Buildbot

Buildbot masters and slaves

Master

From the root of the git checkout (this will use the master.cfg from CoolProp):

virtualenv buildbot-sandbox
source buildbot-sandbox/bin/activate
pip install sqlalchemy==0.7.10 buildbot
cd dev/buildbot
buildbot create-master master
buildbot start master

The file buildbot-private.py (which is a python module with the passwords for the slaves as well as the buildbot website), should also be placed in the master folder next to master.cfg. Alternatively, you can put the buildbot_private.py in another folder on the master’s computer and make a soft-link in the master folder to point to the buildbot_private.py file.

If you want to completely restart the master, you can do:

buildbot restart master

but usually a:

buildbot reconfig master

will do the job since it will just reparse the configuration file without signing you out of the server

To ensure that the buildbot server stays online, you can make a script with the contents:

buildbot start /path/to/master_folder

and add it to a cron job.

The work with the master.cfg has proven to be a little more complicated. Storing changes in the repository and pulling them to the repository on the server is a little too cumbersome, especially when many iterations are needed to fix issues with the configurations. We already discussed this a couple of times <https://github.com/CoolProp/CoolProp/issues/1052>, but here the latest version of the preferred work flow for editing the master.cfg file:

  1. SSH to server and play with configuration file.
  2. Run ./scripts/buildbot.sh reconfig on server to activate changes, ignore git warnings.
  3. Repeat 1. and 2. until you are happy with you configuration
  4. Copy the server file to your local repository.
  5. Commit and push local repository.
  6. Run pushd buildbot/CoolProp.git/ && git reset –hard origin/master && popd on server.
  7. Run ./scripts/buildbot.sh reconfig on server, make sure there are no warnings from git.

Slaves

To start a slave connected to a buildbot master at IP address 10.0.0.2 (default for host for VirtualBox), with a slave named a-slave and passsword pass, run the command:

virtualenv a-slave-sandbox
source a-slave-sandbox/bin/activate
pip install sqlalchemy==0.7.10 buildbot-slave
buildslave create-slave a-slave coolprop.dreamhosters.com:port a-slave pass
buildslave start a-slave

If the master is somewhere else, just change the IP address. As of Sept, 2014, the master was at www.coolprop.dreamhosters.com. The buildbot_private.py on the master holds the required passwords.

OSX Virtualbox host

Thanks to http://stackoverflow.com/questions/1261975/addressing-localhost-from-a-virtualbox-virtual-machine and the comment of spsaucier you basically need to do the following, copied verbatim:

To enable this on OSX I had to do the following:

  1. Shut your virtual machine down.
  2. Go to VirtualBox Preferences -> Network -> Host-only Networks -> click the “+” icon. Click OK.
  3. Select your box and click the “Settings” icon -> Network -> Adapter 2 -> On the “Attached to:” dropdown, select “Host-only Adapter” and your network (vboxnet0) should show up below by default. Click OK.
  4. Once you start your box up again, you should be able to access localhost at http://10.0.2.2/

You can refer to it by localhost and access other localhosted sites by adding their references to the hosts file (C:windowssystem32driversetchosts) like the following:

10.0.2.2    localhost
10.0.2.2    subdomain.localhost

Python slaves

Based on the miniconda Python ecosystem, you can create your own virtual environments for building the Python wheels. This requires the following steps on a Windows machine:

conda create -n CoolProp27 python=2.7 cython pip pywin32 unxutils requests jinja2 pyyaml pycrypto wheel ndg-httpsclient
conda create -n CoolProp33 python=3.3 cython pip pywin32 unxutils requests jinja2 pyyaml pycrypto wheel
conda create -n CoolProp34 python=3.4 cython pip pywin32 unxutils requests jinja2 pyyaml pycrypto wheel
conda create -n CoolProp35 python=3.5 cython pip pywin32 unxutils requests jinja2 pyyaml pycrypto wheel

Please repeat the steps above for both 32bit and 64bit Python environments.

On a Linux system, things only change a little bit:

conda create -n CoolProp27 python=2.7 cython pip requests jinja2 pyyaml pycrypto wheel
conda create -n CoolProp33 python=3.3 cython pip requests jinja2 pyyaml pycrypto wheel
conda create -n CoolProp34 python=3.4 cython pip requests jinja2 pyyaml pycrypto wheel
conda create -n CoolProp35 python=3.5 cython pip requests jinja2 pyyaml pycrypto wheel

Please make sure that the standard shell /bin/sh used by the builbot is bash or zsh. We make use of the source command, which is not part of the POSIX specification. In Debian, dpkg-reconfigure dash can be used.

At the moment, it is not possible to use several slaves for the same build job. We have to find a new way to generate the configuration.

