Hi,
I was talking with Daniel Gruno and wrote the following ideas to him. Note that
these are just ideas and not based on any real momentary issue or concern --
though a more general concern about how Apache should evolve.
Apache should NOT use a binary "podling" / "top-level" model. All projects
should simply have a "health score" and that health score is derived from
measurables. Because of Apache Infrastructure's centralized server model (email
lists, version control, distributions, homepages, etc.), it has the ability to
gather metrics such as, for example, the distribution of pushes to the
repository, the branch factor of the mailing list, the centrality of the
project in the Central Maven repository dependency graph, the number of
non-sequisters (dead-end conversations) in the email chain, the length of
discussions in JIRA, etc. etc. Which metrics are important? Who care -- just
make up things to glean from the wealth of information you already have access
to. Watch...
Next, the Apache members subjectively say which projects they think are "good"
(healthy). This can even be a global vote including everyone in the world and
(should be) dynamic over time as projects evolve with time. Either way, lets
say, the ranking says Apache Hadoop, Apache Solr, Apache Commons, etc. are the
(collective subjective's) "best" Apache projects. Now, there should exist a
multi-dimensional projection of the aforementioned gleaned statistics what will
have Hadoop, Solr, Commons, etc. close to one another in metric-space
(clustered). Likewise, low ranking projects should be close to one another in
this space and far from Hadoop, Solr, Commons, etc. Find that projection and
that is your "healthy metric space."
From here, all Apache projects have a computed "healthy" score(s) and when
users go to download, lets say, Lucene, they go: "Cool. This is a healthy
project." (it has a HEALTH.txt file distributed with it, lets say). What that
means is that Lucene, at that release was in the "healthy" cluster of the
metric space. This model has various benefits:
1. There is no need to have philosophical arguments (not grounded in
measurables) about what rules a project should follow (bounded by law).
- Perhaps a project that is exclusive, but is X is still in the
"healthy" subspace.
- Perhaps having bad documentation is a "unhealthy" even though
Apache doesn't care about documentation.
- Perhaps too much discussion causes a project to become
"unhealthy."
- Perhaps … who knows? … let the statistics do the talking.
- Apache becomes a breeding ground for different models of
open source (bounded by law), not just "The Apache Way."
- And these models are measurable! Let us study the act
of open source.
2. "Top-level" projects can fall from grace.
- Currently, all "top-level" projects are "equal." This should
by dynamic as the mighty do fall.
- It is possible for what are now "podlings" to be "healthy" as
they simply are coming into Apache.
- "The student is the master."
- Hadoop 1.2.1 might be the healthiest version of Hadoop (as I
tend to believe). "Hadoop" is not a thing eternal.
3. Less work for people.
- No more VOTEing on graduation.
- No more amorphous aesthetic arguments about "The Apache Way."
- No more long winded contradictory documentation about how
things should be done (bounded by law).
The Apache Way should be about metrics, not about philosophy as different paths
lead to the same mountain top <--- See! Is that random Buddhist saying that
everyone just "believes" even true? :) Get the human out of the loop!
Thanks for reading,
Marko.
http://markorodriguez.com
P.S. The same should hold true for educational degrees. I graduate and now
forever I'm an expert in computers? Medical doctors too! A 90 year old doctor
can do surgery on me?!?!… Binary graduation is not "real." Metrics, metrics,
metrics --- we live in a world where this is possible. For every "thing" good
comes and goes, up and down…
