<li><ahref="#How_are_the_Classifications_Recorded.3F">How are the Classifications Recorded?</a></li>
<li><ahref="#FAQ">FAQ</a></li></ul>
<section>
<h2><aname="Motivation"></a>Motivation</h2>
<p>The interface taxonomy classification provided here is for guidance to developers and users of interfaces. The classification guides a developer to declare the targeted audience or users of an interface and also its stability.</p>
<ul>
<li>
<p>Benefits to the user of an interface: Knows which interfaces to use or not use and their stability.</p>
</li>
<li>
<p>Benefits to the developer: to prevent accidental changes of interfaces and hence accidental impact on users or other components or system. This is particularly useful in large systems with many developers who may not all have a shared state/history of the project.</p>
<p>Hadoop adopts the following interface classification, this classification was derived from the OpenSolaris taxonomy and, to some extent, from taxonomy used inside Yahoo. Interfaces have two main attributes: Audience and Stability.</p><section>
<h3><aname="Audience"></a>Audience</h3>
<p>Audience denotes the potential consumers of the interface. While many interfaces are internal/private to the implementation, other are public/external interfaces that are meant for wider consumption by applications and/or clients. For example, in posix, libc is an external or public interface, while large parts of the kernel are internal or private interfaces. Also, some interfaces are targeted towards other specific subsystems.</p>
<p>Identifying the audience of an interface helps define the impact of breaking it. For instance, it might be okay to break the compatibility of an interface whose audience is a small number of specific subsystems. On the other hand, it is probably not okay to break a protocol interface that millions of Internet users depend on.</p>
<p>Hadoop uses the following kinds of audience in order of increasing/wider visibility:</p>
<blockquote>
<p>Hadoop doesn’t have a Company-Private classification, which is meant for APIs which are intended to be used by other projects within the company, since it doesn’t apply to opensource projects. Also, certain APIs are annotated as @VisibleForTesting (from com.google.common .annotations.VisibleForTesting) - these are meant to be used strictly for unit tests and should be treated as “Private” APIs.</p>
</blockquote><section>
<h4><aname="Private"></a>Private</h4>
<p>A Private interface is for internal use within the project (such as HDFS or MapReduce) and should not be used by applications or by other projects. Most interfaces of a project are Private (also referred to as project-private). Unless an interface is intentionally exposed for external consumption, it should be marked Private.</p></section><section>
<p>A Limited-Private interface is used by a specified set of projects or systems (typically closely related projects). Other projects or systems should not use the interface. Changes to the interface will be communicated/negotiated with the specified projects. For example, in the Hadoop project, some interfaces are LimitedPrivate{HDFS, MapReduce} in that they are private to the HDFS and MapReduce projects.</p></section><section>
<h4><aname="Public"></a>Public</h4>
<p>A Public interface is for general use by any application.</p></section></section><section>
<p>Changes to an API fall into two broad categories: compatible and incompatible. A compatible change is a change that meets the following criteria:</p>
<ul>
<li>no existing capabilities are removed,</li>
<li>no existing capabilities are modified in a way that prevents their use by clients that were constructed to use the interface prior to the change, and</li>
<li>no capabilities are added that require changes to clients that were constructed to use the interface prior to the change.</li>
</ul>
<p>Any change that does not meet these three criteria is an incompatible change. Stated simply a compatible change will not break existing clients. These examples are compatible changes:</p>
<ul>
<li>adding a method to a Java class,</li>
<li>adding an optional parameter to a RESTful web service, or</li>
<li>adding a tag to an XML document.</li>
<li>making the audience annotation of an interface more broad (e.g. from Private to Public) or the change compatibility annotation more restrictive (e.g. from Evolving to Stable)</li>
</ul>
<p>These examples are incompatible changes:</p>
<ul>
<li>removing a method from a Java class,</li>
<li>adding a method to a Java interface,</li>
<li>adding a required parameter to a RESTful web service, or</li>
<li>renaming a field in a JSON document.</li>
<li>making the audience annotation of an interface less broad (e.g. from Public to Limited Private) or the change compatibility annotation more restrictive (e.g. from Evolving to Unstable)</li>
</ul></section><section>
<h3><aname="Stability"></a>Stability</h3>
<p>Stability denotes how stable an interface is and when compatible and incompatible changes to the interface are allowed. Hadoop APIs have the following levels of stability.</p><section>
<h4><aname="Stable"></a>Stable</h4>
<p>A Stable interface is exposed as a preferred means of communication. A Stable interface is expected not to change incompatibly within a major release and hence serves as a safe development target. A Stable interface may evolve compatibly between minor releases.</p>
<p>Incompatible changes allowed: major (X.0.0) Compatible changes allowed: maintenance (x.y.Z)</p></section><section>
<h4><aname="Evolving"></a>Evolving</h4>
<p>An Evolving interface is typically exposed so that users or external code can make use of a feature before it has stabilized. The expectation that an interface should “eventually” stabilize and be promoted to Stable, however, is not a requirement for the interface to be labeled as Evolving.</p>
<p>Incompatible changes are allowed for Evolving interface only at minor releases.</p>
<p>Incompatible changes allowed: minor (x.Y.0) Compatible changes allowed: maintenance (x.y.Z)</p></section><section>
<h4><aname="Unstable"></a>Unstable</h4>
<p>An Unstable interface is one for which no compatibility guarantees are made. An Unstable interface is not necessarily unstable. An unstable interface is typically exposed because a user or external code needs to access an interface that is not intended for consumption. The interface is exposed as an Unstable interface to state clearly that even though the interface is exposed, it is not the preferred access path, and no compatibility guarantees are made for it.</p>