Information on building the single wrappers can be found on this dedicated page.

For uploading generated binary python files to PYPI, you should create a file ~\.pypirc with the contents:

[distutils]
index-servers=
    pypi
    test

[test]
repository = https://testpypi.python.org/pypi
username = user
password = XXXXXXXXXXXXXXXX

[pypi]
repository = https://pypi.python.org/pypi
username = user
password = XXXXXXXXXXXXXXXX

Buildbot as a service (Windows)

On Windows, you create a batch script that activates your virtual environment and starts the buildslave:

@echo off
call "C:\Program Files (x86)\Miniconda32_27\Scripts\activate.bat" Buildbot
buildslave start "C:\CoolProp-slave"

This script can then be added to the system services via:

sc create <serviceName> binpath= <pathToBatFile> DisplayName= "CoolProp Buildbot" start= auto

You might want to run services.msc to edit the user that runs the service. If you are tired of the error messages from the non-returning script, you could also use a service wrapper like NSSM to start the script.

Buildbot and launchd (Mac OS)

As written in the Buildbot Wiki, you can start your slaves automatically with a so called plist or property list. Place the example content below in a file called /Library/LaunchDaemons/org.coolprop.a-slave.plist and make sure it is owned by the user root and the group wheel:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>StandardOutPath</key>
    <string>org.coolprop.a-slave.log</string>
    <key>StandardErrorPath</key>
    <string>org.coolprop.a-slave-err.log</string>
    <key>Label</key>
    <string>org.coolprop.a-slave</string>
    <key>Program</key>
    <string>/Users/buildbot/bin/a-slave.command</string>
    <key>RunAtLoad</key>
    <true/>
    <key>KeepAlive</key>
    <dict>
        <key>SuccessfulExit</key>
        <false/>
    </dict>
    <key>GroupName</key>
    <string>staff</string>
    <key>UserName</key>
    <string>buildbot</string>
    <key>WorkingDirectory</key>
    <string>/Users/buildbot/slave/logs</string>
    <key>SessionCreate</key>
    <true/>
</dict>
</plist>

Please change the file above according to your needs and pay special attention to username and path definitions. The script a-slave.command that is called by launchd could look like this one:

#!/bin/bash
#
# Description: This file call the control script to start and
#              stop the buildbot slave. It stays open when being
#              called and waits for a signal to terminate running
#              and endless while-loop. After catching a signal
#              to terminate, it shuts down the build slave and
#              returns. It is a wrapper for another Bash script
#              allowing us to use launchd on MacOS.
#
# Author: Jorrit Wronski <jowr@mek.dtu.dk>
#
# Please remove the "Author" lines above and replace them
# with your own name if you copy and modify this script.
#
# If you experience any problems with the PATH variable on OSX,
# this setting might be for you:
if [ -x /usr/libexec/path_helper ]; then
  eval `/usr/libexec/path_helper -s`
fi
#
CTRLSCRI="/Users/username/a-slave.bsh"
#
trap "$CTRLSCRI stop; exit 0; " TERM SIGINT SIGTERM
#
$CTRLSCRI start & wait
# Just idle for one hour and keep the process alive
# waiting for SIGTERM.
while : ; do
  sleep 3600 & wait
done
#
echo "The endless loop terminated, something is wrong here."
exit 1

Note that this script calls another Bash script that does the actual work. We hope to simplify maintenance by using a common control script for Linux and MacOS as shown in Buildbot slave management (Mac OS and Linux).

Or alternatively, you can just launch buildslave directly if you do not use conda environment:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>KeepAlive</key>
    <true/>
    <key>Label</key>
    <string>com.start.buildbot</string>
    <key>ProgramArguments</key>
    <array>
        <string>/Users/Ian/anaconda/bin/buildslave</string>
        <string>restart</string>
        <string>slave</string>
    </array>
    <key>RunAtLoad</key>
    <true/>
    <key>StandardErrorPath</key>
    <string>/Users/Ian/.buildbot_stderr</string>
    <key>StandardOutPath</key>
    <string>/Users/Ian/.buildbot_stdout</string>
    <key>UserName</key>
    <string>Ian</string>
    <key>WorkingDirectory</key>
    <string>/Users/Ian</string>
</dict>
</plist>

Buildbot as a daemon (Linux)

On Linux, you can add the following lines to the end of your ~/.profile file (similar ideas apply on other platforms) to start the slave automatically at user log in:

# Connect to the buildbot master
buildslave start ~/slave

... or even better, you install a service that gets started and shutdown together with your computer. For Debian/Ubuntu, we recommend a script like:

#! /bin/sh
### BEGIN INIT INFO
# Provides:          buildslave
# Required-Start:    $remote_fs $syslog
# Required-Stop:     $remote_fs $syslog
# Default-Start:     2 3 4 5
# Default-Stop:      0 1 6
# Short-Description: A script to start the buildbot slave at boot time
# Description:       This file activates the virtual environment and starts
#                    the buildbot slaves. It also shuts them down if the
#                    system is halted. Place it in /etc/init.d.
### END INIT INFO

# Author: Jorrit Wronski <jowr@ipu.dk>
#
# Please remove the "Author" lines above and replace them
# with your own name if you copy and modify this script.

