hadoop-general mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Eli Collins <...@cloudera.com>
Subject Re: HEP proposal
Date Wed, 14 Jul 2010 17:46:18 GMT
Hey Konstantin,

Thanks for taking a look, comments in-line.

On Tue, Jul 13, 2010 at 1:54 PM, Konstantin Shvachko <shv@yahoo-inc.com> wrote:
> Eli,
> Thanks for a really good proposal.
> Some questions / comments:
> On voting
> 1. Which voting rule?
> http://www.apache.org/foundation/glossary.html#ConsensusApproval
> http://www.apache.org/foundation/glossary.html#MajorityApproval
> I think you mean the MajorityApproval as it does not have veto rule.
> So may be it's just clarifying the reference.

Good point, clarified so it's majority approval.

> 2. Who can vote?
> Usually PMCs have Binding Votes.
> Would be good to have a sentence clarifying this.

Yup, added.

> 3. How long does the vote go?
> Usual 3 days may not be enough. One week is reasonable?

Specified one week.

> 4. Discussion on public lists.
> A HEP can evolve from a jira, then it should be counted as a public
> discussion. I think it makes sense even to continue the discussion
> there if so.

Agreed, changed the wording to "If the scope of the idea is limited to
a specific project the discussion may happen on the project-specific
list or jira."

> 5. How the set of editors is selected?
>   "The editors are apointed and removed by the PMC informally, similar to
>   how the Apache Board appoints shepherds to projects."
> This needs a reference. How does Apache Board appoints shepherds?

Good question, anyone know? Since it's informal I imagine shepherds
volunteer. The editors could be a subset of the PMC that either
volunteers or is rotated periodically.

> 6. The level of design details.
> I think HEP should have a pretty detailed design. When people vote they
> will want to be sure the design can lead to a reasonable implementation.
> Should we say "implementation-ready design", rather than
> "A high-level explanation of the design."
> Or just
> "A _detailed_ explanation of the design."

Rewrote this section, tried to make it more explicit about giving both
a high-level view and complete enough description so the design can
lead to a reasonable implementation. Also added that this section
should cover how to test the design.

> 7. Typos:
> successuflly, apointed, intial


Updated draft follows.


HEP: 1
Title: HEP Purpose and Guidelines
Author: Eli Collins
Status: Draft

What is a HEP?

HEP stands for Hadoop Enhancement Proposal, and is based on Python's
PEP (Python Enhancement Proposal) [1].  A HEP is a document that
describes a new feature, it's rationale, and issues the feature needs
to address in order to be successfully incorporated.

The intent is for HEPs to be the primary mechanism for proposing
significant new features to core Hadoop (common, HDFS and MapReduce),
incorporating community feedback, and recording the proposal.  Going
through the HEP process should improve the chances that a proposal is

While HEPs do not need to come with code, they are a mechanism to
propose features to the community, with the intent of contributing the
feature, rather than request the community implement a feature.

HEPs must be consistent with Apache bylaws [2], for example, the HEP
workflow takes place on the public Apache Hadoop lists.

When is a HEP Required?

HEPs should not impede casual contribution to Hadoop.  Small
improvements and bugs do not require HEPs.  Not all features need
HEPs.  While the decision is subjective, here are some guidelines to
indicate a HEP should be considered:

- The feature impacts backwards compatibility (eg modifies released
public APIs in an incompatible way).

- The feature requires that an existing component be substantially
re-designed (eg NameNode modified to use Bookkeeper).

- The implementation impact multiple parts of the system (eg symbolic
links versus adding a pluggable component like a codec).

- The feature impacts the entire development community (eg converts
the build system to use maven).

HEP Workflow

The author of a HEP should first try to determine if their idea is
HEP-able by sending mail to the general list.  If the scope of the
idea is limited to a specific project the discussion may happen on the
project-specific list or jira.  This gives the author a chance to
flesh out the proposal, address initial concerns, and figure out
whether it has a chance of being accepted.  The author's role is to
build consensus, and gather dissenting opinions.

