Design & Architecture

The ruffus module has the following design goals:

• Simplicity. Can be picked up in 10 minutes
• Elegance
• Lightweight
• Unintrusive
• Flexible/Powerful

Task-based rather than file-based work flow

Computational pipelines, especially in science are best thought of as successively dependent stages (we call them tasks: See the Glossary ).

Traditionally, especially in Unix, pipelined stages have been linked together using files.

Tools like GNU make or scons create a static dependency graph between all the files which are produced by the pipe line. Inference “rules” specifying how files of one type are converted into another add additional conciseness to such systems.

However, the use of files to link pipeline stages makes these tools a poor match for scientific computational pipelines:

1. Though the stages of a pipeline will be specified up-front, the number of (conceptually parallel) operations (jobs) involved in each stage may not be. If the data sets are large, this may be broken up into larger number of smaller pieces.
2. Some stages can depend on databases or data in other formats. It seems clumsy to always shoehorn everything into a “file” paradigm
3. Static File dependencies graphs are very difficult to debug.

This is how ruffus runs its pipelines:

1. Dependencies between tasks are specified up front. These can be displayed as a flow chart.

2. Out-of-date pipeline stages are identified by asking each task if it has any out-of-date operations.

   This is a short-circuiting operation, and ruffus tries to do the minimal amount of querying

3. All out-of-date tasks and subsequent stages in the pipeline will be re-run

4. When each task is run, it is asked (again) which operations are out-of-date and those jobs will be executed

In practice, this means that ruffus re-constructs its dependencies on-the-fly when it gets to each stage, giving much more flexibility.

Alternatives

Tools used to build executables can be used to manage computational pipelines. These include

• GNU make
• scons
• ant

A comparison of more make-like tools is available from Ian Holmes’ group.

It is often necessary to learn a specialised (domain-specific) language. GNU make syntax, for example, is much criticised because of limited support for abstraction compared with modern programming languages like C, Perl, python etc. GNU makefiles can quickly become unmaintainable.

Pipeline specifications are usually written in a “declarative” rather than “imperative” manner. You write a specification that describes the dependencies, and the tool figures out how to perform the computations in the correct order. However, because GNU make and its kin depend entirely on file dependencies, the links between pipeline stages can be difficult to trace, and nigh impossible to debug when there are problems.

There are also complete workload managements systems such as Condor. Various bioinformatics pipelines are also available, including that used by the leading genome annotation website Ensembl, Pegasys, GPIPE, Taverna, Wildfire, MOWserv, Triana, Cyrille2 etc. These all are either hardwired to specific databases, and tasks, or have steep learning curves for both the scientist/developer and the IT system administrators

See also

Bioinformatics pipelines

Condor:

http://www.cs.wisc.edu/condor/description.html

Ensembl Analysis pipeline:

http://www.ncbi.nlm.nih.gov/pubmed/15123589

Pegasys:

http://www.ncbi.nlm.nih.gov/pubmed/15096276

GPIPE:

http://www.biomedcentral.com/pubmed/15096276

Taverna:

http://www.ncbi.nlm.nih.gov/pubmed/15201187

Wildfire:

http://www.biomedcentral.com/pubmed/15788106

MOWserv:

http://www.biomedcentral.com/pubmed/16257987

Triana:

http://dx.doi.org/10.1007/s10723-005-9007-3

Cyrille2:

http://www.biomedcentral.com/1471-2105/9/96

Acknowledgements

• Bruce Eckel’s insightful article on A Decorator Based Build System was the obvious inspiration for the use of decorators in Ruffus.

  The rest of the Ruffus takes uses a different approach. In particular:

  1. Ruffus uses task-based not file-based dependencies

  2. Ruffus tries to have minimal impact on the functions it decorates.

     Bruce Eckel’s design wraps functions in “rule” objects.

     Ruffus tasks are added as attributes of the functions which can be still be called normally. This is how Ruffus decorators can be layered in any order onto the same task.

• Languages like c++ and Java would probably use a “mixin” approach. Python’s easy support for reflection and function references, as well as the necessity of marshalling over process boundaries, dictated the internal architecture of Ruffus.

• The Boost Graph library for text book implementations of directed graph traversals.

• Graphviz. Just works. Wonderful.