MetPX-Sarracenia Developer's Guide

Version: 2.17.03a3
Date: March 2017

Tools you Need

To hack on the sarracenia source, you need:

  • python3. The application is developed in and depends on python versions > 3.
  • some python-amqp bindings (like python-amqplib for current implementations)
  • a bunch of other modules indicated in the dependencies ( or debian/control)
  • paramiko. For SSH/SFTP support you need to install the python-paramiko package (which works with python3 even though the documentation says it is for python2.)
  • (soon?) watchdog ( ) not available as a .deb yet. used to encapsulate directory watching for sr_watch.
  • git. in order to download the source from the repository.
  • a running rabbitmq broker (if you want to actually run any code.)

after you have cloned the source code:

git clone git:// metpx-sarracenia
cd metpx-sarracenia

The rest of the Guide assumes you are there.


The development process is to write up what one intends to to or have done into a restructured text file in the doc/design sub-directory. The files there provide a basis for discussion. Ideally, the information there acts as a pieces which can be edited into documentation for the features as they are implemented.

Each new component sr_whatever, should have relevant man pages implemented. The Guides should also be revised. The form of the documentation is still under discussion. Current thinking:

When there are new sections, they should likely start out in design/ and after review, graduate into the main documentation.


Development occurs on the master branch, which may be in any state at any given time, and should note be relied upon. From time to time releases are tagged, and maintenance results in a branch. Releases are classified as follows:

snapshot releases taken directly from master, with no other qualitative guarantees. no gurantee of functionality, some components may be partially implemented, some breakage may occur. no bug-fixes, issues addressed by subsequent version. Often used for early end-to-end testing (rather than installing custom from tree on each test machine.)
Feature Complete for a given release. Components in their final form for this release. Documentation exists in at least one language. All previously known release block bugs addressed. no bug-fixes, issues addressed by subsequent version.
RC - Release Candidate.
implies it has gone through beta to identify and address major issues. Translated documentation available. no bug-fixes, issues addressed by subsequent version.

Final versions have no suffix and are considered stable and supported. Stable should receive bug-fixes if necessary from time to time. One can build python wheels, or debian packages for local testing purposes during development.

Python Wheel

For testing and development:

python3 bdist_wheel

should build a wheel in the dist sub-directory.


This process builds a local .deb in the parent directory using standard debian mechanisms. - check the build-depends line in debian/control for dependencies that might be needed to build from source. - The following steps will build sarracenia but not sign the changes or the source package:

cd metpx/sarracenia
sudo apt-get install devscripts
debuild -uc -us
sudo dpkg -i ../<the package just built>


Before releasing, as a Quality Assurance measure one should run all available self-tests. It is assumed that the specific changes in the code have already been unit tested. Please add self-tests as appropriate to this process to reflect the new ones.

The configuration one is trying to replicate:


Assumption: test environment is a linux PC, either a laptop/desktop, or a server on which one can start a browser.

0 - Make a local wheel and installing on your workstation

In the git clone tree ... metpx-sarracenia create a wheel by running either:

python3 bdist_wheel

it creates a wheel package under dist/metpx*.whl then as root install that new package:

pip3 install --upgrade ...<path>/dist/metpx*.whl


debuild -us -uc
sudo dpkg -i ../python3-metpx-...

which accomplishes the same thing using debian packaging.

1- Install servers on localhost

Install a minimal localhost broker, configure test users. with credentials stored for localhost:

sudo apt-get install rabbitmq-server
sudo rabbitmq-plugins enable rabbitmq_management
echo "amqp://bunnymaster:MaestroDelConejito@localhost/" >>~/.config/sarra/credentials.conf
echo "amqp://tsource:TestSOUrCs@localhost/" >>~/.config/sarra/credentials.conf
echo "amqp://tsub:TestSUBSCibe@localhost/" >>~/.config/sarra/credentials.conf
echo "amqp://tfeed:TestFeeding@localhost/" >>~/.config/sarra/credentials.conf