Following this discussion the author should draft a HEP proposal
following the HEP template. The proposal should accurately reflect and
address feedback and dissenting opinions.  For example, flesh out
sections on backwards compatibility or testing. The author should send
the draft of the proposal to hep@hadoop.apache.org for review.  This
is a new, public list for editors and those interested in following
the review process.

A set of editors reviews incoming HEPs. Each HEP is assigned a single
primary editor. An editor may volunteer if they feel particular
functional expertise is required or assign HEPs to editors round

The editor reviews the proposal and may request it be updated if it
does not sufficiently address feedback raised during discussion, eg
why the proposal is not redundant with existing functionality, or is
technically sound, sufficiently motivated, covers backwards
compatibility, etc. As updates are necessary, the HEP author can check
in new versions if they have commit permissions, or can email new HEP
versions to the editor for committing. In order to ensure HEP
proposals make progress the editor should respond to proposal drafts
within two weeks of receiving them (or the proposer can request
another editor), and the proposer should generate updates to the draft
within two weeks of receiving feedback from the editor.

The editor's role is to determine if the proposal is complete, so that
the proposal can be voted on, not whether they agree with the proposal
itself.  The editor's involvement should increase the chance that a
HEP proposal makes it to a vote.

Once the editor deems the proposal is complete they add it to a
versioned HEP repository and the author posts the proposal to
general@hadoop.apache.org for vote.  HEP votes, like Apache procedural
votes, use majority approval [3]. Only PMC members have binding votes.
Votes are open for a period of 1 week to allow all active voters time
to consider the proposal. Successful HEPs are assigned a number,
unsuccessful HEPs remain drafts.

The editors are appointed and removed by the PMC informally, similar
to how the Apache Board appoints shepherds to projects.

HEP Contents

Each HEP should contain the following:

1. Preamble -- Including the HEP number, a short descriptive title,
and the names of the authors.

2. Abstract -- A short (~200 word) description of the technical issue
being addressed.

3. Copyright/public domain -- Each HEP must either be explicitly
labelled as placed in the public domain (see this HEP as an example).

4. Design -- This section should give both a high-level view and a
complete description of the feature.  While the design does not need
to cover implementation detail it should be clear to the reader that
the design can lead to a reasonable implementation.  This section
should cover intended use cases, failure scenarios, strategies for
testing, and impact on the existing system.

5. Motivation -- The motivation spells out the use case for the
feature and the benefits it provides.

6. Rationale -- The rationale describes what motivated the design and
why particular design decisions were made.  It should describe
alternate designs that were considered and related work, e.g. how the
feature is designed in other systems. It should also consider whether
the feature could be achieved by layering atop the existing system
rather than modifying it.

The rationale should provide evidence of consensus within the
community and discuss important objections or concerns raised during

7. Backwards Compatibility -- All HEPs that introduce backwards
incompatibilities must include a section describing these
incompatibilities and their severity.  The HEP must explain how the
author proposes to deal with these incompatibilities.  HEP submissions
without a sufficient backwards compatibility treatise may be rejected

HEP Template

HEPs should be plain text with minimal structural markup that adheres
to a rigid style.  You can use this HEP as an example. Each HEP starts
with a header that contains the HEP number (or empty if the number has
not yet been assigned), title, list of authors and status (Draft,
Accepted, Rejected, or Withdrawn).

Auxiliary Files

HEPs may include auxiliary files such as diagrams.  Such files must be
named ``hep-XXXX-Y.ext``, where "XXXX" is the HEP number, "Y" is a
serial number (starting at 1), and "ext" is replaced by the actual
file extension (e.g. "png").


1. http://www.python.org/dev/peps/pep-0001

2. http://www.apache.org/foundation/bylaws.html

3. http://www.apache.org/foundation/glossary.html#MajorityApproval


This document has been placed in the public domain.

View raw message