusion/Workstation
vagrant plugin install vagrant-vmware-desktop

It's highly recommend to install a few more Vagrant plugins for a better user experience:

  • vagrant-env -- for saving commonly used configuration variables in a config file
  • vagrant-hostmanager -- for managing /etc/hosts file on guest machines
  • vagrant-scp -- for easier file copying between the host and VM

DevVM provisioning

After installing required tools from the above section, all that remains is to run vagrant up command in the directory containing this README file, that will do rest of the work. But, there is a better way to achieve the same result which is more convenient:

./scripts/m0vg up

The m0vg helper script is a wrapper around Vagrant and Ansible commands that can be symlinked somewhere into the PATH and be called from any directory. Check out m0vg --help for more info.

It will spawn a VM and configure it using Ansible "playbook" scripts/provisioning/cmu.yml, that specifies all Motr dependencies which should be installed in order to build and run Motr. It will install Lustre 2.10.4 from the official Whamcloud's repository. During provisioning, Vagrant might pause and ask for user password, this is needed for NFS auto-configuration (it will add a new entry in /etc/exports and restart nfsd service).

By default, Vagrant creates a vagrant user inside VM with password-less sudo privileges. The user password is also vagrant.

When provisioning is finished it should be possible to login into the VM with ./scripts/m0vg ssh command. Please, refer to the Vagrant basics section below for the list of other useful Vagrant commands.

If a cluster-like environment is needed, more machines can be provisioned:

./scripts/m0vg up cmu /ssu/ /client/

The additional parameters are also explained in the Vagrant basics section below.

It is possible to control different parameters of the Vagrantfile via environment variables or .env file that should be placed alongside Vagrantfile. For instance, the following two examples do the same thing but with the latter there is no need to specify env variables every time while executing a vagrant command, they will be loaded from the .env file:

# -1- using env variables
M0_SSU_NR=5 M0_CLIENT_NR=3 vagrant up

# -2- using env file
cat .env
M0_SSU_NR=5
M0_CLIENT_NR=3

vagrant up

By the way, there is no need to create .env file manually, m0vg env edit helps with that. A complete list of supported variables is printed by m0vg params command.

All additional nodes can be accessed from the main machine (cmu) by their name in a .local domain. For example, here is how to execute a command on the ssu1 from cmu:

ssh ssu1.local <command>

The host directory containing motr sources directory will be mounted over NFS on each VM under /data.

NOTE: one important aspect of how Vagrant works is that it creates a hidden .vagrant directory, alongside Vagrantfile, where it keeps all configuration data related to provisioned VMs. If that directory is lost the access to the VMs is lost as well. Which can happen unintentionally as a result of git clean -dfx. This is another reason to use m0vg script which takes care of it by moving .vagrant directory outside of motr source tree.

Building and running Motr

Normally, Motr sources should be accessible over NFS (or native VirtualBox/VMware shared folder) on each VM under /data directory:

# build Motr in source tree
cd /data/motr
./scripts/m0 make

If, for some reason, Vagrant hasn't been able to configure the NFS share it is still possible to copy motr sources to VM with the help of vagrant-scp plugin:

# on the host
tar -czf ~/motr.tar.gz $MOTR_SRC
m0vg scp ~/motr.tar.gz :~

# on VM
cd ~
tar -xf motr.tar.gz
cd motr
./autogen.sh && ./configure && make rpms-notests

Resulting rpm files will be available in ~/rpmbuild/RPMS/x86_64/ directory. To verify them they can be installed with:

sudo yum install rpmbuild/RPMS/x86_64/*

Try single-node Motr cluster

To bootstrap a single-node cluster Hare should be installed also. Here is the short script which can be run on the VM to prepare everything:

[[ -d cortx-motr ]] ||
    git clone --recurse https://github.com/Seagate/cortx-motr.git &&
        ln -s cortx-motr motr
cd motr
echo 'Building and installing Motr...'
./autogen.sh && ./configure --disable-expensive-checks && make -j8 &&
    ./scripts/install-motr-service
cd -

[[ -d cortx-hare ]] ||
    git clone --recurse https://github.com/Seagate/cortx-hare.git &&
        ln -s cortx-hare hare
cd hare
echo 'Building and installing Hare...'
make && make devinstall
cd -