cat >~/.config/sarra/default.conf <<EOT

broker amqp://tfeed@localhost/
cluster localhost
admin amqp://bunnymaster@localhost/
feeder amqp://tfeed@localhost/
declare source tsource
declare subscribe tsub

sudo rabbitmqctl delete_user guest
sudo rabbitmqctl add_user bunnymaster MaestroDelConejito
sudo rabbitmqctl set_permissions bunnymaster ".*" ".*" ".*"
sudo rabbitmqctl set_user_tags bunnymaster administrator
cd /usr/local/bin
sudo wget http://localhost:15672/cli/rabbitmqadmin
chmod 755 rabbbitmqadmin
sr_audit --users foreground

Perhaps in a separate window if you want to see output separately, a report message is printed for each GET the server answers. the setup script starts a trivial web server, and defines some fixed test clients that will be used during self-tests:

cd sarracenia/test
. ./

The working test flow setup script ( will install configuration files for:

  • two sr_shovel configurations to copy messages from from
  • an sr_winnow to remove duplicates from the shovelled sources.
  • an sr_sarra to read the winnow output, and post fills mirrored on the trivial web server.
  • an sr_subscribe to down load the files from the local server.

and starts this network of configurations running. if the passes, then one has a reasonable confidence in the overall functionality of the application, but the test coverage is not exhaustive. It is more qualitative sampling of the most common use cases rather than a thorough examination of all functionality. While not thorough, it is good to know wtf is working.

2 - Rerun basic self test

The following script runs some unit self tests of individual .py files in the source code:

3 - Run Working Test Flow Check

The script reads the log files of all the components started, and compares the number of messages, looking for a correspondence within +- 10% It takes a few minutes for the configuration to run before there is enough data to do the proper measurements:


sample output:

blacklab% ./
initial sample building sample size 3421 need at least 1000
test 1: SUCCESS, shovel1 (3421) reading the same as shovel2 (3421) does
test 2: SUCCESS, winnow (6841) reading double what sarra (3421) does
test 3: SUCCESS, subscribe (3421) has the same number of items as sarra (3421)
test 4: SUCCESS, subscribe (3421) has the same number of items as shovel1 (3421)
4 - Run and check results

The following tests are self descriptive, but there is no obvious check of success. One must examine the output of the command and determine if the result is as intended:

When done testing, run:

. ./

Which will kill the running web server, and delete all local queues.

Building a Release

MetPX-Sarracenia is distributed in a few different ways, and each has it's own build process. Packaged releases are always preferable to one off builds, because they are reproducible.

When development requires testing across a wide range of servers, it is preferred to make an alpha release, rather than installing one off packages. So the preferred mechanisms is to build the ubuntu and pip packages at least, and install on the test machines using the relevant public repositories.

To publish a release one needs to:

  • Set the version.
  • upload the release to so that installation with pip succeeds.
  • upload the release to, so that the installation of debian packages using the repository succeeds.
  • upload the packages to sourceforge for other users to download the package directly
  • upload updated documentation to sourceforge.

Versioning Scheme

Each release will be versioned as <protocol version>.<YY>.<MM> <segment>


  • protocol version is the message version. In Sarra messages, they are all prefixed with v02 (at the moment).
  • YY is the last two digits of the year of the initial release in the series.
  • MM is a TWO digit month number i.e. for April: 04.
  • segment is what would be used within a series. from pep0440: X.YaN # Alpha release X.YbN # Beta release X.YrcN # Release Candidate X.Y # Final release


The first alpha release in January 2016 would be versioned as metpx-sarracenia-2.16.01a01

Setting the Version

Each new release triggers a tag in the git repository ( executes git tag -a sarra-v2.16.01a01 -m "release 2.16.01a01" )

A convenience script has been created to automate the release process. Simply run and it will guide you in cutting a new release.

  • Edit sarra/ manually and set the version number.

