[zeta-dev] Requirements document for a Workflow CouchDB Tiein

[Benjamin Eberlei <ko...@beberlei.de>]

Hello everyone,

read on for my first draft of a requirements document for a Workflow CouchDB Tiein.

greetings,
Benjamin

Zeta Components - Workflow CouchDB Tiein Design
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
:Author: Benjamin Eberlei

Introduction
============

Description
-----------

The WorkflowCouchDBTiein provides functionalities to load and save workflows into CouchDB
and to start, save and resume executions by persisting the state into CouchDB.

Benefits of a CouchDB backend over the DatabaseTiein is the non-relational structure of a workflow.
Each node and execution has different variables, state and dependencies. Additionally a relational
implementation requires to split the workflow in many different tables, reducing performance during
save and query operations. CouchDB could save workflow and executions in their own documents
respectively, requiring only between one and three HTTP calls to CouchDB to fetch and save all the workflow
related data again.

Additionally having the workflows (+history) and executions in single documents makes the whole
workflow process much more readable and debugable for developers.

Current implementation
----------------------

I have a current implementation to figure out some of the problems that might come
with using CouchDB for workflow and it helped find some problems in Workflow Base
component that have to be fixed before. However the code is not tested so I will
probably begin from scratch.

Requirements
============

The implementation should obviously work exactly like the base or database workflows,
so that there are no differences using either one.

Design goals
============

-   A thin HTTP Layer is provided with this component to speak to CouchDB
    so that no third dependency needs to be introduced.
-   Workflows, their Nodes and Executions will be assigned CouchDB UUIDs.
-   Workflows and executions are saved into a single document each.
-   Time inconsistencies due to changing workflows will be circumvented
    by saving all versions of the workflow into a substructure and associating
    each active execution with a workflow id and version.
-   Document JSON structures are crafted in a way that developers can implement
    their own queries on various variables of the execution state are possible
    using CouchDB views.