PHOENIX-4466 Relocate Avatica and hadoop-common in thin-client jar (T…

…oshihiro Suzuki)

When using the thin-client in Spark, we encounter problems in that Spark
is placing its own version of avatica on the classpath as well. We can
relocate most of Avatica (all but the protobuf generated messages as
their classnames are required to be 'org.apache.calcite.avatica.proto'
presently) and hadoop-common to avoid future problems.

phoenix-queryserver-client/pom.xml

@@ -123,9 +123,19 @@
                   <shadedPattern>${shaded.package}.com.google.protobuf</shadedPattern>
                 </relocation>
                 <!-- ORG relocation -->
-                <!-- Calcite/Avatica is not relocated because the wire API (as of <=1.8.0) expects
-                     consistent class names on client and server. Relocating these would break
-                     backwards compatibility. -->
+                <relocation>
+                  <pattern>org.apache.calcite.avatica</pattern>
+                  <shadedPattern>${shaded.package}.org.apache.calcite.avatica</shadedPattern>
+                  <!-- The protobuf messages can't be relocated due to a limitation
+                       in the Avatica protocol. -->
+                  <excludes>
+                    <exclude>org.apache.calcite.avatica.proto.*</exclude>
+                  </excludes>
+                </relocation>
+                <relocation>
+                  <pattern>org.apache.hadoop</pattern>
+                  <shadedPattern>${shaded.package}.org.apache.hadoop</shadedPattern>
+                </relocation>
                 <relocation>
                   <pattern>org.apache.commons</pattern>
                   <shadedPattern>${shaded.package}.org.apache.commons</shadedPattern>