<p>Unless there is a reason to offer a compatibility guarantee on an interface, whether it is exposed or not, it should be labeled as Unstable. Private interfaces also should be Unstable in most cases.</p>
<p>Incompatible changes to Unstable interfaces are allowed at any time.</p>
<p>A Deprecated interface could potentially be removed in the future and should not be used. Even so, a Deprecated interface will continue to function until it is removed. When a Deprecated interface can be removed depends on whether it is also Stable, Evolving, or Unstable.</p></section></section></section><section>
<h2><aname="How_are_the_Classifications_Recorded.3F"></a>How are the Classifications Recorded?</h2>
<p>How will the classification be recorded for Hadoop APIs?</p>
<ul>
<li>
<p>Each interface or class will have the audience and stability recorded using annotations in the org.apache.hadoop.classification package.</p>
</li>
<li>
<p>The javadoc generated by the maven target javadoc:javadoc lists only the public API.</p>
</li>
<li>
<p>One can derive the audience of java classes and java interfaces by the audience of the package in which they are contained. Hence it is useful to declare the audience of each java package as public or private (along with the private audience variations).</p>
</li>
</ul>
<p>How will the classification be recorded for other interfaces, such as CLIs?</p>
<ul>
<li>See the <ahref="Compatibility.html">Hadoop Compatibility</a> page for details.</li>
</ul></section><section>
<h2><aname="FAQ"></a>FAQ</h2>
<ul>
<li>
<p>Why aren’t the java scopes (private, package private and public) good enough?</p>
<ul>
<li>Java’s scoping is not very complete. One is often forced to make a class public in order for other internal components to use it. It also does not have friends or sub-package-private like C++.</li>
</ul>
</li>
<li>
<p>But I can easily access a Private interface if it is Java public. Where is the protection and control?</p>
<ul>
<li>The purpose of this classification scheme is not providing absolute access control. Its purpose is to communicate to users and developers. One can access private implementation functions in libc; however if they change the internal implementation details, the application will break and one will receive little sympathy from the folks who are supplying libc. When using a non-public interface, the risks are understood.</li>
</ul>
</li>
<li>
<p>Why bother declaring the stability of a Private interface? Aren’t Private interfaces always Unstable?</p>
<ul>
<li>Private interfaces are not always Unstable. In the cases where they are Stable they capture internal properties of the system and can communicate these properties to its internal users and to developers of the interface.
<ul>
<li>e.g. In HDFS, NN-DN protocol is Private but Stable and can help implement rolling upgrades. The stability annotation communicates that this interface should not be changed in incompatible ways even though it is Private.</li>
<li>e.g. In HDFS, FSImage the Stabile designation provides more flexible rollback.</li>
</ul>
</li>
</ul>
</li>
<li>
<p>What is the harm in applications using a Private interface that is Stable? How is it different from a Public Stable interface?</p>
<ul>
<li>While a Private interface marked as Stable is targeted to change only at major releases, it may break at other times if the providers of that interface also are willing to change the internal consumers of that interface. Further, a Public Stable interface is less likely to break even at major releases (even though it is allowed to break compatibility) because the impact of the change is larger. If you use a Private interface (regardless of its stability) you run the risk of incompatibility.</li>
</ul>
</li>
<li>
<p>Why bother with Limited-Private? Isn’t it giving special treatment to some projects? That is not fair.</p>
<ul>
<li>Most interfaces should be Public or Private. An interface should be Private unless it is explicitly intended for general use.</li>
<li>Limited-Private is for interfaces that are not intended for general use. They are exposed to related projects that need special hooks. Such a classification has a cost to both the supplier and consumer of the interface. Both will have to work together if ever there is a need to break the interface in the future; for example the supplier and the consumers will have to work together to get coordinated releases of their respective projects. This contract should not be taken lightly–use Private if possible; if the interface is really for general use for all applications then use Public. Always remember that making an interface Public comes with large burden of responsibility. Sometimes Limited-Private is just right.</li>
<li>A good example of a Limited-Private interface is BlockLocations. This interface is a fairly low-level interface that is exposed to MapReduce and HBase. The interface is likely to change down the road, and at that time the release effort will have to be coordinated with the MapReduce development team. While MapReduce and HDFS are always released in sync today, that policy may change down the road.</li>
<li>If you have a Limited-Private interface with many projects listed then the interface is probably a good candidate to be made Public.</li>
</ul>
</li>
<li>
<p>Let’s treat all Private interfaces as Limited-Private for all of Hadoop. What is the harm if projects in the Hadoop family have access to private classes?</p>
<ul>
<li>There used to be many cases in the code where one project depended on the internal implementation details of another. A significant effort went into cleaning up those issues. Opening up all interfaces as Limited-Private for all of Hadoop would open the door to reintroducing such coupling issues.</li>
</ul>
</li>
<li>
<p>Aren’t all Public interfaces Stable?</p>
<ul>
<li>One may mark a Public interface as Evolving in its early days. Here one is promising to make an effort to make compatible changes but may need to break it at minor releases.</li>
<li>One example of a Public interface that is Unstable is where one is providing an implementation of a standards-body based interface that is still under development. For example, many companies, in an attempt to be first to market, have provided implementations of a new NFS protocol even when the protocol was not fully completed by IETF. The implementor cannot evolve the interface in a fashion that causes least disruption because the stability is controlled by the standards body. Hence it is appropriate to label the interface as Unstable.</li>
