Merge pull request #3409 from vespa-engine/kkraune/attribute-files

Kkraune/attribute files

en/attributes.html

@@ -343,6 +343,39 @@ <h2 id="attribute-memory-usage">Attribute memory usage</h2>
 
 
 
+<h2 id="attribute-file-usage">Attribute file usage</h2>
+<p>Attribute data is stored in two locations on disk:</p>
+<ul>
+  <li>
+    <p>
+      The attribute store in memory, which is regularly flushed to disk.
+      At startup, the flushed files are used to quickly populate the memory structures,
+      resulting in a much quicker startup compared to generating the attribute store from the source in the document store.
+    </p>
+    <p>
+      The attribute store will temporarily double its disk usage when generating a new flush file,
+      see <a href="/en/proton.html#attribute-flush">attribute flush</a>.
+    </p>
+  </li>
+  <li>
+    <p>
+      The document store on disk.
+      Documents here are used to (re)generate index structures,
+      as well as being the source for replica generation across nodes.
+    </p>
+    <p>
+      Note that the attribute data is stored in the document store regardless of
+      the <a href="/en/document-summaries.html">summary</a> configuration.
+    </p>
+  </li>
+</ul>
+<p>
+  The different field types use various data types for storage, see below,
+  a conservative rule of thumb for steady-state disk usage is hence twice the data size.
+</p>
+
+
+
 <h2 id="sizing">Sizing</h2>
 <p>
   Attribute sizing is not an exact science but rather an  approximation.