Update references to rpc.html

Author: Jessica Lin

docs/stable/rpc/rref.html

@@ -334,27 +334,27 @@
 <span id="id1"></span><h1>Remote Reference Protocol<a class="headerlink" href="#remote-reference-protocol" title="Permalink to this headline">¶</a></h1>
 <p>This note describes the design details of Remote Reference protocol and walks
 through message flows in different scenarios. Make sure you’re familiar with the
-<a class="reference internal" href="rpc.html#distributed-rpc-framework"><span class="std std-ref">Distributed RPC Framework</span></a> before proceeding.</p>
+<a class="reference internal" href="../rpc.html#distributed-rpc-framework"><span class="std std-ref">Distributed RPC Framework</span></a> before proceeding.</p>
 <div class="section" id="background">
 <h2>Background<a class="headerlink" href="#background" title="Permalink to this headline">¶</a></h2>
 <p>RRef stands for Remote REFerence. It is a reference of an object which is
 located on the local or a remote worker, and transparently handles reference
 counting under the hood. Conceptually, it can be considered as a distributed
 shared pointer. Applications can create an RRef by calling
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a>. Each RRef is owned by the callee worker
-of the <a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call (i.e., owner) and can be used
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a>. Each RRef is owned by the callee worker
+of the <a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call (i.e., owner) and can be used
 by multiple users. The owner stores the real data and keeps track of the global
 reference count. Every RRef can be uniquely identified by a global <code class="docutils literal notranslate"><span class="pre">RRefId</span></code>,
 which is assigned at the time of creation on the caller of the
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call.</p>
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call.</p>
 <p>On the owner worker, there is only one <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code> instance, which contains
 the real data, while on user workers, there can be as many <code class="docutils literal notranslate"><span class="pre">UserRRefs</span></code> as
 necessary, and <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> does not hold the data. All usage on the owner will
 retrieve the unique <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code> instance using the globally unique <code class="docutils literal notranslate"><span class="pre">RRefId</span></code>.
 A <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> will be created when it is used as an argument or return value in
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a> or
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> invocation, and the owner will be notified
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a> or
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> invocation, and the owner will be notified
 according to update the reference count. An <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code> and its data will be
 deleted when there is no <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> instances globally and there are no
 reference to the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code> on the owner as well.</p>
@@ -369,9 +369,9 @@ <h2>Assumptions<a class="headerlink" href="#assumptions" title="Permalink to thi
 may take down all workers, revert to the previous checkpoint, and resume
 training.</p></li>
 <li><p><strong>Non-idempotent UDFs</strong>: We assume the user functions (UDF) provided to
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a> or
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> are not idempotent and therefore
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a> or
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> are not idempotent and therefore
 cannot be retried. However, internal RRef control messages will be made
 idempotent and retryable.</p></li>
 <li><p><strong>Out of Order Message Delivery</strong>: We do not assume message delivery order
@@ -395,9 +395,9 @@ <h3>Design Reasoning<a class="headerlink" href="#design-reasoning" title="Permal
 <li><p>Creating a new <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> owned by another worker.</p></li>
 </ol>
 <p>Case 1 is the simplest where the owner passes its RRef to a user, where the
-owner calls <a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a>, or
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> and uses its RRef as an argument. In this
+owner calls <a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_sync" title="torch.distributed.rpc.rpc_sync"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_sync()</span></code></a>,
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.rpc_async" title="torch.distributed.rpc.rpc_async"><code class="xref py py-meth docutils literal notranslate"><span class="pre">rpc_async()</span></code></a>, or
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> and uses its RRef as an argument. In this
 case a new <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> will be created on the user. As the owner is the caller,
 it can easily update its local reference count on the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code>.</p>
 <p>The only requirement is that any
@@ -451,10 +451,10 @@ <h3>Design Reasoning<a class="headerlink" href="#design-reasoning" title="Permal
     <span class="n">A</span> <span class="o">-&gt;</span> <span class="n">Y</span> <span class="o">-&gt;</span> <span class="n">Z</span>
 </pre></div>
 </div>
-<p>If Z calls <a class="reference internal" href="rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a> on the <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code>, the
+<p>If Z calls <a class="reference internal" href="../rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a> on the <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code>, the
 owner at least knows A when Z is deleted, because otherwise,
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a> wouldn’t finish. If Z does not call
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a>, it is possible that the owner
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a> wouldn’t finish. If Z does not call
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a>, it is possible that the owner
 receives all messages from Z before any message from A and Y. In this case, as
 the real data of the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code> has not been created yet, there is nothing to
 be deleted either. It is the same as Z does not exist at all. Hence, it’s still
@@ -487,10 +487,10 @@ <h3>User Share RRef with Owner as Return Value<a class="headerlink" href="#user-
 </div>
 <p>In this case, the <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> is created on the user worker A, then it is
 passed to the owner worker B together with the remote message, and then B
-creates the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code>. The method <a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a>
+creates the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code>. The method <a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a>
 returns immediately, meaning that the <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code> can be forked/used before
 the owner knows about it.</p>
-<p>On the owner, when receiving the <a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call, it
+<p>On the owner, when receiving the <a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> call, it
 will create the <code class="docutils literal notranslate"><span class="pre">OwnerRRef</span></code>, and returns an ACK to acknowledge <code class="docutils literal notranslate"><span class="pre">{100,</span> <span class="pre">1}</span></code>
 (<code class="docutils literal notranslate"><span class="pre">RRefId</span></code>, <code class="docutils literal notranslate"><span class="pre">ForkId</span></code>). Only after receiving this ACK, can A delete its
 <code class="docutils literal notranslate"><span class="pre">UserRRef</span></code>. This involves both <strong>G1</strong> and <strong>G2</strong>. <strong>G1</strong> is obvious. For
@@ -499,8 +499,8 @@ <h3>User Share RRef with Owner as Return Value<a class="headerlink" href="#user-
 <a class="reference internal image-reference" href="https://user-images.githubusercontent.com/16999635/69164772-98181300-0abe-11ea-93a7-9ad9f757cd94.png"><img alt="user_to_owner_ret.png" src="https://user-images.githubusercontent.com/16999635/69164772-98181300-0abe-11ea-93a7-9ad9f757cd94.png" style="width: 500px;" /></a>
 <p>The diagram above shows the message flow, where solid arrow contains user
 function and dashed arrow are builtin messages. Note that the first two messages
-from A to B (<a class="reference internal" href="rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> and
-<a class="reference internal" href="rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a>) may
+from A to B (<a class="reference internal" href="../rpc.html#torch.distributed.rpc.remote" title="torch.distributed.rpc.remote"><code class="xref py py-meth docutils literal notranslate"><span class="pre">remote()</span></code></a> and
+<a class="reference internal" href="../rpc.html#torch.distributed.rpc.RRef.to_here" title="torch.distributed.rpc.RRef.to_here"><code class="xref py py-meth docutils literal notranslate"><span class="pre">to_here()</span></code></a>) may
 arrive at B in any order, but the final delete message will only be sent out
 when:</p>
 <ul class="simple">
@@ -894,4 +894,4 @@ <h2>Resources</h2>
     })
   </script>
 </body>
-</html>
+</html>