  • git commit -a

  • Run `` example:

    ./ "release 2.16.01a01"
  • you will be prompted to enter information about the release.

  • git push


Assuming pypi upload credentials are in place, uploading a new release is a one liner:

python3 bdist_wheel upload

Note that the same version can never be uploaded twice.

A convenience script has been created to build and publish the wheel file. Simply run and it will guide you in that.


Automated Build

  • Ensure the code mirror is updated by checking the Import details by checking this page
  • if the code is out of date, do Import Now , and wait a few minutes while it is updated.
  • once the repository is uptodate, proceed with the build request.
  • Go to the sarracenia release recipe
  • Click on the Request build(s) button to create a new release
  • The built packages will be available in the metpx ppa

Daily Builds

Daily builds are configured using this recipe and are run once per day when changes to the repository occur. These packages are stored in the metpx-daily ppa. One can also Request build(s) on demand if desired.

Manual Process

The process for manually publishing packages to Launchpad ( ) involves a more complex set of steps, and so the convenience script will be the easiest way to do that. Currently the only supported releases are trusty and xenial. So the command used is: sarra-v2.15.12a1 trusty xenial

However, the steps below are a summary of what the script does:

  • for each distribution (precise, trusty, etc) update debian/changelog to reflect the distribution

  • build the source package using:

    debuild -S -uc -us
  • sign the .changes and .dsc files:

    debsign -k<key id> <.changes file>
  • upload to launchpad:

    dput ppa:ssc-hpc-chp-spc/metpx-<dist> <.changes file>

Note: The GPG keys associated with the launchpad account must be configured in order to do the last two steps.

Updating The Project Website

The MetPX website is built from the documentation in the various modules in the project. It builds using all .rst files found in sarracenia/doc as well as some of the .rst files found in sundew/doc.

Building Locally

In order to build the HTML pages, the following software must be available on your workstation:

From a command shell:

cd site

Updating The Website

To publish the site to sourceforge (updating, you must have a account and have the required permissions to modify the site.

From a shell, run:

make SFUSER=myuser deploy

Development Environment

Local Python

Working with a non-packaged version:


python3 build
python3 install


Install winpython from version 3.4 or higher. Then use pip to install from PyPI.


Below are some coding practices that are meant to guide developers when contributing to sarracenia. They are not hard and fast rules, just guidance.

When to Report

sr_report(7) messages should be emitted to indicate final disposition of the data itself, not any notifications or report messages (don't report report messages, it becomes an infinite loop!) For debugging and other information, the local log file is used. For example, sr_shovel does not emit any sr_report(7) messages, because no data is transferred, only messages.

Where to Put Option Documentation

Most options are documented in sr_config(7), because they are common to many components. Any options used by multiple components should be documented there. Options which are unique to a single component, should be documented in the man page for that component.

Where the default value for an option varies among components, sr_config(7) should indicate only that the default varies. Each component's man page should indicate the option's default for that component.

Adding Checksum Algorithms

To add a checksum algorithm, need to add a new class to, and then modify to associate it with a label. Reading of makes this pretty clear. Each algorithm needs: - an initializer (sets it to 0) - an algorithm selector. - an updater to add info of a given block to an existing sum, - get_value to obtain the hash (usually after all blocks have updated it)

These are called by the code as files are downloaded, so that processing and transfer are overlapped.

For example, to add SHA-2 encoding:

from hashlib import sha256

class checksum_s(object):
    checksum the entire contents of the file, using SHA256.
    def __init__(self):
        self.value = '0'

    def get_value(self):
        self.value = self.filehash.hexdigest()
        return self.value

    def update(self,chunk):

    def set_path(self,path):
        self.filehash = sha256()

Then in, in the set_sumalgo routine:

if flgs == 's':
    self.sumalgo = checksum_s()

Might want to add 's' to the list of valid sums in validate_sum( as well.

It is planned for a future version to make a plugin interface for this so that adding checksums becomes an application programmer activity.