EXECUSER=username
NAME="a-slave"
CTRLSCRI="/home/username/$NAME.bsh"

# Load the VERBOSE setting and other rcS variables
. /lib/init/vars.sh

# Define LSB log_* functions.
# Depend on lsb-base (>= 3.2-14) to ensure that this file is present
# and status_of_proc is working.
. /lib/lsb/init-functions

#
# Function that starts the daemon/service
#
do_start(){
  sudo -u $EXECUSER $CTRLSCRI start
  #start-stop-daemon --start --user $EXECUSER --chuid $EXECUSER --startas $CTRLSCRI -- start
  RETVAL="$?"
  return "$RETVAL"
}

#
# Function that stops the daemon/service
#
do_stop() {
  #start-stop-daemon --stop --user $EXECUSER --startas
  sudo -u $EXECUSER $CTRLSCRI stop
  RETVAL="$?"
  return "$RETVAL"
}

case "$1" in
start)
    log_action_msg "Starting $NAME"
    do_start
    ;;
stop)
    log_action_msg "Stopping $NAME"
    do_stop
    ;;
restart)
    log_action_msg "Restarting $NAME"
    do_stop
    do_start
    ;;
*)
    log_action_msg "Usage: $0 {start|stop|restart}"
    exit 2
    ;;
esac
exit 0

Which then can be added to the scheduler with update-rc.d buildslave defaults. This should gracefully terminate the bot at shutdown and restart it again after reboot. To disable the service, run update-rc.d -f buildslave remove. You can enable and disable the daemon by runnning update-rc.d buildslave enable|disable.

Buildbot slave management (Mac OS and Linux)

Note that the two examples above call a user-script to activate the virtual environment and start the buildslave. Such a script could look like this:

#!/bin/bash
#
# Description: This file activates the virtual environment and starts
#              the buildbot slaves. It is also used to shut them down
#              during system shutdown.
#
# Author: Jorrit Wronski <jowr@ipu.dk>
#
# Please remove the "Author" lines above and replace them
# with your own name if you copy and modify this script.
#
VIRTENV="a-slave-sandbox"
SLAVEDIR="/home/username/a-slave"
#
## For virtualenv
#ACTICM="source $VIRTENV/bin/activate"
##DEACCM="source $VIRTENV/bin/deactivate"
#
# For miniconda
MINICO="/home/username/miniconda/bin/activate"
ACTICM="source $MINICO $VIRTENV"
#DEACCM="source deactivate"
#
# Carry out specific functions when asked to by the system
case "$1" in
  start)
    echo "Starting script buildbotslave "
    $ACTICM
    buildslave start $SLAVEDIR
    #$DEACCM
    ;;
  stop)
    echo "Stopping script buildbotslave"
    $ACTICM
    buildslave stop $SLAVEDIR
    #$DEACCM
    ;;
  restart)
    echo "Restarting script buildbotslave"
    $ACTICM
    buildslave restart $SLAVEDIR
    #$DEACCM
    ;;
  *)
    echo "Usage: $0 {start|stop|restart}"
    exit 1
    ;;
esac
exit 0

Setting MIME type handler

To change the MIME types on the server so that unknown file types will map properly to application/octet-stream, modify the buildbot.tac file to add the following block:

from twisted.web.static import File

webdir = File("public_html")
webdir.contentTypes['.mexw32'] = 'application/octet-stream'
webdir.contentTypes['.mexw64'] = 'application/octet-stream'
webdir.contentTypes['.mexmaci64'] = 'application/octet-stream'
webdir.contentTypes['.jnilib'] = 'application/octet-stream'
webdir.contentTypes['.mexa64'] = 'application/octet-stream'
webdir.contentTypes['.oct'] = 'application/octet-stream'
webdir.contentTypes['.whl'] = 'application/octet-stream'
webdir.contentTypes['.dylib'] = 'application/octet-stream'
...