# Create block devices
mkdir -p /var/motr
for i in {0..9}; do
    dd if=/dev/zero of=/var/motr/disk$i.img bs=1M seek=9999 count=1
    losetup /dev/loop$i /var/motr/disk$i.img
done

# Prepare CDF (Cluster Description File)
[[ -f singlenode.yaml ]] || cp hare/cfgen/examples/singlenode.yaml ./
sed 's/localhost/cmu/' -i singlenode.yaml
sed 's/data_iface: eth./data_iface: eth0/' -i singlenode.yaml

After all the above steps completed successfully, the single-node Motr cluster is ready for the 1st start:

hctl bootstrap --mkfs singlenode.yaml

Vagrant basics

Vagrant can be thought of as a scriptable unification API on top of various virtualization providers, like VirtualBox, VMware, KVM etc. From a user perspective all virtual machine configuration is done in a single Vagrantfile, which essentially is just a ruby script. It's processed every time vagrant command is executed, which expects to find it in the current working directory.

Most common Vagrant commands are:

# checking status of VM(s), e.g. running, halted, destroyed
vagrant status

# creating a VM if it doesn't exist or starting it if it's stopped;
# if there are provisioning steps, they are performed only once when VM is
# created/started for the first time
vagrant up

# stopping (powering off) a VM in graceful way
vagrant halt

# forcing a VM to power off, as if unplugging power cable
vagrant halt -f

# loggin into VM
vagrant ssh

# repeating provisioning steps (VM should be running)
vagrant provision

# repeating provisioning steps (VM should be stopped)
vagrant up --provision

# destroying VM and all associated data disks
vagrant destroy

Most of the Vagrant commands accept VM name as argument if there are multiple VMs configured in Vagrantfile. It is also possible to specify a regular expression instead of a single name to operate on several VMs:

# start a 'cmu' VM
vagrant up cmu

# start all SSUs
vagrant up /ssu/

# start all VMs
vagrant up cmu /ssu/ /client/

NOTE: on Windows host, Ansible is running on guest VMs (not on host) and all the Ansible tasks scripts are rsync-ed to guests at /vagrant/ folder. So whenever the scripts are updated on host they can be rsync-ed to guests (in order to pick up the changes) with the following command:

m0vg rsync cmu /ssu/ /client/

Streamlining VMs creation and provisioning with snapshots

It might be useful to save VM state just after provisioning for instant access to a clean VM without re-doing a complete provisioning from scratch. Please notice, that it's better when VMs are powered off when snapshot is made:

# poweroff all VMs
m0vg halt

# create snapshots, 'clean-vm' is just a name of the snapshot and can be
# changed to your liking
m0vg snapshot save cmu clean-vm
m0vg snapshot save ssu1 clean-vm
m0vg snapshot save ssu2 clean-vm
m0vg snapshot save client1 clean-vm

Then later, in order to discard the current state and restore a clean VM one may do:

m0vg snapshot restore --no-provision cmu clean-vm

If --no-provision option is omitted, the Ansible provisioning will be repeated after the restore phase. It may come in handy for getting latest security updates for the VM since snapshot creation.

Managing multiple VM sets with workspaces

Workspaces is a handy little feature of m0vg script, it exploits the fact that Vagrant keeps all configuration data related to Vagrantfile in a single directory. If that special directory is replaced with another one, Vagrant will use it instead. So it's possible to keep around multiple of those directories and switch between them, thus having multiple virtual clusters.

The m0vg supports following actions on workspaces:

m0vg workspace list
m0vg workspace add    <NAME>
m0vg workspace switch <NAME>

The workspace sub-command can be shortened as just ws.

m0vg also maintains a dedicated .env file for each workspace so when switching between workspaces each keeps it's own set of environment variables.

Executing Ansible commands manually

In some rare cases it can be useful to run Ansible commands against Vagrant VMs manually. For this purpose m0vg script supports ansible command. Here are just a few examples:

# list all hosts present in the cluster
m0vg ansible cluster.yml --list-hosts

# list all the tasks that would be performed for 'cmu' machine
m0vg ansible cmu.yml --list-tasks

VirtualBox / VMware / Libvirt specifics

On the other hand it can be achieved by starting a VM with GUI enabled:

M0_VM_ENABLE_GUI=yes m0vg up cmu

This needs to be done only once and VM will appear in VMware's VM Library.