[ZEPPELIN-6709] Document paragraph output worker settings

Author: miinhho

conf/zeppelin-site.xml.template

@@ -442,16 +442,17 @@
 <property>
   <name>zeppelin.interpreter.output.worker.count</name>
   <value>4</value>
-  <description>Number of workers in each of the server output and checkpoint pools. Each note is
-  processed serially; independent notes may run concurrently, and checkpoint saves do not occupy
-  output workers. Tune for listener and repository latency and available server resources.</description>
+  <description>Number of workers in each server pool: one pool delivers paragraph output, and the
+  other saves output checkpoints. Output for a note stays in order, while different notes can be
+  processed concurrently. Checkpoint saves do not occupy output workers.</description>
 </property>
 
 <property>
   <name>zeppelin.interpreter.output.events.per.batch</name>
   <value>1000</value>
-  <description>Maximum output events processed for one note before yielding to other ready notes.
-  This controls scheduling granularity, not queue capacity or output truncation.</description>
+  <description>Maximum output events a worker processes for one note before giving other ready
+  notes a turn. Smaller batches can reduce their wait but add scheduling overhead. This setting
+  does not limit queued events or truncate output.</description>
 </property>
 
 <property>

docs/setup/operation/configuration.md

@@ -370,6 +370,18 @@ Sources descending by priority:
     <td>102400</td>
     <td>Output message from interpreter exceeding the limit will be truncated</td>
   </tr>
+  <tr>
+    <td><h6 class="properties">ZEPPELIN_INTERPRETER_OUTPUT_WORKER_COUNT</h6></td>
+    <td><h6 class="properties">zeppelin.interpreter.output.worker.count</h6></td>
+    <td>4</td>
+    <td>Number of workers in each server pool: one pool delivers paragraph output, and the other saves output checkpoints. Output for a note stays in order, while different notes can be processed concurrently. Checkpoint saves do not occupy output workers.</td>
+  </tr>
+  <tr>
+    <td><h6 class="properties">ZEPPELIN_INTERPRETER_OUTPUT_EVENTS_PER_BATCH</h6></td>
+    <td><h6 class="properties">zeppelin.interpreter.output.events.per.batch</h6></td>
+    <td>1000</td>
+    <td>Maximum output events a worker processes for one note before giving other ready notes a turn. Smaller batches can reduce their wait but add scheduling overhead. This setting does not limit queued events or truncate output.</td>
+  </tr>
   <tr>
     <td><h6 class="properties">ZEPPELIN_INTERPRETER_CONNECT_TIMEOUT</h6></td>
     <td><h6 class="properties">zeppelin.interpreter.connect.timeout</h6></td>