and then do a buildbot restart master

Starting VirtualBox images at boot

You can use the built-in functionality https://www.virtualbox.org/manual/ch09.html#autostart on Linux and Mac or use your own configuration and create a daemon entry in Libray/LaunchDaemons. Make sure you use full paths to VBoxManage:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>GroupName</key>
    <string>staff</string>
    <key>InitGroups</key>
    <true/>
    <key>KeepAlive</key>
    <false/>
    <key>Label</key>
    <string>com.start.windows.vm</string>
    <key>ProgramArguments</key>
    <array>
        <string>/usr/bin/Vboxmanage</string>
        <string>startvm</string>
        <string>xp</string>
    </array>
    <key>RunAtLoad</key>
    <true/>
    <key>StandardErrorPath</key>
    <string>/Users/Ian/.virtualbox_window_stderr</string>
    <key>StandardOutPath</key>
    <string>/Users/Ian/.virtualbox_windows_stdout</string>
    <key>UserName</key>
    <string>Ian</string>
</dict>
</plist>

Documentation Builds

Some parts of the documentation are quite involved. That is why we decided not to rebuild the whole documentation after every commit. There is a special python script that runs once a day and performs the most expensive jobs during documentation rebuild. This covers the generation of validation figures for all fluids and the fitting reports for the incompressible fluids.

If you have some tasks that take a long time, make sure to add them to that special script in Web/scripts/__init__.py. This helps us to keep the continuous integration servers running with an acceptable latency with regard to the commits to the git repository. However, if you are unlucky and your commit coincides with figure generation, you will experience a long delay between your commit and the appearance of the freshly generated documentation on the website. You can follow the progress in the logfiles on the buildbot master though.

Work in Progress - Dockerfile Generator

To make it short, here is what you need to know if you trust us and the docker build system:

  • Make sure to set the correct environment variables in an additional file before you run a container, call it for example Dockerfile.slave.env.list:

    SLAVEDIR=/home/buildbot/slavedir
    MASTERHOST=bots.coolprop.org:port
    SLAVENAME=slavename
    SLAVEPASSWORD=pass
    BOTADMIN=Author Name
    BOTEMAIL=noreply@coolprop.org
    BOTHOST=A short description of the host computer
    
  • You can then run the official coolprop buildbot configuration with:

    docker run -d --env-file ./Dockerfile64.slave.env.list --name=CoolProp64-slave coolprop/slavepython
    docker run -d --env-file ./Dockerfile32.slave.env.list --name=CoolProp32-slave coolprop/slavepython32
    

    The above commands launch background processes using the docker containes for the Python buildslaves in 64bit and 32bit, respectively.

  • Some steps require the upload of files to different servers. In such cases, you should copy your SSH configuration or other login information to the container to make use of the automatic login that is required for rsync to work properly:

    docker cp ${HOME}/.ssh ${SLAVENAME}:/home/buildbot/
    docker cp ${HOME}/.pypirc ${SLAVENAME}:/home/buildbot/
    docker exec --user root ${SLAVENAME} chown -R buildbot /home/buildbot/.ssh /home/buildbot/.pypirc
        docker exec --user root ${SLAVENAME} chgrp -R buildbot /home/buildbot/.ssh /home/buildbot/.pypirc
    

Note

If you cannot copy the SSH keys, you can change the upload function in the master configuration to employ the built-in upload framework of buildbot.

Why the containers? In 2015, some of the buildbot slaves did not perform as expected. Especially the Python builds on the 64bit Linux machine took ages to complete and we could not find any obvious reason for this behaviour.

To make sure that there are no hidden flaws in the configuration of the buildbots or the virtual machines. Special configuration files can be used to build docker containers. Storing all configuration tasks in a structured Dockerfile reduces the risk of data loss and allows us to move the slaves between different machines.

Warning

Remember that each command in the Dockerfile leads to the creation of a new layer of files that cannot be deleted. Be careful here and try to bundle commands to save disk space and to keep garbage out of the image. See http://jrruethe.github.io/blog/2015/09/20/dockerfile-generator/ and https://docs.docker.com/articles/dockerfile_best-practices/ for more good advice on this topic.

Some more useful commands when working with docker are:

docker stop `docker ps -aq`; docker rm `docker ps -aq`; #delete all docker containers
docker rmi `docker images -f "dangling=true" -q`; #delete all dangling docker images

The workflow to generate the images locally could look like:

git clone --recursive https://github.com/CoolProp/Dockerfiles.git CoolProp.Dockerfiles.git
cd CoolProp.Dockerfiles.git
cd slavebase/64bit      ; docker build -t coolprop/slavebase      -f Dockerfile . ; cd ..
cd slavepython/64bit    ; docker build -t coolprop/slavepython    -f Dockerfile . ; cd ..
cd slavelinuxopen/64bit ; docker build -t coolprop/slavelinuxopen -f Dockerfile . ; cd ..

Please also have a look at the CoolProp repository on Docker Hub to see which images are available for download https://hub.docker.com/r/coolprop/ and do not hesitate to contribute to the sources at https://github.com/CoolProp/Dockerfiles